Reviewer #1 (Public review):

Summary:

In this manuscript, Azlan et al. identified a novel maternal factor called Sakura that is required for proper oogenesis in Drosophila. They showed that Sakura is specifically expressed in the female germline cells. Consistent with its expression pattern, Sakura functioned autonomously in germline cells to ensure proper oogenesis. In sakura KO flies, germline cells were lost during early oogenesis and often became tumorous before degenerating by apoptosis. In these tumorous germ cells, piRNA production was defective and many transposons were derepressed. Interestingly, Smad signaling, a critical signaling pathway for the GSC maintenance, was abolished in sakura KO germline stem cells, resulting in ectopic expression of Bam in whole germline cells in the tumorous germline. A recent study reported that Bam acts together with the deubiquitinase Otu to stabilize Cyc A. In the absence of sakura, Cyc A was upregulated in tumorous germline cells in the germarium. Furthermore, the authors showed that Sakura co-immunoprecipitated Otu in ovarian extracts. A series of in vitro assays suggested that the Otu (1-339 aa) and Sakura (1-49 aa) are sufficient for their direct interaction. Finally, the authors demonstrated that the loss of otu phenocopies the loss of sakura, supporting their idea that Sakura plays a role in germ cell maintenance and differentiation through interaction with Otu during oogenesis.

Strengths:

To my knowledge, this is the first characterization of the role of CG14545 genes. Each experiment seems to be well-designed and adequately controlled

Weaknesses:

However, the conclusions from each experiment are somewhat separate, and the functional relationships between Sakura's functions are not well established. In other words, although the loss of Sakura in the germline causes pleiotropic effects, the cause-and-effect relationships between the individual defects remain unclear.

Comments on latest version:

The authors have attempted to address my initial concerns with additional experiments and refutations. Unfortunately, my concerns, especially my specific comments 1-3, remain unaddressed. The present manuscript is descriptive and fails to describe the molecular mechanism by which Sakura exerts its function in the germline. Nevertheless, this reviewer acknowledges that the observed defects in sakura mutant ovaries and the possible physiological significance of the Sakura-Out interaction are worth sharing with the research community, as they may lay the groundwork for future research in functional analysis.

Reviewer #3 (Public review):

In this very thorough study, the authors characterize the function of a novel Drosophila gene, which they name Sakura. They start with the observation that sakura expression is predicted to be highly enriched in the ovary and they generate an anti-sakura antibody, a line with a GFP-tagged sakura transgene, and a sakura null allele to investigate sakura localization and function directly. They confirm the prediction that it is primarily expressed in the ovary and, specifically, that it is expressed in germ cells, and find that about 2/3 of the mutants lack germ cells completely and the remaining have tumorous ovaries. Further investigation reveals that Sakura is required for piRNA-mediated repression of transposons in germ cells. They also find evidence that sakura is important for germ cell specification during development and germline stem cell maintenance during adulthood. However, despite the role of sakura in maintaining germline stem cells, they find that sakura mutant germ cells also fail to differentiate properly such that mutant germline stem cell clones have an increased number of "GSC-like" cells. They attribute this phenotype to a failure in the repression of Bam by dpp signaling. Lastly, they demonstrate that sakura physically interacts with otu and that sakura and otu mutants have similar germ cell phenotypes. Overall, this study helps to advance the field by providing a characterization of a novel gene that is required for oogenesis. The data are generally high-quality and the new lines and reagents they generated will be useful for the field.

Comments on latest version:

With these revisions, the authors have addressed my main concerns.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this manuscript, Azlan et al. identified a novel maternal factor called Sakura that is required for proper oogenesis in Drosophila. They showed that Sakura is specifically expressed in the female germline cells. Consistent with its expression pattern, Sakura functioned autonomously in germline cells to ensure proper oogenesis. In Sakura KO flies, germline cells were lost during early oogenesis and often became tumorous before degenerating by apoptosis. In these tumorous germ cells, piRNA production was defective and many transposons were derepressed. Interestingly, Smad signaling, a critical signaling pathway for GSC maintenance, was abolished in sakura KO germline stem cells, resulting in ectopic expression of Bam in whole germline cells in the tumorous germline. A recent study reported that Bam acts together with the deubiquitinase Otu to stabilize Cyc A. In the absence of sakura, Cyc A was upregulated in tumorous germline cells in the germarium. Furthermore, the authors showed that Sakura co-immunoprecipitated Otu in ovarian extracts. A series of in vitro assays suggested that the Otu (1-339 aa) and Sakura (1-49 aa) are sufficient for their direct interaction. Finally, the authors demonstrated that the loss of otu phenocopies the loss of sakura, supporting their idea that Sakura plays a role in germ cell maintenance and differentiation through interaction with Otu during oogenesis.

Strengths:

To my knowledge, this is the first characterization of the role of CG14545 genes. Each experiment seems to be well-designed and adequately controlled.

Weaknesses:

However, the conclusions from each experiment are somewhat separate, and the functional relationships between Sakura's functions are not well established. In other words, although the loss of Sakura in the germline causes pleiotropic effects, the cause-and-effect relationships between the individual defects remain unclear.

Reviewer #2 (Public review):

In this study, the authors identified CG14545 (and named it Sakura), as a key gene essential for Drosophila oogenesis. Genetic analyses revealed that Sakura is vital for both oogenesis progression and ultimate female fertility, playing a central role in the renewal and differentiation of germ stem cells (GSC).

The absence of Sakura disrupts the Dpp/BMP signaling pathway, resulting in abnormal bam gene expression, which impairs GSC differentiation and leads to GSC loss. Additionally, Sakura is critical for maintaining normal levels of piRNAs. Also, the authors convincingly demonstrate that Sakura physically interacts with Otu, identifying the specific domains necessary for this interaction, suggesting a cooperative role in germline regulation. Importantly, the loss of otu produces similar defects to those observed in Sakura mutants, highlighting their functional collaboration.

The authors provide compelling evidence that Sakura is a critical regulator of germ cell fate, maintenance, and differentiation in Drosophila. This regulatory role is mediated through the modulation of pMad and Bam expression. However, the phenotypes observed in the germarium appear to stem from reduced pMad levels, which subsequently trigger premature and ectopic expression of Bam. This aberrant Bam expression could lead to increased CycA levels and altered transcriptional regulation, impacting piRNA expression. Given Sakura's role in pMad expression, it would be insightful to investigate whether overexpression of Mad or pMad could mitigate these phenotypic defects (UAS-Mad line is available at Bloomington Drosophila Stock Center).

As suggested reviewer 1, we tested whether overexpression of Mad could rescue or mitigate the loss of sakura phenotypic defects, by using nos-Gal4-VP16 > UASp-Mad-GFP in the background of sakura^null. As shown in Fig S11, we did not observe any mitigation of defects.

Then, we also tested whether expressing a constitutive active form of Tkv, by using UAS-Dcr2, NGT-Gal4 > UASp-tkv.Q235D in the background of sakura^RNAi. As shown in Fig S12, we did not observe any mitigation of defects by this approach either.

A major concern is the overstated role of Sakura in regulating Orb. The data does not reveal mislocalized Orb; rather, a mislocalized oocyte and cytoskeletal breakdown, which may be secondary consequences of defects in oocyte polarity and structure rather than direct misregulation of Orb. The conclusion that Sakura is necessary for Orb localization is not supported by the data. Orb still localizes to the oocyte until about stage 6. In the later stage, it looks like the cytoskeleton is broken down and the oocyte is not positioned properly, however, there is still Orb localization in the ~8-stage egg chamber in the oocyte. This phenotype points towards a defect in the transport of Orb and possibly all other factors that need to localize to the oocyte due to cytoskeletal breakdown, not Orb regulation directly. While this result is very interesting it needs further evaluation on the underlying mechanism. For example, the decrease in E-cadherin levels leads to a similar phenotype and Bam is known to regulate E-cadherin expression. Is Bam expressed in these later knockdowns?

We examined Bam and DE-Cadherin expression in later RNAi knockdowns driven by ToskGal4. As shown in Fig S9, Bam was not expressed in these later knockdowns compared with controls. DE-Cadherin staining suggested a disorganized structure in late-stage egg chambers.

We agree that we overstated a role of Sakura in regulating Orb in the initial manuscript. We changed the text to avoid overstating.

The manuscript would benefit from a more balanced interpretation of the data concerning Sakura's role in Orb regulation. Furthermore, a more expanded discussion on Sakura's potential role in pMad regulation is needed. For example, since Otu and Bam are involved in translational regulation, do the authors think that Mad is not translated and therefore it is the reason for less pMad? Currently the discussion presents just a summary of the results and not an extension of possible interpretation discussed in context of present literature.

We changed the text to avoid overstating a role of Sakura in regulating Orb localization.

Based on our newly added results showing that transgenic overexpression of Mad could not rescue or mitigate the phenotypic defects of sakura^null mutant (Fig S11), we do not think the reason for less pMad is less translation of Mad.

Reviewer #3 (Public review):

In this very thorough study, the authors characterize the function of a novel Drosophila gene, which they name Sakura. They start with the observation that sakura expression is predicted to be highly enriched in the ovary and they generate an anti-sakura antibody, a line with a GFP-tagged sakura transgene, and a sakura null allele to investigate sakura localization and function directly. They confirm the prediction that it is primarily expressed in the ovary and, specifically, that it is expressed in germ cells, and find that about 2/3 of the mutants lack germ cells completely and the remaining have tumorous ovaries. Further investigation reveals that Sakura is required for piRNA-mediated repression of transposons in germ cells. They also find evidence that sakura is important for germ cell specification during development and germline stem cell maintenance during adulthood. However, despite the role of sakura in maintaining germline stem cells, they find that sakura mutant germ cells also fail to differentiate properly such that mutant germline stem cell clones have an increased number of "GSC-like" cells. They attribute this phenotype to a failure in the repression of Bam by dpp signaling. Lastly, they demonstrate that sakura physically interacts with otu and that sakura and otu mutants have similar germ cell phenotypes. Overall, this study helps to advance the field by providing a characterization of a novel gene that is required for oogenesis. The data are generally high-quality and the new lines and reagents they generated will be useful for the field. However, there are some weaknesses and I would recommend that they address the comments in the Recommendations for the authors section below.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

General Comments:

(1) The gene nomenclature: As mentioned in the text, Sakura means cherry blossom and is one of the national flowers of Japan. I am not sure whether the phenotype of the CG14545 mutant is related to Sakura or not. I would like to suggest the authors reconsider the naming.

The striking phenotype of sakura mutant­ is tumorous and germless ovarioles. The tumorous phenotype, exhibiting lots of round fusome in germarium visualized by anti-Hts staining, looks like cherry blossom blooming to us. Also, the germless phenotype reminds us falling of the cherry blossom, especially considering that the ratio of tumorous phenotype decreases and that of germless decreases over fly age. Furthermore, “Sakura” symbolizes birth and renewal in Japanese culture (the last author of this manuscript is Japanese). Our findings indicated that the gene sakura is involved in regulation of renewal and differentiation of GSCs (which leads to birth). These are the reasons for the naming, which we would like to keep.

(2) In many of the microscopic photographs in the figures, especially for the merged confocal images, the resolution looks low, and the images appear blurred, making it difficult to judge the authors' claims. Also, the Alpha Fold structure in Figure 10A requires higher contrast images. The magnification of the images is often inadequate (e.g. Figures 3A, 3B, 5E, 7A, etc). The authors should take high-magnification images separately for the germarium and several different stages of the egg chambers and lay out the figures.

We are very sorry for the low-resolution images. This was caused when the original PDF file with high-resolution images was compressed in order to meet the small file size limit in the eLife submission portal. In the revised submission, we used high-resolution images.

Specific Comments

(1) How Sakura can cooperate with Otu remains unanswered. Sakura does not regulate deubiquitinase activity in vitro. Both sakura and otu appear to be involved in the Dpp-Smad signaling pathway and in the spatial control of Bam expression in the germarium, whereas Otu has been reported to act in concert with Bam to deubiquitinate and stabilize Cyc A for proper cystoblast differentiation. Therefore, it is plausible that the stabilization of Cyc A in the Sakura mutant is an indirect consequence of Bam misexpression and independent of the Sakura-Otu interaction. The authors may need to provide much deeper insight into the mechanism by which Sakura plays roles in these seemingly separable steps to orchestrate germ cell maintenance and differentiation during early oogenesis.

Yes, it is possible that the stabilization of CycA in the sakura mutant is an indirect consequence of Bam misexpression and independent of the Sakura-Otu interaction. To test the significance and role of the Sakura-Otu interaction, we have attempted to identify Sakura point mutants that lose interaction with Otu. If such point mutants were successfully obtained, we were planning to test if their transgene expression could rescue the phenotypes of sakura mutant as the wild-type transgene did. However, after designing and testing the interaction of over 30 point mutants with Otu, we could not obtain such mutant version of Sakura yet. We will continue making efforts, but it is beyond the scope of the current study. We hope to address this important point in future studies.

(2) Figure 3A and Figure 4: The authors show that piRNA production is abolished in Sakura KO ovaries. It is known that piRNA amplification (the ping-pong cycle) occurs in the Vasa-positive perinuclear nuage in nurse cells. Is the nuage normally formed in the absence of Sakura? The authors provide high-magnification images in the germarium expressing Vas-GFP. How does Sakura, and possibly Out, contribute to piRNA production? Are the defects a direct or indirect consequence of the loss of Sakura?

We provided higher magnification images of germarium expressing Vasa-EGFP in sakura mutant background (Fig 3A and 3B). The nuage formation does not seem to be dysregulated in sakura mutant. Currently, we do not know if the piRNA defects are direct or indirect consequence of the loss of Sakura. This question cannot be answered easily. We hope to address this in future studies.

(3) Figure 7 and Figure 12: The authors showed that Dpp-Smad signaling was abolished in Sakura KO germline cells. The same defects were also observed in otu mutant ovaries (Figure 12B). How does the Sakura-Otu axis contribute to the Dpp-Smad pathway in the germline?

As we mentioned in the response to comment (1), we attempted to test the significance and role of the Sakura-Otu interaction, including in the Dpp-Smad pathway in the germline, but we have not yet been able to obtain loss-of-interaction mutant(s) of Sakura. We hope to address this in future studies.

(4) Figure 9 and Fig 10: The authors raised antibodies against both Sakura and Otu, but their specificities were not provided. For Western blot data, the authors should provide whole gel images as source data files. Also, the authors argue that the Otu band they observed corresponds to the 98-kDa isoform (lines 302-304). The molecular weight on the Western blot alone would be insufficient to support this argument.

When we submitted the initial manuscript, we also submitted original, uncropped, and unmodified whole Western blot images for all gel images to the eLife journal, as requested. We did the same for this revised submission. I believe eLife makes all those files available for downloading to readers.

In the newly added Fig S13B, we used very young 2-5 hours ovaries and 3-7 days ovaries. 2-5 days ovaries contain only mostly pre-differentiated germ cells. Older ovaries (3-7 days in our case here) contain all 14 stages of oogenesis and later stages predominate in whole ovary lysates.

As reported in previous literature (Sass et al. 1995), we detected a higher abundance of the 104 kDa Otu isoform than the 98 kDa isoform in from 2-5 hours ovaries and predominantly the 98 kDa isoform in 3-7 days ovaries (Fig S13B). These results confirmed that the major Otu isoform we detected in Western blot, all of which uses old ovaries except for the 2-5 hours ovaries in Fig S13B, is the 98 kDa isoform.

(5) Otu has been reported to regulate ovo and Sxl in the female germline. Is Sakura involved in their regulation?

We examined sxl alternative splicing pattern in sakura mutant ovaries. As shown in Fig S6, we detected the male-specific isoform of sxl RNA and a reduced level of the female-specific sxl isoform in sakura mutant ovaries. Thus Sakura seems to be involved in sxl splicing in the female germline, while further studies will be needed to understand whether Sakura has a direct or indirect role here.

(6) Lines 443-447: The GSC loss phenotype in piwi mutant ovaries is thought to occur in a somatic cell-autonomous manner: both piwi-mutant germline clones and germline-specific piwi knockdown do not show the GSC-loss phenotype. In contrast, the authors provide compelling evidence that Sakura functions in the germline. Therefore, the Piwi-mediated GSC maintenance pathway is likely to be independent of the Sakura-Otu axis.

We changed the text accordingly.

Reviewer #2 (Recommendations for the authors):

Overall, this is a cleanly written manuscript, with some sentences/sections that are confusing the way they are constructed (i.e. Line 37-38, 334, section on Flp/FRT experiments).

We rewrote those sections to avoid confusion.

Comment for all merged image data: the quality of the merged images is very poor - the individual channels are better but should also be reprocessed for more resolved image data sets. Also, it would be helpful to have boundaries drawn in an individual panel to identify the regions of the germarium, as cartooned in Figure S1A (which should be brought into Figure 1) F-actin or Vsg staining would have helped throughout the manuscript to enhance the visualization of described phenotypes.

We are very sorry for the low-resolution images. This was caused when the original PDF file with high-resolution images was compressed in order to meet the small file size limit in the eLife submission portal. In the revised submission, we used high-resolution images.

We outlined the germarium in Fig 1E.

We brought the former FigS1 into Fig 1A.

We provided Phalloidin (F-Actin) staining images in Fig S7.

All p-values seem off. I recommend running the data through the student t-test again.

We used the student t-test to calculate p-values and confirmed that they are correct. We don’t understand why the reviewer thinks all p-values seem off.

In the original manuscript, as we mentioned in each figure legends, we used asterisk (*) to indicate p-value <0.05, without distinguishing whether it’s <0.001, <0.01< or <0.05.

Probably reviewer 2 is suggesting us to use ***, **, and *, to indicate p-value of <0.001, <0.01, and <0.05, respectively? If so, we now followed reviewer2’s suggestions.

Figure 1

(1) Within the text, C is mentioned before A.

We updated the text and now we mentioned Fig 1A before Fig 1C.

(2) B should be the supplemental figure.

We moved the former Fig 1B to Supplemental Figure 1.

(3) C - How were the different egg chamber stages selected in the WB? Naming them 'oocytes' is deceiving. Recommend labeling them as 'egg chambers', since an oocyte is claimed to be just the one-cell of that cyst.

We changed the labeling to egg chambers.

(4) Is the antibody not detecting Sakura in IF? There is no mention of this anywhere in the manuscript.

While our Sakura antibody detects Sakura in IF, it seems to detect some other proteins as well. Since we have Sakura-EGFP fly strain (which fully rescues sakura^null phenotypes) to examine Sakura expression and localization without such non-specific signal issues, we relied on Sakura-EGFP rather than anti-Sakura antibodies for IF.

(5) Expand on the reliance of the sakura-EGFP fly line. Does this overexpression cause any phenotypes?

sakura-EGFP does not cause any phenotypes in the background of sakura[+/+] and sakura[+/-].

(6) Line 95 "as shown below" is not clear that it's referencing panel D.

We now referenced Fig 1D.

(7) Re: Figures 1 E and F. There is no mention of Hts or Vasa proteins in the text.<br /> "Sakura-EGFP was not expressed in somatic cells such as terminal filament, cap cells, escort cells, or follicle cells (Figure 1E). In the egg chamber, Sakura-EGFP was detected in the cytoplasm of nurse cells and was enriched in developing oocytes (Figure 1F)". Outline these areas or label these structures/sites in the images. The color of Merge labels is confusing as the blue is not easily seen.

We mentioned Hts and Vasa in the text. We labeled the structures/sites in the images and updated the color labeling.

Figure 2

(1) Entire figure is not essential to be a main figure, but rather supplemental.

We don’t agree with the reviewer. We think that the female fertility assay data, where sakura null mutant exhibits strikingly strong phenotype, which was completely rescued by our Sakura-EGFP transgene, is very important data and we would like to present them in a main figure.

(2) 2A- one star (*) significance does not seem correct for the presented values between 0 and 100+.

In the original manuscript, as we mentioned in each figure legends, we used asterisk (*) to indicate p-value <0.05, without distinguishing whether it’s <0.001, <0.01< or <0.05.

Probably reviewer 2 is suggesting us to use ***, **, and *, to indicate p-value of <0.001, <0.01, and <0.05, respectively? If so, we now followed reviewer2’s suggestions.

(3) 2C images are extremely low quality. Should be presented as bigger panels.

We are very sorry for the low-resolution images. This was caused when the original PDF file with high-resolution images was compressed in order to meet the small file size limit in the eLife submission portal. In the revised submission, we used high-resolution images. We also presented as bigger panels.

Figure 3

(1) "We observed that some sakura^null /null ovarioles were devoid of germ cells ("germless"), while others retained germ cells (Fig 3A)" What is described is, that it is hard to see. Must have a zoomed-in panel.

We provided zoomed-in panels in Fig 3B

(2) C - The control doesn't seem to match. Must zoom in.

We provided matched control and also zoomed in.

(3) For clarity, separate the tumorous and germless images.

In the new image, only one tumorous and one germless ovarioles are shown with clear labeling and outline, for clarity.

(4) Use arrows to help clearly indicate the changes that occur. As they are presented, they are difficult to see.

We updated all the panels to enhance clarity.

(5) Line 158 seems like a strong statement since it could be indirect.

We softened the statement.

Figure 4

(1) Line 188-189 - Conclusion is an overstatement.

We softened the statement.

(2) Is the piRNA reduction due to a change in transcription? Or a direct effect by Sakura?

We do not know the answers to these questions. We hope to address these in future studies.

Figure 5

(1) D - It might make more sense if this graph showed % instead of the numbers.

We did not understand the reviewer’s point. We think using numbers, not %, makes more sense.

(2) Line 213 - explain why RNAi 2 was chosen when RNAi 1 looks stronger.

Fly stock of RNAi line 2 is much healthier than RNAi line 1 (without being driven Gal4) for some reasons. We had a concern that the RNAi line 1 might contain an unwanted genetic background. We chose to use the RNAi 2 line to avoid such an issue.

(3) In Line 218 there's an extra parenthesis after the PGC acronym.

We corrected the error.

(4) TOsk-Gal4 fly is not in the Methods section.

We mentioned TOsk-Gal4 in the Methods.

Figure 6:

(1) The FLP-FRT section must be rewritten.

We rewrote the FLP-FRT section.

(2) A - include statistics.

We included statistics using the chi-square test.

(3) B - is not recalled in the Results text.

We referred Fig 6B in the text.

(4) Line 232 references Figure 3, but not a specific panel.

We referred Fig 3A, 3C, 3D, and 3E, in the text.

Figure 7/8 - can go to Supplemental.

We moved Fig 8 to supplemental. However, we think Fig 7 data is important and therefore we would like to present them as a main figure.

(1) There should be CycA expression in the control during the first 4 divisions.

Yes, there is CycA expression observed in the control during the first 4 divisions, while it’s much weaker than in sakura^null clone.

(2) Helpful to add the dotted lines to delineate (A) as well.

We added a dotted outline for germarium in Fig 7A.

(3) Line 263 CycA is miswritten as CyA.

We corrected the typo.

Figure 9

(1) Otu antibody control?

We validated Otu antibody in newly added Fig 10C and Fig S13A.

(2) Which Sakura-EGFP line was used? sakura het. or null background? This isn't mentioned in the text, nor legend.

We used Sakura-EGFP in the background of sakura[+/+]. We added this information in the methods and figure legend.

(3) C - Why the switch to S2 cells? Not able to use the Otu antibody in the IP of ovaries?

We can use the Otu antibody in the IP of ovaries. However, in anti-Sakura Western after anti-Otu IP, antibody light chain bands of the Otu antibodies overlap with the Sakura band. Therefore, we switched to S2 cells to avoid this issue by using an epitope tag.

Figure 10

(1) A- The resolution of images of the ribbon protein structure is poor.

We are very sorry for the low-resolution images. This was caused when the original PDF file with high-resolution images was compressed in order to meet the small file size limit in the eLife submission portal. In the revised submission, we used high-resolution images.

(2) A table summarizing the interactions between domains would help bring clarity to the data presented.

We added a table summarizing the fragment interaction results.

(3) Some images would be nice here to show that the truncations no longer colocalize.

We did not understand the reviewer’s points. In our study, even for the full-length proteins.

We have not shown any colocalization of Sakura and Otu in S2 cells or in ovaries, except that they both are enriched in developing oocytes in egg chambers.

Figure 12

(1) A - control and RNAi lines do not match.

We provided matched images.

(2) In general, since for Sakura, only its binding to Otu was identified and since they phenocopy each other, doesn't most of the characterization of Sakura just look at Otu phenotypes? Does Sakura knockdown affect Otu localization or expression level (and vice versa)?

We tested this by Western (Fig S15) and IF (Fig 12). Sakura knockdown did not decrease Otu protein level, and Otu knockdown did not decrease Sakura protein level (Fig S15). In sakura^null clone, Otu level was not notably affected (Fig 12). In sakura^null clone, Otu lost its localization to the posterior position within egg chambers.

Figure S6

(1) It is Luciferase, not Lucifarase.

We corrected the typo.

Reviewer #3 (Recommendations for the authors):

(1) It is interesting that germless and tumorous phenotypes coexist in the same population of flies. Additional consideration of these essentially opposite phenotypes would significantly strengthen the study. For example, do they co-exist within the same fly and are the tumorous ovarioles present in newly eclosed flies or do they develop with age? The data in Figure 8 show that bam knockdown partially suppresses the germless phenotype. What effect does it have on the tumorous phenotype? Is transposon expression involved in either phenotype? Do Sakura mutant germline stem cell clones overgrow relative to wild-type cells in the same ovariole? Does sakura RNAi driven by NGT-Gal4 only cause germless ovaries or does it also cause tumorous phenotypes? What happens if the knockdown of Sakura is restricted to adulthood with a Gal80ts? It may not be necessary to answer all of these questions, but more insight into how these two phenotypes can be caused by loss of sakura would be helpful.

We performed new experiments to answer these questions.

do they co-exist within the same fly and are the tumorous ovarioles present in newly eclosed flies or do they develop with age?

Tumorous and germless ovarioles coexist in the same fly (in the same ovary). Tumorous ovarioles are present in very young (0-1 day old) flies, including newly eclosed (Fig S5). The ratio of germless ovarioles increases and that of tumorous ovarioles decreases with age (Fig S5).

The data in Figure 8 show that bam knockdown partially suppresses the germless phenotype. What effect does it have on the tumorous phenotype?

bam knockdown effect on tumorous phenotype is shown in Fig S10. bam knockdown increased the ratio of tumorous ovarioles and the number of GSC-like cells.

Is transposon expression involved in either phenotype?

Since our transposon-piRNA reporter uses germline-specific nos promoter, it is expressed only in germ line cells, so we cannot examine in germless ovarioles.

Do Sakura mutant germline stem cell clones overgrow relative to wild-type cells in the same ovariole?

Yes, Sakura mutant GSC clones overgrow. Please compare Fig 6C and Fig S8.

Does sakura RNAi driven by NGT-Gal4 only cause germless ovaries or does it also cause tumorous phenotypes?

Fig S10 and Fig S12 show the ovariole phenotypes of sakura RNAi driven by NGT-Gal4. It causes both germless and tumorous phenotypes.

What happens if the knockdown of Sakura is restricted to adulthood with a Gal80ts?

Our mosaic clone was induced at the adult stage, so we already have data of adulthood-specific loss of function. Gal80ts does not work well with nos-Gal4.

(2) The idea that the excessive bam expression in tumorous ovaries is due to a failure of bam repression by dpp signaling is not well-supported by the data. Dpp signaling is activated in a very narrow region immediately adjacent to the niche but the images in Figure 7A show bam expression in cells that are very far away from the niche. Thus, it seems more likely to be due to a failure to turn bam expression off at the 16-cell stage than to a failure to keep it off in the niche region. To determine whether bam repression in the niche region is impaired, it would be important to examine cells adjacent to the niche directly at a higher magnification than is shown in Figure 7A.

We provided higher magnification images of cells adjacent to the niche in new Fig 7A.

We found that cells adjacent to the niche also express Bam-GFP.

That said, we agree with the reviewer. A failure to turn bam expression off at the 16-cell stage may be an additional or even a main cause of bam misexpression in sakura mutant. We added this in the Discussion.

(3) In addition, several minor comments should be addressed:

a. Does anti-Sakura work for immunofluorescence?

While our Sakura antibody detects Sakura in IF, it seems to detect some other proteins as well. Since we have Sakura-EGFP fly strain to examine Sakura expression and localization without such non-specific signal issues, we relied on Sakura-EGFP rather than anti-Sakura antibodies.

b. Please provide insets to show the phenotypes indicated by the different color stars in Figure 3C more clearly.

We provided new, higher-magnification images to show the phenotypes more clearly.

c. Please indicate the frequency of the expression patterns shown in Figure 4D (do all ovarioles in each genotype show those patterns or is there variable penetrance?).

We indicated the frequency.

d. An image showing TOskGal4 driving a fluorophore should be provided so that readers can see which cells express Gal4 with this driver combination.

It has been already done in the paper ElMaghraby et al, GENETICS, 2022, 220(1), iyab179, so we did not repeat the same experiment.

eLife Assessment

This study provides important insights into how cryptic pockets play a role in shaping binding preferences of protein-nucleic acid interactions. By combining biochemical assays and state-of-the-art molecular dynamics simulations, mechanism underlying viral protein 35 (VP35) homologs to bind the backbone of double stranded RNA is presented. The evidence is compelling for molecular determinants that suggest two different dsRNA binding modes for VP35 and also underscores the evolutionary importance of these pockets.

Reviewer #1 (Public review):

Summary:

Mallimadugula et al. combined Molecular Dynamics (MD) simulations, thiol-labeling experiments, and RNA-binding assays to study and compare the RNA-binding behavior of the Interferon Inhibitory Domain (IID) from Viral Protein 35 (VP35) of Zaire ebolavirus, Reston ebolavirus, and Marburg marburgvirus. Although the structures and sequences of these viruses are similar, the authors suggest that differences in RNA binding stem from variations in their intrinsic dynamics, particularly the opening of a cryptic pocket. More precisely, the dynamics of this pocket may influence whether the IID binds to RNA blunt ends or the RNA backbone.

Overall, the authors present important findings to reveal how the intrinsic dynamics of proteins can influence their binding to molecules and, hence, their functions. They have used extensive biased simulations to characterize the opening of a pocket which was not clearly seen in experimental results - at least when the proteins were in their unbound forms. Biochemical assays further validated theoretical results and linked them to RNA binding modes. Thus, with the combination of biochemical assays and state-of-the-art Molecular Dynamics simulations, these results are clearly compelling.

Strengths:

The use of extensive Adaptive Sampling combined with biochemical assays clearly point to the opening of the Interferon Inhibitory Domain (IID) as a factor for RNA binding. This type of approach is especially useful to assess how protein dynamics can affect its function.

Weaknesses:

Although a connection between the cryptic pocket dynamics and RNA binding mode is proposed, the precise molecular mechanism linking pocket opening to RNA binding still remains unclear.

Reviewer #2 (Public review):

Summary:

The authors aimed to determine whether a cryptic pocket in the VP35 protein of Zaire ebolavirus has a functional role in RNA binding and, by extension, in immune evasion. They sought to address whether this pocket could be an effective therapeutic target resistant to evolutionary evasion by studying its role in dsRNA binding among different filovirus VP35 homologs. Through simulations and experiments, they demonstrated that cryptic pocket dynamics modulate the RNA binding modes, directly influencing how VP35 variants block RIG-I and MDA5-mediated immune responses.<br /> The authors successfully achieved their aim, showing that the cryptic pocket is not a random structural feature but rather an allosteric regulator of dsRNA binding. Their results not only explain functional differences in VP35 homologs despite their structural similarity but also suggest that targeting this cryptic pocket may offer a viable strategy for drug development with reduced risk of resistance.

This work represents a significant advance in the field of viral immunoevasion and therapeutic targeting of traditionally "undruggable" protein features. By demonstrating the functional relevance of cryptic pockets, the study challenges long-standing assumptions and provides a compelling basis for exploring new drug discovery strategies targeting these previously overlooked regions.

Strengths:

The combination of molecular simulations and experimental approaches is a major strength, enabling the authors to connect structural dynamics with functional outcomes. The use of homologous VP35 proteins from different filoviruses strengthens the study's generality, and the incorporation of point mutations adds mechanistic depth. Furthermore, the ability to reconcile functional differences that could not be explained by crystal structures alone highlights the utility of dynamic studies in uncovering hidden allosteric features.

Weaknesses:

While the methodology is robust, certain limitations should be acknowledged. For example, the study would benefit from a more detailed quantitative analysis of how specific mutations impact RNA binding and cryptic pocket dynamics, as this could provide greater mechanistic insight. This study would also benefit from providing a clear rationale for the selection of the amber03 force field and considering the inclusion of volume-based approaches for pocket analysis. Such revisions will strengthen the robustness and impact of the study.

Comments on revisions:

The authors addressed the concerns raised.

Reviewer #3 (Public review):

Summary:

The authors suggest a mechanism that explains the preference of<br /> viral protein 35 (VP35) homologs to bind the backbone of double stranded RNA versus blunt ends. These preferences have a biological impact in terms of the ability of different viruses to escape the immune response of the host.<br /> The proposed mechanism involves the existence of a cryptic pocket, where VP35 binds the blunt ends of dsRNA when the cryptic pocket is closed and preferentially binds the RNA double stranded backbone when the pocket is open.<br /> The authors performed MD simulation results, thiol labelling experiments, fluorescence polarization assays, as well as point mutations to support their hypothesis.

Strengths:

This is a genuinely interesting scientific questions, which is approached through multiple complementary experiments as well as extensive MD simulations. Moreover, structural biology studies focused on RNA-protein interactions are particularly rare, highlighting the importance of further research in this area.

Weaknesses:

- Sequence similarity between Ebola-Zaire (94% similarity) explains their similar behaviour in simulations and experimental assays. Marburg instead is a more distant homolog (~80% similarity relative to Ebola/Zaire). This difference is sequence and structure can explain the propensities, without the need to involve the existence of a cryptic pocket.<br /> - No real evidence for the presence of a cryptic pocket is presented, but rather a distance probability distribution between two residues obtained from extensive MD simulations. It would be interesting to characterise the modelled RNA-protein interface in more detail

Comments on revisions:

-I still think that the term cryptic pocket is misleading here, unless the cryptic pocket is more thoroughly characterised. I would find it more appropriate to use the term open/closed state.

- Mg ions are known to be crucial in stabilising RNA structure both in vitro and in MD simulations (see e.g. Draper BJ 2008 and many others). While I understand that the authors cannot repeat simulations in presence of ions, I believe that this detail should be more clearly detailed in the manuscript.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

Mallimadugula et al. combined Molecular Dynamics (MD) simulations, thiol-labeling experiments, and RNA-binding assays to study and compare the RNA-binding behavior of the Interferon Inhibitory Domain (IID) from Viral Protein 35 (VP35) of Zaire ebolavirus, Reston ebolavirus, and Marburg marburgvirus. Although the structures and sequences of these viruses are similar, the authors suggest that differences in RNA binding stem from variations in their intrinsic dynamics, particularly the opening of a cryptic pocket. More precisely, the dynamics of this pocket may influence whether the IID binds to RNA blunt ends or the RNA backbone.

Overall, the authors present important findings to reveal how the intrinsic dynamics of proteins can influence their binding to molecules and, hence, their functions. They have used extensive biased simulations to characterize the opening of a pocket which was not clearly seen in experimental results - at least when the proteins were in their unbound forms. Biochemical assays further validated theoretical results and linked them to RNA binding modes. Thus, with the combination of biochemical assays and state-of-the-art Molecular Dynamics simulations, these results are clearly compelling.

Strengths:

The use of extensive Adaptive Sampling combined with biochemical assays clearly points to the opening of the Interferon Inhibitory Domain (IID) as a factor for RNA binding. This type of approach is especially useful to assess how protein dynamics can affect its function.

Weaknesses:

Although a connection between the cryptic pocket dynamics and RNA binding mode is proposed, the precise molecular mechanism linking pocket opening to RNA binding still remains unclear.

Reviewer #2 (Public review):

Summary:

The authors aimed to determine whether a cryptic pocket in the VP35 protein of Zaire ebolavirus has a functional role in RNA binding and, by extension, in immune evasion. They sought to address whether this pocket could be an effective therapeutic target resistant to evolutionary evasion by studying its role in dsRNA binding among different filovirus VP35 homologs. Through simulations and experiments, they demonstrated that cryptic pocket dynamics modulate the RNA binding modes, directly influencing how VP35 variants block RIG-I and MDA5-mediated immune responses.

The authors successfully achieved their aim, showing that the cryptic pocket is not a random structural feature but rather an allosteric regulator of dsRNA binding. Their results not only explain functional differences in VP35 homologs despite their structural similarity but also suggest that targeting this cryptic pocket may offer a viable strategy for drug development with reduced risk of resistance.

This work represents a significant advance in the field of viral immunoevasion and therapeutic targeting of traditionally "undruggable" protein features. By demonstrating the functional relevance of cryptic pockets, the study challenges long-standing assumptions and provides a compelling basis for exploring new drug discovery strategies targeting these previously overlooked regions.

Strengths:

The combination of molecular simulations and experimental approaches is a major strength, enabling the authors to connect structural dynamics with functional outcomes. The use of homologous VP35 proteins from different filoviruses strengthens the study's generality, and the incorporation of point mutations adds mechanistic depth. Furthermore, the ability to reconcile functional differences that could not be explained by crystal structures alone highlights the utility of dynamic studies in uncovering hidden allosteric features.

Weaknesses:

While the methodology is robust, certain limitations should be acknowledged. For example, the study would benefit from a more detailed quantitative analysis of how specific mutations impact RNA binding and cryptic pocket dynamics, as this could provide greater mechanistic insight. This study would also benefit from providing a clear rationale for the selection of the amber03 force field and considering the inclusion of volume-based approaches for pocket analysis. Such revisions will strengthen the robustness and impact of the study.

Reviewer #3 (Public review):

Summary:

The authors suggest a mechanism that explains the preference of viral protein 35 (VP35) homologs to bind the backbone of double-stranded RNA versus blunt ends. These preferences have a biological impact in terms of the ability of different viruses to escape the immune response of the host.

The proposed mechanism involves the existence of a cryptic pocket, where VP35 binds the blunt ends of dsRNA when the cryptic pocket is closed and preferentially binds the RNA double-stranded backbone when the pocket is open.

The authors performed MD simulation results, thiol labelling experiments, fluorescence polarization assays, as well as point mutations to support their hypothesis.

Strengths:

This is a genuinely interesting scientific question, which is approached through multiple complementary experiments as well as extensive MD simulations. Moreover, structural biology studies focused on RNA-protein interactions are particularly rare, highlighting the importance of further research in this area.

Weaknesses:

- Sequence similarity between Ebola-Zaire (94% similarity) explains their similar behaviour in simulations and experimental assays. Marburg instead is a more distant homolog (~80% similarity relative to Ebola/Zaire). This difference is sequence and structure can explain the propensities, without the need to involve the existence of a cryptic pocket.  

- No real evidence for the presence of a cryptic pocket is presented, but rather a distance probability distribution between two residues obtained from extensive MD simulations. It would be interesting to characterise the modelled RNA-protein interface in more detail

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Before assessing the overall quality and significance of this work, this reviewer needs to specify the context of this review. This reviewer's expertise lies in biased and unbiased molecular dynamics simulations and structural biology. Hence, while this reviewer can overall understand the results for thiol-labeling and RNA-binding assays, this review will not assess the quality of these biochemical assays and will mainly focus on the modelling results.

Overall, the authors present important findings to reveal how the intrinsic dynamics of proteins can influence their binding to molecules and, hence, their functions. They have used extensive biased simulations to characterize the opening of a pocket which was not clearly seen in experimental results - at least when the proteins were in their unbound forms. Biochemical assays further validated theoretical results and linked them to RNA binding modes. Thus, with the combination of biochemical assays and state-of-the-art Molecular Dynamics simulations, these results are clearly compelling.

Beyond the clear qualities of this work, I would like to mention a few points that may help to better contextualize and rationalize the results presented here.

- First, both the introduction and discussion sections seem relatively condensed. Extending them to, for example, better describe the methodological context and discuss the methodological limitations and potential future developments related to biased simulations may help the reader get a better idea of the significance of this work.

- The authors presented 3 homologs in this study: IIDs of Reston, Zaire, and Marburg viruses. While Zaire and Reston are relatively similar in terms of sequence (Figure S1). The sequences clearly differ between Marburg and the two other viruses. Can the author indicate a similarity/identity score for each sequence alignment and extend Figure S1 to really compare Marburg sequence with Reston and Zaire? Can they also discuss how these differences may impact the comparison of the three IIDs? This may also help the reader to understand why sometimes the authors compare the three viruses and why sometimes they are focusing only on comparing Zaire and Reston.

We would like to thank the reviewer for raising this point and we agree that additional details about the sequence comparison provide more context for the choices of substitutions we made. Therefore, we have updated Fig S1 to include a detailed pairwise comparison of all the IID sequences including the percentage sequence similarity and identity. We have also added the following sentences to the results section where we first introduced the substitutions between Zaire and Reston IIDs

“While the sequence of Marburg IID differs significantly from Reston and Zaire IIDs with a sequence identity of 42% and 45% respectively (Fig S1), the sequences of Reston and Zaire IID are 88% identical and 94% similar. Particularly, substitutions between these homologs are all distal to the RNA-binding interfaces and all the residues known to make contacts with dsRNA from structural studies are identical. Therefore, we reasoned that comparing these two homologs would help us identify minimal substitutions that control pocket opening probability and allow us to study its effect on dsRNA binding with minimal perturbation of other factors.”

- In this work, the authors mentioned the cryptic pocket but only illustrated the opening of this pocket by using a simple distance between residues (Figure 2) and a SASA of one cysteine (Figure 3). In previous work done by the authors (Cruz et al. , Nature Communications, 2022), they better characterized residues involved in RNA binding and forming the cryptic pocket. Thus, would it be possible to better described this cryptic pocket (residues involved, volume, etc ..) and better explain how, structurally speaking, it can affect RNA binding mode (blunt ends vs backbone) ?

We thank the reviewer for pointing out the need for clarification on the residues involved in RNA binding and pocket opening and the mechanism linking them. We have performed the CARDS analysis on Reston and Marburg IID simulations as we had done on Zaire IID simulations in Cruz et al, 2022. The results are shown in Fig S3 and discussed in the main text in the first results section.

- As a counter-example, the authors used C315 for SASA calculation and thiol labeling (Figure 3). This cysteine is mainly buried as seen by SASA for Reston and Marburg and thiol labelling (Figure 3 E,G,H). Would it be possible to also get thiol labeling rates for Cystein 264 in Reston and its equivalent to see a case where the residue is solvent exposed?

We have shown the SASA for C264 from the simulations in Fig S4 and the thiol labeling rates for all 4 cysteines in Reston IID in Fig S6. Comparing these rates to the rates of all 4 cysteines obtained for Zaire IID (Fig 4 in Cruz et Al, 2022), we observe that the rates for C264, which is expected to be exposed are significantly faster than those of C315 which is largely buried in all variants.  

- I strongly support here the will of the authors to share their data by depositing them in an OSF repository. These data help this reviewer to assess some of the results produced by the authors and help to better understand the dynamics of their respective systems. I have just a few comments that need to be addressed regarding these data: o While there are data for WT Reston and Marburg, there is no data for Zaire. Is this because these data correspond to the previous work (Cruz et al. 2022) (in this case, it would be good to make this clear in the main text) or is it an omission? o There is no center.xtc file in the Marburg-MSM directory o There is no protmasses.pdb in the Reston-MSM directory

- In general, if possible, it would be good to use the same name for each type of file presented in each directory to help a potential user understand a bit more how to use these data.

- If possible, adding a bit more of metadata and explanations on the OSF webpage would be very beneficial to help find these data. To help in this direction, the authors may have a look to the guidelines presented at the end of this article: https://elifesciences.org/articles/90061

We thank the reviewer for pointing out the omissions from the OSF repository. We have added the missing files and followed a uniform naming convention. We have also added documentation in the metadata section of the OSF repository to help others use the data.  

Indeed, the simulation data used for Zaire IID is available on the OSF repository corresponding to Cruz et al. 2022 at https://osf.io/5pg2a. We have also clarified this in the data availability section of the main text.  

Minor point:

In Figure 2, there is a slight bump for the 225-295 distance around 1 nm for Reston. Can the author comment it ? As these results are based on long AS, even if very small, do the authors think this population is significant?

Comparing the probability distributions obtained from bootstrapping the frames used to calculate the MSM equilibrium probabilities (Revised Fig1), we observe that the bump for the Reston IID distribution is persistent in all bootstraps indicating that it might indeed be significant. This is also consistent with our observation that the cysteine 296 does get fully labeled in our thiol labeling experiments, albeit significantly slowly compared to the other homologs.  

Reviewer #2 (Recommendations for the authors):

I recommend that the authors implement moderate revisions prior to the publication of this research article, addressing the identified weaknesses (see below).

The authors should provide a rationale for their selection of the amber03 force field (Duan et al., JCTC 24, 1999-2012, 2003) for molecular dynamics simulations, particularly given the availability of more recent and optimized versions of the AMBER force fields. These newer force fields may offer improved parameterization for biomolecular systems, potentially enhancing the accuracy and reliability of the simulation results.

We chose the Amber03 force field because it has performed well in much of our past work, including the original prediction of the cryptic pocket that we study in this manuscript. The results presented in this manuscript also demonstrate the predictive power of Amber03.

Additionally, while the authors utilized solvent-accessible surface area (SASA) for cryptic pocket analysis, volume-based approaches may be more suitable for this purpose. Several studies (e.g., Sztain et al. J. Chem. Inf. Model. 2021, 61, 7, 3495-3501) have demonstrated the utility of volume analysis in identifying and characterizing cryptic pockets. The authors could consider incorporating such methodologies to provide a more comprehensive assessment of pocket dynamics.

The authors propose that the cryptic pocket is not merely a random structural feature but functions as an allosteric regulator of dsRNA binding. To further substantiate this claim, an in-depth analysis of this allosteric effect using for instance network analysis could significantly enhance the study. Such an approach could identify key residues and interaction networks within the protein that mediate the allosteric regulation. This type of mechanistic insight would not only provide a stronger theoretical framework but also offer valuable information for the rational design of therapeutic interventions targeting the cryptic pocket.  

We thank the reviewer for pointing out the need for clarification on the molecular mechanism linking the opening of the cryptic pocket to RNA binding. We have performed the CARDS analysis on Reston and Marburg IID simulations as was done on Zaire IID simulations in Cruz et al, 2022. The results are shown in Fig S3 and discussed in the main text in the first results section. Briefly, we do find a community (blue) comprising the pocket residues in Reston and Marburg IIDs as we did in Zaire. Similarly, we find that many of the RNA binding residues fall into the orange and green communities as in Zaire. However, there are differences in exactly which residues are clustered into which of these two communities. There are also differences in how strongly connected these communities are in the three homologs. Therefore, while we can conclude that pocket residues likely have varying influence on the RNA binding residues in the homologs, it is hard to say exactly what that variation is from this analysis alone.  

Reviewer #3 (Recommendations for the authors):

- MD simulations: All simulations were initialised from the 3 crystal structures, is it correct? In all cases, RNA ds was not included in simulations, right? Were crystallographic MG ions in the vicinity of the binding site included? these are known to influence structural dynamics to a large extent.

All simulations were indeed initialized using only protein atoms from the crystal structures 3FKE, 4GHL, and 3L2A. Therefore, crystallographic Mg ions were not included in the simulations. However, we do agree with the reviewer and think that the effect of parameters such as salt concentration, specifically Mg ions which are known to be important for the stability of dsRNA, on the pocket opening equilibrium merits detailed study in future work.

- Figure 2: Would it be possible to perform e.g. a block error analysis and show the statistical errors of the distributions?

We agree that showing the statistical variation in the MSM equilibrium probabilities is important for comparing the different distributions. Therefore, we have updated Figs 2 and 5 to show the distributions obtained from MSMs constructed using 100 and 10 random samples of the data respectively to indicate the extent of the statistical variability in the MSM construction.  

- More detailed structural biology experiments (such as NMR or HDX-MS) could potentially shed more light on the differential behaviour of the three different homologs, providing more evidence for the presence of the cryptic pocket.

We agree that NMR and HDX-MS are powerful means to study dynamics and are actively exploring these approaches for our future work.

eLife Assessment

This important study offers convincing evidence that intra-Golgi transport slows from cis to trans and varies between cargos even within the same cisternae, supporting a more stable compartment model. Using nocodazole-induced ministacks, the authors show cargo-specific transport kinetics with distinct velocities and residence times. These findings refine the cisternal progression model and prompt further investigation into alternative mechanisms, such as rapid partitioning or rim progression. This study will be of interest to cell biologists studying membrane trafficking, Golgi organization, and protein secretion, as well as researchers investigating the mechanisms of organelle dynamics and the molecular basis of intracellular transport.

Reviewer #1 (Public review):

Summary:<br /> In the manuscript by Tie et.al., the authors couple the methodology which they have developed to measure LQ (localization quotient) of proteins within the Golgi apparatus along with RUSH based cargo release to quantify the speed of different cargos traveling through Golgi stacks in nocodazole induced Golgi ministacks to differentiate between cisternal progression vs stable compartment model of the Golgi apparatus. The debate between cisternal progression model and stable compartment model has been intense and going on for decades and important to understand the basic way of function/organization of the Golgi apparatus. As per the stable compartment model, cisterna are stable structures, and cargo moves along the Golgi apparatus in vesicular carriers. While as per cisternal progression model, Golgi cisterna themselves mature acquiring new identity from the cis face to the trans face and act as transport carriers themselves. In this work, authors provide a missing part regarding intra-Golgi speed for transport of different cargoes as well as the speed of TGN exit and based on the differences in the transport velocities for different cargoes tested favor a stable compartment model. The argument which authors make is that if there is cisternal progression, all the cargoes should have a similar intra-Golgi transport speed which is essentially the rate at which the Golgi cisterna mature. Furthermore, using a combination of BFA and Nocodazole treatments authors show that the compartments remain stable in cells for at least 30-60 minutes after BFA treatment.

Strengths:<br /> The method to accurately measure localization of a protein within the Golgi stack is rigorously tested in the previous publications from the same authors and in combination with pulse chase approaches has been used to quantify transport velocities of cargoes through the Golgi. This is a novel aspect in this paper and differences in intra-Golgi velocities for different cargoes tested makes a case for a stable compartment model.

Weaknesses:<br /> None noted in the revised version of the manuscript.

Reviewer #2 (Public review):

Summary:<br /> This manuscript describes the use of quantitative imaging approaches, that have been a key element of the labs work over the past years, to address one of the major unresolved discussions in trafficking: intra-Golgi transport. The approach used has been clearly described in the labs previous papers, and is thus clearly described. The authors clearly address the weaknesses in this manuscript, and do not overstate the conclusions drawn from the data. The only weakness not addressed is the concept of blocking COPI transport with BFA, which is a strong inhibitor and causes general disruption of the system. This is an interesting element of the paper, which I think could be improved upon by using more specific COPI inhibitors instead, although I understand that this is not necessarily straightforward.

I commend the authors on their clear and precise presentation of this body of work, incorporating mathematical modelling with a fundamental question in cell biology. In all, I think that this is a very robust body of work, that provides a sound conclusion in support of the stable compartment model for the Golgi.

General points:<br /> The manuscript contains a lot of background in its results sections, and the authors may wish to consider rebalancing the text: The section beginning at Line 175 is about 90% background and 10% data. Could some data currently in supplementary be included here to redress this balance, or this part combined with another?

Minor points:<br /> Equation 2: A should be in front of the ln2. It's already resolved in equation 3, so likely only needs changing in the text

Line 152: Why is there a lack of experimental data? High ER background and low golgi signal make it difficult to select ministacks: would be good to see examples of these images. Is 0 a relevant timepoint as cargo is still at the ER? Instead would a timepoint <5' be better demonstrate initial arrival in fast cargo, and 0' discarded?

Table 1 Line 474: 1-3 independent replicates: is there a better way of incorporating this into the table to make it more streamlined? It would be useful to see each cargo as a mean with error. Is there a more demonstrative way to present the table, for example (but does not have to be) fastest cargo first (Tintra) as in Table 2?

Line 264 / Fig 3B: It's unclear to me why the VHH-anti-GFP-mCherry internalisation approach was used, when the cells were expressing GFP, that could be used for imaging. Also, this introduces a question over trafficking of the VHH itself, to access the same compartments as the GFP-proteins are localised. It would be useful to describe the choice of this approach briefly in the text.

446 Typo "internalization"

Post-Revision

I thank the authors for their work revising the paper in light of our comments. I am satisfied with their response, and I have no other comments.

Reviewer #3 (Public review):

The manuscript by Tie et al. provides a quantitative assessment of intra-Golgi transport of diverse cargos. Quantitative approaches using fluorescence microscopy of RUSH synchronized cargos, namely GLIM and measurement of Golgi residence time, previously developed by the author's team (publications from 20216 to 2022), are being used here.

Most of the results have been already published by the same team in 2016, 2017, 2020 and 2021. In this manuscript, the authors have put together measurement of intra-Golgi transport kinetics and Golgi residence time of many cargos. The quantitative results are supported by a large number of Golgi mini-stacks/cells analyzed. They are discussed with regard to the intra-Golgi transport models being debated in the field, namely the cisternal maturation/progression model and the stable compartments model.

The authors show that different cargos have distinct intra-Golgi transport kinetics and that the Golgi residence time of glycosyltransferases is high. From this and experiment using brefeldinA, the authors suggest that the rim progression model, adapted from the stable compartments model, fits with their experimental data.

Strengths:<br /> The major strength of this manuscript is to put together many quantitative results that the authors previously obtained and to discuss them to advance our understanding of the intra-Golgi transport mechanisms.<br /> The analysis by fluorescence microscopy of intra-Golgi transport is tough and this is a tour de force of the authors even though their approach shows limitations, which are clearly stated. Their work is remarkable in regards of the numbers of Golgi markers and secretory cargos which have been analyzed.

Weaknesses:<br /> Most of the data provided here were already published and thus accessible for the community. The tubular connections between cisternae and the diffusion/biochemical properties of cargos are not taken into account to interpret the results. Indeed, tubular connections and biochemical properties of the cargos may affect their transit through the Golgi and the kinetics with which they reach the TGN for Golgi exit.

The use of nocodazole might affect cellular homeostasis but this is clearly stated by the authors and is acceptable as we need to perturb the system to conduct this analysis.

The manual selection of the Golgi mini-stack being analyzed (where the cargo and the Golgi reference markers are clearly detectable ) might introduce a bias in the analysis.

eLife Assessment

In this important study, the authors conducted atomistic molecular dynamics simulations to probe the interactions between IRE and unfolded peptides. The results help reconcile contradicting experimental findings in the literature and offer mechanistic insights into the activation of the unfolded protein response. The level of evidence is considered solid, although the use of enhanced sampling and more quantitative analysis would further strengthen the conclusions.

Reviewer #1 (Public review):

Summary:

This work provides structural and mechanistic insights into the disordered protein recognition process inside the endoplasmic reticulum by the inositol-requiring enzyme 1. Using state-of-the-art molecular dynamics simulation tools, the authors propose a mechanism of disordered protein recognition that reconciles contradictory findings of biochemical and structural biology experiments.

Strengths:

(1) All MD simulations have been carried out in triplicate, and several different folded conformations were generated using alphafold2. This provides adequate statistics to draw meaningful conclusions from the simulations.

(2) Potential limitations of the disordered protein force fields and water models have been taken into consideration. Particularly, performing the simulation in both TIP3P and TIP4PD water models ensures that the conclusions drawn are not influenced by the force field choice.

(3) The binding of a large number of disordered peptides was investigated, ensuring that the conclusions drawn about disordered peptide recognition are sufficiently general.

Weaknesses:

(1) The timescales of the peptide recognition and unbinding process are much longer than what can be sampled from unbiased simulations. Therefore, the proposed mechanism of recognition should only be considered a hypothesis based on the results presented here. For example, peptides that do not dissociate within one one-microsecond MD simulation are considered to be stable binders. However, they may not have a viable way to bind to the narrow protein cleft in the first place.

(2) Oftentimes, representative structures sampled from MD simulation are used to draw conclusions (e.g., Figure 4 about the role of R161 mutation in binding affinity). This is not appropriate as one unbinding event being observed or not observed in a microsecond-long trajectory does not provide sufficient information about the binding strength of the free energy difference.

Reviewer #2 (Public review):

Summary:

In this manuscript, the authors investigated the interactions between IRE and unfolded peptides using all-atom molecular dynamics simulations. The interactions between a couple of unfolded peptides and IRE might shed light on the activation of the UPR.

Strengths:

(1) Well-written manuscript tailored for a biology audience.

(2) State-of-the-art structural predictions and all-atom simulations.

(3) Validation with existing experimental data

(4) Clear schematic diagram summarizing the mechanisms learned from simulations.

(5) Shared simulation data and code in a public repository.

Weaknesses:

(1) Improving presentation to include more computational details.

(2) More quantitative analysis in addition to visual structures.

Reviewer #3 (Public review):

Summary:

In this important work, the authors use extensive MD simulations to study how the IRE1 protein can detect unfolded peptides. Their study consolidates contradicting experimental results and offers a unique view of the different sensing models that have been proposed in the literature. Overall, it is an excellent study that is quite extensive. The research is solid, meticulous, and carefully performed, leading to convincing conclusions.

Strengths:

The strength of this work is the extensive and meticulous molecular dynamics simulations. The authors use and investigate different structural models, for example, carefully comparing a model based on a PDB structure with reconstructed loops with an AlphaFold 2 Multimer model. The author also investigates a wide range of different protein structural models that probe different aspects of the peptide sensing process. These solid and meticulous MD simulations allow the authors to obtain convincing conclusions concerning the peptide sensing process of the IRE1 protein.

Weaknesses:

A potential weakness of the study is the usage of equilibrium (unbiased) molecular dynamics simulations, so that processes and conformational changes on the microsecond time scale can be probed. Furthermore, there can be inaccuracies and biases in the description of unfolded peptides and protein segments due to the protein force fields. Here, it should be noted that the authors do acknowledge these possible limitations of their study in the conclusions.

Author response:

Reviewer #1:

We appreciate the Reviewer's positive feedback on the strengths of our study.

The timescales of the peptide recognition and unbinding process are much longer than what can be sampled from unbiased simulations. Therefore, the proposed mechanism of recognition should only be considered a hypothesis based on the results presented here. For example, peptides that do not dissociate within one one-microsecond MD simulation are considered to be stable binders. However, they may not have a viable way to bind to the narrow protein cleft in the first place.

We thank the Reviewer for this valuable feedback. We agree with the Reviewer. Our work on the IRE1 cLD activation mechanism is focused on generating hypotheses of the binding mechanism driven by MD simulations. We recognize the limitations in defining a stable binder due to the time scales sampled. However, our primary focus was to sample and characterize a possible binding pose in the center of the cLD dimer. We will contextualize our statements about stable binders and limit our claims to stating that the protein-peptide complex is stable within 1 μs-long simulations. However, we believe that our finding that the cLD dimer groove is not able to accommodate peptides is solid, as the steric impediment described is present in all our replicas, both with and without peptides, in a cumulative sampling time of 72 μs. Additionally, we will include a plot showing the distribution of groove width across all replicas.

Oftentimes, representative structures sampled from MD simulation are used to draw conclusions (e.g., Figure 4 about the role of R161 mutation in binding affinity). This is not appropriate as one unbinding event being observed or not observed in a microsecond-long trajectory does not provide sufficient information about the binding strength of the free energy difference.

We thank the Reviewer for the insightful comment. As explained in the previous point, we believe that our simulations provide useful hypotheses, and we agree that we do not currently have data to comment on binding affinity. We will, therefore, remove all references to this term. We are aware of the limitations due to the timescale and agree that these limitations cannot be overcome with standard equilibrium simulations. To address these limitations, we plan to use orthogonal methods, namely MM/PB(GB)SA calculations for calculating binding free energies from existing trajectories (as performed by https://doi.org/10.1021/acs.jcim.4c00975). We will add predictions of all the peptides using AlphaFold 3, to confirm the binding region.

Reviewer #2:

We thank the Reviewer for their positive feedback.

Improving presentation to include more computational details.

We thank the Reviewer for raising this critical point. We agree that the manuscript is tailored for a biology audience, as the data are particularly relevant for that community. Nevertheless, we also understand the importance of providing sufficient methodological detail for computational readers. We will add appropriate computational information in the main text.

More quantitative analysis in addition to visual structures.

We will add an uncertainty estimate for the HDX calculations using bootstrapping and include additional information on bond distances for Y161. We will also incorporate time-series data showing the distance of the peptide from the groove across all replicas.

Reviewer #3:

We appreciate the Reviewer's positive feedback on our work.

A potential weakness of the study is the usage of equilibrium (unbiased) molecular dynamics simulations so that processes and conformational changes on the microsecond time scale can be probed. Furthermore, there can be inaccuracies and biases in the description of unfolded peptides and protein segments due to the protein force fields. Here, it should be noted that the authors do acknowledge these possible limitations of their study in the conclusions.

We appreciate the Reviewer's thoughtful comment. As noted in our response to Reviewer 1, we plan to address the concern about sampling by applying orthogonal methods. We agree with the Reviewer that some form of enhanced sampling is necessary if we want to assess binding in a more quantitative way, e.g., via free energy calculations. However, we also realize that applying any enhanced sampling scheme to our system is very challenging, given its large size and the complex peptide-protein interactions, which are not easily captured in a few collective variables. After a careful assessment and some preliminary tests, we decided that estimating free energies using enhanced sampling would necessitate a separate paper due to both the conceptual complexity of the project and the size of the necessary sampling campaign.

eLife Assessment

This work provides one of the first important attempts to look at Drosophila immune responses against bacterial, viral, and fungal pathogens in a way that combines the roles of four major arms in immunity (Imd signaling, Toll signaling, phagocytosis, and melanization) rather than studying them separately. The findings are convincing, and the tools provided can be used as they are, or built upon, in various contexts.

Reviewer #1 (Public review):

Summary:

The innate immune system serves as the first line of defense against invading pathogens. Four major immune-specific modules - the Toll pathway, the Imd pathway, melanization, and phagocytosis- play critical roles in orchestrating the immune response. Traditionally, most studies have focused on the function of individual modules in isolation. However, in recent years, it has become increasingly evident that effective immune defense requires intricate interactions among these pathways.

Despite this growing recognition, the precise roles, timing, and interconnections of these immune modules remain poorly understood. Moreover, addressing these questions represents a major scientific undertaking.

Strengths:

In this manuscript, Ryckebusch et al. systematically evaluate both the individual and combined contributions of these four immune modules to host defense against a range of pathogens. Their findings significantly enhance our understanding of the layered architecture of innate immunity.

Weaknesses:

While I have no critical concerns regarding the study, I do have several suggestions to offer that may help further strengthen the manuscript. These include:

(1) Have the authors validated the efficiency of the mutants used in this study? It would be helpful to include supporting data or references confirming that the mutations effectively disrupted the intended immune pathways.

(2) Given the extensive use of double, triple, and quadruple mutants, a more detailed description of the mutant construction process is warranted.

Reviewer #2 (Public review):

Summary:

In this work, the authors take a holistic view of Drosophila immunity by selecting four major components of fly immunity often studied separately (Toll signaling, Imd signaling, phagocytosis, and melanization), and studying their combinatory effects on the efficiency of the immune response. They achieve this by using fly lines mutant for one of these components, or modules, as well as for a combination of them, and testing the survival of these flies upon infection with a plethora of pathogens (bacterial, viral, and fungal).

Strengths:

It is clear that this manuscript has required a large amount of hands-on work, considering the number of pathogens, mutations, and timepoints tested. In my opinion, this work is a very welcome addition to the literature on fly immune responses, which obviously do not occur in one type of response at a time, but in parallel, subsequently, and/or are interconnected. I find that the major strength of this work is the overall concept, which is made possible by the mutations designed to target the specific immune function of each module (at least seemingly) without major effects on other functions. I believe that the combinatory mutants will be of use for the fly community and enable further studies of the interplay of these components of immune response in various settings.

To control for the effects arising from the genetic variation other than the intended mutations, the mutants have been backcrossed into a widely used, isogenized Drosophila strain called w1118. Therefore, the differences accounted for by the genotype are controlled.

I also appreciate that the authors have investigated the two possible ways of dealing with an infection: tolerance and resistance, and how the modules play into those.

Weaknesses:

While controlling for the background effects is vital, the w1118 background is problematic (an issue not limited to this manuscript) because of the wide effects of the white mutation on several phenotypes (also other than eye color/eyesight). It is a possibility that the mutation influences the functionality of the immune response components, for example, via effects of the faulty tryptophan handling on the metabolism of the animal.

I acknowledge that it is not reasonable to ask for data in different backgrounds better representing a "wild type" fly (however, that is defined is another question), but I think this matter should be brought up and discussed.

The whole study has been conducted on male flies. Immune responses show quite extensive sex-specific variation across a variety of species studied, also in the fly. But the reasons for this variation are not fully understood. Therefore, I suggest that the authors conduct a subset of experiments on female flies to see if the findings apply to both sexes, especially the infection-specificity of the module combinations.

eLife Assessment

This important study reports an advancement in the diagnosis of Animal African Trypanosomosis (AAT), which adapts a CRISPR-based diagnostic tool (SHERLOCK4AAT) to detect different trypanosome species responsible for AAT. The evidence supporting the conclusions is convincing and in line with the current state-of-the-art diagnostics. This study will be of interest to the fields of Epidemiology, Public Health, and Veterinary Medicine.

Reviewer #1 (Public review):

Summary:

This study addresses a critical gap in veterinary diagnostics by developing a CRISPR-based diagnostic toolbox (SHERLOCK4AAT) for detecting animal African trypanosomosis. It describes the development and field deployment of SHERLOCK4AAT, a CRISPR-Cas13-based diagnostic toolbox for the eco-epidemiological surveillance of animal African trypanosomosis (AAT) in West Africa.

The authors successfully created and validated species-specific assays for multiple trypanosomes, including T. congolense, T. vivax, T. theileri, T. simiae, and T. suis, alongside pan-trypanosomatid and pan-Trypanozoon assays. The field validation in pigs from Guinea and Côte d'Ivoire revealed high trypanosome prevalence (62.7%), frequent co-infections, and importantly identified T. b. gambiense in one animal at each site, suggesting pigs may serve as potential reservoirs for this human-infective parasite.

A major strength of the study lies in its methodological innovation. By adapting SHERLOCK to target both conserved and species-discriminating sequences, the authors achieved high sensitivity and specificity in detecting Trypanosoma species. Their use of dried blood spots, validated thresholds through ROC analyses, and statistical robustness (e.g., Bayesian latent class modeling) provides a strong foundation for their conclusions.

The results are significant: over 60% of pigs tested positive for at least one trypanosome species, with co-infections observed frequently and T. b. gambiense detected in pigs at both sites. These findings have direct implications for the role of animal reservoirs in human disease transmission and underscore the value of pigs as sentinel hosts in gHAT elimination efforts.

The limitations are well acknowledged, particularly the suboptimal sensitivity of the T. vivax assay and the reliance on synthetic controls for T. suis and T. simiae. However, these limitations do not undermine the overall conclusions, and the paper provides a clear roadmap for further assay refinement and implementation.

This study offers a timely, impactful, and well-substantiated contribution to the field. The SHERLOCK4AAT toolbox holds promise for improving AAT diagnostics in resource-limited settings and advancing One Health surveillance frameworks.

Strengths:

(1) The adaptation of SHERLOCK technology for AAT represents a significant technical advancement, offering higher sensitivity than traditional parasitological methods and the ability to detect multiple species simultaneously.

(2) Rigorously performed with validation using appropriate controls, ROC curve analyses, and Bayesian latent class modelling, establishing clear analytical sensitivity and specificity for most assays.

(3) Testing 424 pig samples across two countries provides robust evidence of the tool's utility and reveals important epidemiological insights about trypanosome diversity and prevalence.

(4) The identification of T. b. gambiense in pigs at both sites has significant implications for HAT elimination strategies and highlights the need for integrated One Health approaches.

(5) The use of dried blood spots and RNA detection for active infections makes the approach practical for field surveillance in resource-limited settings.

Weaknesses:

(1) The manuscript would benefit from more detailed discussion of practical considerations such as cost, equipment requirements, and training needs for implementing SHERLOCK in endemic areas and rural settings which would improve applicability.

(2) Limited discussion of pig selection criteria: More justification for choosing pigs as sentinel animals and discussion of potential limitations of this approach would strengthen the manuscript.

(3) More details on why certain genes were targeted would strengthen the methods.

(4) Table formatting could be improved for readability.

(5) Some figures are complex and would benefit from additional explanations in the legends.

Reviewer #2 (Public review):

Summary:

The manuscript is important due to the significance of the findings. The strength of evidence is convincing.

Strengths:

(1) Using a Novel SHERLOCK4AAT toolkit for diagnosis.

(2) Identification of various sub-species of Trypanosomes.

(3) Differentiating the animal subspecies from the human one.

Weaknesses:

(1) The title is too long, and the use of definite articles should be reduced in the title.

(2) The route of blood sample collection in the animals should be well defined and explained.

Reviewer #3 (Public review):

Summary:

The study adapts CRISPR-based detection toolkit (SHERLOCK assay) using conserved and species-specific targets for the detection of some members of the Trypanosomatidae family of veterinary importance and species-specific assays to differentiate between the six most common animal trypanosome species responsible for AAT (SHERLOCK4AAT). The assays were able to discriminate between Trypanozoon (T. b. brucei, T. evansi, and T. equiperdum), T. congolense (Savanah, Forest Kilifi, and Dzanga sangha), T. vivax, T. theileri, T. simiae, and T. suis. The design of both broad and species-specific assays was based primarily on sequences of the 18S rRNA, GAPDH (Glyceraldehyde-3-phosphate dehydrogenase), and invariant flagellum antigen (IFX) genes for species identification. Most importantly, the authors showed varying limits of detection for the different SHERLOCK assays, which is somewhat comparable to PCR-derived molecular techniques currently used for detecting animal trypanosomes, even though some of these methodologies have used other primers that target genes such as ITS1 and 7SL sRNA.

The data presented in the study are particularly useful and of significant interest for the diagnosis of AAT in affected areas.

Strengths:

The assays convincingly allow for the analysis and detection of most trypanosomes in AAT.

Weaknesses:

Inability for the assay to distinguish T. b. brucei, T. evansi, and T. equiperdum using the 18S rRNA gene, as well as the IFX gene, not achieving the sensitivity requirements for detection of T. vivax. Both T. brucei brucei and T. vivax are the most predominant infective species in animals (in addition to T. congolense), therefore, a reliable assay should be able to convincingly detect these to allow for proper use of the diagnostic assay.

eLife Assessment

This important study investigates frequency-dependent effects of transcutaneous tibial nerve stimulation (TTNS) on bladder function in healthy humans and, through a computational model, shows that low-frequency stimulation accelerates, and high-frequency delays, the urge to void. The integration of experimental and modeling approaches provides a solid foundation for clinical trials targeting urinary retention. However, concerns were raised about over-interpretation of modest effects and the limited physiological validity of the computational model, especially its mismatch with typical bladder behaviour and lack of quantitative validation.

Reviewer #1 (Public review):

Summary:

The research investigates the frequency-dependent effects of transcutaneous tibial nerve stimulation (TTNS) on bladder function in healthy humans and via a computational model. The authors report that low-frequency (1 Hz) TTNS accelerates the urge to void, while high-frequency (20 Hz) TTNS delays it, corroborated by a computational model suggesting brainstem-mediated mechanisms. The work bridges experimental and theoretical approaches to propose a novel framework for TTNS applications in urinary retention.

Strengths:

(1) The integration of human experiments and computational modeling is a major strength. The model successfully replicates bladder dynamics and provides mechanistic insights into frequency-dependent effects.

(2) Identifies potential therapeutic applications for urinary retention, a condition with limited non-invasive treatments.

(3) Figures are clear and illustrative, and supplementary materials provide essential methodological depth.

(4) Controlled experimental design (eg., single-blinded, fluid/caffeine restrictions, etc), detailed computational model parameters and validation against animal data, transparency in data exclusion criteria and statistical adjustments.

Weaknesses:

(1) The study uses healthy participants; extrapolation to clinical populations (e.g., urinary retention patients) requires validation.

(2) The simulated bladder capacity (100-150 mL) is lower than physiological ranges (300-400 mL). While the authors note this, the impact on model validity should be further addressed.

(3) The model omits nociceptive afferents, limiting its applicability to pathological conditions like overactive bladder.

(4) The lack of significant differences in urge intensity between groups (despite timing differences) warrants deeper discussion. Is the primary effect on efferent activity (as suggested) rather than sensory perception?

(5) One of the highlights of this study is the identification of the effect of low-frequency (1 Hz) tibial nerve stimulation (TNS) on facilitating bladder contraction. Although the authors have clarified this effect in healthy participants, it would strengthen the conclusion if a UAB animal model (e.g., PMCID: PMC7927909, PMC8163611, PMC7847056, PMC8799394) were used to evaluate the same effect.

Reviewer #2 (Public review):

Summary:

Tibial nerve (electrical) stimulation (TNS) has emerged over the past 15 years as a non-invasive method to treat bladder overactivity, but interestingly, new animal work has suggested that TNS could actually be used to excite the bladder when appropriately tuning the stimulation frequency, effectively inverting its effect, perhaps opening the door to treat different conditions (e.g., UAB). The present study tests how healthy people respond to low and high frequency TNS, with the authors showing that they can substantially delay people's first sensation of bladder fullness with high frequencies (20Hz, shown many times before) but also that they can slightly hasten people's first sensation with low frequencies (1Hz, new result in humans). Moreover, the authors develop a computational model of interconnected conductance-based simulated neurons arranged in a physiologically plausible circuit that reproduces some aspects of the frequency-dependent effects of TNS. Their simulations suggest that we might expect low-frequency TNS to also increase the duration of bladder contractions in humans. The study highlights a potential new research direction, optimizing TNS stimulation parameters to increase basal bladder excitability.

Strengths:

The main strength of the work is to call attention to a new possibility of inverting the effect of TNS in humans by manipulating stimulation frequency, opening new indications for the therapy. This is highly relevant because of the recent popularity of TNS and its non-invasiveness, which lends itself to rapid testing and evaluation for new conditions and a high willingness to adopt. The authors convincingly demonstrate a modest excitatory effect on bladder sensation with low-frequency TNS, which clearly warrants further investigation.

The high-level design of the hypotheses, concepts, and experiments is clearly articulated in both the methods and in particularly clear diagrams, letting the reader focus their attention on the most important findings.

It is rare to develop a new computational model of the lower urinary tract at a systems level, and even more so for it to incorporate circuits in the spinal cord and brainstem centers, and this work undoubtedly advances the field's ability to engineer such systems. Further, because the model is comprised of linked conductance-based point-neurons, it is an excellent tool to investigate how an arguably plausible wiring diagram for neural control of the LUT could result in stimulation frequency-dependent effects on pelvic efferents. It is a proof of concept demonstrating how their mechanistic hypothesis of TNS could be implemented neurophysiologically by the nervous system.

Weaknesses:

The main drawback of the work is the frequent overinterpretation of the results. The human study and computational model are both proof-of-principle studies because the experimental effect size and sample size are modest, and the computational model is poorly validated and does not generate physiologically typical cystometric responses in simulations that are designed to recapitulate nominal LUT behavior.

Despite the stated caveats about the small effect in the human study, it should be emphasized throughout that this result is most reasonably interpreted as showing the possibility that TNS can have a low-frequency excitatory effect that merits follow-up, rather than a conclusive demonstration. The effect size is small (as the authors note) and should be placed in context with some minimally clinically important difference, if possible. The result is statistically significant, but even this may be subject to revision due to the small sample and the effect of post-hoc outlier removal and data analysis choices.

Given the apparent mismatch between the model and the cystometric behavior at the systems level in the "normal" case (e.g., low capacity, low voiding efficiency, omitted pressure profiles, frequency, etc.) and the absence of quantitative model validation (e.g., it was not compared directly with any experimental data from human urodynamics or rodent cystometry, beyond the initial fit to the neural data, no sensitivity analyses were performed, no goodness of fit computed, etc.) the discussion should be much more circumspect about interpreting the results at a systems level and should probably contain a paragraph explicitly detailing the limitations of the model. The subsequent interpretation should focus narrowly on the neural circuitry, rather than things like contraction duration, where the model is at its strongest. As written, the authors over-interpret what the in silico study can reasonably be used to infer about LUT function.

More justification is needed for why the contraction duration of the model is the central focus of analysis, when it connects only tentatively to the human study results, which focus on urgency. While not necessarily incorrect, a clearer link or motivation should be offered for how this informs our understanding of frequency-dependent TNS afferent or efferent inhibition during filling (which was the focus of the human studies and the abstract). In other words, why doesn't the model reproduce the 1Hz excitation effect of expediting void onset (or urgency in the human study), and why is it justified to look at contraction duration as a surrogate measure?

The authors claim that "voiding behavior occurred earlier [at 1Hz stim in the model]", pointing to Figure 6A as evidence, but this panel appears to show a single example model run where 1Hz voiding occurs only ~1s earlier (display makes this very hard to estimate). This is insufficient evidence to support the claim. Later, it is stated that "TNS did not ... void much earlier". The claims should be made compatible, and all such claims should have reasonable supporting evidence.

There are a number of reporting concerns that can be easily addressed:

(1) Human Study:

(a) To interpret the human study analysis, a fuller description of the "optional 10m inute extension" is necessary. How were participants presented with this option, how was blinding preserved, what fraction of participants accepted, and did phase 1 results influence their decisions to continue?

(b) For reproducibility, details about the TNS parameters should be articulated, such as the method of determining "motor thresholds" (unless this is synonymous with "urge to urinate"), the shape of the stimulation pulses (e.g., biphasic, charge balanced), typical applied current, etc.

(2) The Computational Model

(a) The code availability statement for this type of work is inadequate. The model used for simulations in this work, as well as the code used to initialize (and randomize synaptic connections), needs to be hosted publicly because i) a model this intricate is extremely hard to reproduce/verify without code, ii) simulations are an essential piece of the argument, iii) hosting code requires very little overhead. Although there is an appropriate level of detail in the model description, it would not be possible to reproduce the model in any reasonable amount of time (or at all) because of the implementation-level details that are, understandably, omitted from the methods (e.g., what is a "unit", what 'exactly' do the connections in the PMC and PAG diagrams relate to, what were the final parameters used for all conductances, which parameters were "matched" to the original papers and which were not, etc.).

b) Critical cystometric/urodynamic values that are typically analyzed to assess healthy LUT function are detrusor pressure (timeseries) and/or post-void residual or voiding efficiency (scalars). These should be included to verify that the model is representative of the "normal" case. This is especially important because the model's "normal" behavior appears to have extremely low voiding efficiency (Figure 6A).

eLife Assessment

This valuable study provides insights into the structure and function of bacterial contractile injection systems that are present in the cytoplasm of many Streptomyces strains. A convincing high-resolution model of the structure of extended forms of the cytoplasmic contractile injection system assembly from Streptomyces coelicolor is presented, with some investigation of the membrane protein CisA in attachment of the extended assembly to the inner face of the cytoplasmic membrane and the firing of the system. The work expands the current understanding of these diverse bacterial nanomachines.

Reviewer #2 (Public review):

Summary:

The paper addresses how the S. coelicolor contractile injection system (CISSc) interacts with the membrane, how it contracts and fires, and how it affects both cell viability and differentiation, which it has been implicated to do in previous work from this group and others. The Streptomyces CIS systems have been enigmatic in the sense that they are free-floating in the cytoplasm in an extended form and are seen in contracted conformation (i.e. after having been triggered) mainly in dead and partially lysed cells, suggesting involvement in some kind of regulated cell death. So, how do the structure and function of the CISSc system compare to other types of CIS from other bacteria and phages, does it interact with the cytoplasmic membrane, how does it do that, and is the membrane interaction involved in the suggested role in stress-induced, regulated cell death? The authors address these questions by investigating the role of a membrane protein, CisA, that is encoded by a gene in the CIS gene cluster in S. coelicolor. Further, they show for the first time the structure of the assembled CISSc, purified from the cytoplasm of S. coelicolor, analysed using single-particle cryo-electron microscopy.

Strengths:

The beautiful visualisation of the CIS system both by cryo-electron tomography of intact bacterial cells and by single-particle electron microscopy of purified CIS assemblies are clearly the strengths of the paper, both in terms of methods and results. Further, the paper provides genetic evidence that the membrane protein CisA is required for the contraction of the CISSc assemblies that are seen in partially lysed or ghost cells of the wild type. The conclusion that CisA is a transmembrane protein and the inferred membrane topology are well supported by experimental data. The cryo-EM data suggest that CisA is not a stable part of the extended form of the CISSc assemblies. These findings raise the question of what CisA does. Interestingly, Alphafold modelling suggests that the cytoplasmic part of CisA interacts directly with the base plate protein Cis11.

Weaknesses:

The investigations of the role of CisA in function, membrane interaction, and triggering of contraction of CIS assemblies are key parts of the paper and are highlighted in the title. However, the data presented to answer these questions are partially incomplete and have some limitations.

As an example, although the modelling that suggests interaction between CisA and the base plate protein Cis11 appears compelling, the interaction has not yet been possible to test and verify experimentally. Further, it remains unclear whether or how CisA recruits the CISSc system to the membrane. Overall, the mechanism by which CisA may act on CISSc and cause firing remains largely unclear.

Further, the paper does not provide new insights into the role of the CISSc system in growth or developmental biology of streptomycetes. The assay of how CisA affects the function of the system involves monitoring stress-induced loss of viability based on loss of cytoplasmic GFP signal, as described in a previous paper. The assay looks only at single hyphal fragments released from mycelial networks or mycelial pellets, and it could have been interesting to observe effects also under other growth conditions. Similarly, the effect on the developmental life cycle is limited to showing accelerated sporulation in the CisA mutant, similar to what was previously shown for mutants lacking other parts of the system. The paper shows that CisA is needed for the observed phenotypic effects of the CISSc system, but the overall biological roles of the CISSc and CisA remain elusive.

Concluding remarks:

This paper provides new insights into the structure of the unusual subclass of bacterial contractile injection systems (CIS) that is constituted by the cytoplasmically located systems found in streptomycetes. Importantly, the work also describes a membrane protein, CisA, that likely links the CISSc to the cytoplasmic membrane and is required for its function and likely its triggering. The paper will be of large interest in the field, and it will likely be the basis for further and more mechanistic and functional investigations of the Streptomyces CIS systems.

Reviewer #3 (Public review):

Summary

In this work, Casu et al. have reported the characterization of a previously uncharacterized membrane protein CisA encoded in a non-canonical contractile injection system of Streptomyces coelicolor, CISSc, which is a cytosolic CISs significantly distinct from both intracellular membrane-anchored T6SSs and extracellular CISs. The authors have presented the first high-resolution structure of the extended CISSc structure. It revealed important structural insights of the extended state of this non-canonical CIS.

To further explore how CISSc interacted with cytoplasmic membrane, they further set out to investigate a membrane protein CisA encoded in the CISSc cluster and previously hypothesized to be the membrane adaptor for CISSc; however, the structure revealed that it was not associated with CISSc. Using a fluorescence microscope and cell fractionation assay, the authors verified that CisA is indeed a membrane-associated protein. They further determined experimentally that CisA had a cytosolic N-terminal domain and a periplasmic C-terminus. The functional analysis of cisA mutant revealed that it is not required for CISSc assembly but is essential for the contraction, as a result, the deletion significantly affects CISSc-mediated cell death upon stress, timely differentiation, as well as secondary metabolite production. Although the work did not resolve the mechanistic detail how CisA interacts with CISSc structure, they used in-silico prediction of protein-protein interactions between monomeric CisA and CISSc components using Alphafold2-Multimer, which identified baseplate protein Cis11 as a potential interaction partner. Such prediction sets out a strong basis for future investigations to explore the molecular mechanistic details how CisA mediates the contraction via interactions with the CIS structural components such as Cis11. Using AlphaFold3, the authors also estimated the oligomerization state of CisA, which can be present as a pentamer. Authors further suggested that such oligomerization is mediated by the interaction of C-terminal solute-binding like domain.

In general, the work provides solid data and a strong foundation for future investigation toward understanding the mechanism of CISSc contraction, and potentially, the relation between the membrane association of CISSc, the sheath contraction and the cell death.

Major Strength:

The paper is well-structured, and the conclusion of the study is supported by solid data and careful data interpretation were presented. The authors provided strong evidence on (1) the high-resolution structure of extended CISSc determined by cryo-EM, and the subsequent comparison with known eCIS structures, which sheds light on both its similarity and different features from other subtypes of eCISs in detail; (2) the topological features of CisA using fluorescence microscopic analysis, cell fractionation and PhoA-LacZα reporter assays, (3) functions of CisA in CISSc-mediated cell death and secondary metabolite production, likely via the regulation of sheath contraction, (4) structural prediction of the oligomerization state of CisA and potential interaction partners of CIS structure.

Weakness:

Due to technical limitations, authors are not able to experimentally demonstrate the direct interaction between CisA with baseplate complex of CISSc, since they could not express cisA in E. coli due to its potential toxicity. Therefore, there is a lack of biochemical analysis of direct interaction between CisA and baseplate wedge. However, they have provided solid AlphaFold2-multimer prediction data and identified baseplate protein Cis11 as a potential interaction partner. Such predictions will guide future work towards biochemical analysis to verify such interaction.

While there is no direct evidence showing that CisA is responsible for tethering CISSc to the membrane upon stress, and the spatial and temporal relation between membrane association and contraction remains unclear, I recognize that this is above the scope of the current work, so I would expect further investigation to address these questions in future.

Conclusion

Overall, the work provides a valuable contribution to our understanding on the structure of a much less understood subtype of CISs, which is unique compared to both membrane-anchored T6SSs and host-membrane targeting eCISs. Authors have successfully demonstrated the role of CisA in the contraction of CISSc, along with solid and detailed analysis of the contraction state of the particles with or without CisA using cryo-ET. Using structural modeling, authors also identified the potential oligomerization state and possible interaction partner within the CIS particle.

Importantly, the work serves as a strong foundation to further investigate how the sheath contraction works here. The work contributes to expanding our understanding of the diverse CIS superfamilies, with significant novelty.

eLife Assessment

This important study provides information on the TMEM16 family of membrane proteins, which play roles in lipid scrambling and ion transport. By simulating 27 structures representing five distinct family members, the authors captured hundreds of lipid scrambling events, offering insights into the mechanisms of lipid translocation and the specific protein regions involved in these processes. While the data on comparison of scrambling competence is compelling, the evidence for outside-the-groove scramblase activity without experimental validation is missing and is based on a limited set of observed events.

Reviewer #1 (Public review):

Summary:

The manuscript investigates lipid scrambling mechanisms across TMEM16 family members using coarse-grained molecular dynamics (MD) simulations. While the study presents a statistically rigorous analysis of lipid scrambling events across multiple structures and conformations, several critical issues undermine its novelty, impact, and alignment with experimental observations.

Review on revised version:

The referee notes that the authors, in their response letter, have concurred with most of the concerns originally raised. Specifically, the authors acknowledge the referee's view that the manuscript primarily confirms previously reported findings and does not present a significantly novel advance, particularly regarding the central observation of groove-mediated lipid scrambling in the open Ca²⁺-bound TMEM16 structures. The authors have also acknowledged the potential discrepancies with existing experimental studies and have addressed this point candidly through additional discussion. Furthermore, the referee appreciates that the authors have echoed the concern regarding the limited statistical robustness of the observed scrambling events.<br /> Given that the authors have essentially affirmed the key points raised in the initial review, the referee believes that these acknowledgements reinforce the basis of the original assessment. Therefore, the referee maintains the original opinion that, despite its technical merits and useful discussion made in the revised version, the manuscript does not offer sufficient novelty or mechanistic depth.

Reviewer #2 (Public review):

Summary:

Stephens et al. present a comprehensive study of TMEM16-members via coarse-grained MD simulations (CGMD). They particularly focus on the scramblase ability of these proteins and aim to characterize the "energetics of scrambling". Through their simulations, the authors interestingly relate protein conformational states to membrane's thickness and link those to the scrambling ability of TMEM members, measured as the trespassing tendency of lipids across leaflets. They validate their simulation with a direct qualitative comparison with Cryo-EM maps.

Strengths:

The study demonstrates an efficient use of CGMD simulations to explore lipid scrambling across various TMEM16 family members. By leveraging this approach, the authors are able to bypass some of the sampling limitations inherent in all-atom simulations, providing a more comprehensive and high-throughput analysis of lipid scrambling. Their comparison of different protein conformations, including open and closed groove states, presents a detailed exploration of how structural features influence scrambling activity, adding significant value to the field. A key contribution of this study is the finding that groove dilation plays a central role in lipid scrambling. The authors observe that for scrambling-competent TMEM16 structures, there is substantial membrane thinning and groove widening. The open Ca2+-bound nhTMEM16 structure (PDB ID 4WIS) was identified as the fastest scrambler in their simulations, with scrambling rates as high as 24.4 {plus minus} 5.2 events per μs. This structure also shows significant membrane thinning (up to 18 Å), which supports the hypothesis that groove dilation lowers the energetic barrier for lipid translocation, facilitating scrambling.

The study also establishes a correlation between structural features and scrambling competence, though analyses often lack statistical robustness and quantitative comparisons. The simulations differentiate between open and closed conformations of TMEM16 structures, with open-groove structures exhibiting increased scrambling activity, while closed-groove structures do not. This finding aligns with previous research suggesting that the structural dynamics of the groove are critical for scrambling. Furthermore, the authors explore how the physical dimensions of the groove qualitatively correlate with observed scrambling rates. For example, TMEM16K induces increased membrane thinning in its open form, suggesting that membrane properties, along with structural features, play a role in modulating scrambling activity.

Another significant finding is the concept of "out-of-the-groove" scrambling, where lipid translocation occurs outside the protein's groove. This observation introduces the possibility of alternate scrambling mechanisms that do not follow the traditional "credit-card model" of groove-mediated lipid scrambling. In their simulations, the authors note that these out-of-the-groove events predominantly occur at the dimer interface between TM3 and TM10, especially in mammalian TMEM16 structures. While these events were not observed in fungal TMEM16s, they may provide insight into Ca2+-independent scrambling mechanisms, as they do not require groove opening.

Weaknesses:

A significant challenge of the study is the discrepancy between the scrambling rates observed in CGMD simulations and those reported experimentally. Despite the authors' claim that the rates are in line experimentally, the observed differences can mean large energetic discrepancies in describing scrambling (larger than 1kT barrier in reality). For instance, the authors report scrambling rates of 10.7 events per μs for TMEM16F and 24.4 events per μs for nhTMEM16, which are several orders of magnitude faster than experimental rates. While the authors suggest that this discrepancy could be due to the Martini 3 force field's faster diffusion dynamics, this explanation does not fully account for the large difference in rates. A more thorough discussion on how the choice of force field and simulation parameters influence the results, and how these discrepancies can be reconciled with experimental data, would strengthen the conclusions. Likewise, rate calculations in the study are based on 10 μs simulations, while experimental scrambling rates occur over seconds. This timescale discrepancy limits the study's accuracy, as the simulations may not capture rare or slow scrambling events that are observed experimentally and therefore might underestimate the kinetics of scrambling. It's however, important to recognize that it's hard (borderline unachievable) to pinpoint reasonable kinetics for systems like this using the currently available computational power and force field accuracy. The faster diffusion in simulations may lead to overestimated scrambling rates, making the simulation results less comparable to real-world observations. Thus, I would therefore read the findings qualitatively rather than quantitatively. An interesting observation is the asymmetry observed in the scrambling rates of the two monomers. Since MARTINI is known to be limited in correctly sampling protein dynamics, the authors, in order to preserve the fold, have applied a strong (500 kJ mol-1 nm-2) elastic network. However, I am wondering how the ENM applies across the dimer and if any asymmetry can be noticed in the application of restraints for each monomer and at the dimer interface. How can this have potentially biased the asymmetry in the scrambling rates observed between the monomers? Is this artificially obtained from restraining the initial structure, or is the asymmetry somehow gatekeeping the scrambling mechanism to occur majorly across a single monomer? Answering this question would have far-reaching implications to better describe the mechanism of scrambling.

Notably, the manuscript does not explore the impact of membrane composition on scrambling rates. While the authors use a specific lipid composition (DOPC) in their simulations, they acknowledge that membrane composition can influence scrambling activity. However, the study does not explore how different lipids or membrane environments or varying membrane curvature and tension, could alter scrambling behaviour. I appreciate that this might have been beyond the scope of this particular paper and the authors plan to further chase these questions, as this work sets a strong protocol for this study. Contextualizing scrambling in the context of membrane composition is particularly relevant since the authors note that TMEM16K's scrambling rate increases tenfold in thinner membranes, suggesting that lipid-specific or membrane-thickness-dependent effects could play a role.

Comments on revisions:

I have carefully reviewed the replies of the author, which address the points I raised and improved the manuscript by making the changes outlined in their response. Particularly, I am pleased to see that the authors report ensemble averages in Figure 1-supplement 1 and add relevant information in a newly created table. I welcome the refinement of the discussion towards a cautionary approach in describing quantitatively the findings of experiments and computations for what concerns scrambling rates. I still feel that proper statistical analysis to compare the distributions in Figure 3-figure supplement 6 would have made the points claimed even stronger, but - at the same time - I do see the points of the authors in commenting the differences between these distributions more qualitatively. Overall, I support the publication of this manuscript, it has been a pleasure to read it.

Reviewer #3 (Public review):

Summary:

The paper investigates the TMEM16 family of membrane proteins, which play roles in lipid scrambling and ion transport. A total of 27 experimental structures from five TMEM16 family members were analyzed, including mammalian and fungal homologs (e.g., TMEM16A, TMEM16F, TMEM16K, nhTMEM16, afTMEM16). The identified structures were in both Ca²⁺-bound (open) and Ca²⁺-free (closed) states to compare conformations and were preprocessed (e.g., modeling missing loops) and equilibrated. Coarse-grain simulations were performed in DOPC membranes for 10 microseconds to capture the scrambling events. These events were identified by tracking lipids transitioning between the two membrane leaflets and they analysed correlation between scrambling rates, in addition, structural properties such as groove dilation and membrane thinning were calculated. They report 700 scrambling events across structures and the figure 2 elaborates on how open structures show higher activity, also as expected. The authors also address how structures may require open groove, this and other mechanisms around scrambling is a bit controversial in the field.

Strengths:

The strength of this study emerges from comparative analysis of multiple structural starting points and understand global/local motions of the protein with respect to lipid movement. Although the protein is well-studied, both experimentally and computationally, the understanding of conformational events in different family members, especially membrane thickness less compared to fungal scramblases offers good insights.

Weaknesses:

The weakness of the work is to fully reconcile with experimental evidence of Ca²⁺-independent scrambling rates observed in prior studies, but this part is also challenging using coarse-grain molecular simulations. Previous reports have identified lipid crossing, packing defects and other associated events, so it is difficult to place this paper in that context. However, the absence of validation leaves certain claims, like alternative scrambling pathways, speculative.

Author response:

The following is the authors’ response to the current reviews.

We wanted to clarify Reviewer #1’s latest comment in the last round of review, “Furthermore, the referee appreciates that the authors have echoed the concern regarding the limited statistical robustness of the observed scrambling events.” We appreciate the follow up information provided from Reviewer #1 that their comment is specifically about the low count alternative pathway events that we view at the dimer interface, and not the statistics of the manuscript overall as they believe that “the study presents a statistically rigorous analysis of lipid scrambling events across multiple structures and conformations (Reviewer #1)”. We agree with the Reviewer and acknowledge that overall our coarse-grained study represents the most comprehensive single manuscript of the entire TMEM16 family to date.

The following is the authors’ response to the original reviews.

Public Review:

Reviewer #1 (Public review):

Summary:

The manuscript investigates lipid scrambling mechanisms across TMEM16 family members using coarse-grained molecular dynamics (MD) simulations. While the study presents a statistically rigorous analysis of lipid scrambling events across multiple structures and conformations, several critical issues undermine its novelty, impact, and alignment with experimental observations.

Critical issues:

(1) Lack of Novelty:

The phenomenon of lipid scrambling via an open hydrophilic groove is already well-established in the literature, including through atomistic MD simulations. The authors themselves acknowledge this fact in their introduction and discussion. By employing coarse-grained simulations, the study essentially reiterates previously known findings with limited additional mechanistic insight. The repeated observation of scrambling occurring predominantly via the groove does not offer significant advancement beyond prior work.

We agree with the reviewer’s statement regarding the lack of novelty when it comes to our observations of scrambling in the groove of open Ca2+-bound TMEM16 structures. However, we feel that the inclusion of closed structures in this study, which attempts to address the yet unanswered question of how scrambling by TMEM16s occurs in the absence of Ca2+, offers new observations for the field. In our study we specifically address to what extent the induced membrane deformation, which has been theorized to aid lipids cross the bilayer especially in the absence of Ca2+, contributes to the rate of scrambling (see references 36, 59, and 66). There are also several TMEM16F structures solved under activating conditions (bound to Ca2+ and in the presence of PIP2) which feature structural rearrangements to TM6 that may be indicative of an open state (PDB 6P48) and had not been tested in simulations. We show that these structures do not scramble and thereby present evidence against an out-of-the-groove scrambling mechanism for these states. Although we find a handful of examples of lipids being scrambled by Ca2+-free structures of TMEM16 scramblases, none of our simulations suggest that these events are related to the degree of deformation.

(2) Redundancy Across Systems:

The manuscript explores multiple TMEM16 family members in activating and non-activating conformations, but the conclusions remain largely confirmatory. The extensive dataset generated through coarse-grained MD simulations primarily reinforces established mechanistic models rather than uncovering fundamentally new insights. The effort, while statistically robust, feels excessive given the incremental nature of the findings.

Again, we agree with the reviewer’s statement that our results largely confirm those published by other groups and our own. We think there is however value in comparing the scrambling competence of these TMEM16 structures in a consistent manner in a single study to reduce inconsistencies that may be introduced by different simulation methods, parameters, environmental variables such as lipid composition as used in other published works of single family members. The consistency across our simulations and high number of observed scrambling events have allowed us to confirm that the mechanism of scrambling is shared by multiple family members and relies most obviously on groove dilation.

(3) Discrepancy with Experimental Observations:

The use of coarse-grained simulations introduces inherent limitations in accurately representing lipid scrambling dynamics at the atomistic level. Experimental studies have highlighted nuances in lipid permeation that are not fully captured by coarse-grained models. This discrepancy raises questions about the biological relevance of the reported scrambling events, especially those occurring outside the canonical groove.

We thank the reviewer for bringing up the possible inaccuracies introduced by coarse graining our simulations. This is also a concern for us, and we address this issue extensively in our discussion. As the reviewer pointed out above, our CG simulations have largely confirmed existing evidence in the field which we think speaks well to the transferability of observations from atomistic simulations to the coarse-grained level of detail. We have made both qualitative and quantitative comparisons between atomistic and coarse-grained simulations of nhTMEM16 and TMEM16F (Figure 1, Figure 4-figure supplement 1, Figure 4-figure supplement 5) showing the two methods give similar answers for where lipids interact with the protein, including outside of the canonical groove. We do not dispute the possible discrepancy between our simulations and experiment, but our goal is to share new nuanced ideas for the predicted TMEM16 scrambling mechanism that we hope will be tested by future experimental studies.

(4) Alternative Scrambling Sites:

The manuscript reports scrambling events at the dimer-dimer interface as a novel mechanism. While this observation is intriguing, it is not explored in sufficient detail to establish its functional significance. Furthermore, the low frequency of these events (relative to groove-mediated scrambling) suggests they may be artifacts of the simulation model rather than biologically meaningful pathways.

We agree with the reviewer that our observed number of scrambling events in the dimer interface is too low to present it as strong evidence for it being the alternative mechanism for Ca2+-independent scrambling. This will require additional experiments and computational studies which we plan to do in future research. However, we are less certain that these are artifacts of the coarse-grained simulation system as we observed a similar event in an atomistic simulation of TMEM16F.

Conclusion:

Overall, while the study is technically sound and presents a large dataset of lipid scrambling events across multiple TMEM16 structures, it falls short in terms of novelty and mechanistic advancement. The findings are largely confirmatory and do not bridge the gap between coarse-grained simulations and experimental observations. Future efforts should focus on resolving these limitations, possibly through atomistic simulations or experimental validation of the alternative scrambling pathways.

Reviewer #2 (Public review):

Summary:

Stephens et al. present a comprehensive study of TMEM16-members via coarse-grained MD simulations (CGMD). They particularly focus on the scramblase ability of these proteins and aim to characterize the "energetics of scrambling". Through their simulations, the authors interestingly relate protein conformational states to the membrane's thickness and link those to the scrambling ability of TMEM members, measured as the trespassing tendency of lipids across leaflets. They validate their simulation with a direct qualitative comparison with Cryo-EM maps.

Strengths:

The study demonstrates an efficient use of CGMD simulations to explore lipid scrambling across various TMEM16 family members. By leveraging this approach, the authors are able to bypass some of the sampling limitations inherent in all-atom simulations, providing a more comprehensive and high-throughput analysis of lipid scrambling. Their comparison of different protein conformations, including open and closed groove states, presents a detailed exploration of how structural features influence scrambling activity, adding significant value to the field. A key contribution of this study is the finding that groove dilation plays a central role in lipid scrambling. The authors observe that for scrambling-competent TMEM16 structures, there is substantial membrane thinning and groove widening. The open Ca2+-bound nhTMEM16 structure (PDB ID 4WIS) was identified as the fastest scrambler in their simulations, with scrambling rates as high as 24.4 {plus minus} 5.2 events per μs. This structure also shows significant membrane thinning (up to 18 Å), which supports the hypothesis that groove dilation lowers the energetic barrier for lipid translocation, facilitating scrambling.

The study also establishes a correlation between structural features and scrambling competence, though analyses often lack statistical robustness and quantitative comparisons. The simulations differentiate between open and closed conformations of TMEM16 structures, with open-groove structures exhibiting increased scrambling activity, while closed-groove structures do not. This finding aligns with previous research suggesting that the structural dynamics of the groove are critical for scrambling. Furthermore, the authors explore how the physical dimensions of the groove qualitatively correlate with observed scrambling rates. For example, TMEM16K induces increased membrane thinning in its open form, suggesting that membrane properties, along with structural features, play a role in modulating scrambling activity.

Another significant finding is the concept of "out-of-the-groove" scrambling, where lipid translocation occurs outside the protein's groove. This observation introduces the possibility of alternate scrambling mechanisms that do not follow the traditional "credit-card model" of groove-mediated lipid scrambling. In their simulations, the authors note that these out-of-the-groove events predominantly occur at the dimer interface between TM3 and TM10, especially in mammalian TMEM16 structures. While these events were not observed in fungal TMEM16s, they may provide insight into Ca2+-independent scrambling mechanisms, as they do not require groove opening.

Weaknesses:

A significant challenge of the study is the discrepancy between the scrambling rates observed in CGMD simulations and those reported experimentally. Despite the authors' claim that the rates are in line experimentally, the observed differences can mean large energetic discrepancies in describing scrambling (larger than 1kT barrier in reality). For instance, the authors report scrambling rates of 10.7 events per μs for TMEM16F and 24.4 events per μs for nhTMEM16, which are several orders of magnitude faster than experimental rates. While the authors suggest that this discrepancy could be due to the Martini 3 force field's faster diffusion dynamics, this explanation does not fully account for the large difference in rates. A more thorough discussion on how the choice of force field and simulation parameters influence the results, and how these discrepancies can be reconciled with experimental data, would strengthen the conclusions. Likewise, rate calculations in the study are based on 10 μs simulations, while experimental scrambling rates occur over seconds. This timescale discrepancy limits the study's accuracy, as the simulations may not capture rare or slow scrambling events that are observed experimentally and therefore might underestimate the kinetics of scrambling. It's however important to recognize that it's hard (borderline unachievable) to pinpoint reasonable kinetics for systems like this using the currently available computational power and force field accuracy. The faster diffusion in simulations may lead to overestimated scrambling rates, making the simulation results less comparable to real-world observations. Thus, I would therefore read the findings qualitatively rather than quantitatively. An interesting observation is the asymmetry observed in the scrambling rates of the two monomers. Since MARTINI is known to be limited in correctly sampling protein dynamics, the authors - in order to preserve the fold - have applied a strong (500 kJ mol-1 nm-2) elastic network. However, I am wondering how the ENM applies across the dimer and if any asymmetry can be noticed in the application of restraints for each monomer and at the dimer interface. How can this have potentially biased the asymmetry in the scrambling rates observed between the monomers? Is this artificially obtained from restraining the initial structure, or is the asymmetry somehow gatekeeping the scrambling mechanism to occur majorly across a single monomer? Answering this question would have far-reaching implications to better describe the mechanism of scrambling.

The main aim of our computational survey was to directly compare all relevant published TMEM16 structures in both open and closed states using the Martini 3 CGMD force field. Our standardized simulation and analysis protocol allowed us to quantitatively compare scrambling rates across the TMEM16 family, something that has never been done before. We do acknowledge that direct comparison between simulated versus experimental scrambling rates is complicated and is best to be interpreted qualitatively. In line with other reports (e.g., Li et al, PNAS 2024), lipid scrambling in CGMD is 2-3 orders of magnitude faster than typical experimental findings. In the CG simulation field, these increased dynamics due to the smoother energy landscape are a well known phenomenon. In our view, this is a valuable trade-off for being able to capture statistically robust scrambling dynamics and gain mechanistic understanding in the first place, since these are currently challenging to obtain otherwise. For example, with all-atom MD it would have been near-impossible to conclude that groove openness and high scrambling rates are closely related, simply because one would only measure a handful of scrambling events in (at most) a handful of structures.

Considering the elastic network: the reviewer is correct in that the elastic network restrains the overall structure to the experimental conformation. This is necessary because the Martini 3 force field does not accurately model changes in secondary (and tertiary) structure. In fact, by retaining the structural information from the experimental structures, we argue that the elastic network helped us arrive at the conclusion that groove openness is the major contributing factor in determining a protein’s scrambling rate. This is best exemplified by the asymmetric X-ray structure of TMEM16K (5OC9), in which the groove of one subunit is more dilated than the other. In our simulation, this information was stored in the elastic network, yielding a 4x higher rate in the open groove than in the closed groove, within the same trajectory.

Notably, the manuscript does not explore the impact of membrane composition on scrambling rates. While the authors use a specific lipid composition (DOPC) in their simulations, they acknowledge that membrane composition can influence scrambling activity. However, the study does not explore how different lipids or membrane environments or varying membrane curvature and tension, could alter scrambling behaviour. I appreciate that this might have been beyond the scope of this particular paper and the authors plan to further chase these questions, as this work sets a strong protocol for this study. Contextualizing scrambling in the context of membrane composition is particularly relevant since the authors note that TMEM16K's scrambling rate increases tenfold in thinner membranes, suggesting that lipid-specific or membrane-thickness-dependent effects could play a role.

Considering different membrane compositions: for this study, we chose to keep the membranes as simple as possible. We opted for pure DOPC membranes, because it has (1) negligible intrinsic curvature, (2) forms fluid membranes, and (3) was used previously by others (Li et al, PNAS 2024). As mentioned by the reviewer, we believe our current study defines a good, standardized protocol and solid baseline for future efforts looking into the additional effects of membrane composition, tension, and curvature that could all affect TMEM16-mediated lipid scrambling.

Reviewer #3 (Public review):

Strengths:

The strength of this study emerges from a comparative analysis of multiple structural starting points and understanding global/local motions of the protein with respect to lipid movement. Although the protein is well-studied, both experimentally and computationally, the understanding of conformational events in different family members, especially membrane thickness less compared to fungal scramblases offers good insights.

We appreciate the reviewer recognizing the value of the comparative study. In addition to valuable insights from previous experimental and computational work, we hope to put forward a unifying framework that highlights various TMEM16 structural features and membrane properties that underlie scrambling function.

Weaknesses:

The weakness of the work is to fully reconcile with experimental evidence of Ca²⁺-independent scrambling rates observed in prior studies, but this part is also challenging using coarse-grain molecular simulations. Previous reports have identified lipid crossing, packing defects, and other associated events, so it is difficult to place this paper in that context. However, the absence of validation leaves certain claims, like alternative scrambling pathways, speculative.

Answer: It is generally difficult to quantitatively compare bulk measurements of scrambling phenomena with simulation results. The advantage of simulations is to directly observe the transient scrambling events at a spatial and temporal resolution that is currently unattainable for experiments. The current experimental evidence for the precise mechanism of Ca2+-independent scrambling is still under debate. We therefore hope to leverage the strength of MD and statistical rigor of coarse-grained simulations to generate testable hypotheses for further structural, biochemical, and computational studies.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

The findings are largely confirmatory and do not bridge the gap between coarse-grained simulations and experimental observations. Future efforts should focus on resolving these limitations, possibly through atomistic simulations or experimental validation of the alternative scrambling pathways.

While we agree with what the reviewer may be hinting at regarding limitations of coarse-grained MD simulations, we believe that our study holds much more merit than this comment suggests. We have provided something that has yet to be done in the field: a comprehensive study that directly compares the scrambling rates of multiple TMEM16 family members in different conformations using identical simulation conditions. Our work clearly shows that a sufficiently dilated grooves is the major structural feature that enables robust scrambling for all TMEM16 scramblases members with solved structures. While all TMEM16s cause significant distortion and thinning of the membrane, we assert that the extreme thinning observed around open grooves is significantly enhanced by the lipid scrambling itself as the two leaflets merge through lipid exchange.  We saw no evidence that membrane thinning/distortion alone, in the absence of an open groove, could support scrambling at the rates observed under activating conditions or even the low rates observed in Ca2+-independent scrambling. Moreover, our handful of observations of scrambling events outside of the groove, which has not yet been reported in any study, opens an exciting new direction for studying alternative scrambling mechanisms. That said, we are currently following up on many of the observations reported here such as: scrambling events outside the groove, the kinetics of scrambling, the possibility that lipids line the groove of non-scramblers like TMEM16A, etc. This is being done experimentally with our collaborators through site directed mutagenesis and with all-atom MD in our lab. Unfortunately, it is well beyond the scope of the current study to include all of this in the current paper.

Reviewer #2 (Recommendations for the authors):

Major comments and questions:

(1) Line 214 and Figure 1- Figure Supplement 1: why have you only compared the final frame of the trajectory to the cryo-EM structure? Even if these comparisons are qualitative, they should be representative of the entire trajectory, not a single frame.

We thank the reviewer for this suggestion and replaced the single-frame snapshots in Figure 1-figure supplement 1 for ensemble-averaged head groups densities. The overall agreement between membrane shapes in CGMD and cryo-EM was not affected by this change.

(2) Lines 228-231: You comment 'Residues in this site on nhTMEM16 and TMEMF also seem to play a role in scrambling but the mechanism by which they do so is unclear.' This is something you could attempt to quantify in the simulations by calculating the correlation between scrambling and protein-membrane interactions/contacts in this site. Can you speculate on a mechanism that might be a contributing factor?

We probed the correlation between these residues and scrambling lipids, as suggested by the reviewer, and interestingly not all scrambling lipids interact with these residues. Yet there is strong lipid density in this vicinity (see insets in Figure 1 and Figure 4-figure supplement 2). These observations lead us to suspect these residues impact scrambling indirectly through influencing the conformation of the protein or flexibility and shape of the membrane. This interpretation fits with mutagenesis studies highlighting a role for these residues in scrambling (see refs 59, 62, and 67). Specifically, Falzone et al. 2022 (ref 59) suggested that they may thin the membrane near the groove, but this has not been tested via structure determination and a detailed model of how they impact scrambling is missing. We could address this question with in silico mutations; however, CG simulation is not an appropriate method to study large scale protein dynamics, and AA simulations are likely best, but beyond the scope of this paper.

(3) Lines 240-245 and Figure 1B: This section discusses the coupling between membrane distortions and the sinusoidal curve around the protein, however, Figure 1B only shows snapshots of the membrane distortions. Is it possible to understand how these two collective variables are correlated quantitatively (as opposed to the current qualitative analysis)?

We believe that it may be possible to quantitatively capture these two key features of the membrane, as we did previously with nhTMEM16 using our continuum elasticity-based model of the membrane (Bethel and Grabe 2016). Our model agreed with all atom MD surfaces to within ~1 Å, hence showing good quantitative agreement throughout the entire membrane. However, we doubt that we could distill the essence of our model down to a simple functional relationship between the sinusoidal wave and pinching, which we think the reviewer is asking. Rather, we believe that the large-scale sinusoidal distortion (collective variable 1) and pinching/distortion (collective variable 2) near the groove arise from the interplay of the specific protein surface chemistry for each protein (patterning of polar and non-polar residues) and the membrane. This is why we chose to simply report the distinct patterns that the family members impose on the surrounding membrane, which we think is fascinating. Specifically, Fig. 1B shows that different TMEM16 family members distort the membrane in different ways. Most notably, fungal TMEM16s feature a more pronounced sinusoidal deformation, whereas the mammalian members primarily produce local pinching. Then, in Fig. 3A we show that the thinning at the groove happens in all structures and is more pronounced in open, scrambling-competent conformations. In other words, proteins can show very strong thinning (e.g. TMEM16K, 5OC9) even though the membrane generally remains flat.

(4) Lines 257-258: Authors comment that TMEM16A lacks scramblase activity yet can achieve a fully lipid-lined groove (note the typo - should be lipid-lined, not lipid-line). Is a fully lipid-lined groove a prerequisite for scramblase activity? Are lipid-lined grooves the only requirement for scramblase activity? Could the authors clarify exactly what the prerequisite for scramblase activity is to avoid any confusion; this will be useful for later descriptions (i.e. line 295) where scrambling competence is again referred to. Additionally, the associated figure panel (Figure 1D) shows a snapshot of this finding but lacks any statistical quantifications - is a fully lipid-lined groove a single event? Perhaps the additional analyses, such as the groove-lipid contacts, may be useful here.

The definition of lipid scrambling is that a lipid fully transitions from one membrane leaflet to the other. While a single lipid could transition through the groove on its own, it is well documented in both atomistic and CG MD simulations, that lipid scrambling typically happens through a lipid-lined groove, as shown in Fig. 1A-B. The lipids tend to form strong choline-to-phosphate interactions with nearest neighbors that make this energetically favorable. That said, lipid-lined grooves are not sufficient for robust scrambling, which is what we show in Fig. 1D where the non-scrambler TMEM16A did in fact feature a lipid-lined groove. As suggested, we performed contact analysis and found that residue K645 on TM6 in the middle of the groove contacts lipids in 9.2% of the simulation frames.

To get a better understanding of how populated the TM4-TM6 pathway is with lipids across all simulated structures, we determined for every simulation frame how many headgroup beads resided in the groove. This indicates that the ion-conductive state of TMEM16A (5OYB*, Fig. 1D) only had 1 lipid in the pathway, on average, meaning that the configuration shown Fig. 1D is indeed exceptional. As a reference, our strongest scrambler nhTMEM16 4WIS, had an average of 2.8 lipids in the groove. We added a table containing the means and standard deviations that resulted from this analysis as Figure 1-Table supplement 1.

(5) Lines 295-298 : The scrambling rates of the Ca²⁺-bound and Ca²⁺-free structures fall within overlapping error margins, it becomes difficult to definitively state that Ca²⁺ binding significantly enhances scrambling activity. This undermines the claim that the Ca²⁺-bound structure is the strongest scrambler. The authors should conduct statistical analyses to determine if the difference between the two conditions is statistically significant.

In contrast to the reviewer’s comment, we do not claim that Ca2+-binding itself enhances lipid scrambling. Instead, what we show is that WT structures that are solved in an open confirmation (all of which are Ca2+-bound, except 6QM6) are robust scramblers. For nhTMEM16, we did not observe any scrambling events for the closed-groove proteins, making further statistical analysis redundant.

(6) The authors claim that the scrambling rates derived from their MD simulations are in "excellent agreement" with experimental findings (lines 294-295), despite significant discrepancy between simulated and experimentally measured rates. For example, the simulated rate of 24.4 {plus minus} 5.2 events/µs for the open, Ca²⁺-bound fungal nhTMEM16 (PDB ID 4WIS) corresponds to approximately 24 million events per second, which is vastly higher than experimental rates. Experimental studies have reported scrambling rate constants of ~0.003 s⁻¹ for TMEM16 family members in the absence of Ca²⁺, measured under physiological conditions (https://doi.org/10.1038/s41467-019-11753-1 ). Even with Ca²⁺ activation, scrambling rates remain several orders of magnitude lower than the rates observed in simulations. Moreover, this highlights a larger problem: lipid scrambling rates occur over timescales that are not captured by these simulations. While the authors elude to these discrepancies (lines 605-606), they should be emphasised in the text, as opposed to the table caption. These should also be reconducted to differences between the membrane compositions of different studies.

We agree with the spirit of the reviewer’s comment, and because of that, we were very careful not to claim that we reproduce experimental scrambling rates, just that the trends (scrambling-competent, or not) are correct. On lines 294-295, we actually said that the scrambling rates in our simulations excellently agree with “the presumed scrambling competence of each experimental structure”, which is true. 

As explained extensively in the discussion section of our paper (and by many others), direct comparison between MD (e.g., Martini 3, but also atomistic force fields) dynamics and experimental measurements is challenging. The primary goal of our paper is to quantify and compare the scrambling capacity of different TMEM16 family members and different states, within a CGMD context.

That said, we agree with the reviewer that we may have missed rare or long-timescale events (as is the case in any MD experiment) and added this point to the discussion.

(7) To address these discrepancies, the authors should: i) emphasize that simulated rates serve as qualitative indicators of scrambling competence rather than absolute values comparable to experimental findings and ii) discuss potential reasons for the divergence, such as simulation timescale limitations or lipid bilayer compositions that may favor scrambling and force field inaccuracies.

Please see our answer to question 6. Within the context of our CGMD survey, we confidently call our results quantitative. However, we agree with the reviewer that comparison with experimental scrambling rates is qualitative and should be interpreted with caution. To reflect this, we rewrote the first sentence of the relevant paragraph in the discussion section.

(8) Line 310: Can the authors provide a rationale as to why one monomer has a wider groove than the other? Perhaps a contact analysis could be useful. See the comment above about ENM.

The simulation of Ca2+-bound TMEM16K was initiated from an asymmetric X-ray structure in which chain B features a more dilated groove than chain A (PDB 5OC9). The backbones of TM4 and TM6 in the closed groove (A) are close enough together to be directly interconnected by the elastic network. In contrast, TM4 and TM6 in the more dilated subunit (B) are not restricted by the elastic network and, as a consequence, display some “breathing” behavior (Fig. 3B and Fig. 3-Suppl. 6A), giving rise to a ~4x higher scrambling rate. We explicitly added the word “cryo-EM” and the PDB ID to the sentence to emphasize that the asymmetry stems from the original experimental structure.

When answering this question, we also corrected a mislabeled chain identifier which was in the original manuscript ‘chain A’ when it is actually ‘chain B’ in Fig.2-Suppl. 3A.

(9) Line 312: Authors speculate that increased groove width likely accounts for increased scrambling rates. For statistical significance, authors should attempt to correlate scrambling rates and groove width over the simulation period.

The Reviewer is referring to our description of scrambling rates we measured for TMEM16K where we noted that on average the groove with the highest scrambling rate is also on average wider than the opposite subunit which is below 6 Å. We do not suggest that the correlation between scrambling and groove width is continuous, as the Reviewer may have interpreted from our original submission, but we think it is a binary outcome – lipids cannot easily enter narrow grooves (< 6 Å) and hence scrambling can only occur once this threshold is reached at which point it occurs at a near constant rate. We showed this for 4 different family members in the original Fig. 3B, where scrambling events (black dots) were much more likely during, or right after, groove dilation to distances > 6 Å. 

(10) Line 359: Authors have plotted the minimum distance between residues TM4 and TM6 in Fig. 3A/B, claiming that a wide groove is required for scrambling. Upon closer examination, it is clear that several of these distributions overlap, reducing the statistical significance of these claims. Statistical tests (i.e. KS-tests) should be performed to determine whether the differences in distributions are significant.

The Reviewer appears to be asking for a statistical test between the six distance distributions represented by the data in Fig. 3A for the scrambling competent structures (6QP6*, 8B8J, 6QM6, 7RXG, 4WIS, 5OC9), and we think this is being asked because it is believed that we are making a claim that the greater the distance, the greater the scrambling rate. If we have interpreted this comment correctly, we are not making this claim. Rather, we are simply stating that we only observe robust scrambling when the groove width regularly separates beyond 6 Å. The full distance distributions can now be found in Figure 3-figure supplement 6B, and we agree there is significant overlap between some of these distributions. However, the distinguishing characteristic of the 6 distributions from scrambling competent proteins is that they all access large distances, while the others do not. Notably, TMEM16F proteins (6QP6*, 8B8J) are below the 6 Å threshold on average, but they have wide standard deviations and spend well over ¼ of their time in the permissive regime (the upper error bar in the whisker plots in Fig. 3A is the 75% boundary).

(11) Line 363-364: The authors state that all TMEM16 structures thin the membrane. Could the authors include a description of how membrane thinning is calculated, for instance, is the entire membrane considered, or is thinning calculated on a membrane patch close to the protein? Do membrane patches closer to the transmembrane protein increase or decrease thickness due to hydrophobic packing interactions? The latter question is of particular concern since Martini3 has been shown to induce local thinning of the membrane close to transmembrane helices, yielding thicknesses 2-3 Å thinner than those reported experimentally (https://doi.org/10.1016/j.cplett.2023.140436). This could be an important consideration in the authors' comparison to the bulk membrane thickness (line 364). Finally, how is the 'bulk membrane thickness' measured (i.e., from the CG simulations, from AA simulations, or from experiments)?

Regarding the calculation of thinning and bulk membrane thickness, as described in Method “Quantification of membrane deformations”, the minimal membrane thickness, or thinning, is defined as the shortest distance between any two points from the interpolated upper and lower leaflet surfaces constructed using the glycerol beads (GL1 and GL2). Bulk membrane thickness is calculated by taking the vertical distance between the averaged glycerol surfaces at the membrane edge.

The concern of localized membrane deformation due to force field artifacts is well-founded. However, the sinusoidal deformations shown here are much greater than 2-3 Å Martini3 imperfections, and they extend for up to 10 Å radially away from the protein into the bulk membrane (see Figure 3-figure supplement 1-5 for more of a description). Most importantly, the sinusoidal wave patterns set up by the proteins is very similar to those described in the previous continuum calculation and all-atom MD for nhTMEM16 (https://www.pnas.org/doi/full/10.1073/pnas.1607574113).

(12) Line 374: The authors state a 'positive correlation' between membrane thinning/groove opening and scrambling rates. To support this claim, the authors should report. the correlation coefficients.

We have removed any discussion concerning correlations between the magnitude of the scrambling rate and the degree of membrane thinning/groove opening. Rather we simply state that opening beyond a threshold distance is required for robust scrambling, as shown in our analysis in Fig. 3A.

Concerning the relation between thinning and scrambling: Instantaneous membrane thinning is poorly defined (because it is governed by fluctuations of single lipids), and therefore difficult to correlate with the timing of individual scrambling events in a meaningful way.  Moreover, as we state later in that same section, “we argue that the extremely thin membranes are likely correlated with groove opening, rather than being an independent contributing factor to lipid scrambling”.

(13) Line 396: It is stated that TMEM16A is not a scramblase but the simulating scrambling activity is not zero. How can you be sure that you are monitoring the correct collective variable if you are getting a false positive with respect to experiments?

We only observe 2 scrambling events in 10 ms, which is a very small rate compared to the scrambling competent states. In a previous large survey Martini CG simulation study that inspired our protocol (Li et al, PNAS 2024), they employed a 1 event/ms cut-off to distinguish scramblers from non-scramblers. Hence, they would have called TMEM16A a non-scrambler as well. We expect that false negatives in this context might be an artifact of the CG forcefield, or it could be that TMEM16A can scramble but too slowly to be experimentally detected. Regarding the collective variable for lipid flipping, it is correct, and we know that this lipid actually flipped.

(14) Line 402: Distance distributions for the electrostatic interactions between E633 and K645 should be included in the manuscript. This is also the case for the interactions between E843-K850 (lines 491-492).

Our description of interactions between lipid headgroups and E633 and K645 in TMEM16A (5OYB*) are based on qualitative observations of the MD trajectory, and we highlight an example of this interaction in Figure 3-video 4. The video clearly shows that the lipid headgroups in the center of the groove orient themselves such that the phosphate bead (red) rests just above K645 (blue) and at other times the choline bead (blue) rests just below E633 (red). We do not think an additional plot with the distance distributions between lipids and these residues will add to our understanding of how lipids interact residues in the TMEM16A pore.

We made a similar qualitative observation for the interaction between the POPC choline to E843 and POPC phosphate to K850 while watching the AAMD simulation trajectory of TMEM16F (PDB ID 6QP6). Given that this was a single observation, and the same interactions does not appear in CG simulation of the same structure (see simulation snapshots in Figure 4-figure supplement 5) we do not think additional analysis would add significantly to our understanding of which residues may stabilize lipids in the dimer interface.

(15) Lines 450-451: 'As the groove opens, water is exposed to the membrane core and lipid headgroups insert themselves into the water-filled groove to bridge the leaflets.' Is this a qualitative observation? Could the authors report the correlation between groove dilation and the number of water permeation events?

Yes, this is qualitative, and it sketches the order of events during scrambling, and we revised the main text starting at line 450 to indicate this. As illustrated by the density isosurfaces in Appendix 1-Figure 2A, the amount of water found in the closed versus open grooves is striking – there is a significant flood of water that connects the upper and lower solutions upon groove opening. Moreover, Appendix 1-Figure 2B shows much greater water permeation for open structures (4WIS, 7RXG, 5OC9, 8B8J, …) compared to closed structures (6QMB, 6QMA, 8B8Q, and many of the non-labeled data in the figure that all have closed grooves and near 0 water permeation). A notable exception is TMEM16A (7ZK3*8), which has water permeation but a closed groove and little-to-no lipid scrambling.

Minor Comments:

(1) Inconsistent use of '10' and 'ten' throughout.

We like to kindly point out that we do not find examples of inconsistent use.

(2) Line 32: 'TM6 along with 3, 4 and 5...' should be 'TM6 along with TM3, TM4 and TM5...'. Same in line 142. Naming should stay consistent.

Changes are reflected in the updated manuscript.

(3) Line 141: do you mean traverse (i.e. to travel across)? Or transverse (i.e. to extend across the membrane)?

This is a typo. We meant “traverse”. Thanks for pointing it out.

(4) Line 142: 'greasy' should be 'strongly hydrophobic'.

Changes are reflected in the updated manuscript.

(5) Line 143-144: "credit card mechanism" requires quotation marks.

Changes are reflected in the updated manuscript.

(6) Line 144: state if Nectria haematococca is mammalian or fungal, this is not obvious for all readers.

Changes are reflected in the updated manuscript.

(7) Line 147-148: Is TMEM16A/TMEM16K fungal or mammalian? What was the residue before the mutation and which residue is mutated? Perhaps the nomenclature should read as TMEM16X10Y where X=the residue prior to the mutation, 10 is a placeholder for the residue number that is mutated and Y=the new residue following mutation.

“TMEM16” is the protein family. “A” denotes the specific homolog rather than residue.  

(8) Lines 157-158: same as 10, it is unclear if these are fungal or mammalian.

Clarifications added.

(9) Line 184: "...CGMD simulation" should be "...CGMD simulations".

Changes made.

(10) Line 191-192: It would help to create a table of all of the mutants (including if they are mammalian or fungal) summarizing the salt concentrations, lipid and detergent environments, the presence of modulators/activators, etc.

We added this information to Appendix 1-Table 1 in the supplemental information. We did not specify NaCl concentrations, because they all experimental procedures used standard physiological values for this (100-150 mM).

(11) Line 210: inconsistencies with 'CG' and 'coarse-grain'.

Changes made.

(12) Figure 1 caption: '...totaling ~2μs (B)...' is missing the fullstop after 2μs.

Changes made.

(13) Figure 1B: it may be useful to label where the Ca2+ ion binds or include a schematic.

We updated Fig. 1A to illustrate where Ca2+ binds.

(14) Line 311: Are these mean distances? The authors should add standard deviations.

Yes, they are. We added the standard deviations to the text.

(15) Line 321-322: Perhaps a schematic in Figure 2 would be useful to visualize the structural features described here.

We would kindly refer interested readers to reference [60].

(16) Line 377: '...are likely a correlate of groove opening...' should read as: '...are likely correlated to groove opening...'.

Thank you for pointing it out. Changes made.

(17) Line 398: the '...empirically determined 6Å threshold for scrambling.' Was this determined from the simulations or from experiments? What does "empirically" mean here? Please state this.

This value was determined from the simulations. Based on our analysis of the correlation between scrambling rate and groove dilation, we found that the minimal TM4/6 distance of 6 Å can distinguish between the high and low activity scramblers. The exact numerical value is somewhat arbitrary as there is a range of values around 6 Å that serve to distinguish scramblers from non-scramblers.

(18) Figure 4: This figure should be labelled as A, B, C and D, with the figure caption updated accordingly.

We updated Figure 4 and its caption.

Reviewer #3 (Recommendations for Authors):

The authors must do additional simulations to further validate their claim with different lipids and further substantiate dimer interface independent of Ca2+ ions.

Thank you for the suggestion. We completely agree that studying scrambling in the context of a diverse lipid environment is an exciting area to explore. We are indeed actively working on a project that shares the similar idea. We decided not to include that study because we think the additional discussion involved would be excessive for the current manuscript. We, however, look forward to publishing our findings in a separate manuscript in the near future. In terms of Ca2+-independent scrambling, we are planning with our experimental collaborator for mutagenesis studies that target the residues we identified along the dimer interface.

Since calcium ions are critical for the stability of these structures, authors should show that they were placed throughout the simulations consistently.

As stated in the method section “Coarse-grained system preparation and simulation detail”, all Ca2+ ions are manually placed into the coarse-grained structure from the beginning of the simulation at their identical corresponding position in the experimental structure and harmonically bonded to adjacent acidic residues throughout the duration of simulation. We have also added a label to Fig 1A to indicate where the two Ca2+ ions are located.

The comparison with experimental structures should be consistent with complete simulation, and not the last structure of the trajectory. Depending on the conformational variability, this might be misleading.

We agree and updated Fig. 1-supplement figure 1 accordingly. The overall agreement between membrane shapes in CGMD and cryo-EM was not affected by this change.

Reviewer #1 (Public review):

Summary:

Meteorin proteins were initially described as secreted neurotrophic factors. In this manuscript, Eggeler et al. demonstrate a novel role for Meteorins in establish left-right axis formation in the zebrafish embryo. The authors generated null mutations in each of the three zebrafish meteorin genes - metrn, metrnla, and metrnlab. Triple mutant embryos displayed phenotypes strongly associated with left-right defects such as heart looping and visceral organ placement, and disrupted expression of Nodal-responsive genes, as did single mutants for metrn and metrnla. The authors then go on to demonstrate that these defects in left-right asymmetry are likely to due to defects in Kupffer's Vesicle and the progenitor dorseal forerunner cells including impaired lumen formation and reduced fluid flow, reduced clustering among DFCs, impaired DFC migration, mislocalization of apical proteins ZO-1 and aPKC, and detachment of DFCs from the EVL. Notably, the authors found that expression of marker genes sox32 and sox17 were not affected, suggesting Meteorins are required for DFC/KV morphogenesis but not necessarily fate specification. Finally, the authors show genetic interaction between Meteorins and integrin receptors, which were previously implicated in left-right patterning. In a supplemental figure, the manuscript also presents data showing expression of meteorin genes around the chick Hensen's node, suggesting that the left-right patterning functions may be conserved among vertebrates.

Strengths:

Strengths of this study include the generation of a triple mutant line that targets all known zebrafish meteorin family members. The experiments presented in this study were rigorous especially with respect to quantification and statistical analysis.

Weaknesses:

Although the authors convincingly demonstrate a role for Meteorins in zebrafish left-right patterning, data supporting a conserved role in other vertebrates is compelling but limited to one supplemental figure. This aspect would be interesting to follow up in future studies.

Comments on revisions:

I thank the authors for their thoughtful responses to the reviewers. They have adequately addressed all of my concerns.

Author response:

The following is the authors’ response to the original reviews

Public Review:

Reviewer #1 (Public review):

Summary:

Meteorin proteins were initially described as secreted neurotrophic factors. In this manuscript, Eggeler et al. demonstrate a novel role for Meteorins in establish left-right axis formation in the zebrafish embryo. The authors generated null mutations in each of the three zebrafish meteorin genes - metrn, metrnla, and metrnlab. Triple mutant embryos displayed phenotypes strongly associated with left-right defects such as heart looping and visceral organ placement, and disrupted expression of Nodal-responsive genes, as did single mutants for metrn and metrnla. The authors then go on to demonstrate that these defects in left-right asymmetry are likely to due to defects in Kupffer's Vesicle and the progenitor dorseal forerunner cells including impaired lumen formation and reduced fluid flow, reduced clustering among DFCs, impaired DFC migration, mislocalization of apical proteins ZO-1 and aPKC, and detachment of DFCs from the EVL. Notably, the authors found that expression of marker genes sox32 and sox17 were not affected, suggesting Meteorins are required for DFC/KV morphogenesis but not necessarily fate specification. Finally, the authors show genetic interaction between Meteorins and integrin receptors, which were previously implicated in left-right patterning. In a supplemental figure, the manuscript also presents data showing expression of meteorin genes around the chick Hensen's node, suggesting that the left-right patterning functions may be conserved among vertebrates.

Strengths:

Strengths of this study include the generation of a triple mutant line that targets all known zebrafish meteorin family members. The experiments presented in this study were rigorous, especially with respect to quantification and statistical analysis.

Weaknesses:

Although the authors convincingly demonstrate a role for Meteorins in zebrafish left-right patterning, data supporting a conserved role in other vertebrates is compelling but limited to one supplemental figure.

We thank the reviewer for their thoughtful summary of our study and for highlighting the strengths of our work, including the generation of the triple mutant line and the rigor of our experimental design and quantitative analyses. We also appreciate the constructive feedback regarding the limited functional data supporting the conservation of Meteorin function in other vertebrates. We agree that this is an important aspect that could be further explored. While functional studies in additional species are beyond the current scope, we will consider such experiments in future work.

We would like to highlight the phylogenetic analysis of Meteorin proteins we have already performed and included in the manuscript (Fig. S7D), which illustrates the evolutionary conservation of this protein family and supports the possibility of a conserved role in left-right patterning.

Additionally, we have expanded the methods and discussion to include: (1) details on zebrafish viability in contrast to reported embryonic lethality in metrn mutant mice, (2) the background strains used in our study, (3) observed variability in DFC number and potential batch effects and (4) clarification of our 'convergence ratio' quantification approach.

Reviewer #2 (Public review):

Summary:

In this manuscript the authors describe their study on the role of meteorins in establishing the left-right organizer. The left-right organizer is a transient organ in vertebrate embryos in which rotating cilia cause a fluid flow that breaks the left-right symmetry and coordinates lateralization of internal organs such as gut and heart. In zebrafish, the left-right organizer (also named Kupffer's vesicle) is formed by dorsal forerunner cells, but very little is known about how dorsal forerunner cells coalles and form this ciliated vesicle in the embryo. The authors mutated the three meteorin-coding genes in zebrafish and observed that mutations in each one of these causes laterality defects with the strongest defects observed in the triple mutant. Loss of meteorins affects nodal gene expression, which play essential roles in establishing organ laterality. Meteorins are widely expressed in developing embryos and expression in lateral plate mesoderm and dorsal forerunner cells was observed. The meteorin triple mutant embryos display defects in the migration and clustering of the dorsal forerunner cells impairing kupffer's vesicle formation and cilia rotation. Finally, the authors show that meteorins genetically interact with integrins.

Strengths:

- These authors went through the lengthy process of generating triple mutants affecting all three meteorin genes. This provides robust genetic evidence on the role of meteorins in establishing organ laterality and circumvented that interpretation of the results would be hard due to redundant functions of meteorins.

- The use of life imaging on triple mutants is appreciated

- High-quality imaging of dorsal forerunner to quantify cell migrations and its relation to Kupffer's vesicle formation.

Weaknesses:

- Lack of a model how meteorins regulate dorsal forerunner cell migration.

- Only genetic data to suggest a link between meteorins and integrins

- Besides its role in DFC migration, meteorins may also play a more direct role in regulating Nodal signaling, which is not addressed here.

We appreciate the recognition of the strengths of our study, particularly the generation of the triple meteorin mutants and the use of high-resolution imaging to quantify DFC behavior and Kupffer’s vesicle formation—both of which were central to providing robust evidence for Meteorins' role in left-right patterning.

We also value the reviewer’s comments on areas that need further exploration, including the need for a mechanistic model explaining how Meteorins regulate DFC migration, the genetic interaction with integrins, and the potential direct involvement of Meteorins in Nodal signaling.

We agree that deeper mechanistic insights would strengthen the study. While our findings suggest that Meteorins influence DFC migration and clustering through integrin pathways, a detailed mechanistic dissection, particularly regarding the yet unidentified Meteorin receptor, lies beyond the current scope. However, we consider this a key aspect for future research and have discussed it further in the revised discussion section.

In response to the reviewer’s suggestions, we have expanded the discussion to address the limitations of the current data linking Meteorins and integrins, including relevant citations to studies that implicate integrins in similar contexts. Additionally, we have added a more detailed discussion of the potential for Meteorins to directly influence Nodal signaling, and we cite a relevant study to support this possibility.

Once again, we thank the reviewer for their insightful and constructive comments. These points raise important directions for future investigation that will further advance our understanding of Meteorin function in left-right axis formation.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

In the Results section (p. 9), the authors state, "...a reduced ZO-1 enrichment at the apical junctions of triplMUT GFP-positive DFCs could be detected." However, in Fig. 4F-G, the areas of ZO-1 enrichment indicated by arrowheads appear quite far from the DFCs themselves, making it unclear if these ZO-1-enriched areas are apical DFC junctions (as stated in the text) or instead are part of the EVL. Is it possible to include an additional cell membrane marker or other landmarks? In addition, the differences in ZO-1 accumulation between mutants and WT appear relatively modest. Is it possible to provide quantification of this effect?

We appreciate the reviewer’s request for additional stainings and further clarification and we would like to highlight the requested quantifications of ZO-1 accumulation, including statistical analysis, are already provided in Fig. S5E.

In mouse, loss of Meteorin is embryonic lethal yet the zebrafish triple mutants are viable. Could the authors discuss this discrepancy?

We have expanded the discussion to address this point, suggesting that species-specific differences in compensatory mechanisms may explain the observed differences in viability. We would like to reiterate that while one study has reported embryonic lethality in metrn mutant mice, this specific mouse line has not been further investigated in any recent publications. Additionally, in collaboration with the lab of Alain Chédotal, we generated independent metrn and metrnl mutant mouse lines, which did not exhibit the phenotype described in the previously mentioned study.

It has been reported that TL and AB strains exhibit variable numbers of DFCs and thus laterality defects (Moreno-Ayala et al., 2021, Cell Reports 34(2):108606). Would it be possible for the authors to report background stains used in this study and those used to generate the meteorin knock-outs?

We appreciate the comment highlighting the importance of specifying the background strains used in our study. We have now included this information in the methods section, detailing the zebrafish strains utilized throughout our experiments.

For statistical analysis, would be possible for the authors to report the number of clutches examined to control for batch effects (especially given the wide variability in DFC numbers as noted above)?

For further clarification, we have now included additional explanation on number of clutches in the methods section.

In the Methods section (p. 19), the description of how the convergence ratio was computed was somewhat unclear. Could the authors provide a citation or include a diagram/schematic?

We have revised the Methods section to provide a clearer definition of the convergence ratio and have included a schematic (Fig. 4D) to illustrate how it was calculated.

Reviewer #2 (Recommendations for the authors):

- Meteorins are widely expressed in the embryo. Can the authors comment on whether meteorin expression is required in the dorsal forerunner cells (DFCs) or in other cells? This could be addressed by knockdown experiments in DFCs as described by others (PMID: 15716348)

We thank the reviewer for this important comment. In our study, we have shown that Meteorins are not required for the identity of DFCs, as several DFC-specific markers remain expressed in the respective cells within the meteorin mutant background (see Fig. S4).

- In fig1d and 1e the authors use heterotaxy to describe visceral organ placement. The embryo shown in 1d seems to display situs inversus instead of heterotaxy, which is defined as discordance in organ position. The authors should clarify this.

We agree with the reviewer and have revised the figures and figure legends to clarify the distinction between situs inversus and heterotaxy.

- In Fig2 the authors show that nodal pathway genes are reduced, suggesting reduced Nodal signaling. How do they explain this as loss of cilia rotation generally leads to randomization of Nodal signaling but not a reduction in signaling.

Following this suggestion we have now added a further discussion on the possibility that Meteorins could directly regulate Nodal signaling in addition to their role in DFC migration and have cited a relevant study.

- Reduced Nodal signaling in the LPM leads to organ laterality defects. Most anterior tissues like the heart are more sensitive to perturbation in Nodal signaling in the LPM compared to more posterior organs like gut (see also PMID: 25684355). Since in triple mutants the position of the heart is more affected than the position of the visceral organs this suggests that meteorins play an additional role in Nodal signaling in the LPM. As others have shown that meteorins regulate nodal activity (PMID: 24558432), the authors should address this further.

As described above, we have now added a further discussion on the possibility that Meteorins could directly regulate Nodal signaling in addition to their role in DFC migration and have cited a relevant study. Further investigation into a possible direct role of Meteorins in Nodal signaling will be pursued in future work.

- The term 'convergence ratio' is not clearly described and confusing as convergence is also used for the movement of LPM cells towards the midline.

As noted in response to Reviewer #1, we have revised the Methods section and included a schematic in Fig. 4D to better explain this parameter.

We are grateful for the thoughtful critiques from both reviewers, which have been very constructive and improved the clarity of our study. We believe that the revisions we have made address the concerns raised, and we look forward to your evaluation of our revised manuscript.

eLife Assessment

This study presents important insights into the regulation of left-right organ formation. By combining genetic perturbation of all three Meteorin genes in zebrafish and timelapse imaging, the authors identify an essential role for this protein family in the establishment of left-right patterning. They provide convincing evidence that Meteorins are required for the morphogenesis of dorsal forerunner cells, the precursors of the left-right organizer (also named Kupffer's vesicle) in zebrafish. In line with this, Meteorins were shown to genetically interact with integrins ItgaV and Itgb1b to regulate dorsal forerunner cell clustering.

Reviewer #1 (Public review):

Summary:

Meteorin proteins were initially described as secreted neurotrophic factors. In this manuscript, Eggeler et al. demonstrate a novel role for Meteorins in establish left-right axis formation in the zebrafish embryo. The authors generated null mutations in each of the three zebrafish meteorin genes - metrn, metrnla, and metrnlab. Triple mutant embryos displayed phenotypes strongly associated with left-right defects such as heart looping and visceral organ placement, and disrupted expression of Nodal-responsive genes, as did single mutants for metrn and metrnla. The authors then go on to demonstrate that these defects in left-right asymmetry are likely to due to defects in Kupffer's Vesicle and the progenitor dorseal forerunner cells including impaired lumen formation and reduced fluid flow, reduced clustering among DFCs, impaired DFC migration, mislocalization of apical proteins ZO-1 and aPKC, and detachment of DFCs from the EVL. Notably, the authors found that expression of marker genes sox32 and sox17 were not affected, suggesting Meteorins are required for DFC/KV morphogenesis but not necessarily fate specification. Finally, the authors show genetic interaction between Meteorins and integrin receptors, which were previously implicated in left-right patterning. In a supplemental figure, the manuscript also presents data showing expression of meteorin genes around the chick Hensen's node, suggesting that the left-right patterning functions may be conserved among vertebrates.

Strengths:

Strengths of this study include the generation of a triple mutant line that targets all known zebrafish meteorin family members. The experiments presented in this study were rigorous especially with respect to quantification and statistical analysis.

Weaknesses:

Although the authors convincingly demonstrate a role for Meteorins in zebrafish left-right patterning, data supporting a conserved role in other vertebrates is compelling but limited to one supplemental figure. This aspect would be interesting to follow up in future studies.

Comments on revisions:

I thank the authors for their thoughtful responses to the reviewers. They have adequately addressed all of my concerns.

eLife Assessment

The study provides a valuable analysis of escape from X-inactivation based on three rare female GTEX-donors with non-mosaic X-inactivation. The methods and analyses are solid and broadly support the authors' claims. Their data are more comprehensive than those presented previously and add significant weight to evidence for which genes are inactivated or escape from X inactivation in humans.

Reviewer #1 (Public review):

Summary:

This manuscript investigates genes that escape X-Chromosome Inactivation (XCI) across human tissues, using females that exhibit skewed or non-random XCI. The authors identified 2 female individuals with skewed XCI in the GTex database, in addition to the 1 female skewed sample in this database that has been described in a previous publication (Ref.16). The authors also determined the genes which escape XCI for 380 X-linked genes across 30 different tissues.

Strengths:

The novelty of this manuscript is that the authors have identified the XCI expression status for a total of 380 genes across 30 different human tissues, and also discovered the XCI status (escape, variable escape, or silenced) for 198 X-linked genes, whose status was previously not determined. This report is a good resource for the field of XCI, and would benefit from additional analyses and clarification of their comparisons of XCI status.

Reviewer #2 (Public review):

Summary:

Gylemo et al. present a manuscript focused on identifying the X-inactivation or X-inactivation escape status for 380 genes across 30 normal human tissues. X-inactivation status of X-linked genes across tissues is important for understanding sex-specific differences in X-linked gene expression and therefore traits, and the likely effect of X-linked pathogenic variants in females. These new data are significant as they double the number of genes that have been classified in the human, and double the number of tissues studied previously.

Strengths:

The strengths of this work are that they analyse 3 individuals from the GTex dataset (2 newly identified, 1 previously identified and published) that have highly/ completely skewed X inactivation, which allows the study of escape from X inactivation in bulk RNA-sequencing. The number of individuals and breadth of tissues analysed adds significantly to both the number of genes that have been classified and the weight of evidence for their claims. The additional 198 genes that have been classified and the reclassification of genes that previously had only limited support for their status is useful for the field.

In analysing the data they find that tissue-specific escape from X inactivation appears relatively rare. Rather, if genes escape, even variably, it tends to occur across tissues. Similarly if a gene is inactivated, it is stable across tissues.

Comments on revised version:

The authors have answered all of my queries. While they have not been able to pinpoint the genetic cause of the highly skewed XCI cases in their cohort, I agree this is beyond the scope of this study. I have no further requests.

Reviewer #3 (Public review):

Summary:

Nestor and colleagues identify genes escaping X chromosome inactivation (XCI) in rare individuals with non-mosaic XCI (nmXCI) whose tissue-specific RNA-seq datasets were obtained from the GTEX database. Because XCI is non-mosaic, read counts representing a second allele are tested for statistical significant escape, in this case > 2.5% of active X expression. Whereas a prior GTEX analysis found only one nmXCI female, this study finds two additional donors in GTEX, therefore expanding the number of assessed X-linked genes to 380. Although this is fewer than half of X-linked genes, the study demonstrates that although rare, nmXCI females are represented in RNA-seq databases such as GTEX. Therefore this analytical approach is worthwhile pursuing in other (larger) databases as well, to provide deeper insight into escape from XCI which is relevant to X-linked diseases and sex differences.

Strengths:

The analysis is well-documented, straight-forward and valuable. The supplementary tables are useful, and the claims in the main text well-supported.

Weaknesses:

There are very few, except that this escape catalogue is limited to 3 donors, based on a single (representative) tissue screen in 285 female donors, mostly using muscle samples. However, if only pituitary samples had been screened, nmXCI-1 would have been missed. Additional donors in the 285 representative samples cross a lower threshold of AE = 0.4. It would be worthwhile to query all tissues of the 285 donors to discover more nmXCI cases, as currently fewer than half of X-linked genes received a call using this very worthwhile approach.

Comments on revised version:

The authors incorporated some textual changes, but deferred any new analysis, or expansion from these two new skewed donors to include more individuals/tissues, or going more in depth for individual genes to future manuscripts. They appear to have that option at eLife.

Reviewer #4 (Public review):

Summary:

This study by Gylemo et al. investigates genes that escape X-Chromosome Inactivation (XCI) by analyzing RNA-sequencing data from three female individuals with highly skewed XCI identified in the GTEx database-two newly reported and one previously described. Utilizing these rare non-mosaic XCI cases, the authors assess allelic expression patterns across 30 normal human tissues to classify the XCI status of 380 X-linked genes, including 198 not previously annotated. The study provides a broader and more comprehensive catalog of XCI escape, contributing valuable insights into sex-specific gene expression and the potential implications of X-linked variants in disease.

Strengths:

The primary strength of this work lies in its expanded scope: it doubles the number of tissues and significantly increases the number of X-linked genes with known XCI status compared to previous studies. By focusing on rare individuals with non-random XCI, the authors provide a unique opportunity to observe allelic expression and classify escape status with more confidence. Their findings that escape from XCI is relatively consistent across tissues (rather than tissue-specific) enhance the understanding of XCI mechanisms. The methodology is robust, the data are well-documented, and the supplementary resources are comprehensive. This study thus represents a valuable resource for the XCI field and a promising basis for future investigations.

Weaknesses:

Despite its strengths, the study is limited by its reliance on only three individuals, which restricts statistical power and generalizability. Concerns were raised regarding the comparability of XCI status across tissue types and cell lines, particularly given that previous classifications may have used cancer or immortalized cells. Additionally, more could be done to explore the genetic basis behind the observed skewed XCI, which might affect the conclusions about escape patterns. Finally, the authors are encouraged to expand their approach to additional RNA-seq datasets or single-cell analyses to validate their findings and potentially discover more individuals with skewed XCI, which would deepen understanding of this important biological phenomenon.

eLife Assessment

This important study uncovers the mechanism of inhibition of a membrane pyrophosphatase by non-hydrolyzable phosphonate substrate analogs. Convincing crystallography, EPR spectroscopy, and functional measurements support the presence of a distinct conformational equilibrium of TmPPase in solution, and further supports the notion of asymmetric inhibitor binding at the active site, while maintaining a symmetric conformation at the periplasmic interface.

Reviewer #1 (Public review):

Summary:<br /> This work examines the binding of several phosphonate compounds to a membrane-bound pyrophosphatase using several different approaches, including crystallography, electron paramagnetic resonance spectroscopy, and functional measurements of ion pumping and pyrophosphatase activity. The work synthesizes these different approaches into a model of inhibition by phosphonates in which the two subunits of the functional dimer interact differently with the phosphonate. This asymmetry in the two subunits of the dimer is consistent with past studies of this system.

Strengths:<br /> This study integrates a variety of approaches, including structural biology, spectroscopic measurements of protein dynamics, and functional measurements. Overall, data analysis was thoughtful, with careful analysis of the substrate binding sites (for example calculation of POLDOR omit maps). This study agrees with previous studies that have detected functional asymmetry in the membrane PPase dimer.

Reviewer #3 (Public review):

Summary:<br /> Membrane-bound pyrophosphatases (mPPases) are homodimeric proteins that hydrolyze pyrophosphate and pump H+/Na+ across membranes. They are an attractive drug target against protist pathogens. Non-hydrolysable PPi analogue bisphosphonates such as risedronate (RSD) and pamidronate (PMD) serve as primary drugs currently used. Bisphosphonates have a P-C-P bond, with their central carbon can accommodate up to two substituents, allowing a large compound variability. Here authors solved two TmPPase structures in complex with the bisphosphonates etidronate (ETD) and zoledronate (ZLD) and monitored their conformational ensemble using DEER spectroscopy in solution. These results reveal the inhibition mechanism by these compounds, which is crucial for developing future small-molecule inhibitors.

Strengths:<br /> Authors show that seven different bisphosphonates can inhibit TmPPase with IC50 values in the micromolar range. Branched aliphatic and aromatic modifications showed weaker inhibition. High-resolution structures for TmPPase with ETD (3.2 Å) and ZLD (3.3 Å) are determined. These structures reveal the binding mode and shed light on the inhibition mechanism. The nature of modification on the bisphosphonate alters the conformation of the binding pocket. The conformational heterogeneity is further investigated using EPR/DEER spectroscopy under several conditions. Altogether, this provides convincing evidence for a distinct conformational equilibrium of TmPPase in solution and further supports the notion of asymmetric inhibitor binding at the active site, while maintaining a symmetric conformation at the periplasmic interface.

eLife Assessment

This study presented valuable findings regarding the basic molecular pathways leading to the cystogenesis of Autosomal Dominant Polycystic Kidney Disease, suggesting BICC1 functions as both a minor causative gene for PKD and a modifier of PKD severity. Although some solid data were supplied to show the functional and structural interactions between BICC-1 and PKD2 and their relevance to the pathogenesis of ADPKD, the characterization of such interactions appear to be incomplete, which renders the specific relevance of these findings for disease etiology unclear.

Reviewer #1 (Public review):

In this manuscript, Tran et al. investigate the interaction between BICC1 and ADPKD genes in renal cystogenesis. Using biochemical approaches, they reveal a physical association between Bicc1 and PC1 or PC2 and identify the motifs in each protein required for binding. Through genetic analyses, they demonstrate that Bicc1 inactivation synergizes with Pkd1 or Pkd2 inactivation to exacerbate PKD-associated phenotypes in Xenopus embryos and potentially in mouse models. Furthermore, by analyzing a large cohort of PKD patients, the authors identify compound BICC1 variants alongside PKD1 or PKD2 variants in trans, as well as homozygous BICC1 variants in patients with early-onset and severe disease presentation. They also show that these BICC1 variants repress PC2 expression in cultured cells.

Overall, the concept that BICC1 variants modify PKD severity is plausible, the data are robust, and the conclusions are largely supported. However, several aspects of the study require clarification and discussion:

(1) The authors devote significant effort to characterizing the physical interaction between Bicc1 and Pkd2. However, the study does not examine or discuss how this interaction relates to Bicc1's well-established role in posttranscriptional regulation of Pkd2 mRNA stability and translation efficiency.

(2) Bicc1 inactivation appears to downregulate Pkd1 expression, yet it remains unclear whether Bicc1 regulates Pkd1 through direct interaction or by antagonizing miR-17, as observed in Pkd2 regulation. This should be further examined or discussed.

(3) The evidence supporting Bicc1 and ADPKD gene cooperativity, particularly with Pkd1, in mouse models is not entirely convincing, likely due to substantial variability and the aggressive nature of Bpk/Bpk mice. Increasing the number of animals or using a milder Bicc1 strain, such as jcpk heterozygotes, could help substantiate the genetic interaction.

Reviewer #2 (Public review):

Tran and colleagues report evidence supporting the expected yet undemonstrated interaction between the Pkd1 and Pkd2 gene products Pc1 and Pc2 and the Bicc1 protein in vitro, in mice, and collaterally, in Xenopus and HEK293T cells. The authors go on to convincingly identify two large and non-overlapping regions of the Bicc1 protein important for each interaction and to perform gene dosage experiments in mice that suggest that Bicc1 loss of function may compound with Pkd1 and Pkd2 decreased function, resulting in PKD-like renal phenotypes of different severity. These results led to examining a cohort of very early onset PKD patients to find three instances of co-existing mutations in PKD1 (or PKD2) and BICC1. Finally, preliminary transcriptomics of edited lines gave variable and subtle differences that align with the theme that Bicc1 may contribute to the PKD defects, yet are mechanistically inconclusive.

These results are potentially interesting, despite the limitation, also recognized by the authors, that BICC1 mutations seem exceedingly rare in PKD patients and may not "significantly contribute to the mutational load in ADPKD or ARPKD". The manuscript has several intrinsic limitations that must be addressed.

The manuscript contains factual errors, imprecisions, and language ambiguities. This has the effect of making this reviewer wonder how thorough the research reported and analyses have been.

Reviewer #3 (Public review):

Summary:

This study investigates the role of BICC1 in the regulation of PKD1 and PKD2 and its impact on cytogenesis in ADPKD. By utilizing co-IP and functional assays, the authors demonstrate physical, functional, and regulatory interactions between these three proteins.

Strengths:

(1) The scientific principles and methodology adopted in this study are excellent, logical, and reveal important insights into the molecular basis of cystogenesis.

(2) The functional studies in animal models provide tantalizing data that may lead to a further understanding and may consequently lead to the ultimate goal of finding a molecular therapy for this incurable condition.

(3) In describing the patients from the Arab cohort, the authors have provided excellent human data for further investigation in large ADPKD cohorts. Even though there was no patient material available, such as HUREC, the authors have studied the effects of BICC1 mutations and demonstrated its functional importance in a Xenopus model.

Weaknesses:

This is a well-conducted study and could have been even more impactful if primary patient material was available to the authors. A further study in HUREC cells investigating the critical regulatory role of BICC1 and potential interaction with mir-17 may yet lead to a modifiable therapeutic target.

Conclusion:<br /> The authors achieve their aims. The results reliably demonstrate the physical and functional interaction between BICC1 and PKD1/PKD2 genes and their products.

The impact is hopefully going to be manifold:

(1) Progressing the understanding of the regulation of the expression of PKD1/PKD2 genes.

(2) Role of BiCC1 in mir/PKD1/2 complex should be the next step in the quest for a modifiable therapeutic target.

Author response:

Reviewer #1 (Public Review):

In this manuscript, Tran et al. investigate the interaction between BICC1 and ADPKD genes in renal cystogenesis. Using biochemical approaches, they reveal a physical association between Bicc1 and PC1 or PC2 and identify the motifs in each protein required for binding. Through genetic analyses, they demonstrate that Bicc1 inactivation synergizes with Pkd1 or Pkd2 inactivation to exacerbate PKD-associated phenotypes in Xenopus embryos and potentially in mouse models. Furthermore, by analyzing a large cohort of PKD patients, the authors identify compound BICC1 variants alongside PKD1 or PKD2 variants in trans, as well as homozygous BICC1 variants in patients with early-onset and severe disease presentation. They also show that these BICC1 variants repress PC2 expression in cultured cells.

Overall, the concept that BICC1 variants modify PKD severity is plausible, the data are robust, and the conclusions are largely supported. However, several aspects of the study require clarification and discussion:

(1) The authors devote significant effort to characterizing the physical interaction between Bicc1 and Pkd2. However, the study does not examine or discuss how this interaction relates to Bicc1's well-established role in posttranscriptional regulation of Pkd2 mRNA stability and translation efficiency.

The reviewer is correct that the present study has not addressed the downstream consequences of this interaction considering that Bicc1 is a posttranscriptional regulator of Pkd2 (and potentially Pkd1). We think that the complex of Bicc1/Pkd1/Pkd2 retains Bicc1 in the cytoplasm and thus restrict its activity in participating in posttranscriptional regulation. As we do not have yet experimental data to support this model, we have not included this model in the manuscript. Yet, we will update the discussion of the manuscript to further elaborate on the potential mechanism of the Bicc1/Pkd1/Pkd2 complex.

(2) Bicc1 inactivation appears to downregulate Pkd1 expression, yet it remains unclear whether Bicc1 regulates Pkd1 through direct interaction or by antagonizing miR-17, as observed in Pkd2 regulation. This should be further examined or discussed.

This is a very interesting comment. The group of Vishal Patel published that PKD1 is regulated by a mir-17 binding site in its 3’UTR (PMID: 35965273). We, however, have not evaluated whether BICC1 participates in this regulation. A definitive answer would require us utilize some of the mice described in above reference, which is beyond the scope of this manuscript. We, however, will revise the discussion to elaborate on this potential mechanism.

(3) The evidence supporting Bicc1 and ADPKD gene cooperativity, particularly with Pkd1, in mouse models is not entirely convincing, likely due to substantial variability and the aggressive nature of Bpk/Bpk mice. Increasing the number of animals or using a milder Bicc1 strain, such as jcpk heterozygotes, could help substantiate the genetic interaction.

We have initially performed the analysis using our Bicc1 complete knockout, we previously reported on (PMID 20215348) focusing on compound heterozygotes. Yet, like the Pkd1/Pkd2 compound heterozygotes (PMID 12140187) no cyst development was observed until we sacrificed the mice at P21. Our strain is similar to the above mentioned jcpk, which is characterized by a short, abnormal transcript thought to result in a null allele (PMID: 12682776). We thank the reviewer for pointing use to the reference showing the heterozygous mice show glomerular cysts in the adults (PMID: 7723240). This suggestion is an interesting idea we will investigate. In general, we agree with the reviewer that the better understanding the contribution of Bicc1 to the adult PKD phenotype will be critical. To this end, we are currently generating a floxed allele of Bicc1 that will allow us to address the cooperativity in the adult kidney, when e.g. crossed to the Pkd1^RC/RC mice. Yet, these experiments are unfortunately beyond the scope of this manuscript.

Reviewer #2 (Public Review):

Tran and colleagues report evidence supporting the expected yet undemonstrated interaction between the Pkd1 and Pkd2 gene products Pc1 and Pc2 and the Bicc1 protein in vitro, in mice, and collaterally, in Xenopus and HEK293T cells. The authors go on to convincingly identify two large and non-overlapping regions of the Bicc1 protein important for each interaction and to perform gene dosage experiments in mice that suggest that Bicc1 loss of function may compound with Pkd1 and Pkd2 decreased function, resulting in PKD-like renal phenotypes of different severity. These results led to examining a cohort of very early onset PKD patients to find three instances of co-existing mutations in PKD1 (or PKD2) and BICC1. Finally, preliminary transcriptomics of edited lines gave variable and subtle differences that align with the theme that Bicc1 may contribute to the PKD defects, yet are mechanistically inconclusive.

These results are potentially interesting, despite the limitation, also recognized by the authors, that BICC1 mutations seem exceedingly rare in PKD patients and may not "significantly contribute to the mutational load in ADPKD or ARPKD". The manuscript has several intrinsic limitations that must be addressed.

As mentioned above, the study was designed to explore whether there is an interaction between BICC1 and the PKD1/PKD2 and whether this interaction is functionally important. How this translates into the clinical relevance will require additional studies (and we have addressed this in the discussion of the manuscript).

The manuscript contains factual errors, imprecisions, and language ambiguities. This has the effect of making this reviewer wonder how thorough the research reported and analyses have been.

We respectfully disagree with the reviewer on the latter interpretation. The study was performed with rigor. We have carefully assessed the critiques raised by the reviewer. Most of the criticisms raised by the reviewer will be easily addressed in the revised version of the manuscript. Yet, none of the critiques raised by the reviewer seems to directly impact the overall interpretation of the data.

Reviewer #3 (Public Review):

Summary:

This study investigates the role of BICC1 in the regulation of PKD1 and PKD2 and its impact on cytogenesis in ADPKD. By utilizing co-IP and functional assays, the authors demonstrate physical, functional, and regulatory interactions between these three proteins.

Strengths:

(1) The scientific principles and methodology adopted in this study are excellent, logical, and reveal important insights into the molecular basis of cystogenesis.

(2) The functional studies in animal models provide tantalizing data that may lead to a further understanding and may consequently lead to the ultimate goal of finding a molecular therapy for this incurable condition.

(3) In describing the patients from the Arab cohort, the authors have provided excellent human data for further investigation in large ADPKD cohorts. Even though there was no patient material available, such as HUREC, the authors have studied the effects of BICC1 mutations and demonstrated its functional importance in a Xenopus model.

Weaknesses:

This is a well-conducted study and could have been even more impactful if primary patient material was available to the authors. A further study in HUREC cells investigating the critical regulatory role of BICC1 and potential interaction with mir-17 may yet lead to a modifiable therapeutic target.

This is an excellent suggestion. We agree with the reviewer that it would have been interesting to analyze HUREC material from the affected patients. Unfortunately, besides DNA and the phenotypic analysis described in the manuscript neither human tissue nor primary patient-derived cells collected before the two patients with the BICC1 p.Ser240Pro mutation passed away. To address this missing link, we have – as a first pass - generated HEK293T cells carrying the BICC1 p.Ser240Pro variant. While these admittingly are not kidney epithelial cells, they indeed show a reduced level of PC2 expression. These data are shown in the manuscript. We have not yet addressed how this relates to its crosstalk with miR-17.

Conclusion:

The authors achieve their aims. The results reliably demonstrate the physical and functional interaction between BICC1 and PKD1/PKD2 genes and their products.

The impact is hopefully going to be manifold:

(1) Progressing the understanding of the regulation of the expression of PKD1/PKD2 genes.

(2) Role of BiCC1 in mir/PKD1/2 complex should be the next step in the quest for a modifiable therapeutic target.

eLife Assessment

The ratio of nuclei to cell volume is a well-controlled parameter in eukaryotic cells. This important study now substantially advances our understanding of the regulatory relationship between cell size and the number of nuclei by identifying novel players in this process. The evidence supporting the conclusions is compelling, with biochemical assays and state-of-the-art microscopy. The paper will be of broad interest for cell biologists and fungal biotechnologists seeking to understand mechanisms determining cell size and number of nuclei, and why this knowledge is also of significant importance for the production of enzymes, and thus production strains not only of Aspergillus oryzae, but also other industrially used fungi.

Reviewer #1 (Public review):

Filamentous fungi are established workhorses in biotechnology, with Aspergillus oryzae as a prominent example with a thousand-year history. Still, the cell biology and biochemical properties of the production strains is not well understood. The paper of the Takeshita group describes the change in nuclear numbers and correlates it to different production capacities. They used microfluidic devices to really correlate the production with nuclear numbers. In addition, they used microdissection to understand expression profile changes and found an increase in ribosomes. The analysis of two genes involved in cell volume control in S. pombe did not reveal conclusive answers to explain the phenomenon. It appears that it is a multi-trait phenotype. Finally, they identified SNPs in many industrial strains and tried to correlate them to the capability of increasing their nuclear numbers.

The methods used in the paper range from high-quality cell biology, Raman spectroscopy, to atomic force and electron microscopy, and from laser microdissection to the use of microfluidic devices to study individual hyphae.

This is a very interesting, biotechnologically relevant paper with the application of excellent cell biology. I have only minor suggestions for improvement.

Reviewer #2 (Public review):

Summary:

In the study presented by Itani and colleagues, it is shown that some strains of Aspergillus oryzae - especially those used industrially for the production of sake and soy sauce - develop hyphae with a significantly increased number of nuclei and cell volume over time. These thick hyphae are formed by branching from normal hyphae and grow faster and therefore dominate the colonies. The number of nuclei positively correlates with the thicker hyphae and also the amount of secreted enzymes. The addition of nutrients such as yeast extract or certain amino acids enhanced this effect. Genome and transcriptome analyses identified genes, including rseA, that are associated with the increased number of nuclei and enzyme production. The authors conclude from their data involvement of glycosyltransferases, calcium channels, and the tor regulatory cascade in the regulation of cell volume and number of nuclei. Thicker hyphae and an increased number of nuclei were also observed in high-production strains of other industrially used fungi such as Trichoderma reesei and Penicillium chrysogenum, leading to the hypothesis that the mentioned phenotypes are characteristic of production strains, which is of significant interest for fungal biotechnology.

Strengths:

The study is very comprehensive and involves the application of diverse state-of-the-art cell biological, biochemical, and genetic methods. Overall, the data are properly controlled and analyzed, figures and movies are of excellent quality.<br /> The results are particularly interesting with regard to the elucidation of molecular mechanisms that regulate the size of fungal hyphae and their number of nuclei. For this, the authors have discovered a very good model: (regular) strains with a low number of nuclei and strains with a high number of nuclei. Also, the results can be expected to be of interest for the further optimization of industrially relevant filamentous fungi.

Weaknesses:

There are only a few open questions concerning the activity of the many nuclei in production strains (active versus inactive), their number of chromosomes (haploid/diploid), and whether hyper-branching always leads to propagation of nuclei.

Reviewer #3 (Public review):

Summary:

The authors seek to determine the underlying traits that support the exceptional capacity of Aspergillus oryzae to secrete enzymes and heterologous proteins. To do so, they leverage the availability of multiple domesticated isolates of A. oryzae along with other Aspergillus species to perform comparative imaging and genomic analysis.

Strengths:

The strength of this study lies in the use of multifaceted approaches to identify significant differences in hyphal morphology that correlate with enzyme secretion, which is then followed by the use of genomics to identify candidate functions that underlie these differences.

Weaknesses:

There are aspects of the methods that would benefit from the inclusion of more detail on how experiments were performed and data interpreted.

Overall, the authors have achieved their aims in that they are able to clearly document the presence of two distinct hyphal forms in A. oryzae and other Aspergillus species, and to correlate the presence of the thicker, rapidly growing form with enhanced enzyme secretion. The image analysis is convincing. The discovery that the addition of yeast extract and specific amino acids can stimulate the formation of the novel hyphal form is also notable. Although the conclusions are generally supported by the results, this is perhaps less so for the genetic analysis as it remains unclear how direct the role of RseA and the calcium transporters might be in supporting the formation of the thicker hyphae.

The results presented here will impact the field. The complexity of hyphal morphology and how it affects secretion is not well understood despite the importance of these processes for the fungal lifestyle. In addition, the description of approaches that can be used to facilitate the study of these different hyphal forms (i.e., stimulation using yeast extract or specific amino acids) will benefit future efforts to understand the molecular basis of their formation.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Filamentous fungi are established workhorses in biotechnology, with Aspergillus oryzae as a prominent example with a thousand-year history. Still, the cell biology and biochemical properties of the production strains is not well understood. The paper of the Takeshita group describes the change in nuclear numbers and correlates it to different production capacities. They used microfluidic devices to really correlate the production with nuclear numbers. In addition, they used microdissection to understand expression profile changes and found an increase in ribosomes. The analysis of two genes involved in cell volume control in S. pombe did not reveal conclusive answers to explain the phenomenon. It appears that it is a multi-trait phenotype. Finally, they identified SNPs in many industrial strains and tried to correlate them to the capability of increasing their nuclear numbers.

The methods used in the paper range from high-quality cell biology, Raman spectroscopy, to atomic force and electron microscopy, and from laser microdissection to the use of microfluidic devices to study individual hyphae.

This is a very interesting, biotechnologically relevant paper with the application of excellent cell biology. I have only minor suggestions for improvement.

We sincerely appreciate your fair and positive evaluation of our work. Thank you for your suggestions for improvement. We respond to each of them appropriately.

Reviewer #2 (Public review):

Summary:

In the study presented by Itani and colleagues, it is shown that some strains of Aspergillus oryzae - especially those used industrially for the production of sake and soy sauce - develop hyphae with a significantly increased number of nuclei and cell volume over time. These thick hyphae are formed by branching from normal hyphae and grow faster and therefore dominate the colonies. The number of nuclei positively correlates with the thicker hyphae and also the amount of secreted enzymes. The addition of nutrients such as yeast extract or certain amino acids enhanced this effect. Genome and transcriptome analyses identified genes, including rseA, that are associated with the increased number of nuclei and enzyme production. The authors conclude from their data involvement of glycosyltransferases, calcium channels, and the tor regulatory cascade in the regulation of cell volume and number of nuclei. Thicker hyphae and an increased number of nuclei were also observed in high-production strains of other industrially used fungi such as Trichoderma reesei and Penicillium chrysogenum, leading to the hypothesis that the mentioned phenotypes are characteristic of production strains, which is of significant interest for fungal biotechnology.

Strengths:

The study is very comprehensive and involves the application of diverse state-of-the-art cell biological, biochemical, and genetic methods. Overall, the data are properly controlled and analyzed, figures and movies are of excellent quality.

The results are particularly interesting with regard to the elucidation of molecular mechanisms that regulate the size of fungal hyphae and their number of nuclei. For this, the authors have discovered a very good model: (regular) strains with a low number of nuclei and strains with a high number of nuclei. Also, the results can be expected to be of interest for the further optimization of industrially relevant filamentous fungi.

Weaknesses:

There are only a few open questions concerning the activity of the many nuclei in production strains (active versus inactive), their number of chromosomes (haploid/diploid), and whether hyper-branching always leads to propagation of nuclei.

We are very grateful for your recognition of our findings, the proposed model, and their significance for future applications. We are grateful for the questions, which contribute to a more accurate understanding.

Our responses to each are provided below. Necessary experiments are in progress.

Reviewer #3 (Public review):

Summary:

The authors seek to determine the underlying traits that support the exceptional capacity of Aspergillus oryzae to secrete enzymes and heterologous proteins. To do so, they leverage the availability of multiple domesticated isolates of A. oryzae along with other Aspergillus species to perform comparative imaging and genomic analysis.

Strengths:

The strength of this study lies in the use of multifaceted approaches to identify significant differences in hyphal morphology that correlate with enzyme secretion, which is then followed by the use of genomics to identify candidate functions that underlie these differences.

Weaknesses:

There are aspects of the methods that would benefit from the inclusion of more detail on how experiments were performed and data interpreted.

Overall, the authors have achieved their aims in that they are able to clearly document the presence of two distinct hyphal forms in A. oryzae and other Aspergillus species, and to correlate the presence of the thicker, rapidly growing form with enhanced enzyme secretion. The image analysis is convincing. The discovery that the addition of yeast extract and specific amino acids can stimulate the formation of the novel hyphal form is also notable. Although the conclusions are generally supported by the results, this is perhaps less so for the genetic analysis as it remains unclear how direct the role of RseA and the calcium transporters might be in supporting the formation of the thicker hyphae.

The results presented here will impact the field. The complexity of hyphal morphology and how it affects secretion is not well understood despite the importance of these processes for the fungal lifestyle. In addition, the description of approaches that can be used to facilitate the study of these different hyphal forms (i.e., stimulation using yeast extract or specific amino acids) will benefit future efforts to understand the molecular basis of their formation.

We are very grateful for your fair and thoughtful evaluation of our work. We agree that the genetic analysis in the latter part is relatively weaker compared to the imaging analysis in the first half. Rather than a single mutation causing a dramatic phenotypic change, we believe that the accumulation of various mutations through breeding leads to the observed phenotype, making it difficult to clearly demonstrate causality. Since transcriptome and SNP analyses have revealed key pathways and phenotypes, it would be gratifying if these insights could contribute to future applications utilizing filamentous fungi.

eLife Assessment

The manuscript presents a valuable finding that CCDC32, beyond its reported role in AP2 assembly, follows AP2 to the plasma membrane and regulates clathrin-coated pit assembly and dynamics. The authors further suggest that the alpha-helical region of CCDC32 interacts with AP2 via the alpha appendage domain to mediate this function. While live-cell and ultrastructural imaging data are solid, future biochemical studies will be needed to confirm the proposed CCDC32-AP2 interaction.

[Editors' note: this paper was reviewed by Review Commons.]

Reviewer #1 (Public review):

This is a revision of a manuscript previously submitted to Review Commons. The authors have partially addressed my comments, mainly by expanding the introduction and discussion sections. Sandy Schmid, a leading expert on the AP2 adaptor and CME, has been added as a co-corresponding author. The main message of the manuscript remains unchanged. Through overexpression of fluorescently tagged CCDC32, the authors propose that, in addition to its established role in AP2 assembly, CCDC32 also follows AP2 to the plasma membrane and regulates CCP maturation. The manuscript presents some interesting ideas, but there are still concerns regarding data inconsistencies and gaps in the evidence.

(1) eGFP-CCDC32 was expressed at 5-10 times higher levels than endogenous CCDC32. This high expression can artificially drive CCDC32 to the cell surface via binding to the alpha appendage domain (AD)-an interaction that may not occur under physiological conditions.

(2) Which region of CCDC32 mediates alpha AD binding? Strangely, the only mutant tested in this work, Δ78-98, still binds AP2, but shifts to binding only mu and beta. If the authors claim that CCDC32 is recruited to mature AP2 via the alpha AD, then a mutant deficient in alpha AD binding should not bind AP2 at all. Such a mutant is critical for establish the model proposed in this work.

(3) The concept of hemicomplexes is introduced abruptly. What is the evidence that such hemicomplexes exist? If CCDC32 binds to hemicomplexes, this must occur in the cytosol, as only mature AP2 tetramers are recruited to the plasma membrane. The authors state that CCDC32 binds the AD of alpha but not beta, so how can the Δ78-98 mutant bind mu and beta?

(4) The reported ability of CCDC32 to pull down AP2 beta is puzzling. Beta is not found in the CCDC32 interactome in two independent studies using 293 and HCT116 cells (BioPlex). In addition, clathrin is also absent in the interactome of CCDC32, which is difficult to reconcile with a proposed role in CCPs. Can the authors detect CCDC32 binding to clathrin?

(5) Figure 5B appears unusual-is this a chimera? Figure 5C likely reflects a mixture of immature and mature AP2 adaptor complexes.

(6) CCDC32 is reduced by about half in siRNA knockdown. Why not use CRISPR to completely eliminate CCDC32 expression?

Reviewer #2 (Public review):

Yang et al. describes CCDC32 as a new clathrin mediated endocytosis (CME) accessory protein. The authors show that CCDC32 binds directly to AP2 via a small alpha helical region and cells depleted for this protein show defective CME. Finally, the authors show that the CCDC32 nonsense mutations found in patients with cardio-facial-neuro-developmental syndrome (CFNDS) disrupt the interaction of this protein to the AP2 complex. The results presented suggest that CCDC32 may act as both a chaperone (as recently published) and a structural component of the AP2 complex.

Strengths:<br /> The conclusions presented are generally well supported by experimental data and the authors carefully point out the differences between their results and the results by Wan et al. (PNAS 2024).

Weaknesses:<br /> The experiments regarding the role of CCDC32 in CFNDS still require some clarifications to make them clearer to scientists working on this disease. The authors fail to describe that the CCDC32 isoform they use in their studies is different from the one used when CFNDS patient mutations were described. This may create some confusion. Also, the authors did not discuss that the frame-shift mutations in patients may be leading to nonsense mediated decay.

Reviewer #3 (Public review):

In this manuscript, Yang et al. characterize the endocytic accessory protein CCDC32, which has implications in cardio-facio-neuro-developmental syndrome (CFNDS). The authors clearly demonstrate that the protein CCDC32 has a role in the early stages of endocytosis, mainly through the interaction with the major endocytic adaptor protein AP2, and they identify regions taking part in this recognition. Through live cell fluorescence imaging and electron microscopy of endocytic pits, the authors characterize the lifetimes of endocytic sites, the formation rate of endocytic sites and pits and the invagination depth, in addition to transferrin receptor (TfnR) uptake experiments. Binding between CCDC32 and CCDC32 mutants to the AP2 alpha appendage domain is assessed by pull down experiments. While interaction between CCDC32 and the alpha appendage domain of AP2 is clearly described, a discussion of potential association with other AP2 domains would be beneficial to understand the impact of CCDC32 in endocytosis.

Together, these experiments allow deriving a phenotype of CCDC32 knock-down and CCDC32 mutants within endocytosis, which is a very robust system, in which defects are not so easily detected. A mutation of CCDC32, mimicking CFNDS mutations, is also addressed in this study and shown to have endocytic defects.

In summary, the authors present a strong combination of techniques, assessing the impact of CCDC32 in clathrin mediated endocytosis and its binding to AP2.

Author Response:

We sincerely thank the reviewers and the editorial team for their thoughtful and constructive evaluation of our manuscript. We are very pleased that both reviewers and the Reviewing Editor found the work to be compelling and of interest to the community studying membrane-associated condensates. Below we outline our planned revisions in response to the public reviews.

Reviewer #1

We appreciate Reviewer #1’s positive evaluation of the study’s significance and the utility of our theoretical framework.

1. Understandably, the authors used one system to test their theory (ZO-1). However, to establish a theoretical framework, this is sufficient.

Response: We acknowledge this limitation. While we agree that additional systems would strengthen the generality of our theory, we note that the focus of this work is to introduce and validate a theoretical framework. As the reviewer notes, this is sufficient for establishing the framework. Nonetheless, we are open to further collaborations or future studies to test the model with other systems.

Reviewer #2

We are grateful for Reviewer #2’s detailed comments and will address each of the points as follows:

1. In the theoretical section, what has previously been known, compared to which equations are new, should be made more clear.

Response: We will revise the theory section to clearly distinguish previously established formulations from novel contributions.

1. Some assumptions in the model are made purely for convenience and without sufficient accompanying physical justification. E.g., the authors should justify, on physical grounds, why binding rate effects are/could be larger than the other fluxes.

Response: We will expand the discussion to provide key physical justification, especially to explain why binding rate effects are/could be larger than the other fluxes.

1. I feel that further mechanistic explanation as to why bulk phase separation widens the regime of surface phase separation is warranted.

Response: We will elaborate on the mechanism underlying this coupling.

1. The major advantage of the non-dilute theory as compared with a best parameterized dilute (or homogenous) theory requires further clarification/evidence with respect to capturing the experimental data.

Response: We will clarify this comparison more explicitly and highlight how the non-dilute model captures key nonlinear behaviors and concentration-dependent adsorption phenomena that the dilute model fails to reproduce.

1. Discrete (particle-based) molecular modelling could help to delineate the quantitative improvements that the non-dilute theory has over the previous state-of-the-art. Also, this could help test theoretical statements regarding the roles of bulk-phase separation, which were not explored experimentally.

Response:  We appreciate the suggestion and agree that such modeling would be valuable. However, this is beyond the scope of the current study. We will add a discussion on how discrete simulations could be used to further test our theory in future work.

1. Discussion of the caveats and limitations of the theory and modelling is missing from the text.

Response:  We will add a paragraph outlining caveats and limitations of the modelling.

We believe these changes will significantly improve the clarity and impact of our manuscript, and we thank the reviewers again for their valuable input.

eLife Assessment

This important study presents a compelling theoretical framework for understanding phase separation of membrane-bound proteins, with a focus on the organization of tight junction components. By incorporating non-dilute binding effects into thermodynamic models and validating the model's predictions with in vitro experiments on the tight junction protein ZO-1, the authors provide a quantitative tool that will be of interest for biologists interested in membrane-associated condensates. While further clarification of model assumptions and broader mechanistic context would strengthen the work even further, the combination of theory and experiment here is robust and a key advancement in the field.

Reviewer #1 (Public review):

Summary:

Biomolecular condensates are an essential part of cellular homeostatic regulation. In this manuscript, the authors develop a theoretical framework for the phase separation of membrane-bound proteins. They show the effect of non-dilute surface binding and phase separation on tight junction protein organization.

Strengths:

It is an important study, considering that the phase separation of membrane-bound molecules is taking the center stage of signaling, spanning from immune signaling to cell-cell adhesion. A theoretical framework will help biologists to quantitatively interpret their findings.

Weaknesses:

Understandably, the authors used one system to test their theory (ZO-1). However, to establish a theoretical framework, this is sufficient.

Reviewer #2 (Public review):

Summary:

The authors present a clear expansion of biophysical (thermodynamic) theory regarding the binding of proteins to membrane-bound receptors, accounting for higher local concentration effects of the protein. To partially test the expanded theory, the authors perform in vitro experiments on the binding of ZO1 proteins to Claudin2 C-terminal receptors anchored to a supported lipid bilayer, and capture the effects that surface phase separation of ZO1 has on its adsorption to the membrane.

Strengths:

(1) The derived theoretical framework is consistent and largely well-explained.

(2) The experimental and numerical methodologies are transparent.

(3) The comparison between the best parameterized non-dilute theory is in reasonable agreement with experiments.

Weaknesses:

(1) In the theoretical section, what has previously been known, compared to which equations are new, should be made more clear.

(2) Some assumptions in the model are made purely for convenience and without sufficient accompanying physical justification. E.g., the authors should justify, on physical grounds, why binding rate effects are/could be larger than the other fluxes.

(3) I feel that further mechanistic explanation as to why bulk phase separation widens the regime of surface phase separation is warranted.

(4) The major advantage of the non-dilute theory as compared with a best parameterized dilute (or homogenous) theory requires further clarification/evidence with respect to capturing the experimental data.

(5) Discrete (particle-based) molecular modelling could help to delineate the quantitative improvements that the non-dilute theory has over the previous state-of-the-art. Also, this could help test theoretical statements regarding the roles of bulk-phase separation, which were not explored experimentally.

(6) Discussion of the caveats and limitations of the theory and modelling is missing from the text.

Author response:

We thank the reviewers for their thoughtful and constructive feedback. As the reviewers noted, dissecting the contributions of Gtr1/2 and Pib2 to TORC1 signaling across diverse nutrient states is a technically and conceptually challenging problem. Indeed, many of the issues raised—including the interpretation of non-canonical TORC1 readouts (e.g., Rps6, Par32), the influence of strain auxotrophy and media composition, and the limitations of phosphoproteomic analysis performed under a single growth condition—underscore the challenges of working with the TORC1 signaling system.

In response to the reviewers’ comments, we have undertaken a broader and more systematic analysis of TORC1 regulation across defined nitrogen transitions, building directly on the signaling framework established in Figures 6 and 8 of this manuscript. This work, which includes expanded phosphoproteomic profiling and the use of refined genetic tools, supports and extends the key conclusions of Cecil et. al. Specifically, it reinforces the existence of a Pib2-dependent TORC1 output under nitrogen-limited conditions and further clarifies the physiological relevance of the intermediate TORC1 activity state. Due to the scope and depth of this expanded work, we are reporting those findings in a separate publication. Nonetheless, we view the data presented here as a key foundational step in establishing a non-redundant framework for Gtr1/2- and Pib2-dependent control of TORC1.

We have therefore made minor changes to the manuscript to clarify our use of different growth media and to temper our conclusions where appropriate. These changes, together with the context of ongoing work, should reinforce the value of Cecil et. al. in advancing our understanding of TORC1 and nutrient signaling in eukaryotes.

eLife Assessment

This study provides an in-depth exploration of the impact of X-linked ZDHHC9 gene mutations on cognitive deficits and epilepsy, with a particular focus on the expression and function of ZDHHC9 in myelin-forming oligodendrocytes (OLs). These valuable findings offer insights into ZDHHC9-related X-linked intellectual disability (XLID) and shed light on the regulatory mechanisms of palmitoylation in myelination. The experimental design and analysis of results are solid, providing a reference for further research in this field.

Reviewer #1 (Public review):

Summary:<br /> Having shown that acyltransferase ZDHHC9 expression is far higher in myelinating oligodendrocytes (OLs) than in other CNS cell types, Jeong and colleagues focus on exploring the role of ZDHHC9 in myelinating OLs in particular in the palmitoylation of several myelin proteins. This study is relevant in the context of X-linked intellectual disability as it suggests a more relevant role for myelinating glia than previously thought. It also provides useful insights the mechanisms of ZDHHC9-associated XLID and on the palmitoylation-dependent control of myelination.

Strengths:<br /> Well written paper<br /> In general good data quality<br /> Use of transgenics strategies (in addition to the ZDHHC9 KO) strengthen the data and claims

Weaknesses:<br /> A few claims might have needed better experimental support but new data and revised discussion sections addressed some of these weaknesses

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public Review):

In this work Jeong and colleagues focus on exploring the role of the acyltransferase ZDHHC9 in myelinating OLs in particular in the palmitoylation of several myelin proteins. After confirming the specific enrichment of the Zdhhc9 transcript in mouse and human OLs, the authors examine the subcellular localization of the protein in vitro and observed that in comparison with other isoforms, ZDHHC9 localizes at OLs cell bodies and at discrete puncta in the processes. These observations (Figures 1 and 2) led the authors to hypothesize that ZDHHC9 plays an important role in myelination. No gross changes were detected in OL development in Zdhhc9 KO mice and analyses from P28 Zdhhc9 KO mice crossed with Mobp-EGFP reporter mice did not show changes in EGFP+ OL differentiation (Figure 3).

However, and given the observed subcellular localization of ZDHHC9 in OL processes (Figure 2) and the observation that the percentage of unmyelinated axons is increased in Zdhhc9 KO (Figure 6), early time points to examine the differentiated pools of OLs and their capacity to extend processes/contact axons need to be considered.

We appreciate this point, but due to the order in which experiments were performed, the ZDHHC9 KO mouse colony that we maintained after initial submission of this work contains homozygous MOBP-EGFP, but not the mT/mG transgene that would be most optimal for the proposed experiment. We hope the reviewer appreciates that it would take considerable time and effort regarding mouse breeding to cross out the MOBP and add back the mT/mG. We nonetheless appreciate the importance of the point raised and therefore examined an earlier developmental time point (P21, 3 weeks) to quantify OLs and NG2+ OPCs. In our updated Fig 3C1-C3, we use Mobp-EGFP mice to show that Zdhhc9 KO does not significantly affect the number of EGFP+ OLs at this time point in the cortex, corpus callosum and spinal cord. We also show that in corpus callosum, Zdhhc9 KO does not significantly affect the number of NG2+ OPCs at this earlier time point (Fig 3D, E). Furthermore, immunostaining to detect BCAS1, a marker of pre-mature OLs, also revealed no qualitative difference with ZDHHC9 loss at P21. We show representative images from these BCAS1 experiments in an updated Fig S3. While these new experiments do not address the morphology of OLs in Zdhhc9 KO, they do provide further evidence that deficits in myelination in young Zdhhc9 KO mice (Figure 6) are not likely due to gross differences in OPC or OL numbers during development.

Maturation of OL in Zdhhc9 KO was examined by crossing Zdhhc9 KO with Pdgfra-CreER;R26- EGFP and following the newly EGFP-labelled OPCs following tamoxifen administration. No changes in the numbers of EGFP+ OL were detected. The authors concluded that the loss of ZDHHC9 does not alter oligodendrogenesis in either the young or mature CNS. The authors observed defects in Zdhhc9 KO OL protrusions that they attributed to abnormal OL membrane expansion (Fig 4 and 5). Can they show evidence for this?

This is an important point, and we appreciate the opportunity to explain the reasoning behind our initial statement more fully, while noting that other explanations are possible. Fig 5B (an Imaris-assisted reconstruction using the EGFP cell fill/morphology marker) highlights large spheroid-like distensions along OL processes. We reason that these spheroids are enclosed by the OL lipid membrane because if the membrane were ruptured, the EGFP signal would likely diffuse. This in turn suggests that the caliber of the OL process at the position of the spheroid is grossly abnormal i.e. the membrane has hyper-expanded. Given that OL membrane growth during myelination extends in two directions, i.e., spiral growth to the axonal surface and longitudinal growth along the axon, it is possible that spheroid-like structures are formed by uneven myelin growth. We recognize that we cannot yet conclude whether and how spheroid formation might be linked to the myelination deficit that we observe in Zdhhc9 KO mice. However, defining the subcellular mechanism for spheroid formation may provide further insights into this issue. We have therefore largely retained the original statement but have added the reasoning above to our revised Discussion.

The authors report that Zdhhc9 KO primary and secondary branches in OL were longer, some contained spheroid-like swellings and the OL protrusion complexity was higher. However, these data is partially contradictory to what they show in OL differentiation experiments in vitro (Fig 7). There is also no evidence for increased membrane expansion in Zdhhc9 knockdown myelin forming cells in culture. How to reconcile this? 

We appreciate the reviewer’s interest in this issue. Several non-mutually exclusive factors could account for the differences in OL morphology in vitro versus in vivo caused by Zdhhc9 loss. First, morphology in vivo may well be influenced by the axons and/or other extrinsic components around each OL that are not present in our primary cultures. Second, OL growth in vivo is highly 3-dimensional, whereas growth in culture is largely 2-dimensional – it may be difficult to support formation of spheroids (by definition, a 3-dimensional structure) in the latter situation. Finally, Zdhhc9 is absent in vivo from the beginning of development until the time points examined, whereas in our cultured OL experiments, Zdhhc9 shRNA is virally delivered to OPC cultures at DIV2 and likely acutely affects Zdhhc9 expression predominantly in committed OLs (following the switch to differentiation medium at DIV3). These differences may also affect the ability of other PATs or, potentially, palmitoylation-independent subcellular processes, to compensate for Zdhhc9 loss. We have more fully explained these points in our revised Discussion. 

Reviewer #2 (Public Review):

This study provides an in-depth exploration of the impact of X-linked ZDHHC9 gene mutations on cognitive deficits and epilepsy, with a particular focus on the expression and function of ZDHHC9 in myelin-forming oligodendrocytes (OLs). These findings offer crucial insights into understanding ZDHHC9-related X-linked intellectual disability (XLID) and shed light on the regulatory mechanisms of palmitoylation in myelination. The experimental design and analysis of results are convincing, providing a valuable reference for further research in this field. However, upon careful review, I believe the article still needs further improvement and supplementation in the following aspects:

(1) Regarding the subcellular localization experiment of ZDHHC9 mutants in OL, it is currently limited to in vitro cultured OL, lacking validation in vivo OL or myelin sheath. Additionally, it is necessary to investigate whether the abnormal subcellular localization of ZDHHC9 mutants affects their enzyme activity and palmitoylation modification of substrate proteins.

This is an important point but is technically challenging to address in vivo as it would likely require delivery of AAV to express ZDHHC9wt and XLID mutants specifically in OLs, preferably in the absence of endogenous ZDHHC9. We hope the reviewers would agree that this experiment is beyond the scope of the current study. However, we did compare the ability of ZDHHC9wt and XLID mutants to palmitoylate MBP, and to autopalmitoylate (sometimes used as a surrogate measure of PAT activity) in transfected heterologous cells. Although we recognize that this over-expression system is less physiological than a native OL, it has the benefit of being able to readily compare transfected wt vs mutant forms of ZDHHC9 with minimal contribution from endogenous ZDHHC9. Intriguingly, using this system, we found that autopalmitoylation activity of the XLID ZDHHC9-P150S mutant does not differ significantly from that of ZDHHC9wt, and that this mutant is still capable of palmitoylating MBP. Moreover, the R96W mutant, while impaired in autopalmitoylation, still palmitoylated MBP approximately 50% as effectively as ZDHHC9wt in our cell-based assay. These findings suggest that ZDHHC9-P150S and, probably, ZDHHC9-R96W mutants might still be able to palmitoylate substrates in OLs if they were properly localized. This possibility in turn suggests that impaired subcellular targeting in addition to, or instead of, impaired catalytic activity, may be a key factor in certain cases of ZDHHC9-associated XLID. We have expanded our Figure 8 (new panels 8E-G) to show these additional experiments and have summarized the conclusions above in our revised Discussion. We thank the reviewer for suggesting that we further investigate this issue.

(2) The experimental period (P21+21 days) using genetic labeling to track the development of myelinating cells may not be long enough. It is recommended to extend the observation time and analyze at more time points to more comprehensively reflect the impact of Zdhhc9 KO.

We appreciate this point from the reviewer but, regrettably, we did not maintain the PdgfraCreER; R26-EGFP; Zdhhc9 KO mouse line and hope the reviewer appreciates that it would take considerable time and effort to rederive this line and then perform the suggested extended time course experiments. However, we note for the reviewer that our preliminary studies did not reveal any effect of Zdhhc9 KO on the number of MOBP-EGFP+ OLs in 6-month-old mice (not shown), consistent with a model in which Zdhhc9 loss does not affect OPC-OL commitment per se.

(3) The author speculates that Zdhhc9 may regulate myelination by affecting the membrane localization of specific myelin proteins, but lacks direct experimental evidence to support this. It is suggested to detect the expression and distribution of relevant proteins in the myelin of Zdhhc9 KO mice.

We share the reviewer’s interest in this point but realized that it is more technically challenging to address than might be initially thought. The main protein we would implicate and seek to test is MBP, but we already found that there is no gross change in MBP distribution in vivo in Zdhhc9 KO mice (Fig 3A). However, an anti-MBP antibody recognizes all forms of MBP, not just the specific splice variants whose palmitoylation is affected by ZDHHC9 loss. Specifically assessing nanoscale distribution of these splice variants would require a way (e.g. anti-MBP splice form-specific antibodies that are compatible with immuno-EM) to distinguish these variants from other, non-palmitoylated forms of MBP. Although such an antibody could be an important tool, we hope the reviewers would agree that developing and characterizing such a reagent is beyond the scope of the current study.

We do, however, note that the lack of gross change in MBP distribution and levels in Zdhhc9 KO mice is consistent with the relatively mild phenotype of these mice, compared with shiverer (shi/shi) mice, in which MBP is completely lost. In shiverer, CNS compact myelin is almost absent (PMID: 671037; PMID: 88695; PMID: 460693) and, as the name suggests, mice display a shivering gait, and exhibit seizures and early death. In contrast, Zdhhc9 mice show only subtle behavioral deficits (PMID: 29944857). These differences are all consistent with a model in which Zdhhc9 KO mice, despite their significantly reduced MBP palmitoylation (Fig 8) have grossly normal distribution and levels of MBP when all splice variants are assessed (Fig 3, Fig 8). It is not inconceivable that Zdhhc9 KO mice have a nanoscale change in the distribution of MBP, particularly of specific palmitoylated splice variants, within myelin that profoundly affects myelin ultrastructure, without grossly altering MBP distribution. However, an alternative and not mutually exclusive possibility is that aberrant palmitoylation of other Zdhhc9 substrates accounts for, or contributes to, the abnormalities in myelin at the ultrastructural level. Addressing this issue would require a multi-pronged approach, not just to assess palmitoylation and distribution of such proteins in Zdhhc9 KO, but also to test whether they are direct Zdhhc9 substrates, in order to rule out indirect effects. We hope reviewers would agree that this is best left to a separate study. However, in our revised Discussion we now summarize what can be inferred regarding Zdhhc9-dependent effects on total and splicevariant specific distribution and levels of MBP.  

(4) Although the article mentions the association of Zdhhc9 with intellectual disabilities, it does not involve behavioral analysis of Zdhhc9 KO mice. It is recommended to supplement some behavioral experimental data to support the important role of Zdhhc9 in maintaining normal cognitive function, enhancing the clinical relevance of the article.

We appreciate this point from the reviewer. The behavior of the same ZDHHC9 KO mouse line that we used was reported in PMID: 31747610 and in PMID: 29944857. In the former study, Zdhhc9 KO mice were reported to display seizures reminiscent of phenotypes in human patients with ZDHHC9 mutation. The latter study assessed performance of Zddhc9 KO mice in several tasks that test cognitive function. Specifically the KO mice were reported to display “altered behaviour in the open-field test, elevated plus maze and acoustic startle test that is consistent with a reduced anxiety level; a reduced hang time in the hanging wire test that suggests underlying hypotonia but which may also be linked to reduced anxiety [and] deficits in the Morris water maze test of hippocampal-dependent spatial learning and memory.”. We have incorporate these findings in our revised Discussion, where we summarize how these phenotypes are common, not just to human patients with ZDHHC9 mutation, but also to other human neurodevelopmental conditions and mouse models in which ID is a common feature.

(5) For the abnormal myelination observed in Zdhhc9 KO mice, including unmyelinated large-diameter axons and excessively myelinated small-diameter axons, the article lacks indepth research and explanation on the exact mechanism and mode of action of ZDHHC9 in regulating myelination.

We share the reviewer’s interest in this point but again note that gaining definitive insights into this issue is far from trivial. Convincing evidence of a causative mechanism would require an exhaustive identification of ZDHHC9 in vivo substrates, followed by point mutation of substrate palmitoylation site(s) to determine the extent to which palmitoylation of such protein(s) phenocopies ZDHHC9 loss. Nonetheless, it is possible to break this question down and to summarize what we do and do not know. For example, our experiments in cultured OLs show that ZDHHC9 loss causes call-autonomous deficits in morphological maturation of these cells. We also know that ZDHHC9 loss results in impaired palmitoylation of MBP, a direct substrate for ZDHHC9. Moreover, loss of ZDHHC9 at Golgi outposts in OLs (a phenotype observed with several XLID-associated mutant forms of ZDHHC9, even those with no significant loss of catalytic activity) correlates with intellectual disability. Together, these findings are consistent with a model in which ZDHHC9 action at OL Golgi outposts is critical for normal myelination. However, it is yet to be determined whether the key substrates of ZDHHC9 include MBP, other palmitoyl-proteins that are key constituents of CNS myelin, or proteins whose palmitoylation is important for myelin protein trafficking and targeting. Another non-mutually exclusive possibility is that ZDHHC9 acts at Golgi outposts but indirectly, for example to drive the expression of myelin protein genes. Future experiments, including but not limited to palmitoyl-proteomics in ZDHHC9 (OL-specific) KO mice, will be needed to provide more definitive insights into this issue. We have expanded our Discussion of links between ZDHHC9 mutation and impaired myelination to summarize the above points.

(6) The function of ZDHHC9 in OL may be related to the Golgi apparatus, but its exact role in these structures is still unclear. It is suggested to discuss in more detail the role of ZDHHC9 in the Golgi apparatus in the discussion section.

We appreciate this point, which we considered as related to point (5) above. In our revised Discussion we highlight how ZDHHC9 action at Golgi outposts may involve direct palmitoylation of myelin proteins, palmitoylation of proteins that direct myelin proteins to the myelin membrane and/or activation of gene expression programs that serve to drive myelination. We further note that these possibilities are not mutually exclusive.

(7) More experimental support and in-depth research are needed on the detailed mechanism of how ZDHHC9 and Golga7 cooperatively regulate MBP palmitoylation, and how this decrease in palmitoylation level leads to myelination defects.

This is another important point – our new experiments suggest that, although some XLID mutations markedly affect ZDHHC9’s ability to palmitoylate MBP, others do not, yet all of the mutant forms fail to localize to Golgi outposts. These findings are consistent with a model in which the subcellular location at which ZDHHC9 palmitoylates MBP, and potentially other substrates, is critical for normal myelination. Interestingly, despite their marked differences in basal catalytic activity (as assessed by autopalmitoylation), wt and all XLID forms of ZDHHC9 appear to show enhanced activity (measured by both auto- and MBP palmitoylation) in the presence of ZDHHC9, suggesting that the association with Golga7 (which also localizes to Golgi outposts) is central to ZDHHC9 activity. This model is also highly consistent with the biased expression of Golga7 in OLs, compared to other CNS cell types (Fig 1E, 1F). Moreover, XLID-associated mutant forms of ZDHHC9 also show reduced protein stability and are impaired in their ability to form complexes with Golga7 (also known as Golgi Complex Protein 16kDa; GCP16; PMID: 37035671). Failure of ZDHHC9 XLID mutants to localize to Golgi outposts may thus be due to aberrant trafficking of mutant ZDHHC9 per se, but may also involve impaired association/stabilization of ZDHHC9/Golga7 complexes at these locations. Again, it is possible that either or both of these mechanisms, which are not mutually exclusive, contribute to impaired MBP palmitoylation and/or myelination deficits. We summarize these points in our revised Discussion.

In summary, it is recommended that the authors address the above issues through additional experiments and improved discussions to further strengthen the credibility and clinical relevance of the article.

Recommendations for the authors:  

Reviewer #1 (Recommendations For The Authors):

No gross changes were detected in OL development in Zdhhc9 KO mice and analyses from P28 Zdhhc9 KO mice crossed with Mobp-EGFP reporter mice did not show changes in EGFP+ OL differentiation (Figure 3). However, and given the observed subcellular localization of ZDHHC9 in OL processes (Figure 2) and the observation that the percentage of unmyelinated axons is increased in Zdhhc9 KO (Figure 6), ***early time points to examine the differentiated pools of OLs and their capacity to extend processes/contact axons need to be considered***.

We appreciate this point, but due to the order in which experiments were performed, the ZDHHC9 KO mouse colony that we maintained after initial submission of this work contains homozygous MOBP-EGFP, but not the mT/mG transgene that would be most optimal for the proposed experiment. We hope the reviewer appreciates that it would take considerable time and effort regarding mouse breeding to cross out the MOBP and add back the mT/mG. We nonetheless appreciate the importance of the point raised and therefore examined an earlier developmental time point (P21, 3 weeks) to quantify OLs and NG2+ OPCs. In our updated Fig 3C1-C3, we use Mobp-EGFP mice to show that Zdhhc9 KO does not significantly affect the number of EGFP+ OLs at this time point in the cortex, corpus callosum and spinal cord. We also show that in corpus callosum, Zdhhc9 KO does not significantly affect the number of NG2+ OPCs at this earlier time point (Fig 3D, E). Furthermore, immunostaining to detect BCAS1, a marker of pre-mature OLs, also revealed no qualitative difference with ZDHHC9 loss at P21. We show representative images from these BCAS1 experiments in an updated Fig S3. While these new experiments do not address the morphology of OLs in Zdhhc9 KO, they do provide further evidence that deficits in myelination in young Zdhhc9 KO mice (Figure 6) are not likely due to gross differences in OPC or OL numbers during development.

The authors observed defects in Zdhhc9 KO OL protrusions that they attributed to abnormal OL membrane expansion (Fig 4 and 5). Can they show evidence for this?

This is an important point, and we appreciate the opportunity to explain the reasoning behind our initial statement more fully, while noting that other explanations are possible. Fig 5B (an Imaris-assisted reconstruction using the EGFP cell fill/morphology marker) highlights large spheroid-like distensions along OL processes. We reason that these spheroids are enclosed by the OL lipid membrane because if the membrane were ruptured, the EGFP signal would likely diffuse. This in turn suggests that the caliber of the OL process at the position of the spheroid is grossly abnormal i.e. the membrane has hyper-expanded. Given that OL membrane growth during myelination extends in two directions, i.e., spiral growth to the axonal surface and longitudinal growth along the axon, it is possible that spheroid-like structures are formed by uneven myelin growth. We recognize that we cannot yet conclude whether and how spheroid formation might be linked to the myelination deficit that we observe in Zdhhc9 KO mice.

However, defining the subcellular mechanism for spheroid formation may provide further insights into this issue. We have therefore largely retained the original statement but have added the reasoning above to our revised Discussion.

The authors report that Zdhhc9 KO primary and secondary branches in OL were longer, some contained spheroid-like swellings and the OL protrusion complexity was higher. However, these data is partially contradictory to what they show in OL differentiation experiments in vitro (Fig 7). There is also no evidence for increased membrane expansion in Zdhhc9 knockdown myelin forming cells in culture. How do they reconcile these different findings?

We appreciate the reviewer’s interest in this issue. Several non-mutually exclusive factors could account for the differences in OL morphology in vitro versus in vivo caused by Zdhhc9 loss. First, morphology in vivo may well be influenced by the axons and/or other extrinsic components around each OL that are not present in our primary cultures. Second, OL growth in vivo is highly 3-dimensional, whereas growth in culture is largely 2-dimensional – it may be difficult to support formation of spheroids (by definition, a 3-dimensional structure) in the latter situation. Finally, Zdhhc9 is absent in vivo from the beginning of development until the time points examined, whereas in our cultured OL experiments, Zdhhc9 shRNA is virally delivered to OPC cultures at DIV2 and likely acutely affects Zdhhc9 expression predominantly in committed OLs (following the switch to differentiation medium at DIV3). These differences may also affect the ability of other PATs or, potentially, palmitoylation-independent subcellular processes, to compensate for Zdhhc9 loss. We have more fully explained these points in our revised Discussion. 

Page 7: "The OL processes in this culture condition correspond to large lipid-rich membranous sheets that form spiral membrane expansion on axons in vivo (49)." At which stage are authors referring to? OL processes are extended in culture before membrane formation and this is not clear here. In a 3-days differentiation culture, most OLs have not yet formed a myelin sheath (eg., Figure 2 in Zuchero et al., 2015, Dev Cell).

We appreciate the reviewer highlighting this point. We first note that our oligodendrocyte (OL) culture conditions differ from the immunopanning method used by Zuchero et al., 2015 (original reference (Emery and Dugas, 2013)), which may affect the time course and progression of OL process elaboration and/or myelin sheath formation. We further note that in our cultures most EGFP+ processes are also MBP+ at the time point examined (strictly 3 days plus 9 hours post-differentiation). It thus seems likely that these MBP+ structures largely correspond to the MBP+ wrapping sheaths that occur in vivo, so we have therefore retained our original statement but have added this further explanation.

Minor: Figure 6 (Legend): Time points should be indicated throughout the panels.

We have added this information as requested

Reviewer 2 Recommendations for the Authors:

(1) Regarding the subcellular localization experiment of ZDHHC9 mutants in OL, it is currently limited to in vitro cultured OL, lacking validation in vivo OL or myelin sheath. Additionally, it is necessary to investigate whether the abnormal subcellular localization of ZDHHC9 mutants affects their enzyme activity and palmitoylation modification of substrate proteins.

We thank the reviewer for raising this point. New data in our revised Figure 8 compares autopalmitoylation (sometimes used as a surrogate measure of PAT activity) of ZDHHC9wt and XLID mutants, and their ability to palmitoylate MBP in transfected cells. Intriguingly, we found that autopalmitoylation activity of the ZDHHC9-P150S mutant does not differ significantly from that of ZDHHC9wt, and that this mutant is still capable of palmitoylating MBP. Moreover, the R96W mutant, while impaired in autopalmitoylation, still palmitoylated MBP approximately 50% as effectively as ZDHHC9wt in our cell-based assay. These findings suggest that ZDHHC9-P150S and, probably, ZDHHC9-R96W mutants might still be able to palmitoylate substrates in OLs if they were properly localized. This possibility in turn suggests that impaired subcellular targeting in addition to, or instead of, impaired catalytic activity, may be a key factor in certain cases of ZDHHC9-associated XLID. We have expanded our Figure 8 to show these new experiments and have summarized the conclusions above in our revised Discussion. We thank the reviewer for suggesting that we further investigate this issue.

(2) The experimental period (P21+21 days) using genetic labeling to track the development of myelinating cells may not be long enough. It is recommended to extend the observation time and analyze at more time points to more comprehensively reflect the impact of Zdhhc9 KO.

We appreciate this point from the reviewer but, regrettably, we did not maintain the PdgfraCreER; R26-EGFP; Zdhhc9 KO mouse line and hope the reviewer appreciates that it would take considerable time and effort to rederive this line and then perform the suggested extended time course experiments. However, we note for the reviewer that our preliminary studies did not reveal any effect of Zdhhc9 KO on the number of MOBP-EGFP+ OLs in 6-month-old mice (not shown), consistent with a model in which Zdhhc9 loss does not affect OPC-OL commitment per se.

(3) The author speculates that Zdhhc9 may regulate myelination by affecting the membrane localization of specific myelin proteins, but lacks direct experimental evidence to support this. It is suggested to detect the expression and distribution of relevant proteins in the myelin of Zdhhc9 KO mice.

We share the reviewer’s interest in this point but realized that it is more technically challenging to address than might be initially thought. The main protein we would implicate and seek to test is MBP, but we already found that there is no gross change in MBP distribution in vivo in Zdhhc9 KO mice (Fig 3A). However, an anti-MBP antibody recognizes all forms of MBP, not just the specific splice variants whose palmitoylation is affected by ZDHHC9 loss. Specifically assessing nanoscale distribution of these splice variants would require a way (e.g. am anti-MBP splice form-specific antibody that is compatible with immuno-EM) to distinguish these variants from other, non-palmitoylated forms of MBP. Although such an antibody could be an important tool we hope the reviewers would agree that developing and characterizing such a reagent is beyond the scope of the current study.

We do, however, note that the lack of gross change in MBP distribution and levels in Zdhhc9 KO mice is consistent with the relatively mild phenotype of these mice, compared with shiverer (shi/shi) mice, in which MBP is completely lost. In shiverer, CNS compact myelin is almost absent (PMID: 671037; PMID: 88695; PMID: 460693) and, as the name suggests, mice display a shivering gait, and exhibit seizures and early death. In contrast, Zdhhc9 mice show only subtle behavioral deficits (PMID: 29944857). These differences are all consistent with a model in which Zdhhc9 KO mice, despite their significantly reduced MBP palmitoylation (Fig 8) have grossly normal distribution and levels of MBP when all splice variants are assessed (Fig 3, Fig 8). It is not inconceivable that Zdhhc9 KO mice have a nanoscale change in the distribution of MBP, particularly of specific palmitoylated splice variants, within myelin that profoundly affects myelin ultrastructure, without grossly altering MBP distribution. However, an alternative and not mutually exclusive possibility is that aberrant palmitoylation of other

Zdhhc9 substrates accounts for, or contributes to, the abnormalities in myelin at the ultrastructural level. Addressing this issue would require a multi-pronged approach, not just to assess palmitoylation and distribution of such proteins in Zdhhc9 KO, but also to test whether they are direct Zdhhc9 substrates, in order to rule out indirect effects. We hope reviewers would agree that this is best left to a separate study. However, in our revised Discussion we now summarize what can be inferred regarding Zdhhc9-dependent effects on total and splicevariant specific distribution and levels of MBP.  

(4) Although the article mentions the association of Zdhhc9 with intellectual disabilities, it does not involve behavioral analysis of Zdhhc9 KO mice. It is recommended to supplement some behavioral experimental data to support the important role of Zdhhc9 in maintaining normal cognitive function, enhancing the clinical relevance of the article.

We appreciate this point from the reviewer. The behavior of the same ZDHHC9 KO mouse line that we used was reported in PMID: 31747610 and in PMID: 29944857. In the former study, Zdhhc9 KO mice were reported to display seizures reminiscent of phenotypes in human patients with ZDHHC9 mutation. The latter study assessed performance of Zddhc9 KO mice in several tasks that test cognitive function. Specifically the KO mice were reported to display “altered behaviour in the open-field test, elevated plus maze and acoustic startle test that is consistent with a reduced anxiety level; a reduced hang time in the hanging wire test that suggests underlying hypotonia but which may also be linked to reduced anxiety [and] deficits in the Morris water maze test of hippocampal-dependent spatial learning and memory.”. We have incorporate these findings in our revised Discussion, where we summarize how these phenotypes are common, not just to human patients with ZDHHC9 mutation, but also to other human neurodevelopmental conditions and mouse models in which ID is a common feature.

(5) For the abnormal myelination observed in Zdhhc9 KO mice, including unmyelinated large-diameter axons and excessively myelinated small-diameter axons, the article lacks indepth research and explanation on the exact mechanism and mode of action of ZDHHC9 in regulating myelination.

We share the reviewer’s interest in this point but again note that gaining definitive insights into this issue is far from trivial. Convincing evidence of a causative mechanism would require an exhaustive identification of ZDHHC9 in vivo substrates, followed by point mutation of substrate palmitoylation site(s) to determine the extent to which palmitoylation of such protein(s) phenocopies ZDHHC9 loss. Nonetheless, it is possible to break this question down and to summarize what we do and do not know. For example, our experiments in cultured OLs show that ZDHHC9 loss causes call-autonomous deficits in morphological maturation of these cells. We also know that ZDHHC9 loss results in impaired palmitoylation of MBP, a direct substrate for ZDHHC9. Moreover, loss of ZDHHC9 at Golgi outposts in OLs (a phenotype observed with several XLID-associated mutant forms of ZDHHC9, even those with no significant loss of catalytic activity) correlates with intellectual disability. Together, these findings are consistent with a model in which ZDHHC9 action at OL Golgi outposts is critical for normal myelination. However, it is yet to be determined whether the key substrates of ZDHHC9 include MBP, other palmitoyl-proteins that are key constituents of CNS myelin, or proteins whose palmitoylation is important for myelin protein trafficking and targeting. Another non-mutually exclusive possibility is that ZDHHC9 acts at Golgi outposts but indirectly, for example to drive the expression of myelin protein genes. Future experiments, including but not limited to palmitoyl-proteomics in ZDHHC9 (OL-specific) KO mice, will be needed to provide more definitive insights into this issue. We have expanded our Discussion of links between ZDHHC9 mutation and impaired myelination to summarize the above points.

(6) The function of ZDHHC9 in OL may be related to the Golgi apparatus, but its exact role in these structures is still unclear. It is suggested to discuss in more detail the role of ZDHHC9 in the Golgi apparatus in the discussion section.

We appreciate this point, which we considered as related to point (5) above. In our revised Discussion we highlight how ZDHHC9 action at Golgi outposts may involve direct palmitoylation of myelin proteins, palmitoylation of proteins that direct myelin proteins to the myelin membrane and/or activation of gene expression programs that serve to drive myelination. We further note that these possibilities are not mutually exclusive.

(7) More experimental support and in-depth research are needed on the detailed mechanism of how ZDHHC9 and Golga7 cooperatively regulate MBP palmitoylation, and how this decrease in palmitoylation level leads to myelination defects.

This is another important point – our new experiments suggest that, although some XLID mutations markedly affect ZDHHC9’s ability to palmitoylate MBP, others do not, yet all of the mutant forms fail to localize to Golgi outposts. These findings are consistent with a model in which the subcellular location at which ZDHHC9 palmitoylates MBP, and potentially other substrates, is critical for normal myelination. Interestingly, despite their marked differences in basal catalytic activity (as assessed by autopalmitoylation), wt and all XLID forms of ZDHHC9 appear to show enhanced activity (measured by both auto- and MBP palmitoylation) in the presence of ZDHHC9, suggesting that the association with Golga7 (which also localizes to Golgi outposts) is central to ZDHHC9 activity. This model is also highly consistent with the biased expression of Golga7 in OLs, compared to other CNS cell types (Fig 1E, 1F). Moreover, XLID-associated mutant forms of ZDHHC9 also show reduced protein stability and are impaired in their ability to form complexes with Golga7 (also known as Golgi Complex Protein 16kDa; GCP16; PMID: 37035671). Failure of ZDHHC9 XLID mutants to localize to Golgi outposts may thus be due to aberrant trafficking of mutant ZDHHC9 per se, but may also involve impaired association/stabilization of ZDHHC9/Golga7 complexes at these locations. Again, it is possible that either or both of these mechanisms, which are not mutually exclusive, contribute to impaired MBP palmitoylation and/or myelination deficits. We summarize these points in our revised Discussion.

eLife Assessment

This manuscript determines how PA28g, a proteasome regulator that is overexpressed in tumors, and C1QBP, a mitochondrial protein for maintaining oxidative phosphorylation that plays a role in tumor progression, interact in tumor cells to promote their growth, migration and invasion. Additional experiments and analyses that supported the theoretical models for the interaction have been performed in response to the reviews. The overall findings and conceptual framework are important and the evidence is solid. A logical extrapolation of this work is to test the C1QBP mutants using functional assays to determine whether the mutations can decrease the protein stability mediated by the interaction with PA28g.

Reviewer #2 (Public review):

Summary:

The authors tried to determine how PA28g functions in oral squamous cell carcinoma (OSCC) cells. They hypothesized it may act through metabolic reprogramming in the mitochondria.

Strengths:

They found that the genes of PA28g and C1QBP are in an overlapping interaction network after an analysis of a genome database. They also found that the two proteins interact in coimmunoprecipitation and pull-down assays using the lysate from OSCC cells with or without expression of the exogenous genes. They used truncated C1QBP proteins to map the interaction site to the N-terminal 167 residues of C1QBP protein. They observed the levels of the two proteins are positively correlated in the cells. They provided evidence for the colocalization of the two proteins in the mitochondria and the effect on mitochondrial form and function in vitro and in vivo OSCC models, and the correlation of the protein expression with the prognosis of cancer patients.

Comments on revision:

The third revision added data from two point mutations of C1QBP that would disrupt a hydrogen bond network with PA28g protein. As one would expect from the structural models obtained with AlphaFold, the interaction between the two proteins as detected by co-immunoprecipitation of cell lysate was reduced by both mutations. Therefore, the theoretical models for the interaction were supported by the experimental data. Moving forward, the home run experiments would be to test the C1QBP mutants in functional assays to determine whether the mutations can decrease the protein stability afforded by the interaction with PA28g, which in turn decrease the effect of PA28g on mitochondria and tumor cells via C1QBP. Success of these experiments will conclude this manuscript that presents a novel finding for tumor cell biology which could be a launch pad for therapeutic intervention of tumor development.

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #2 (Public review):

This manuscript determines how PA28g, a proteasome regulator that is overexpressed in tumors, and C1QBP, a mitochondrial protein for maintaining oxidative phosphorylation that plays a role in tumor progression, interact in tumor cells to promote their growth, migration and invasion. Evidence for the interaction and its impact on mitochondrial form and function was provided although it is not particularly strong.

The revised manuscript corrected mislabeled data in figures and provides more details in figure legends. Misleading sentences and typos were corrected. However, key experiments that were suggested in previous reviews were not done, such as making point mutations to disrupt the protein interactions and assess the consequence on protein stability and function. Results from these experiments are critical to determine whether the major conclusions are fully supported by the data.

The second revision of the manuscript included the proximity ligation data to support the PA28g-C1QBP interaction in cells. However, the method and data were not described in sufficient detail for readers to understand. The revision also includes the structural models of the PA28g-C1QBP complex predicted by AlphaFold. However, the method and data were not described with details for readers to understand how this structural modeling was done, what is the quality of the resulting models, and the physical nature of the protein-protein interaction such as what kind of the non-covalent interactions exist in the interface of the protein complexes. Furthermore, while the interactions mediated by the protein fragments were tested by pull-down experiments, the interactions mediated by the three residues were not tested by mutagenesis and pull-down experiments. In summary, the revision was improved, but further improvement is needed.

Thank you very much for your comments.

(1) Based on your suggestion, we predicted the possible interaction sites using AlphaFold 3 and found that mutations in amino acids 76 and 78 of C1QBP affect the interaction with PA28γ (Revised Appendix Figure 1J). Subsequently, pulldown experiment also found that after mutating the amino acids at the two aforementioned sites (T76A, G78N), C1QBP that could bind to PA28γ decreased (Revised Figure 1J). The above results confirm that PA28γ could interacts with C1QBP, in a manner dependent on the N-terminus of C1QBP. These findings are now included in the revised manuscript “In addition, we employed AlphaFold 3 to perform energy minimization and predict hydrogen bonds between the C1QBP N-terminus (amino acids 1-167) and the PA28γ protein interaction region. The results suggest that the T76 and G78 residues of C1QBP may be key contributors to the interaction. Consistently, coimmunoprecipitation analysis demonstrated that mutations at these sites (C1QBPT76A and C1QBPG78N) significantly reduced the binding ability to PA28γ (Fig. 1J and Appendix Fig. 1J)”, specifically in results section. We believe this additional validation strengthens the robustness of our findings.

(2) According to your suggestion, we have added a description of the results of PLA in the figure legend (Revised Figure 1C) and the method of PLA in the appendix file (Revised Appendix file, Part “Proximity Ligation Assay”). The revised text reads as follows: (C) PLA image of UM1 cells shows the interaction between C1QBP and PA28γ in both cytoplasm and nucleus (red fluorescence).

(3) In the light of your suggestion, we have enriched the description of AlphaFold 3 analysis in the appendix file (Revised Appendix file, Page 10-11). The revised text reads as follows:

“Prediction and Analysis of Protein Interactions

Protein Sequence Retrieval and Structure Prediction

The protein sequences of C1QBP and PA28γ were obtained from the AlphaFold Protein Structure Database. Structural predictions of the protein-protein interaction between C1QBP and PA28γ were conducted using AlphaFold 3. The plDDT (predicted local distance difference test) values were utilized to assess the confidence of the predicted models. Models with a plDDT score above 70 were considered confident, while those with a score above 90 were categorized as very high confidence. These values were annotated in the figures to indicate the reliability of the structural predictions.”

“Protein Preparation and Structure Optimization

The best-scored model for the C1QBP-PA28γ interaction predicted by AlphaFold 3 was selected for further analysis. The model was imported into MOE 2022 (Molecular Operating Environment) software for protein preparation. This process included the removal of water molecules and other heteroatoms, followed by the addition of hydrogen atoms to the structure. This step was essential for optimizing the protein’s 3D conformation and ensuring the correctness of the protonation states at physiological pH.”

“Energy Minimization and Hydrogen Bond Prediction

The protein structure was subjected to energy minimization using the Amber10: EHT (Effective Hamiltonian Theory) force field, with R-field 1: 80 settings to refine the model’s geometry. The minimization process was performed to optimize the protein’s internal energy and ensure stable conformation, followed by calculation of hydrogen bond interactions. The interaction energies and hydrogen bonds were analyzed to identify potential binding sites and stabilize the predicted protein-protein complex.”

eLife Assessment

This important study reports solid evidence for the significant role of mother-child neural synchronization and relationship quality in the development of Theory of Mind (ToM) and social cognition. The findings effectively bridge brain development with children's behavior and parenting practices, and will be of interest to researchers studying brain development and social cognition, as well as the general public.

Reviewer #1 (Public review):

The authors have undertaken a significant revision of the manuscript and addressed the vast majority of our original comments. The manuscript is significantly improved as a result and will make a nice contribution to the literature. The new framing is especially impactful.

We have a few remaining comments to improving the manuscript:

Q1: The authors clarified the multiple comparison correction appropriately, and included a comprehensive of the study limitations related to causality and SEM. We think there could be a few further improvements to the manuscript to fully address our initial comment.

Under the results section where the authors describe the use of structural equation modeling, we think that it would be helpful to readers to further emphasize that the current design doesn't allow for delineation of temporal sequences in development and do cannot reflect true mediation. These are important caveats that the readers describe beautifully in their response.

In addition to think about the mediating variables, can the authors conduct a sensitivity analysis that re-orders the IV, mediator, and DV? That way, a formal comparison can be made between model fits. It would provide an empirical basis for how to temper the discussion of these findings.

Q7: We think that this analysis (lack of significant correlations between ISS, child age, and neural maturity) and corresponding discussion by the authors would be very interesting for readers. It does not appear as though they've added this information to the text (even in a supplementary file would suffice), but I think their conclusions about the data are strengthened related to context specific neural dynamics.

Reviewer #2 (Public review):

Summary:<br /> This study investigates the impact of mother-child neural synchronization and the quality of parent-child relationships on the development of Theory of Mind (ToM) and social cognition. Utilizing a naturalistic fMRI movie-viewing paradigm, the authors analyzed inter-subject neural synchronization in mother-child dyads and explored the connections between neural maturity, parental caregiving, and social cognitive outcomes. The findings indicate age-related maturation in ToM and social pain networks, emphasizing the importance of dyadic interactions in shaping ToM performance and social skills, thereby enhancing our understanding of the environmental and intrinsic influences on social cognition.

Strengths:<br /> This research addresses a significant question in developmental neuroscience, by linking social brain development with children's behaviors and parenting. It also uses a robust methodology by incorporating neural synchrony measures, naturalistic stimuli, and a substantial sample of mother-child dyads to enhance its ecological validity. Furthermore, the SEM approach provides a nuanced understanding of the developmental pathways associated with Theory of Mind (ToM). The manuscript also addressed many concerns raised in the initial review. The adoption of the neuroconstructivist framework effectively frames neural and cognitive development as reciprocal, addressing prior concerns about causality. The justification for methodological choices, such as omitting resting-state baselines due to scanning challenges in children and using unit-weighted scoring for ToM tasks, further strengthens the study's credibility.

Weaknesses:<br /> (1) The revised introduction has improved, particularly in framing the first goal-developmental changes in ToM and SPM networks-as a "developmental anchor" for goals 2 and 3. However, given prior research on age-related changes in these networks (e.g., Richardson et al., 2018), the authors should clarify whether this goal seeks to replicate prior findings or to extend them under new contexts. Specifying how this part differs from existing work and articulating specific hypotheses would enhance the focus.<br /> (2) I still have some reservations about retaining the slightly causal term "shape" in the title. While the manuscript now carefully avoids causal claims, the title may still be interpreted as implying directionality, especially by non-specialist audiences.<br /> (3) One more question about Figure 2A and 2B: adults and children showed highly similar response curves for video frames, yet some peaks (e.g., T02, T05, T06) are identified as ToM or SPM events only in adults. Whether statistical methods account for the differences? Or whether the corresponding video frames contain subtle social cues that only adults can process?

Reviewer #3 (Public review):

Summary:<br /> The article explores the role of mother-child interactions in the development of children's social cognition, focusing on Theory of Mind (ToM) and Social Pain Matrix (SPM) networks. Using a naturalistic fMRI paradigm involving movie viewing, the study examines relationships among children's neural development, mother-child neural synchronization, and interaction quality. The authors identified a developmental pattern in these networks, showing that they become more functionally distinct with age. Additionally, they found stronger neural synchronization between child-mother pairs compared to child-stranger pairs, with this synchronization and neural maturation of the networks associated with the mother-child relationship and parenting quality.

Strengths:<br /> This is a well-written paper, and using dyadic fMRI and naturalistic stimuli enhances its ecological validity, providing valuable insights into the dynamic interplay between brain development and social interactions.

Weaknesses:<br /> The current sample size (N = 34 dyads) is a limitation, particularly given the use of SEM, which generally requires larger samples for stable results. Although the model fit appears adequate, this does not guarantee reliability with the current sample size.

eLife Assessment

This fundamental study examines whether synaptic cell adhesion molecules neuroligin 1-3 resident on astrocytes, rather than neurons, exert effects on synaptic structure and function. With compelling evidence, including rigorous validation of neuroligin deletion efficiency in astrocytes and independent confirmation using human neuron-mouse glia co-cultures, the authors report that deletion of neuroligins 1-3 specifically in astrocytes does not alter synapse formation or astrocyte morphology in the hippocampus or visual cortex. This study provides definitive evidence highlighting the specific role of neuronal neuroligins rather than their astrocytic counterparts in synaptogenesis.

Reviewer #1 (Public review):

Astrocytes are known to express neuroligins 1-3. Within neurons, these cell adhesion molecules perform important roles in synapse formation and function. Within astrocytes, a significant role for neuroligin 2 in determining excitatory synapse formation and astrocyte morphology was shown in 2017. However, there has been no assessment of what happens to synapses or astrocyte morphology when all three major forms of neuroligins within astrocytes (isoforms 1-3) are deleted using a well characterized, astrocyte specific, and inducible cre line. By using such selective mouse genetic methods, the authors here show that astrocytic neuroligin 1-3 expression in astrocytes is not consequential for synapse function or for astrocyte morphology. They reach these conclusions with careful experiments employing quantitative western blot analyses, imaging and electrophysiology. They also characterize the specificity of the cre line they used. Overall, this is a very clear and strong paper that is supported by rigorous experiments. The discussion considers the findings carefully in relation to past work. This paper is of high importance, because it now raises the fundamental question of exactly what neuroligins 1-3 are actually doing in astrocytes. In addition, it enriches our understanding of the mechanisms by which astrocytes participate in synapse formation and function. The paper is very clear, well written and well illustrated with raw and average data.

Comments on revisions:

My previous comments have been addressed. I have no additional points to make and congratulate the authors.

Reviewer #2 (Public review):

In the present manuscript, Golf et al. investigate the consequences of astrocyte-specific deletion of Neuroligin (Nlgn) family cell adhesion proteins on synapse structure and function in the brain. Decades of prior research had shown that Neuroligins mediate their effects at synapses through their role in the postsynaptic compartment of neurons and their transsynaptic interaction with presynaptic Neurexins. More recently, it was proposed for the first time that Neuroligins expressed by astrocytes can also bind to presynaptic Neurexins to regulate synaptogenesis (Stogsdill et al. 2017, Nature). However, several aspects of the model proposed by Stogsdill et al. on astrocytic Neuroligin function conflict with prior evidence on the role of Neuroligins at synapse, prompting Golf et al. to further investigate astrocytic Neuroligin function in the current study. Using postnatal conditional deletion of Nlgn1-3 specifically from astrocytes in mice, Golf et al. show that virtually no changes in the expression of synaptic proteins or in the properties of synaptic transmission at either excitatory or inhibitory synapses are observed. Moreover, no alterations in the morphology of astrocytes themselves were found. To further extend this finding, the authors additionally analyzed human neurons co-cultured with mouse glia lacking expression of Nlgn1-4. No difference in excitatory synaptic transmission was observed between neurons cultured in the presence of wildtype vs. Nlgn1-4 conditional knockout glia. The authors conclude that while Neuroligins are indeed expressed in astrocytes and are hence likely to play some role there, this role does not include any direct consequences on synaptic structure and function, in direct contrast to the model proposed by Stogsdill et al.

Overall, this is a strong study that addresses a fundamental and highly relevant question in the field of synaptic neuroscience. Neuroligins are not only key regulators of synaptic function, they have also been linked to numerous psychiatric and neurodevelopmental disorders, highlighting the need to precisely define their mechanisms of action. The authors take a wide range of approaches to convincingly demonstrate that under their experimental conditions, Nlgn1-3 are efficiently deleted from astrocytes in vivo, and that this deletion does not lead to major alterations in the levels of synaptic proteins or in synaptic transmission at excitatory or inhibitory synapses, or in the morphology of astrocytes. The authors have conducted an elegant and compelling analysis demonstrating efficient deletion of astrocytic Nlgn1-3, with deletion rates of 83-96% for Nlgn2 and Nlgn3, and 65-72% for Nlgn1. While the co-culture experiments provide additional support, they are not essential as the in vivo data on astrocytic Nlgn1-3 deletion are compelling on their own. Together, the data from this study provide compelling and important evidence that, whatever the role of astrocytic Neuroligins may be, they do not contribute substantially to synapse formation or function under the conditions investigated.

Comments on revisions:

All of my concerns have been satisfactorily addressed.<br /> The authors have fully addressed my concerns, and have in particular conducted a very elegant and compelling analysis of the degree of deletion of astrocytic Nlgn1-3/4 in their models. This greatly strengthens the main claims of their study and the fundamental nature of their conclusions for the field of synapse biology.<br /> Regarding the co-culture experiments, while I was initially concerned about the lack of controls demonstrating that glia affect synapse formation in human neurons, the authors have appropriately addressed this by clarifying the missing references and explaining that their culture system has been extensively validated in previous studies. Since the data on astrocytic Nlgn1-3 deletion in vivo are compelling on their own, the co-culture experiment provides useful additional support for the main conclusions.<br /> The authors have also added the mouse strain background information to the methods section as requested, which is important for interpreting potential differences with other studies.

Author response:

The following is the authors’ response to the previous reviews

Public Reviews:

Reviewer #1 (Public Review):

Astrocytes are known to express neuroligins 1-3. Within neurons, these cell adhesion molecules perform important roles in synapse formation and function. Within astrocytes, a significant role for neuroligin 2 in determining excitatory synapse formation and astrocyte morphology was shown in 2017. However, there has been no assessment of what happens to synapses or astrocyte morphology when all three major forms of neuroligins within astrocytes (isoforms 1-3) are deleted using a well characterized, astrocyte specific, and inducible cre line. By using such selective mouse genetic methods, the authors here show that astrocytic neuroligin 1-3 expression in astrocytes is not consequential for synapse function or for astrocyte morphology. They reach these conclusions with careful experiments employing quantitative western blot analyses, imaging and electrophysiology. They also characterize the specificity of the cre line they used. Overall, this is a very clear and strong paper that is supported by rigorous experiments. The discussion considers the findings carefully in relation to past work. This paper is of high importance, because it now raises the fundamental question of exactly what neuroligins 1-3 are actually doing in astrocytes. In addition, it enriches our understanding of the mechanisms by which astrocytes participate in synapse formation and function. The paper is very clear, well written and well illustrated with raw and average data.

Comments on revisions:

My previous comments have been addressed. I have no additional points to make and congratulate the authors.

Thank you for your acceptance.

Reviewer #2 (Public Review):

In the present manuscript, Golf et al. investigate the consequences of astrocyte-specific deletion of Neuroligin (Nlgn) family cell adhesion proteins on synapse structure and function in the brain. Decades of prior research had shown that Neuroligins mediate their effects at synapses through their role in the postsynaptic compartment of neurons and their transsynaptic interaction with presynaptic Neurexins. More recently, it was proposed for the first time that Neuroligins expressed by astrocytes can also bind to presynaptic Neurexins to regulate synaptogenesis (Stogsdill et al. 2017, Nature). However, several aspects of the model proposed by Stogsdill et al. on astrocytic Neuroligin function conflict with prior evidence on the role of Neuroligins at synapses, prompting Golf et al. to further investigate astrocytic Neuroligin function in the current study. Using postnatal conditional deletion of Nlgn1-3 specifically from astrocytes in mice, Golf et al. show that virtually no changes in the expression of synaptic proteins or in the properties of synaptic transmission at either excitatory or inhibitory synapses are observed. Moreover, no alterations in the morphology of astrocytes themselves were found. To further extend this finding, the authors additionally analyzed human neurons co-cultured with mouse glia lacking expression of Nlgn1-4. No difference in excitatory synaptic transmission was observed between neurons cultured in the present of wildtype vs. Nlgn1-4 conditional knockout glia. The authors conclude that while Neuroligins are indeed expressed in astrocytes and are hence likely to play some role there, this role does not include any direct consequences on synaptic structure and function, in direct contrast to the model proposed by Stogsdill et al.

Overall, this is a strong study that addresses a fundamental and highly relevant question in the field of synaptic neuroscience. Neuroligins are not only key regulators of synaptic function, they have also been linked to numerous psychiatric and neurodevelopmental disorders, highlighting the need to precisely define their mechanisms of action. The authors take a wide range of approaches to convincingly demonstrate that under their experimental conditions, Nlgn1-3 are efficiently deleted from astrocytes in vivo, and that this deletion does not lead to major alterations in the levels of synaptic proteins or in synaptic transmission at excitatory or inhibitory synapses, or in the morphology of astrocytes. While the co-culture experiments are somewhat more difficult to interpret due to lack of a control for the effect of wildtype mouse astrocytes on human neurons, they are also consistent with the notion that deletion of Nlgn1-4 from astrocytes has no consequences for the function of excitatory synapses. Together, the data from this study provide compelling and important evidence that, whatever the role of astrocytic Neuroligins may be, they do not contribute substantially to synapse formation or function under the conditions investigated.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

The authors have fully addressed my concerns, and have in particular conducted a very elegant and compelling analysis of the degree of deletion of astrocytic Nlgn1-3/4 in their models. This greatly strengthens the main claims of their study and the fundamental nature of their conclusions for the field of synapse biology.

I am somewhat less convinced by the newly added experiment to investigate deletion of Nlgns1-4 from glia in glia-neuron co-cultures. The authors provide no evidence to show that either WT or cKO glia have any effect on synapse formation or function in human neurons, and therefore, the current lack of a difference could equally result from the fact that both WT and cKO glia were non-functional altogether. The authors cite two studies to state that human neurons do not form synapses in the absence of astrocytes, Zhang et al. 2013 and Huang et al. 2017, but neither seem to be listed in the references (unless Zhang et al. 2014 was meant), making it difficult to assess the relevance of these data. However, since the data on astrocytic Nlgn1-3 deletion in vivo are compelling on their own, I do not see the co-culture experiment as essential for the main conclusions of the study.

Minor comment:

Please add the information on the strain background of the mice to the methods section of the manuscript. Strain background can have a significant impact on many aspects of neuronal function, and this information is therefore essential for the interpretation of potential differences to other studies.

We deeply apologize for forgetting to include the two important references mentioned by the reviewer in the reference list. We understand that the reviewer as a result could not assess the validity of our statement that co-culture of glia is required for efficient synapse formation by human neurons that are induced from ES or iPS cells. Note that this conclusion does not postulate that all synapse formation requires glia, since the cited papers demonstrate that human neurons induced by our protocol still form scarce synapses without glia. This observation has been confirmed in many different experiments that were performed after the data presented in the cited papers. As a result of this extensive prior documentation that human neurons produced by forced expression of Ngn2 require coculture of glia for efficient synapse formation, we do not feel that we need to repeat this basic characterization of our culture system again to validate multiple previous papers and hope the reviewer will concur. We have additionally added the relevant mouse strain information to the methods section.

eLife Assessment

This fundamental work extends our understanding of the role of TGFβ2 as a modulator of mechanosensing in the eye and identifies the TRPV4 ion channel as a common regulator of Trabecular Meshwork (TM) contractility and pathological OHT and the data and evidence provided are convincing. This work will clearly be of interest to researchers investigating the role of mechanosensors in the TM and may underpin future research into treatments that aim to lower intra ocular pressure. This work will additionally be of interest to the growing field of researchers investigating the regulation of force sensing via ion channels and their roles in health and disease, in particular the ion channel TRPV4.

Reviewer #1 (public review):

Summary:

This comprehensive study employed molecular, optical, electrophysiological and tonometric strategies to establish the role of TGFβ2 in transcription and functional expression of mechanosensitive channel isoforms alongside studies of TM contractility in biomimetic hydrogels, and intraocular pressure regulation in a mouse model of TGFβ2 -induced ocular hypertension. TGFβ2 upregulated expression of TRPV4 and PIEZO1 transcripts and time-dependently augmented functional TRPV4 activation. TRPV4 activation induced TM contractility whereas pharmacological inhibition suppressed TGFβ2-induced hypercontractility and abrogated ocular hypertension in eyes overexpressing TGFβ2. Trpv4-/- mice resisted TGFβ2-driven increases in IOP. These data establish a fundamental role of TGFβ as a modulator of mechanosensing and identifies TRPV4 channel as a common mechanism for TM contractility and pathological ocular hypertension.

The manuscript is very well written and details the important function of TRPV4 in TM cell function. These data provide novel therapeutic targets and potential for disease-altering therapeutics.

Reviewer #2 (public review):

The manuscript by Christopher N. Rudzitis et al. describes the role of TGFβ2 in the transcription and functional expression of mechanosensitive channel isoforms, alongside studies on TM contractility in biomimetic hydrogels and intraocular pressure. Overall, it is a very interesting study, nicely designed, and will contribute to the available literature on TRPV4 sensitivity to mechanical forces.

eLife Assessment

This important study presents a pipeline for discovering cooperative transcription factor (TF) interactions that regulate development, and applies this pipeline in a systematic investigation of TF co-regulation in 11 human embryonic tissues. The authors provide overall solid bioinformatics and experimental support for their findings: although they make a convincing argument for the role of TEAD factors as co-repressors of regulatory activity with tissue-specific TFs, other aspects of the study would benefit from additional validation. This work would be of interest to cell biologists focused on development or on discovery of TF relationships.

Reviewer #1 (Public review):

Summary:

In this manuscript, the authors present a pipeline for the identification of transcription factor (TF) co-occurrence in regulatory regions. This pipeline aims to generate a catalogue of combinations of TFs working together, and the authors apply this during human embryonic development. In particular, they identified co-occurrences of TFs starting from H3K27ac ChIP-seq and RNA-seq input data to select active enhancers and transcribed TFs. The pipeline is applied to explore TF motifs co-occurrence at tissue-specific developmental enhancers across 11 human embryonic tissues. The application of the pipeline suggests the presence of regulatory patterns in different human developmental tissue-specific enhancers in association with ubiquitous TFs. The authors further explore the role of TEAD1 (an ubiquitously expressed TF) as a repressor. They test the role of TEAD1 as a co-repressor using a luciferase assay and tissue-specific enhancers, either alone or combined with a YAP coactivator. Overall, this paper presents an important aspect in mammalian gene regulation, the cooperative binding of TFs, and provides an important resource for TF pairs.

Strengths:

I appreciated the number of datasets analysed and the validation of a subset of enhancers.

Weaknesses:

Not many, but probably validation at more enhancers could have made the paper stronger.

Reviewer #2 (Public review):

Summary:

Garcia-Mora et al. presented a two-step bioinformatics pipeline using H3K27ac ChIP-seq and RNA-seq data from 11 human embryonic tissues published by the same groups of senior authors. "First Search" identifies motifs for TFs that are both tissue-restricted in expression and enriched in tissue-specific enhancers. "Second Search" then looks for additional motifs that co-occur near each "First Search" motif. The authors here went further than previous motif co-occurrence/co-enrichment analyses by identifying TEAD motifs as (1) representing a ubiquitously expressed family and (2) showing high co-occurrence with tissue-specific motifs at tissue-specific enhancers. They then elaborate on this finding and speculate that "TEAD, in concert with cardiac-restricted transcriptional regulators, may contribute to the recruitment of CHD4 and may play a role in attenuating the activity of enhancers involved in cardiomyocyte differentiation." They also discussed validation experiments using the luciferase assay.

Strengths:

The manuscript is well-written and easy to follow for the most part.

Weaknesses:

My main concerns and criticisms are about the sensitivity of the method and the validation of experiment designs and conclusions. Some examples where validation could be improved are as follows:

(1) The authors propose a mechanism of a TF trio (TEAD - CHD4 - tissue-specific TFs). However, only one validation experiment checked CHD4. CHD4 binding was not mentioned at all in the other cases.

(2) The authors integrated E12.5 TEAD binding with E11.5 acetylation data, and it would be important to show that this experimental approach is valid or otherwise qualify its limitations.

(3) Motif co-occurrence analysis was extended to claiming TF interactions without further validation.

Reviewer #3 (Public review):

Summary:

Mora et al employ published ChIP-seq and RNA-seq from embryonic tissues to nominate transcription factors that work combinatorially during development. This manuscript addresses an important gap in knowledge regarding the complexities of gene regulation. However, as written, the manuscript is focused on confirming mostly known associations and does not unveil principles that can be broadly applied, given multiple technical caveats that are outlined below.

Strengths:

(1) Instead of focusing on a single transcription factor motif enriched within peaks, the authors search the flanking regions of enriched motifs to nominate additional transcription factors that may work cooperatively to provide organ specificity. This type of analysis is a crucial next step in the gene regulation field, as transcription factors rarely work independently.

(2) Figure 6 is a good demonstration of the preliminary experiments that can be done to test the activity of co-occurring motifs.

(3) This is a really nice resource of organ-specific motif associations that can be used to generate many testable hypotheses.

(4) The rationale and writing are very clear and easy to read.

Weaknesses:

(1) Much of this manuscript focuses on confirming transcription factor relationships that have been reported previously. For example, it is well known that GATA4 interacts with MEF2 in the ventricle. There are limited new or unexpected associations discussed and tested.

(2) Embryonic tissues are highly heterogeneous, limiting the utility of the bulk ChIP-seq employed in these analyses. Does the cellular heterogeneity explain the discrepancy between TEAD binding and histone acetylation? Similarly, how does conservation between species affect the TF predictions?

(3) Some of the interpretations should also be fleshed out a bit more to clarify the advantage of the analyses presented here. For example, if Gata4 and Foxa2 transcripts are expressed during different stages of development, then it's likely that (as stated by the authors) these motifs are not used during the same stage of development. But examining the flanking regions wasn't necessary to make that statement. This type of conclusion seems tangential to the benefit of this analysis, which is to understand which TFs work together in a single organ at a single time point.

(4) This manuscript hinges on luciferase assays whose results can be difficult to translate to complex gene regulation networks. Many motifs are often clustered together, which makes designing experiments at endogenous loci important in studies such as this one.

Author response:

Reviewer #1:

Point 1

Not many weaknesses, but probably validation at more enhancers could have made the paper stronger.

We experimentally validated two sets of enhancers from two distinct tissues and observed similar effects. While this supports the idea that the TEAD-tissue-specific TF interaction we observe is not restricted to a single tissue, we agree that testing additional enhancers from a third tissue would strengthen our conclusions. We will acknowledge in the discussion that including a third tissue could provide additional support for the generality of our findings.

Reviewer #2:

Point 1

The authors propose a mechanism of a TF trio (TEAD - CHD4 - tissue-specific TFs). However, only one validation experiment checked CHD4. CHD4 binding was not mentioned at all in the other cases.

Indeed, CHD4 binding was experimentally validated at only one enhancer. This was a deliberate decision based on two key considerations:

(1) Consistent functional response across enhancers: We tested multiple enhancers (n =8) for functional response to the TEAD+YAP and GATA4/6 combination. All enhancers tested exhibited the same trend—attenuation of GATA-mediated activation upon co-expression of TEAD or TEAD/YAP. This consistent pattern supports a shared mechanism across these elements.

(2) Substantial prior evidence supporting CHD4 recruitment by both GATA4 and YAP: Specifically, CHD4 recruitment by GATA4 has been described in the context of cardiovascular development[1], and CHD4 can also be recruited by TEAD coactivator YAP2. Furthermore, published genomic occupancy data from embryonic heart tissue show widespread co-binding of GATA4, TEAD, and CHD4[1,3], including at most of the cardiac enhancers we functionally tested (4 out of 5).

Given the consistent enhancer responses and the supporting literature and genomic data indicating TEAD-CHD4 co-occupancy, we chose to validate CHD4 binding at a representative enhancer as a proof of concept.

We will clarify this rationale in the revised manuscript to better address this concern.

Reviewer #2:

Point 2

The authors integrated E12.5 TEAD binding with E11.5 acetylation data, and it would be important to show that this experimental approach is valid or otherwise qualify its limitations.

We will provide additional evidence in support of this approach in the revised manuscript or alternatively acknowledge its limitations.

Reviewer #2:

Point 3

Motif co-occurrence analysis was extended to claiming TF interactions without further validation.

We thank the reviewer for pointing out this important distinction. We reviewed the manuscript and identified seven instances where TF interactions were mentioned. Four of these correctly refer to previously established protein-protein interactions. For the remaining instances, we will adjust the wording to reflect the level of evidence, e.g.  describe combinatorial binding based on motif co-occurrence, rather than implying direct interaction.

Reviewer #3:

Point 1

Much of this manuscript focuses on confirming transcription factor relationships that have been reported previously. For example, it is well known that GATA4 interacts with MEF2 in the ventricle. There are limited new or unexpected associations discussed and tested.

We thank the reviewer for this important observation and see the recurrence of known interactions, such as GATA4-MEF2, not as a drawback, but as an important validation of our methodology.

The identification of novel TF-TF combinations was geared toward uncovering shared regulatory principles across diverse human developmental tissues. While analysing 13 heterogeneous embryonic tissues introduced limitations, such as cellular complexity that may obscure rare interactions, it also allowed the identification of robust, recurrent patterns across tissues.  Indeed, using this approach, we identified the widespread combinatorial effect of TEAD in partnership with lineage-specific TFs, which is explored more in depth in the manuscript.

Another main goal of the study was to develop and demonstrate a generalizable strategy for identifying combinatorial TF binding patterns that underlie tissue-specific gene regulation. Given the inherent heterogeneity of the embryonic organs analysed, the approach is naturally biased toward recovering the most prevalent, and often well-characterized, TF combinations. While we fully acknowledge this limitation, we believe that the ability to robustly recover well-established TF partnerships across multiple organs provides a valuable proof of concept. The next step will be to apply this strategy to single-cell RNA datasets, in order to define TF relationships at higher resolution, for example, resolving associations down to specific family members that cooperate within distinct lineages or cell types, and identifying less frequent or underrepresented TF-TF relationships.

In this context, we believe that our strategy has successfully highlighted shared enhancer logic and offers a framework for future high-resolution dissection of TF cooperativity at the single-cell level. The rationale for analysing heterogeneous tissues, along with its limitations, will be addressed in the revised version.

Reviewer #3:

Point 2

Embryonic tissues are highly heterogeneous, limiting the utility of the bulk ChIP-seq employed in these analyses. Does the cellular heterogeneity explain the discrepancy between TEAD binding and histone acetylation? Similarly, how does conservation between species affect the TF predictions?

We thank the reviewer for raising these important points. We acknowledge the limitations of using bulk ChIP-seq data in the context of complex embryonic tissues (see also previous point). We cannot exclude that the discrepancy between TEAD binding and histone acetylation is an effect of cellular heterogeneity. Indeed, we mention in the results “Our ventricle-specific enhancers were sampled at a single time point and likely represent enhancers that are selectively active in different cell types and developmental stages, given the heterogeneity of cell types in the ventricle”. The limitation of bulk ChIP-seq will be addressed in the discussion. In the specific case of the enhancers selected for validation, the binding site sequences are conserved between species, suggesting that the cis-regulatory activity is likely to be similar in both.

Reviewer #3:

Point 3

Some of the interpretations should also be fleshed out a bit more to clarify the advantage of the analyses presented here. For example, if Gata4 and Foxa2 transcripts are expressed during different stages of development, then it's likely that (as stated by the authors) these motifs are not used during the same stage of development. But examining the flanking regions wasn't necessary to make that statement. This type of conclusion seems tangential to the benefit of this analysis, which is to understand which TFs work together in a single organ at a single time point.

We appreciate the reviewer’s comment and the opportunity to clarify our interpretation. The reviewer refers to the finding that GATA4 and FOXA2 motifs are flanked by different sets of motifs in liver enhancers, suggesting that these TFs operate within distinct regulatory contexts.

Our aim was not to state that GATA4 and FOXA2 do not function simultaneously—this can indeed be inferred from their non-overlapping expression patterns. Rather, we intended to highlight the potential of our approach, even when applied to bulk data, to resolve distinct regulatory modules that may act in different subpopulations of cells or developmental windows within the same tissue.

We will revise the relevant section of the manuscript to make this interpretative point clearer.

Reviewer #3:

Point 4

This manuscript hinges on luciferase assays whose results can be difficult to translate to complex gene regulation networks. Many motifs are often clustered together, which makes designing experiments at endogenous loci important in studies such as this one.

We agree with the Reviewer that luciferase assays represent an oversimplified model of gene regulation and do not fully capture the complexity of endogenous regulatory networks. We will explicitly acknowledge this limitation in the discussion.

Mutagenesis of TEAD and tissue-specific TF motifs at endogenous loci would provide more conclusive evidence. However, our goal was to test the generality of TEAD effect across multiple enhancers and tissues. Despite its limitations, a luciferase-based assay was the most feasible approach, as an endogenous strategy would not have allowed us to assess a broader set of enhancers efficiently. Additionally, the presence of recurrent motifs and the potential functional redundancy among enhancers targeting the same gene can complicate the interpretation of single-locus perturbations.

References

(1) Robbe ZL, Shi W, Wasson LK, Scialdone AP, Wilczewski CM, Sheng X, et al. CHD4 is recruited by GATA4 and NKX2-5 to repress noncardiac gene programs in the developing heart. Genes Dev. 2022 Apr 1;36(7–8):468–82.

(2) Kim M, Kim T, Johnson RL, Lim DS. Transcriptional Co-repressor Function of the Hippo Pathway Transducers YAP and TAZ. Cell Rep. 2015 Apr;11(2):270–82.

(3) Akerberg BN, Gu F, VanDusen NJ, Zhang X, Dong R, Li K, et al. A reference map of murine cardiac transcription factor chromatin occupancy identifies dynamic and conserved enhancers. Nat Commun. 2019 Oct 28;10(1):4907.

eLife Assessment

There is a growing interest in understanding the individuality of animal behaviours. In this article, the authors build and use an impressive array of high throughput phenotyping paradigms to examine the 'stability' (consistency) of behavioural characteristics in a range of contexts and over time. They find that certain behaviours are individualistic and persist robustly across external stimuli while others are less robust to these changing parameters. The data, while extensive are incompletely analysed/explained. With more appropriate statistical methods adopted, the findings would have important implications for the study of individual variability.

Reviewer #1 (Public review):

Summary:

The authors state the study's goal clearly: "The goal of our study was to understand to what extent animal individuality is influenced by situational changes in the environment, i.e., how much of an animal's individuality remains after one or more environmental features change." They use visually guided behavioral features to examine the extent of correlation over time and in a variety of contexts. They develop new behavioral instrumentation and software to measure behavior in Buridan's paradigm (and variations thereof), the Y-maze, and a flight simulator. Using these assays, they examine the correlations between conditions for a panel of locomotion parameters. They propose that inter-assay correlations will determine the persistence of locomotion individuality.

Strengths:

The OED defines individuality as "the sum of the attributes which distinguish a person or thing from others of the same kind," a definition mirrored by other dictionaries and the scientific literature on the topic. The concept of behavioral individuality can be characterized as: (1) a large set of behavioral attributes, (2) with inter-individual variability, that are (3) stable over time. A previous study examined walking parameters in Buridan's paradigm, finding that several parameters were variable between individuals, and that these showed stability over separate days and up to 4 weeks (DOI: 10.1126/science.aaw718). The present study replicates some of those findings and extends the experiments from temporal stability to examining correlation of locomotion features between different contexts.

The major strength of the study is using a range of different behavioral assays to examine the correlations of several different behavior parameters. It shows clearly that the inter-individual variability of some parameters is at least partially preserved between some contexts, and not preserved between others. The development of high-throughput behavior assays and sharing the information on how to make the assays is a commendable contribution.

Weaknesses:

The definition of individuality considers a comprehensive or large set of attributes, but the authors consider only a handful. In Supplemental Fig. S8, the authors show a large correlation matrix of many behavioral parameters, but these are illegible and are only mentioned briefly in Results. Why were five or so parameters selected from the full set? How were these selected? Do the correlation trends hold true across all parameters? For assays in which only a subset of parameters can be directly compared, were all of these included in the analysis, or only a subset?

The correlation analysis is used to establish stability between assays. For temporal re-testing, "stability" is certainly the appropriate word, but between contexts it implies that there could be 'instability'. Rather, instead of the 'instability' of a single brain process, a different behavior in a different context could arise from engaging largely (or entirely?) distinct context-dependent internal processes, and have nothing to do with process stability per se. For inter-context similarities, perhaps a better word would be "consistency".

The parameters are considered one-by-one, not in aggregate. This focuses on the stability/consistency of the variability of a single parameter at a time, rather than holistic individuality. It would appear that an appropriate measure of individuality stability (or individuality consistency) that accounts for the high-dimensional nature of individuality would somehow summarize correlations across all parameters. Why was a multivariate approach (e.g. multiple regression/correlation) not used? Treating the data with a multivariate or averaged approach would allow the authors to directly address 'individuality stability', along with the analyses of single-parameter variability stability.

The correlation coefficients are sometimes quite low, though highly significant, and are deemed to indicate stability. For example, in Figure 4C top left, the % of time walked at 23{degree sign}C and 32{degree sign}C are correlated by 0.263, which corresponds to an R2 of 0.069 i.e. just 7% of the 32{degree sign}C variance is predictable by the 23{degree sign}C variance. Is it fair to say that 7% determination indicates parameter stability? Another example: "Vector strength was the most correlated attention parameter... correlations ranged... to -0.197," which implies that 96% (1 - R2) of Y-maze variance is not predicted by Buridan variance. At what level does an r value not represent stability?

The authors describe a dissociation between inter-group differences and inter-individual variation stability, i.e. sometimes large mean differences between contexts, but significant correlation between individual test and retest data. Given that correlation is sensitive to slope, this might be expected to underestimate the variability stability (or consistency). Is there a way to adjust for the group differences before examining correlation? For example, would it be possible to transform the values to in-group ranks prior to correlation analysis?

What is gained by classifying the five parameters into exploration, attention, and anxiety? To what extent have these classifications been validated, both in general, and with regard to these specific parameters? Is increased walking speed at higher temperature necessarily due to increased 'explorative' nature, or could it be attributed to increased metabolism, dehydration stress, or a heat-pain response? To what extent are these categories subjective?

The legends are quite brief and do not link to descriptions of specific experiments. For example, Figure 4a depicts a graphical overview of the procedure, but I could not find a detailed description of this experiment's protocol.

Using the current single-correlation analysis approach, the aims would benefit from re-wording to appropriately address single-parameter variability stability/consistency (as distinct from holistic individuality). Alternatively, the analysis could be adjusted to address the multivariate nature of individuality, so that the claims and the analysis are in concordance with each other.

The study presents a bounty of new technology to study visually guided behaviors. The Github link to the software was not available. To verify successful transfer or open-hardware and open-software, a report would demonstrate transfer by collaboration with one or more other laboratories, which the present manuscript does not appear to do. Nevertheless, making the technology available to readers is commendable.<br /> The study discusses a number of interesting, stimulating ideas about inter-individual variability and presents intriguing data that speaks to those ideas, albeit with the issues outlined above.

While the current work does not present any mechanistic analysis of inter-individual variability, the implementation of high-throughput assays sets up the field to more systematically investigate fly visual behaviors, their variability, and their underlying mechanisms.

Comments on revisions:

I want to express my appreciation for the authors' responsiveness to the reviewer feedback. They appear to have addressed my previous concerns through various modifications including GLM analysis, however, some areas still require clarification for the benefit of an audience that includes geneticists.

(1) GLM Analysis Explanation (Figure 9)<br /> While the authors state that their new GLM results support their original conclusions, the explanation of these results in the text is insufficient. Specifically:

- The interpretation of coefficients and their statistical significance needs more detailed explanation. The audience includes geneticists and other non-statistical people, so the GLM should be explained in terms of the criteria or quantities used to assess how well the results conform with the hypothesis, and to what extent they diverge.<br /> - The criteria used to judge how well the GLM results support their hypothesis are not clearly stated.<br /> - The relationship between the GLM findings and their original correlation-based conclusions needs better integration and connection, leading the reader through your reasoning.

(2) Documentation of Changes<br /> One struggle with the revised manuscript is that no "tracked changes" version was included, so it is hard to know exactly what was done. Without access to the previous version of the manuscript, it is difficult to fully assess the extent of revisions made. The authors should provide a more comprehensive summary of the specific changes implemented, particularly regarding:

(3) Statistical Method Selection<br /> The authors mention using "ridge regression to mitigate collinearity among predictors" but do not adequately justify this choice over other approaches. They should explain:

- Why ridge regression was selected as the optimal method<br /> - How the regularization parameter (λ) was determined<br /> - How this choice affects the interpretation of environmental parameters' influence on individuality

Reviewer #2 (Public review):

Summary:

The authors repeatedly measured the behavior of individual flies across several environmental situations in custom-made behavioral phenotyping rigs.

Strengths:

The study uses several different behavioral phenotyping devices to quantify individual behavior in a number of different situations and over time. It seems to be a very impressive amount of data. The authors also make all their behavioral phenotyping rig design and tracking software available, which I think is great, and I'm sure other folks will be interested in using and adapting to their own needs.

Weaknesses/Limitations:

I think an important limitation is that while the authors measured the flies under different environmental scenarios (i.e. with different lighting, temperature) they didn't really alter the "context" of the environment. At least within behavioral ecology, context would refer to the potential functionality of the expressed behaviors so for example, an anti-predator context, or a mating context, or foraging. Here, the authors seem to really just be measuring aspects of locomotion under benign (relatively low risk perception) contexts. This is not a flaw of the study, but rather a limitation to how strongly the authors can really say that this demonstrates that individuality is generalized across many different contexts. It's quite possible that rank-order of locomotor (or other) behaviors may shift when the flies are in a mating or risky context.

I think the authors are missing an opportunity to use much more robust statistical methods It appears as though the authors used pearson correlations across time/situations to estimate individual variation; however far more sophisticated and elegant methods exist. The problem is that pearson correlation coefficients can be anti-conservative and additionally, the authors have thus had to perform many many tests to correlate behaviors across the different trials/scenarios. I don't see any evidence that the authors are controlling for multiple testing which I think would also help. Alternatively, though, the paper would be a lot stronger, and my guess is, much more streamlined if the authors employ hierarchical mixed models to analyse these data, which are the standard analytical tools in the study of individual behavioral variation. In this way, the authors could partition the behavioral variance into its among- and within-individual components and quantify repeatability of different behaviors across trials/scenarios simultaneously. This would remove the need to estimate 3 different correlations for day 1 & day 2, day 1 & 3, day 2 & 3 (or stripe 0 & stripe 1, etc) and instead just report a single repeatability for e.g. the time spent walking among the different strip patterns (eg. figure 3). Additionally, the authors could then use multivariate models where the response variables are all the behaviors combined and the authors could estimate the among-individual covariance in these behaviors. I see that the authors state they include generalized linear mixed models in their updated MS, but I struggled a bit to understand exactly how these models were fit? What exactly was the response? what exactly were the predictors (I just don't understand what Line404 means "a GLM was trained using the environmental parameters as predictors (0 when the parameter was not changed, 1 if it was) and the resulting individual rank differences as the response"). So were different models run for each scenario? for different behaviors? Across scenarios? What exactly? I just harp on this because I'm actually really interested in these data and think that updating these methods can really help clarify the results and make the main messages much clearer!

I appreciate that the authors now included their sample sizes in the main body of text (as opposed to the supplement) but I think that it would still help if the authors included a brief overview of their design at the start of the methods. It is still unclear to me how many rigs each individual fly was run through? Were the same individuals measured in multiple different rigs/scenarios? Or just one?

I really think a variance partitioning modeling framework could certainly improve their statistical inference and likely highlight some other cool patterns as these methods could better estimate stability and covariance in individual intercepts (and potentially slopes) across time and situation. I also genuinely think that this will improve the impact and reach of this paper as they'll be using methods that are standard in the study of individual behavioral variation

Reviewer #3 (Public review):

This manuscript is a continuation of past work by the last author where they looked at stochasticity in developmental processes leading to inter-individual behavioural differences. In that work, the focus was on a specific behaviour under specific conditions while probing the neural basis of the variability. In this work, the authors set out to describe in detail how stable individuality of animal behaviours is in the context of various external and internal influences. They identify a few behaviours to monitor (read outs of attention, exploration, and 'anxiety'); some external stimuli (temperature, contrast, nature of visual cues, and spatial environment); and two internal states (walking and flying).

They then use high-throughput behavioural arenas - most of which they have built and made plans available for others to replicate - to quantify and compare combinations of these behaviours, stimuli, and internal states. This detailed analysis reveals that:

(1) Many individualistic behaviours remain stable over the course of many days.<br /> (2) That some of these (walking speed) remain stable over changing visual cues. Others (walking speed and centrophobicity) remain stable at different temperatures.<br /> (3) All the behaviours they tested fail to remain stable over spatially varying environment (arena shape).<br /> (4) and only angular velocity (a read out of attention) remains stable across varying internal states (walking and flying)

Thus, the authors conclude that there is a hierarchy in the influence of external stimuli and internal states on the stability of individual behaviours.

The manuscript is a technical feat with the authors having built many new high-throughput assays. The number of animals are large and many variables have been tested - different types of behavioural paradigms, flying vs walking, varying visual stimuli, different temperature among others.

Comments on revisions:'

The authors have addressed my previous concerns.

Author response:

The following is the authors’ response to the original reviews

Public Reviews: 

Reviewer #1 (Public Review): 

Summary: 

The authors state the study's goal clearly: "The goal of our study was to understand to what extent animal individuality is influenced by situational changes in the environment, i.e., how much of an animal's individuality remains after one or more environmental features change." They use visually guided behavioral features to examine the extent of correlation over time and in a variety of contexts. They develop new behavioral instrumentation and software to measure behavior in Buridan's paradigm (and variations thereof), the Y-maze, and a flight simulator. Using these assays, they examine the correlations between conditions for a panel of locomotion parameters. They propose that inter-assay correlations will determine the persistence of locomotion individuality.

Strengths: 

The OED defines individuality as "the sum of the attributes which distinguish a person or thing from others of the same kind," a definition mirrored by other dictionaries and the scientific literature on the topic. The concept of behavioral individuality can be characterized as: 

(1) a large set of behavioral attributes, 

(2) with inter-individual variability, that are 

(3) stable over time. 

A previous study examined walking parameters in Buridan's paradigm, finding that several parameters were variable between individuals, and that these showed stability over separate days and up to 4 weeks (DOI: 10.1126/science.aaw718). The present study replicates some of those findings and extends the experiments from temporal stability to examining the correlation of locomotion features between different contexts.

The major strength of the study is using a range of different behavioral assays to examine the correlations of several different behavior parameters. It shows clearly that the inter-individual variability of some parameters is at least partially preserved between some contexts, and not preserved between others. The development of highthroughput behavior assays and sharing the information on how to make the assays is a commendable contribution.

We thank the reviewer for his exceptionally kind assessment of our work!

Weaknesses: 

The definition of individuality considers a comprehensive or large set of attributes, but the authors consider only a handful. In Supplemental Fig. S8, the authors show a large correlation matrix of many behavioral parameters, but these are illegible and are only mentioned briefly in Results. 

We have now uploaded a high-resolution PDF to the Github Address: https://github.com/LinneweberLab/Mathejczyk_2024_eLife_Individuality/blob/main/S8.pdf, and this is also mentioned in the figure legend for Fig. S8

Why were five or so parameters selected from the full set? How were these selected? 

The five parameters (% of time walked, walking speed, vector strength, angular velocity, and centrophobicity) were selected because they describe key aspects of the investigated behaviors that can be compared directly across assays. Importantly, several parameters we typically use (e.g., Linneweber et al., 2020) cannot be applied under certain conditions, such as darkness or the absence of visual cues. Furthermore, these five parameters encompass three critical aspects of navigation across standard visual behavioral arenas: (1) The “exploration” category is characterized by parameters describing the fly’s activity. (2) Parameters related to “attention” reflect heightened responses to visual cues, but unlike commonly used metrics such as angle or stripe deviations (e.g., Coulomb, 2012; Linneweber et al., 2020), they can also be measured in absence of visual cues and are therefore suitable for cross-assay comparisons. (3) The parameter “centrophobicity,” used as a potential indicator of anxiety, is conceptually linked to the open-field test in mice, where the ratio of wall-to-open-field activity is frequently calculated as a measurement of anxiety (see for example Carter, Sheh, 2015, chapter 2. https://www.sciencedirect.com/book/9780128005118/guide-to-researchtechniques-in-neuroscience). Admittedly, this view is frequently challenged in mice, but it has a long history which is why we use it.

Do the correlation trends hold true across all parameters? For assays in which only a subset of parameters can be directly compared, were all of these included in the analysis, or only a subset? 

As noted above, we only included a subset of parameters in our final analysis, as many were unsuitable for comparison across assays while still providing valuable assayspecific information which are important to relate these results to previous publications.

The correlation analysis is used to establish stability between assays. For temporal retesting, "stability" is certainly the appropriate word, but between contexts, it implies that there could be 'instability'. Rather, instead of the 'instability' of a single brain process, a different behavior in a different context could arise from engaging largely (or entirely?) distinct context-dependent internal processes, and have nothing to do with process stability per se. For inter-context similarities, perhaps a better word would be "consistency". 

Thank you for this suggestion. During the preparation of the manuscript, we indeed frequently alternated between the terms “stability” and “consistency.” And decided to go with “stability” as the only descriptor, to keep it simple. We now fully agree with the reviewer’s argument and have replaced “stability” by “consistency” throughout the current version of the manuscript in order to increase clarity and coherence.

The parameters are considered one by one, not in aggregate. This focuses on the stability/consistency of the variability of a single parameter at a time, rather than holistic individuality. It would appear that an appropriate measure of individuality stability (or individuality consistency) that accounts for the high-dimensional nature of individuality would somehow summarize correlations across all parameters. Why was a multivariate approach (e.g. multiple regression/correlation) not used? Treating the data with a multivariate or averaged approach would allow the authors to directly address 'individuality stability' and analyses of single-parameter variability stability.

We agree with the reviewer that a multivariate analysis adds clear advantages in terms of statistical power, in addition to our chosen approach. On one hand, we believe that the simplicity of our initial analysis, both for correlational and mean data, makes easy for readers to understand and reproduce our data. While preparing the previous version of the manuscript we were skeptical since more complex analyses often involve numerous choices, which can complicate reproducibility. For instance, a recent study in personality psychology (Paul et al., 2024) highlighted the risks of “forking paths” in statistical analysis, showing that certain choices of statistical methods could even reverse findings—a concern mitigated by our simplistic straightforward approach. Still, in preparation of this revised version of the manuscript, we accepted the reviewer’s advice and reanalyzed the data using a generalized linear model. This analysis nicely recapitulates our initial findings and is now summarized in a single figure (Fig. 9).

The correlation coefficients are sometimes quite low, though highly significant, and are deemed to indicate stability. For example, in Figure 4C top left, the % of time walked at 23{degree sign}C and 32{degree sign}C are correlated by 0.263, which corresponds to an R2 of 0.069 i.e. just 7% of the 32{degree sign}C variance is predictable by the 23{degree sign}C variance. Is it fair to say that a 7% determination indicates parameter stability? Another example: "Vector strength was the most correlated attention parameter... correlations ranged... to -0.197," which implies that 96% (1 - R2) of Y-maze variance is not predicted by Buridan variance. At what level does an r value not represent stability?

We agree that this is an important question. Our paper clearly demonstrates that individuality always plays a role in decision-making (and, in this context, any behavioral output can be considered a decision). However, the non-linear relationship between certain situations and the individual’s behavior often reduces the predictive value (or correlation) across contexts, sometimes quite drastically.

For instance, temperature has a relatively linear effect on certain behavioral parameters, leading to predictable changes across individuals. As a result, correlations across temperature conditions are often similar to those observed across time within the same situation. In contrast, this predictability diminishes when comparing conditions like the presence or absence of visual stimuli, the use of different arenas, or different modalities.

For this reason, we believe that significance remains the best indicator for describing how measurable individuality persists, even across vastly different situations.

The authors describe a dissociation between inter-group differences and interindividual variation stability, i.e. sometimes large mean differences between contexts, but significant correlation between individual test and retest data. Given that correlation is sensitive to slope, this might be expected to underestimate the variability stability (or consistency). Is there a way to adjust for the group differences before examining the correlation? For example, would it be possible to transform the values to in-group ranks prior to correlation analysis?  

We thank the reviewer for this suggestion, and we have now addressed this point. To account for slope effects, we have now introduced in-group ranks for our linear model computation (see Fig. 9). 

What is gained by classifying the five parameters into exploration, attention, and anxiety? To what extent have these classifications been validated, both in general and with regard to these specific parameters? Is the increased walking speed at higher temperatures necessarily due to an increased 'explorative' nature, or could it be attributed to increased metabolism, dehydration stress, or a heat-pain response? To what extent are these categories subjective?

We agree that grouping our parameters into traits like exploration, attention, and anxiety always includes subjective decisions. The classification into these three categories is even considered partially controversial in the mouse specific literature, which uses the term “anxiety” in similar experiments (see for exampler Carter, Sheh, 2015, chapter 2 . https://www.sciencedirect.com/book/9780128005118/guide-to-research-techniquesin-neuroscience). Nevertheless, we believe that readers greatly benefit from these categories, since they make it easier to understand (beyond mathematical correlations) which aspects of the flies’ individuality can be considered consistent across situations. Furthermore, these categories serve as a bridge to compare insight from very distinct models.

The legends are quite brief and do not link to descriptions of specific experiments. For example, Figure 4a depicts a graphical overview of the procedure, but I could not find a detailed description of this experiment's protocol.

We assume the reviewer is referring to Figure 3a. The detailed experimental protocol can be found in the Materials and Methods section under Setup 2: IndyTrax Multi-Arena Platform. We have now clarified this in the mentioned figure legend.

Using the current single-correlation analysis approach, the aims would benefit from rewording to appropriately address single-parameter variability stability/consistency (as distinct from holistic individuality). Alternatively, the analysis could be adjusted to address the multivariate nature of individuality, so that the claims and the analysis are in concordance with each other.

The reviewer raises an important point about hierarchies within the concept of animal individuality or personality. We agree that this is best addressed by first focusing on single behavioral traits/parameters and then integrating several trait properties into a cohesive concept of animal personality (holistic individuality). To ensure consistency throughout the text, we have now thoroughly reviewed the entire manuscript clearly distinguish between single-parameter variability stability/consistency and holistic individuality/personality.

The study presents a bounty of new technology to study visually guided behaviors. The GitHub link to the software was not available. To verify the successful transfer of open hardware and open-software, a report would demonstrate transfer by collaboration with one or more other laboratories, which the present manuscript does not appear to do. Nevertheless, making the technology available to readers is commendable.

We have now uploaded all codes and materials to GitHub and made them available as soon as we received the reviewers’ comments. All files and materials can be accessed at https://github.com/LinneweberLab/Mathejczyk_2024_eLife_Individuality, which is now frequently mentioned throughout the revised manuscript.

The study discusses a number of interesting, stimulating ideas about inter-individual variability, and presents intriguing data that speaks to those ideas, albeit with the issues outlined above.

While the current work does not present any mechanistic analysis of inter-individual variability, the implementation of high-throughput assays sets up the field to more systematically investigate fly visual behaviors, their variability, and their underlying mechanisms. 

We thank the reviewer again for the extensive and constructive feedback.

Reviewer #2 (Public Review): 

Summary: 

The authors repeatedly measured the behavior of individual flies across several environmental situations in custom-made behavioral phenotyping rigs.

Strengths: 

The study uses several different behavioral phenotyping devices to quantify individual behavior in a number of different situations and over time. It seems to be a very impressive amount of data. The authors also make all their behavioral phenotyping rig design and tracking software available, which I think is great and I'm sure other folks will be interested in using and adapting it to their own needs.

We thank the reviewer for highlighting the strengths of our study.

Weaknesses/Limitations: 

I think an important limitation is that while the authors measured the flies under different environmental scenarios (i.e. with different lighting and temperature) they didn't really alter the "context" of the environment. At least within behavioral ecology, context would refer to the potential functionality of the expressed behaviors so for example, an anti-predator context, a mating context, or foraging. Here, the authors seem to really just be measuring aspects of locomotion under benign (relatively low-risk perception) contexts. This is not a flaw of the study, but rather a limitation to how strongly the authors can really say that this demonstrates that individuality is generalized across many different contexts. It's quite possible that rank order of locomotor (or other) behaviors may shift when the flies are in a mating or risky context. 

We agree with the reviewer that the definition of environmental context can differ between fields and that behavioral context is differently defined, particularly in ecology. Nevertheless, we highlight that our alternations of environmental context are highly stereotypic, well-defined, and unbiased from any interpretation (we only modified what we stated in the experimental description without designing a specific situation that might be again perceived individually differently. E.g., comparing a context with a predator and one without might result in a binary response because one fraction of the tested individuals might perceive the predator in the predator situation, and the other half does not. 

The analytical framework in terms of statistical methods is lacking. It appears as though the authors used correlations across time/situations to estimate individual variation; however, far more sophisticated and elegant methods exist. The paper would be a lot stronger, and my guess is, much more streamlined if the authors employ hierarchical mixed models to analyse these data these models could capture and estimate differences in individual behavior across time and situations simultaneously. Along with this, it's currently unclear whether and how any statistical inference was performed. Right now, it appears as though any results describing how individuality changes across situations are largely descriptive (i.e. a visual comparison of the strengths of the correlation coefficients?). 

The reviewer raises an important point, also raised by reviewer #1. On one hand, we agree with both reviewers that a more aggregated analysis has clear advantages like more statistical power and has the potential to streamline our manuscript, which is why we added such an analysis (see below). On the other hand, we would also like to defend the initial approach we took, since we think that the simplicity of the analysis for both correlational and mean data is easy to understand and reproduce. More complex analyses necessarily include the selection of a specific statistical toolbox by the experimenters and based on these decisions, different analyses become less comparable and more and more complicated to reproduce, unless the entire decision tree is flawlessly documented. For instance, a recent personality psychology paper investigated the relationship between statistical paths within the decision tree (forking analysis) and their results, leading to very surprising results (Paul et al., 2024), since some paths even reversed their findings. Such a variance in conclusions is hardly possible with the rather simplistic and easily reproducible analysis we performed. One of the major strengths of our study is the simple experimental design, allowing for rather simple and easy to understand analyses.

We nevertheless took the reviewer’s advice very seriously and reanalyzed the data using a generalized linear model, which largely recapitulated the findings of our previously performed “low-tech” analysis in a single figure (Fig. 9).

Another pretty major weakness is that right now, I can't find any explicit mention of how many flies were used and whether they were re-used across situations. Some sort of overall schematic showing exactly how many measurements were made in which rigs and with which flies would be very beneficial. 

We apologize for this inconvenience. A detailed overview of male and female sample sizes has been listed in the supplemental boxplots next to the plots (e.g, Fig S6). Apparently, this was not visible enough. Therefore, we have now also uniformly added the sample sizes to the main figure legends.

I don't necessarily doubt the robustness of the results and my guess is that the author's interpretations would remain the same, but a more appropriate modeling framework could certainly improve their statistical inference and likely highlight some other cool patterns as these methods could better estimate stability and covariance in individual intercepts (and potentially slopes) across time and situation.

As described above, we have now added the suggested analyses. We hope that the reviewer will appreciate the new Fig. 9, which, in our opinion, largely confirms our previous findings using a more appropriate generalized linear modelling framework.

Reviewer #3 (Public Review): 

This manuscript is a continuation of past work by the last author where they looked at stochasticity in developmental processes leading to inter-individual behavioural differences. In that work, the focus was on a specific behaviour under specific conditions while probing the neural basis of the variability. In this work, the authors set out to describe in detail how stable the individuality of animal behaviours is in the context of various external and internal influences. They identify a few behaviours to monitor (read outs of attention, exploration, and 'anxiety'); some external stimuli (temperature, contrast, nature of visual cues, and spatial environment); and two internal states (walking and flying).

They then use high-throughput behavioural arenas - most of which they have built and made plans available for others to replicate - to quantify and compare combinations of these behaviours, stimuli, and internal states. This detailed analysis reveals that:

(1) Many individualistic behaviours remain stable over the course of many days. 

(2) That some of these (walking speed) remain stable over changing visual cues. Others (walking speed and centrophobicity) remain stable at different temperatures.

(3) All the behaviours they tested failed to remain stable over the spatially varying environment (arena shape).

(4) Only angular velocity (a readout of attention) remains stable across varying internal states (walking and flying).

Thus, the authors conclude that there is a hierarchy in the influence of external stimuli and internal states on the stability of individual behaviours.

The manuscript is a technical feat with the authors having built many new highthroughput assays. The number of animals is large and many variables have been tested - different types of behavioural paradigms, flying vs walking, varying visual stimuli, and different temperatures among others. 

We thank the reviewer for this extraordinary kind assessment of our work!

Recommendations for the authors:  

Reviewing Editor (Recommendations For The Authors): 

While appreciating the effort and quality of the work that went into this manuscript, the reviewers identified a few key points that would greatly benefit this work.

(1) Statistical methods adopted. The dataset produced through this work is large, with multiple conditions and comparisons that can be made to infer parameters that both define and affect the individualistic behaviour of an animal. Hierarchical mixed models would be a more appropriate approach to handle such datasets and infer statistically the influence of different parameters on behaviours. We recommend that the authors take this approach in the analyses of their data.

(2) Brevity in the text. We urge the authors to take advantage of eLife's flexible template and take care to elaborate on the text in the results section, the methods adopted, the legends, and the guides to the legends embedded in the main text. The findings are likely to be of interest to a broad audience, and the writing currently targets the specialist.

Reviewer #2 (Recommendations For The Authors): 

I want to start by saying this seems like a really cool study! It's an impressive amount of work and addressing a pretty basic question that is interesting (at least I think so!)

We thank the reviewer again for this assessment!

That said, I would really strongly recommend the authors embrace using mixed/hierarchical models to analyze their data. They're producing some really impressive data and just doing Pearson correlation coefficients across time points and situations is very clunky and actually losing out on a lot of information. The most up-todate, state-of-the-art are mixed models - these models can handle very complex (or not so complex) random structures which can estimate variance and importantly, covariance, in individual intercepts both over time and across situations. I actually think this could add some really cool insights into the data and allow you to characterize the patterns you're seeing in far more detail. It's datasets exactly like this that are tailormade for these complex variance partitioning models! 

As mentioned before, we have now adopted a more appropriate GLM-based data analysis (see above).

Regardless of which statistical methods you decide to use, please explicitly state in your methods exactly what analyses you did. That is completely lacking now and was a bit frustrating. As such, it's completely unclear whether or how statistical inference was performed. How did you do the behavioral clustering? 

We apologize that these points were not clearly represented in the previous version of the manuscript. We have now significantly extended the methods section to include a separate paragraph on the statistical methods used, in order to address this critique and hope that the revised version is clear now.

Also, I could not for the life of me figure out how many flies had been measured. Were they reused across the situation? Or not?

We reused the same flies across situations whenever possible. However, having one fly experience all assays consecutively was not feasible due to their fragility. Instead, individual flies were exposed to at least 2 of the 3 groups of assays used here: in the Indytrax setup ,  the Buridan arenas and variants thereof, and the virtual arenas Hence, we have compared flies across entirely different setups, but the number of times flies can be retested is limited (as otherwise, sample sizes will drop over time, and the flies will have gone through too many experimental alternations). To make this more clear, we have elaborated on this point in the main text, and we added group sample sizes to figure legends r.

What are these "groups" and "populations" that are referred to in the results (e.g. lines 384, 391, 409)?

We apologize for using these two terms somewhat interchangeably without proper introduction/distinction. We have now made this more clear in at the beginning of the results in the main text, by focusing on the term ‘group’. By ‘group’ we refer to the average of all individuals tested in the same situation. Sample sizes in the figure legends now indicate group/population sizes to make this clearer.

Some of the rationale for the development of the behavioral rigs would have actually been nice to include in the intro, rather than in the results.

This rationale is introduced at the beginning of the last paragraph of the introduction. We hope that this now becomes clear in the revised version of the manuscript.

Reviewer #3 (Recommendations For The Authors): 

This manuscript would do well to take advantage of eLife's flexible word limit. I sense that it has been written in brevity for a different journal but I would urge the authors to revisit this and unpack the language here - in the text, in the figure legends, in references to the figures within the text. The way it's currently written, though not misleading, will only speak to the super-specialist or the super-invested :). But the findings are nice, and it would be nice to tailor it to a broader audience.

We appreciate this suggestion. Initially, we were hoping that we had described our results as clearly and brief as possible. We apologize if that was not always the case. The comments and requests of all three reviewers now led to a series of additions to both main text and methods, leading to a significantly expanded manuscript. We hope that these additons are helpful for the general, non-expert audience.

eLife Assessment

This important study tests the development of motor reinforcement learning from toddlerhood to adulthood, using a large online sample. They show that learning improves with age in a task that, like real-life movement, involves a continuous range of response options and probabilistic rewards, and link this shift to reduced movement variability and more efficient feedback-based learning through behavioural modeling. Simplifying the task with discrete actions and deterministic outcomes boosted younger children's performance, suggesting early learning is limited by spatial and probabilistic processing. The evidence is convincing, although future work may investigate more naturalistic movement.

Reviewer #1 (Public review):

Summary:

Here the authors address how reinforcement-based sensorimotor adaptation changes throughout development. To address this question, they collected many participants in ages that ranged from small children (3 years old) to adulthood (18+ years old). The authors used four experiments to manipulate whether binary and positive reinforcement was provided probabilistically (e.g., 30 or 50%) versus deterministically (e.g.,100%), and continuous (infinite possible locations) versus discrete (binned possible locations) when the probability of reinforcement varied along the span of a large redundant target. The authors found that both movement variability and the extent of adaptation changed with age.

Strengths:

The major strength of the paper is the number of participants collected (n = 385). The authors also answer their primary question, that reinforcement-based sensorimotor adaptation changes throughout development, which was shown by utilizing established experimental designs and computational modelling. They have compared an extensive number of potential models, finding the one that best fits the data while penalizing the number of free parameters.

Reviewer #2 (Public review):

Summary:

In this study, Hill and colleagues use a novel reinforcement-based motor learning task ("RML"), asking how aspects of RML change over the course of development from toddler years through adolescence. Multiple versions of the RML task were used in different samples, which varied on two dimensions: whether the reward probability of a given hand movement direction was deterministic or probabilistic, and whether the solution space had continuous reach targets or discrete reach targets. Using analyses of both raw behavioral data and model fits, the authors report four main results: First, developmental improvements reflected 3 clear changes, including increases in exploration, an increase in the RL learning rate, and a reduction of intrinsic motor noise. Second, changes to the task that made it discrete and/or deterministic both rescued performance in the youngest age groups, suggesting that observed deficits could be linked to continuous/probabilistic learning settings. Overall, the results shed light on how RML changes throughout human development, and the modeling characterizes the specific learning deficits seen in the youngest ages.

Strengths:

(1) This impressive work addresses an understudied subfield of motor control/psychology - the developmental trajectory of motor learning. It is thus timely and will interest many researchers.

(2) The task, analysis, and modeling methods are very strong. The empirical findings are rather clear and compelling, and the analysis approaches are convincing. Thus, at the empirical level, this study has very few weaknesses.

(3) The large sample sizes and in-lab replications further reflect the laudable rigor of the study.

(4) The main and supplemental figures are clear and concise.

Reviewer #3 (Public review):

Summary:

The study investigates the development of reinforcement learning across the lifespan with a large sample of participants recruited for an online game. It finds that children gradually develop their abilities to learn reward probability, possibly hindered by their immature spatial processing and probabilistic reasoning abilities. Motor noise and exploration after a failure all contribute to children's subpar performance.  

Strengths:

Experimental manipulations of both the continuity of movement options and the probabilistic nature of the reward function enable the inference of what cognitive factors differ between age groups. <br /> A large sample of participants is studied.<br /> The model-based analysis provides further insights into the development of reinforcement learning ability. 

Weaknesses:

The conclusion that immature spatial processing and probabilistic reasoning abilities limit reinforcement learning here still needs more direct evidence.

Author response:

The following is the authors’ response to the original reviews

Overview of changes in the revision

We thank the reviewers for the very helpful comments and have extensively revised the paper. We provide point-by-point responses below and here briefly highlight the major changes:

(1) We expanded the discussion of the relevant literature in children and adults.

(2) We improved the contextualization of our experimental design within previous reinforcement studies in both cognitive and motor domains highlighting the interplay between the two.

(3) We reorganized the primary and supplementary results to better communicate the findings of the studies.

(4) The modeling has been significantly revised and extended. We now formally compare 31 noise-based models and one value-based model and this led to a different model from the original being the preferred model. This has to a large extent cleaned up the modeling results. The preferred model is a special case (with no exploration after success) of the model proposed in Therrien et al. (2018). We also provide examples of individual fits of the model, fit all four tasks and show group fits for all, examine fits vs. data for the clamp phases by age, provide measures of relative and absolute goodness of fit, and examine how the optimal level of exploration varies with motor noise.

Reviewer #1 (Public review):

Summary:

Here the authors address how reinforcement-based sensorimotor adaptation changes throughout development. To address this question, they collected many participants in ages that ranged from small children (3 years old) to adulthood (1 8+ years old). The authors used four experiments to manipulate whether binary and positive reinforcement was provided probabilistically (e.g., 30 or 50%) versus deterministically (e.g., 100%), and continuous (infinite possible locations) versus discrete (binned possible locations) when the probability of reinforcement varied along the span of a large redundant target. The authors found that both movement variability and the extent of adaptation changed with age.

Thank you for reviewing our work. One note of clarification. This work focuses on reinforcementbased learning throughout development but does not evaluate sensorimotor adaptation. The four tasks presented in this work are completed with veridical trajectory feedback (no perturbation).

The goal is to understand how children at different ages adjust their movements in response to reward feedback but does not evaluate sensorimotor adaptation. We now explain this distinction on line 35.

Strengths:

The major strength of the paper is the number of participants collected (n = 385). The authors also answer their primary question, that reinforcement-based sensorimotor adaptation changes throughout development, which was shown by utilizing established experimental designs and computational modelling.

Thank you.

Weaknesses:

Potential concerns involve inconsistent findings with secondary analyses, current assumptions that impact both interpr tation and computational modelling, and a lack of clearly stated hypotheses.

(1) Multiple regression and Mediation Analyses.

The challenge with these secondary analyses is that:

(a) The results are inconsistent between Experiments 1 and 2, and the analysis was not performed for Experiments 3 and 4,

(b) The authors used a two-stage procedure of using multiple regression to determine what variables to use for the mediation analysis, and

(c)The authors already have a trial-by-trial model that is arguably more insightful.

Given this, some suggested changes are to:

(a) Perform the mediation analysis with all the possible variables (i.e., not informed by multiple regression) to see if the results are consistent.

(b) Move the regression/mediation analysis to Supplementary, since it is slightly distracting given current inconsistencies and that the trial-by-trial model is arguably more insightful.

Based on these comments, we have chosen to remove the multiple regression and mediation analyses. We agree that they were distracting and that the trial-by-trial model allows for differentiation of motor noise from exploration variability in the learning block.

(2) Variability for different phases and model assumptions:

A nice feature of the experimental design is the use of success and failure clamps. These clamped phases, along with baseline, are useful because they can provide insights into the partitioning of motor and exploratory noise. Based on the assumptions of the model, the success clamp would only reflect variability due to motor noise (excludes variability due to exploratory noise and any variability due to updates in reach aim). Thus, it is reasonable to expect that the success clamps would have lower variability than the failure clamps (which it obviously does in Figure 6), and presumably baseline (which provides success and failure feedback, thus would contain motor noise and likely some exploratory noise).

However, in Figure 6, one visually observes greater variability during the success clamp (where it is assumed variability only comes from motor noise) compared to baseline (where variability would come from: (a) Motor noise.

(b) Likely some exploratory noise since there were some failures.

(c) Updates in reach aim.

Thanks for this comment. It made us realize that some of our terminology was unintentionally misleading. Reaching to discrete targets in the Baseline block was done to a) determine if participants could move successfully to targets that are the same width as the 100% reward zone in the continuous targets and b) determine if there are age dependent changes in movement precision. We now realize that the term Baseline Variability was misleading and should really be called Baseline Precision.

This is an important distinction that bears on this reviewer's comment. In clamp trials, participants move to continuous targets. In baseline, participants move to discrete targets presented at different locations. Clamp Variability cannot be directly compared to Baseline Precision because they are qualitatively different. Since the target changes on each baseline trial, we would not expect updating of desired reach (the target is the desired reach) and there is therefore no updating of reach based on success or failure. The SD we calculate over baseline trials is the endpoint variability of the reach locations relative to the target centers. In success clamp, there are no targets so the task is qualitatively different.

We have updated the text to clarify terminology, expand upon our operational definitions, and motivate the distinct role of the baseline block in our task paradigm (line 674).

Given the comment above, can the authors please:

(a) Statistically compare movement variability between the baseline, success clamp, and failure clamp phases.

Given our explanation in the previous point we don't think that comparing baseline to the clamp makes sense as the trials are qualitatively different.

(b) The authors have examined how their model predicts variability during success clamps and failure clamps, but can they also please show predictions for baseline (similar to that of Cashaback et al., 2019; Supplementary B, which alternatively used a no feedback baseline)?

Again, we do not think it makes sense to predict the baseline which as we mention above has discrete targets compared to the continuous targets in the learning phase.

(c) Can the authors show whether participants updated their aim towards their last successful reach during the success clamp? This would be a particularly insightful analysis of model assumptions.

We have now compared 31 models (see full details in next response) which include the 7 models in Roth et al. (2023). Several of these model variants have updating even after success with so called planning noise). We also now fit the model to the data that includes the clamp phases (we can't easily fit to success clamp alone as there are only 10 trials). We find that the preferred model is one that does not include updating after success.

(d) Different sources of movement variability have been proposed in the literature, as have different related models. One possibility is that the nervous system has knowledge of 'planned (noise)' movement variability that is always present, irrespective of success (van Beers, R.J. (2009). Motor learning is optimally tuned to the properties of motor noise. Neuron, 63(3), 406-417). The authors have used slightly different variations of their model in the past. Roth et al (2023) directly Rill compared several different plausible models with various combinations of motor, planned, and exploratory noise (Roth A, 2023, "Reinforcement-based processes actively regulate motor exploration along redundant solution manifolds." Proceedings of the Royal Society B 290: 20231475: see Supplemental). Their best-fit model seems similar to the one the authors propose here, but the current paper has the added benefit of the success and failure clamps to tease the different potential models apart. In light of the results of a), b), and c), the authors are encouraged to provide a paragraph on how their model relates to the various sources of movement variability and ther models proposed in the literature.

Thank you for this. We realized that the models presented in Roth et al. (2023) as well as in other papers, are all special cases of a more general model. Moreover, in total there are 30 possible variants of the full model so we have now fit all 31 models to our larger datasets and performed model selection (Results and Methods). All the models can be efficiently fit by Kalman smoother to the actual data (rather than to summary statistics which has sometimes been done). For model selection, we fit only the 100 learning trials and chose the preferred model based on BIC on the children's data (Figure 5—figure Supplement 1). After selecting the preferred model we then refit this model to all trials including the clamps so as to obtain the best parameter estimates.

The preferred model was the same whether we combined the continuous and discrete probabilistic data or just examin d each task separately either for only the children or for the children and adults combined. The preferred model is a pecial case (no exploration after success) of the one proposed in Therrien et al. (2018) and has exploration variability (after failure) and motor noise with full updating with exploration variability (if any) after success. This model differs from the model in the original submission which included a partial update of the desired reach after exploration this was considered the learning rate. The current model suggests a unity learning rate.

In addition, as suggested by another reviewer, we also fit a value-based model which we adapted from the model described in Giron et al. (2023). This model was not preferred.

We have added a paragraph to the Discussion highlighting different sources of variability and links to our model comparison.

(e) line 155. Why would the success clamp be composed of both motor and exploratory noise? Please clarify in the text

This sentence was written to refer to clamps in general and not just success clamps. However, in the revision this sentence seemed unnecessary so we have removed it.

(3) Hypotheses:

The introduction did not have any hypotheses of development and reinforcement, despite the discussion above setting up potential hypotheses. Did the authors have any hypotheses related to why they might expect age to change motor noise, exploratory noise, and learning rates? If so, what would the experimental behaviour look like to confirm these hypotheses? Currently, the manuscript reads more as an exploratory study, which is certainly fine if true, it should just be explicitly stated in the introduction. Note: on line 144, this is a prediction, not a hypothesis. Line 225: this idea could be sharpened. I believe the authors are speaking to the idea of having more explicit knowledge of action-target pairings changing behaviour.

We have included our hypotheses and predictions at two points in the paper In the introduction we modified the text to:

"We hypothesized that children's reinforcement learning abilities would improve with age, and depend on the developmental trajectory of exploration variability, learning rate (how much people adjust their reach after success), and motor noise (here defined as all sources of noise associated with movement, including sensory noise, memory noise, and motor noise). We think that these factors depend on the developmental progression of neural circuits that contribute to reinforcement learning abilities (Raznahan et al., 2014; Nelson et al., 2000; Schultz, 1998)."

In results we modified the sentence to:

"We predicted that discrete targets could increase exploration by encouraging children to move to a different target after failure.”

Reviewer #2 (Public review):

Summary:

In this study, Hill and colleagues use a novel reinforcement-based motor learning task ("RML"), asking how aspects of RML change over the course of development from toddler years through adolescence. Multiple versions of the RML task were used in different samples, which varied on two dimensions: whether the reward probability of a given hand movement direction was deterministic or probabilistic, and whether the solution space had continuous reach targets or discrete reach targets. Using analyses of both raw behavioral data and model fits, the authors report four main results: First, developmental improvements reflected 3 clear changes, including increases in exploration, an increase in the RL learning rate, and a reduction of intrinsic motor noise. Second, changes to the task that made it discrete and/or deterministic both rescued performance in the youngest age groups, suggesting that observed deficits could be linked to continuous/probabilistic learning settings. Overall, the results shed light on how RML changes throughout human development, and the modeling characterizes the specific learning deficits seen in the youngest ages.

Strengths:

(1) This impressive work addresses an understudied subfield of motor control/psychology - the developmental trajectory of motor learning. It is thus timely and will interest many researchers.

(2) The task, analysis, and modeling methods are very strong. The empirical findings are rather clear and compelling, and the analysis approaches are convincing. Thus, at the empirical level, this study has very few weaknesses.

(3) The large sample sizes and in-lab replications further reflect the laudable rigor of the study.

(4) The main and supplemental figures are clear and concise.

Thank you.

Weaknesses:

(1) Framing.

One weakness of the current paper is the framing, namely w/r/t what can be considered "cognitive" versus "non-cognitive" ("procedural?") here. In the Intro, for example, it is stated that there are specific features of RML tasks that deviate from cognitive tasks. This is of course true in terms of having a continuous choice space and motor noise, but spatially correlated reward functions are not a unique feature of motor learning (see e.g. Giron et al., 2023, NHB). Given the result here that simplifying the spatial memory demands of the task greatly improved learning for the youngest cohort, it is hard to say whether the task is truly getting at a motor learning process or more generic cognitive capacities for spatial learning, working memory, and hypothesis testing. This is not a logical problem with the design, as spatial reasoning and working memory are intrinsically tied to motor learning. However, I think the framing of the study could be revised to focus in on what the authors truly think is motor about the task versus more general psychological mechanisms. Indeed, it may be the case that deficits in motor learning in young children are mostly about cognitive factors, which is still an interesting result!

Thank you for these comments on the framing of our study. We now clearly acknowledge that all motor tasks have cognitive components (new paragraph at line 65). We also explain why we think our tasks has features not present in typical cognitive tasks.

(2) Links to other scholarship.

If I'm not mistaken a common observation in tudies of the development of reinforcement learning is a decrease in exploration over-development (e.g., Nussenbaum and Hartley, 2019; Giron et al., 2023; Schulz et al., 2019); this contrasts with the current results which instead show an increase. It would be nice to see a more direct discussion of previous findings showing decreases in exploration over development, and why the current study deviates from that. It could also be useful for the authors to bring in concepts of different types of exploration (e.g. "directed" vs "random"), in their interpretations and potentially in their modeling.

We recognize that our results differ from prior work. The optimal exploration pattern differs from task to task. We now discuss that exploration is not one size fits all, it's benefits vary depending upon the task. We have added the following paragraphs to the Discussion section:

"One major finding from this study is that exploration variability increases with age. Some other studies of development have shown that exploration can decrease with age indicating that adults explore less compared to children (Schulz et al., 2019; Meder et al., 2021; Giron et al., 2023). We believe the divergence between our work and these previous findings is largely due to the experimental design of our study and the role of motor noise. In the paradigm used initially by Schulz et al. (2019) and replicated in different age groups by Meder et al. (2021) and Giron et al. (2023), participants push buttons on a two-dimensional grid to reveal continuous-valued rewards that are spatially correlated. Participants are unaware that there is a maximum reward available and therefore children may continue to explore to reduce uncertainty if they have difficulty evaluating whether they have reached a maxima. In our task by contrast, participants are given binary reward and told that there is a region in which reaches will always be rewarded. Motor noise is an additional factor which plays a key role in our reaching task but minimal if any role in the discretized grid task. As we show in simulations of our task, as motor noise goes down (as it is known to do through development) the optimal amount of exploration goes up (see Figure 7—figure Supplement 2 and Appendix 1). Therefore, the behavior of our participants is rational in terms of R230 increasing exploration as motor noise decreases.

A key result in our study is that exploration in our task reflects sensitivity to failure. Older children make larger adjustments after failure compared to younger children to find the highly rewarded zone more quickly. Dhawale et al. (2017) discuss the different contexts in which a participant may explore versus exploit (i.e., stick at the same position). Exploration is beneficial when reward is low as this indicates that the current solution is no longer ideal, and the participant should search for a better solution. Konrad et al. (2025) have recently shown this behavior in a real-world throwing task where 6 to 12 year old children increased throwing variability after missed trials and minimized variability after successful trials. This has also been shown in a postural motor control task where participants were more variable after non-rewarded trials compared to rewarded trials (Van Mastrigt et al., 2020). In general, these studies suggest that the optimal amount of exploration is dependent on the specifics of the task."

(3) Modeling.

First, I may have missed something, but it is unclear to me if the model is actually accounting for the gradient of rewards (e.g., if I get a probabilistic reward moving at 45°, but then don't get one at 40°, I should be more likely to try 50° next then 35°). I couldn't tell from the current equations if this was the case, or if exploration was essentially "unsigned," nor if the multiple-trials-back regression analysis would truly capture signed behavior. If the model is sensitive to the gradient, it would be nice if this was more clear in the Methods. If not, it would be interesting to have a model that does "function approximation" of the task space, and see if that improves the fit or explains developmental changes.

The model we use (similar to Roth et al. (2023) and Therrien et al. (2016, 2018)) does not model the gradient. Exploration is always zero-mean Gaussian. As suggested by the reviewer, we now also fit a value-based model (described starting at line 810) which we adapted from the model presented in Giron et al. (2023). We show that the exploration and noise-based model is preferred over the value-based model.

The multiple-trials-back regression was unsigned as the intent was to look at the magnitude and not the direction of the change in movement. We have decided to remove this analysis from the manuscript as it was a source of confusion and secondary analysis that did not add substantially to the findings of these studies.

Second, I am curious if the current modeling approach could incorporate a kind of "action hysteresis" (aka perseveration), such that regardless of previous outcomes, the same action is biased to be repeated (or, based on parameter settings, avoided).

In some sense, the learning rate in the model in the original submission is highly related to perseveration. For example if the learning rate is 0, then there is complete perseveration as you simply repeat the same desired movement. If the rate is 1, there is no perseveration and values in between reflect different amounts of perseveration. Therefore, it is not easy to separate learning rate from perseveration. Adding perseveration as another parameter would likely make it and the learning unidentifiable. However, we now compare 31 models and those that have a non-unity learning rate are not preferred suggesting there is little perseveration.

(4) Psychological mechanisms. There is a line of work that shows that when children and adults perform RL tasks they use a combination of working memory and trial-by-trial incremental learning processes (e.g., Master et al., 2020; Collins and Frank 2012). Thus, the observed increase in the learning rate over development could in theory reflect improvements in instrumental learning, working memory, or both. Could it be that older participants are better at remembering their recent movements in short-term memory (Hadjiosif et al., 2023; Hillman et al., 2024)?

We agree that cognitive processes, such as working memory or visuospatial processing, play a role in our task and describe cognitive elements of our task in the introduction (new paragraph at line 65). However, the sensorimotor model we fit to the data does a good job of explaining the variation across age, which suggests that that age-dependent cognitive processes probably play a smaller role.

Reviewer #3 (Public review):

Summary:

The study investigates reinforcement learning across the lifespan with a large sample of participants recruited for an online game. It finds that children gradually develop their abilities to learn reward probability, possibly hindered by their immature spatial processing and probabilistic reasoning abilities. Motor noise, reinforcement learning rate, and exploration after a failure all contribute to children's subpar performance.

Strengths:

(1) The paradigm is novel because it requires continuous movement to indicate people's choices, as opposed to discrete actions in previous studies.

(2) A large sample of participants were recruited.

(3) The model-based analysis provides further insights into the development of reinforcement learning ability.

Thank you.

Weaknesses:

(1 ) The adequacy of model-based analysis is questionable, given the current presentation and some inconsistency in the results.

Thank you for raising this concern. We have substantially revised the model from our first submission. We now compare 31 noise-based models and 1 value-based model and fit all of the tasks with the preferred model. We perform model selection using the two tasks with the largest datasets to identify the preferred model. From the preferred model, we found the parameter fits for each individual dataset and simulated the trial by trial behavior allowing comparison between all four tasks. We now show examples of individual fits as well as provide a measure of goodness of fit. The expansion of our modeling approach has resolved inconsistencies and sharpened the conclusions drawn from our model.

(2) The task should not be labeled as reinforcement motor learning, as it is not about learning a motor skill or adapting to sensorimotor perturbations. It is a classical reinforcement learning paradigm.

We now make it clear that our reinforcement learning task has both motor and cognitive demands, but does not fall entirely within one of these domains. We use the term motor learning because it captures the fact that participants maximize reward by making different movements, corrupted by motor noise, to unmarked locations on a continuous target zone. When we look at previous ublications, it is clear that our task is similar to those that also refer to this as reinforcement motor learning Cashaback et al. (2019) (reaching task using a robotic arm in adults), Van Mastrigt et al. (2020) (weight shifting task in adults), and Konrad et al. (2025) (real-world throwing task in children). All of these tasks involve trial-by-trial learning through reinforcement to make the movement that is most effective for a given situation. We feel it is important to link our work to these previous studies and prefer to preserve the terminology of reinforcement motor learning.

Recommendations for the authors:

Reviewing Editor Comments:

Thank you for this summary. Rather than repeat the extended text from the responses to the reviewers here, we point the Editor to the appropriate reviewer responses for each issue raised.

The reviewers and editors have rated the significance of the findings in your manuscript as "Valuable" and the strength of evidence as "Solid" (see eLife evalutation). A consultancy discussion session to integrate the public reviews and recommendations per reviewer (listed below), has resulted in key recommendations for increasing the significance and strength of evidence:

To increase the Significance of the findings, please consider the following:

(1) Address and reframe the paper around whether the task is truly getting at a motor learning process or more generic cognitive decision-making capacities such as spatial memory, reward processing, and hypothesis testing.

We have revised the paper to address the comments on the framing of our work. Please see responses to the public review comments of Reviewers #2 and #3.

(2) It would be beneficial to specify the differences between traditional reinforcement algorithms (i.e., using softmax functions to explore, and build representations of state-action-reward) and the reinforcement learning models used here (i.e., explore with movement variability, update reach aim towards the last successful action), and compare present findings to previous cognitive reinforcement learning studies in children.

Please see response to the public review comments of Reviewer #1 in which we explain the expansion of our modeling approach to fit a value-based model as well as 31 other noise-based models. In our response to the public review comments of Reviewer #2, we comment on our expanded discussion of how our findings compare with previous cognitive reinforcement learning studies.

To move the "Strength of Evidence" to "Convincing", please consider doing the following:

(1 ) Address some apparently inconsistent and unrealistic values of motor noise, exploration noise, and learning rate shown for individual participants (e.g., Figure 5b; see comments reviewers 1 and take the following additional steps: plotting r squares for individual participants, discussing whether individual values of the fitted parameters are plausible and whether model parameters in each age group can extrapolate to the two clamp conditions and baselines.

We have substantially updated our modeling approach. Now that we compare 31 noise-based models, the preferred model does not show any inconsistent or unrealistic values (see response to Reviewer #3). Additionally, we now show example individual fits and provide both relative and absolute goodness of fit (see response to Reviewer #3).

(2) Relatedly, to further justify if model assumptions are met, it would be valuable to show that the current learning model fits the data better than alternative models presented in the literature by the authors themselves and by others (reviewer 1). This could include alternative development models that formalise the proposed explanations for age-related change: poor spatial memory, reward/outcome processing, and exploration strategies (reviewer 2).

Please see response to public review comments of Reviewer #1 in which we explain that we have now fit a value-based model as well as 31 other noise-based models providing a comparison of previous models as well as novel models. This led to a slightly different model being preferred over the model in the original submission (updated model has a learning rate of unity). These models span many of the processes previously proposed for such tasks. We feel that 32 models span a reasonable amount of space and do not believe we have the power to include memory issues or heuristic exploration strategies in the model.

(3) Perform the mediation analysis with all the possible variables (i.e., not informed by multiple regression) to see if the results are more consistent across studies and with the current approach (see comments reviewer 1).

Please see response to public review comments of Reviewer #1. We chose to focus only on the model based analysis because it allowed us to distinguish between exploration variability and motor noise.

Please see below for further specific recommendations from each reviewer.

Reviewer #1 (Recommendations for the author):

(1) In general, there should be more discussion and contextualization of other binary reinforcement tasks used in the motor literature. For example, work from Jeroen Smeets, Katinka van der Kooij, and Joseph Galea.

Thank you for this comment. We have edited the Introduction to better contextualize our work within the reinforcement motor learning literature (see line 67 and line 83).

(2) Line 32. Very minor. This sentence is fine, but perhaps could be slightly improved. “select a location along a continuous and infinite set of possible options (anywhere along the span of the bridge)"

Thank you for this comment. We have edited the sentence to reflect this suggestion.

(3) Line 57. To avoid some confusion in successive sentences: Perhaps, "Both children over 12 and adolescents...".

Thank you for this comment. We have edited the sentence to reflect this suggestion.

(4) Line 80. This is arguably not a mechanistic model, since it is likely not capturing the reward/reinforcement machinery used by the nervous system, such as updating the expected value using reward predic tion errors/dopamine. That said, this phenomenological model, and other similar models in the field, do very well to capture behaviour with a very simple set of explore and update rules.

We use mechanistic in the standard use in modeling, as in Levenstein et al. (2023), for example. The contrast is not with neural modeling, but with normative modeling, in which one develops a model to optimize a function (or descriptive models as to what a system is trying to achieve). In mechanistic modeling one proposes a mechanism and this can be at a state-space level (as in our case) or a neural level (as suggested my the reviewer) but both are considered mechanistic, just at different levels. Quoting Levenstein "... mechanistic models, in which complex processes are summarized in schematic or conceptual structures that represent general properties of components and their interactions, are also commonly used." We now reference the Levenstein paper to clarify what we mean by mechanistic.

(5) Figure 1. It would be useful to state that the x-axis in Figure 1 is in normalized units, depending on the device.

Thank you for this comment. We have added a description of the x-axis units to the Fig. 1 caption.

(6) Were there differences in behaviour for these different devices? e.g., how different was motor noise for the mouse, trackpad, and touchscreen?

Thank you for this question. We did not find a significant effect of device on learning or precision in the baseline block. We have added these one way ANOVA results for each task in Supplementary Table 1.

(7) Line 98. Please state that participants received reinforcement feedback during baseline.

Thank you for this comment. We have updated the text to specify that participants receive reward feedback during the baseline block.

(8) Line 99. Did the distance from the last baseline trial influence whether the participant learned or did not learn? For example, would it place them too far from the peak success location such that it impacted learning?

Thank you for this question. We looked at whether the position of movement on the last baseline block trial was correlated with the first movement position in the learning block. We did not find any correlations between these positions for any of the tasks. Interestingly, we found that the majority of participants move to the center of the workspace on the first trial of the learning block for all tasks (either in the presence of the novel continuous target scene or the presentation of 7 targets all at once). We do not think that the last movement in the baseline block "primed" the participant for the location of the success zone in the learning block. We have added the following sentence to the Results section:

"Note that the reach location for the first learning trial was not affected by (correlated with) the target position on the last baseline trial (p > 0.3 for both children and adults, separately)."

(9) The term learning distance could be improved. Perhaps use distance from target.

Thank you for this comment. We appreciate that learning distance defined with 0 as the best value is counter intuitive. We have changed the language to be "distance from target" as the learning metric.

(10) Line 188. This equation is correct, but to estimate what the standard deviation by the distribution of changes in reach position is more involved. Not sure if the authors carried out this full procedure, which is described in Cashaback et al., 2019; Supplemental 2.

There appear to be no Supplemental 2 in the referenced paper so we assume the reviewer is referring to Supplemental B which deals with a shuffling procedure to examine lag-1 correlations.

In our tasks, we are limited to only 9 trials to analyze in each clamp phase so do not feel a shuffling analysis is warranted. In these blocks, we are not trying to 'estimate what the standard deviation by the distribution of changes in reach position' but instead are calculating the standard deviation of the reach locations and comparing the model fit (for which the reviewer says the formula is correct) with the data. We are unclear what additional steps the reviewer is suggesting. In our updated model analysis, we fit the data including the clamp phases for better parameter estimation. We use simulations to estimate s.d. in the clamp phase (as we ensure in simulations the data does not fall outside the workspace) making the previous analytic formulas an approximation that are no longer used.

(11) Line 197-199. Having done the demo task, it is somewhat surprising that a 3-year-old could understand these instructions (whose comprehension can be very different from even a 5-year old).

Thank you for raising this concern. We recognize that the younger participants likely have different comprehension levels compared to older participants. However, we believe that the majority of even the youngest participants were able to sufficiently understand the goal of the task to move in a way to get the video clip to play. We intentionally designed the tasks to be simple such that the only instructions the child needed to understand were that the goal was to get the video clip to play as much as possible and the video clip played based on their movement. Though the majority of younger children struggled to learn well on the probabilistic tasks, they were able to learn well on the deterministic tasks where the task instructions were virtually identical with the exception of how many places in the workspace could gain reward. On the continuous probabilistic task, we did have a small number (n = 3) of 3 to 5 year olds who exhibited more mature learning ability which gives us confidence that the younger children were able to understand the task goal.

(12) Line 497: Can the authors please report the F-score and p-value separately for each of these one-way ANOVA (the device is of particular interest here).

Thank you for this request. We have added ina upplementarytable (Supplementary Table 1) with the results of these ANOVAs.

(13) Past work has discussed how motivation influences learning, which is a function of success rate (van der Kooij, K., in 't Veld, L., & Hennink, T. (2021). Motivation as a function of success frequency. Motivation and Emotion, 45, 759-768.). Can the authors please discuss how that may change throughout development?

Thank you for this comment. While motivation most probably plays a role in learning, in particular in a game environment, this was out of the scope of the direct focus of this work and not something that our studies were designed to test. We have added the following sentence to the discussion section to address this comment:

"We also recognize that other processes, such as memory and motivation, could affect performance on these tasks however our study was not designed to test these processes directly and future work would benefit from exploring these other components more explicitly."

(14) Supplement 6. This analysis is somewhat incomplete because it does not consider success.

Pekny and collegues (2015) looked at 3 trials back but considered both success and reward. However, their analysis has issues since successive time points are not i.i.d., and spurious relationships can arise. This issue is brought up by Dwahale (Dhawale, A. K., Miyamoto, Y. R., Smith, M. A., & R475 Ölveczky, B. P. (2019). Adaptive regulation of motor variability. Current Biology, 29(21), 3551-3562.). Perhaps it is best to remove this analysis from the paper.

Thank you for this comment. We have decided to remove this secondary analysis from the paper as it was a source of confusion and did not add to the understanding and interpretation of our behavioral results.

Reviewer #2 (Recommendations for the author):

(1 ) the path length ratio analyses in the supplemental are interesting but are not mentioned in the main paper. I think it would be helpful to mention these as they are somewhat dramatic effects

Thank you for this comment. Path length ratios are defined in the Methods and results are briefly summarized in the Results section with a point to the supplementary figures. We have updated the text to more explicitly report the age related differences in path length ratios.

(2) The second to last paragraph of the intro could use a sentence motivating the use ofthe different task features (deterministic/probabilistic and discrete/continuous).

Thank you for this comment. We have added an additional motivating sentence to the introduction.

Reviewer #3 (Recommendations for the author):

The paper labeled the task as one for reinforcement motor learning, which is not quite appropriate in my opinion. Motor learning typically refers to either skill learning or motor adaptation, the former for improving speed-accuracy tradeoffs in a certain (often new) motor skill task and the latter for accommodating some sensorimotor perturbations for an existing motor skill task. The gaming task here is for neither. It is more like a

decision-making task with a slight contribution to motor execution, i.e., motor noise. I would recommend the authors label the learning as reinforcement learning instead of reinforcement motor learning.

Thank you for this comment. As noted in the response to the public review comments, we agree that this task has components of classical reinforcement learning (i.e. responding to a binary reward) but we specifically designed it to require the learning of a movement within a novel game environment. We have added a new paragraph to the introduction where we acknowledge the interplay between cognitive and motor mechanisms while also underscoring the features in our task that we think are not present in typical cognitive tasks.

My major concern is whether the model adequately captures subjects' behavior and whether we can conclude with confidence from model fitting. Motor noise, exploration noise, and learning rate, which fit individual learning patterns (Figure 5b), show some quite unrealistic values. For example, some subjects have nearly zero motor noise and a 100% learning rate.

We have now compared 31 models and the preferred model is different from the one in the first submission. The parameter fits of the new model do not saturate in any way and appear reasonable to us. The updates to the model analysis have addressed the concern of previously seen unrealistic values in the prior draft.

Currently, the paper does not report the fitting quality for individual subjects. It is good to have an exemplary subject's fit shown, too. My guess is that the r-squared would be quite low for this type of data. Still, given that the children's data is noisier, it might be good to use the adult data to show how good the fitting can be (individual fits, r squares, whether the fitted parameters make sense, whether it can extrapolate to the two clamp phases). Indeed, the reliability of model fitting affects how we should view the age effect of these model parameters.

We now show fits to individual subjects. But since this is a Kalman smoother it fits the data perfectly by generating its best estimate of motor noise and exploration variability on each trial to fully account for the data — so in that sense R² is always 1 so that is not helpful.

While the BIC analysis with the other model variants provides a relative goodness of fit, it is not straightforward to provide an absolute goodness of fit such as standard R² for a feedforward simulation of the model given the parameters (rather than the output of the Kalman smoother). There are two problems. First, there is no single model output. Each time the model is simulated with the fit parameters it produces a different output (due to motor noise, exploration variability and reward stochasticity). Second, the model is not meant to reproduce the actual motor noise, exploration variability and reward stochasticity of a trial. For example, the model could fit pure Gaussian motor noise across trials (for a poor learner) by accurately fitting the standard deviation of motor noise but would not be expected to actually match each data point so would have a traditional R² of O.

To provide an overall goodness of fit we have to reduce the noise component and to do so we exam ined the traditional R² between the average of all the children's data and the average simulation of the model (from the median of 1000 simulations per participant) so as to reduce the stochastic variation. The results for the continuous probabilistic and discrete probabilistic task are R² of 0.41 and 0.72, respectively.

Not that variability in the "success clamp" doe not change across ages (Figure 4C) and does not contribute to the learning effect (Figure 4F). However, it is regarded as reflecting motor noise (Figure SC), which then decreases over age from the model fitting (Figure 5B). How do we reconcile these contradictions? Again, this calls the model fitting into question.

For the success clamp, we only have 9 trials to calculate variability which limits our power to detect significance with age. In contrast, the model uses all 120 trials to estimate motor noise. There is a downward trend with age in the behavioral data which we now show overlaid on the fits of the model for both probabilistic conditions (Figure 5—figure Supplement 4) and Figure 6—figure Supplement 4). These show a reasonable match and although the variance explained is 1 6 and 56% (we limit to 9 trials so as to match the fail clamp), the correlations are 0.52 and 0.78 suggesting we have reasonable relation although there may be other small sources of variability not captured in the model.

Figure 5C: it appears one bivariate outlier contributes a lot to the overall significant correlation here for the "success clamp".

Recalculating after removing that point in original Fig 5C was still significant and we feel the plots mentioned in the previous point add useful information to this issue. With the new model this figure has changed.

It is still a concern that the young children did not understand the instructions. Nine 3-to-8 children (out of 48) were better explained by the noisy only model than the full model. In contrast, ten of the rest of the participants (out of 98) were better explained by the noisy-only model. It appears that there is a higher percentage of the "young" children who didn't get the instruction than the older ones.

Thank you for this comment. We did take participant comprehension of the task into consideration during the task design. We specifically designed it so that the instructions were simple and straight forward. The child simply needs to understand the underlying goal to make the video clip play as often as possible and that they must move the penguin to certain positions to get it to play. By having a very simple task goal, we are able to test a naturalistic response to reinforcement in the absence of an explicit strategy in a task suited even for young children.

We used the updated reinforcement learning model to assess whether an individual's performance is consistent with understanding the task. In the case of a child who does not understand the task, we expect that they simply have motor noise on their reach, and crucially, that they would not explore more after failure, nor update their reach after success. Therefore, we used a likelihood ratio test to examine whether the preferred model was significantly better at explaining each participant's data compared to the model variant which had only motor noise (Model 1). Focusing on only the youngest children (age 3-5), this analysis showed that that 43, 59, 65 and 86% of children (out of N = 21, 22, 20 and 21 ) for the continuous probabilistic, discrete probabilistic, continuous deterministic, and discrete deterministic conditions, respectively, were better fit with the preferred model, indicating non-zero exploration after failure. In the 3-5 year old group for the discrete deterministic condition, 18 out of 21 had performance better fit by the preferred model, suggesting this age group understands the basic task of moving in different directions to find a rewarding location.

The reduced numbers fit by the preferred model for the other conditions likely reflects differences in the task conditions (continuous and/or probabilistic) rather than a lack of understanding of the goal of the task. We include this analysis as a new subsection at the end of the Results.

Supplementary Figure 2: the first panel should belong to a 3-year-old not a 5-year-old? How are these panels organized? This is kind of confusing.

Thank you for this comment. Figure 2—figure Supplement 1 and Figure 2—figure Supplement 2 are arranged with devices in the columns and a sample from each age bin in the rows. For example in Figure 2—figure Supplement 1, column 1, row 1 is a mouse using participant age 3 to 5 years old while column 3, row 2 is a touch screen using participant age 6 to 8 years old. We have edited the labeling on both figures to make the arrangement of the data more clear.

Line 222: make this a complete sentence.

This sentence has been edited to a complete sentence.

Line 331: grammar.

This sentence has been edited for grammar.

eLife Assessment

This study provides an important set of analyses and theoretical derivations to understand the mechanisms used by recurrent neural networks (RNNs) to perform context-dependent accumulation of evidence. The results regarding the dimensionality and neural dynamical signatures of RNNs are convincing and provide new avenues to study the mechanisms underlying context-dependent computations. This manuscript will be of interest to a broad audience in systems and computational neuroscience.

Reviewer #1 (Public review):

Summary:

This paper investigates how recurrent neural networks (RNNs) can perform context-dependent decision-making (CDM). The authors use low-rank RNN modeling and focus on a CDM task where subjects are presented with sequences of auditory pulses that vary in location and frequency, and they must determine either the prevalent location or frequency based on an external context signal. In particular, the authors focus on the problem of differentiating between two distinct selection mechanisms: input modulation, which involves altering the stimulus input representation, and selection vector modulation, which involves altering the "selection vector" of the dynamical system.

First, the authors show that rank-one networks can only implement input modulation, and that higher-rank networks are required for selection vector modulation. Then, the authors use pathway-based information flow analysis to understand how information is routed to the accumulator based on context. This analysis allows the authors to introduce a novel definition of selection vector modulation that explicitly links it to changes in the effective coupling along specific pathways within the network.

The study further generates testable predictions for differentiating selection vector modulation from input modulation based on neural dynamics. In particular, the authors find that: 1) A larger proportion of selection vector modulation is expected in networks with high-dimensional connectivity. 2) Single-neuron response kernels exhibiting specific profiles (peaking between stimulus onset and choice onset) are indicative of neural dynamics in extra dimensions, supporting the presence of selection vector modulation. 3) The percentage of explained variance (PEV) of extra dynamical modes extracted from response kernels at the population level can serve as an index to quantify the amount of selection vector modulation.

Strengths:

The paper is clear and well written, and it draws bridges between two recent important approaches in the study of CDM: circuit-level descriptions of low-rank RNNs, and differentiation across alternative mechanisms in terms of neural dynamics. The most interesting aspect of the study involves establishing a link between selection vector modulation, network dimensionality and dimensionality of neural dynamics. The high correlation between the networks' mechanisms and their dimensionality (Fig. 7d) is surprising since differentiating between selection mechanisms is generally a difficult task, and the strength of this result is further corroborated by its consistency across multiple RNN hyperparameters (Figure 7-figure supplement 1 and Figure 7-figure supplement 2). Interestingly, the correlation between the selection mechanism and the dimensionality of neural dynamics is also high (Fig. 7g), potentially providing a promising future avenue for the study of neural recordings in this task.

Weaknesses:

As acknowledged by the authors, the results linking selection vector modulation and dimensionality might not generalize to neural representations where a significant fraction of the variance encodes information unrelated to the task. Therefore, these tools might not be applicable to neural recordings or to artificial neural networks with additional high-dimensional activity unrelated to the task (e.g. RNNs trained to perform many other tasks).

Reviewer #2 (Public review):

This manuscript examines network mechanisms that allow networks of neurons to perform context-dependent decision-making.<br /> In a recent study, Pagan and colleagues identified two distinct mechanisms by which recurrent neural networks can perform such computations. They termed these two mechanisms input-modulation and selection-vector modulation. Pagan and colleagues demonstrated that recurrent neural networks can be trained to implement combinations of these two mechanisms, and related this range of computational strategies with inter-individual variability in rats performing the same task. What type of structure in the recurrent connectivity favors one or the other mechanism however remained an open question.

The present manuscript addresses this specific question by using a class of mechanistically interpretable recurrent neural networks, low-rank RNNs.<br /> The manuscript starts by demonstrating that unit-rank RNNs can only implement the input-modulation mechanism, but not the selection-vector modulation. The authors then build rank three networks which implement selection-vector modulation, and show how the two mechanisms can be combined. Finally, they relate the amount of selection-vector modulation with the effective rank, ie the dimensionality of activity, of a trained full-rank RNN.

Strength:

- The manuscript is written in an obvious manner<br /> - The analytic approach adopted in the manuscript is impressive<br /> - Very clear identification of the mechanisms leading to the two types of context-dependent modulation<br /> - Altogether, this manuscript reports remarkable insights on a very timely question

eLife Assessment

The manuscript contains important findings regarding inflammatory macrophage subsets that have theoretical and/or practical applications beyond the field of rheumatology. The authors demonstrate with compelling evidence the effects of PGE2 on TNF signaling. This work will be of broad interest to immunologists and cell biologists.

Reviewer #1 (Public review):

Summary:

This article investigates the phenotype of macrophages with a pathogenic role in arthritis, particularly focusing on arthritis induced by immune checkpoint inhibitor (ICI) therapy.

Building on prior data from monocyte-macrophage coculture with fibroblasts, the authors hypothesized a unique role for the combined actions of prostaglandin PGE2 and TNF. The authors studied this combined state using an in vitro model with macrophages derived from monocytes of healthy donors. They complemented this with single-cell transcriptomic and epigenetic data from patients with ICI-RA, specifically, macrophages sorted out of synovial fluid and tissue samples. The study addressed critical questions regarding the regulation of PGE2 and TNF: Are their actions co-regulated or antagonistic? How do they interact with IFN-γ in shaping macrophage responses?

This study is the first to specifically investigate a macrophage subset responsive to the PGE2 and TNF combination in the context of ICI-RA, describes a new and easily reproducible in vitro model, and studies the role of IFNgamma regulation of this particular Mф subset.

Strengths:

Methodological quality: The authors employed a robust combination of approaches, including validation of bulk RNA-seq findings through complementary methods. The methods description is excellent and allows for reproducible research. Importantly, the authors compared their in vitro model with ex vivo single-cell data, demonstrating that their model accurately reflects the molecular mechanisms driving the pathogenicity of this macrophage subset.

Comments on latest version:

The revisions made to this manuscript followed the suggestions and improved the manuscript. The authors have thoroughly addressed my previous concerns, making several key improvements:

The expanded comparison between rheumatoid arthritis (RA) and immune checkpoint inhibitor-induced RA (ICI-RA) in both cellular and molecular pathology is excellent. These additions to the literature review and discussion sections significantly strengthen the manuscript and provide valuable context.

I particularly appreciate the added effort in mapping a particular cell subset onto previously published single-cell RNA-Seq embeddings. The enhanced UMAPs with cell subset projection analyses are methodologically compelling, informative and visually are easy to understand for any reader. The new Figure 3 represents a substantial improvement.

More detailed comparisons with previously published single-cell datasets from 2019, 2020, and 2023 effectively contextualize this research within the broader field of rheumatoid arthritis pathogenesis. This enhances the manuscript's value for specialists in autoimmunity and myeloid immunology.

I find the authors' suggestion to use the defined myeloid pathogenic phenotypes as biomarkers for therapy response prediction or dose optimization particularly insightful and clinically relevant.

Overall, the authors have significantly improved both the analysis and presentation of results. The manuscript has been substantially enhanced.

Reviewer #2 (Public review):

Summary/Significance of the findings:

The authors have done a great job by extensively carrying out transcriptomic and epigenomic analyses in the primary human/mouse monocytes/macrophages to investigate TNF-PGE2 (TP) crosstalk and their regulation by IFN-γ in the Rheumatoid arthritis (RA) synovial macrophages. They proposed that TP induces inflammatory genes via a novel regulatory axis whereby IFN-γ and PGE2 oppose each other to determine the balance between two distinct TNF-induced inflammatory gene expression programs relevant to RA and ICI-arthritis.

Strengths:

The authors have done a great job on RT-qPCR analysis of gene expression in primary human monocytes stimulated with TNF and showing the selective agonists of PGE2 receptors EP2 and EP4 22 that signal predominantly via cAMP. They have beautifully shown IFN-γ opposes the effects of PGE2 on TNF-induced gene expression. They found that TP signature genes are activated by cooperation of PGE2-induced AP-1, CEBP, and NR4A with TNF-induced NF-κB activity. On the other hand, they found that IFN-γ suppressed induction of AP-1, CEBP, and NR4A activity to ablate induction of IL-1, Notch, and neutrophil chemokine genes but promoted expression of distinct inflammatory genes such as TNF and T cell chemokines like CXCL10 indicating that TP induces inflammatory genes via IFN-γ in the RA and ICI-arthritis.

Comments on latest version:

The authors have answered my questions and i recommend this manuscript for publication.

Author response:

The following is the authors’ response to the original reviews

Public Reviews: 

Reviewer #1 (Public review): 

Summary: 

This article investigates the phenotype of macrophages with a pathogenic role in arthritis, particularly focusing on arthritis induced by immune checkpoint inhibitor (ICI) therapy. 

Building on prior data from monocyte-macrophage coculture with fibroblasts, the authors hypothesized a unique role for the combined actions of prostaglandin PGE2 and TNF. The authors studied this combined state using an in vitro model with macrophages derived from monocytes of healthy donors. They complemented this with single-cell transcriptomic and epigenetic data from patients with ICI-RA, specifically, macrophages sorted out of synovial fluid and tissue samples. The study addressed critical questions regarding the regulation of PGE2 and TNF: Are their actions co-regulated or antagonistic? How do they interact with IFN-γ in shaping macrophage responses? 

This study is the first to specifically investigate a macrophage subset responsive to the PGE2 and TNF combination in the context of ICI-RA, describes a new and easily reproducible in vitro model, and studies the role of IFNgamma regulation of this particular Mф subset. 

Strengths: 

Methodological quality: The authors employed a robust combination of approaches, including validation of bulk RNA-seq findings through complementary methods. The methods description is excellent and allows for reproducible research. Importantly, the authors compared their in vitro model with ex vivo single-cell data, demonstrating that their model accurately reflects the molecular mechanisms driving the pathogenicity of this macrophage subset. 

Weaknesses: 

Introduction: The introduction lacks a paragraph providing an overview of ICI-induced arthritis pathogenesis and a comparison with other types of arthritis. Including this would help contextualize the study for a broader audience.

Thank you for this suggestion, we have added a paragraph on ICI-arthritis to intro (pg. 4, middle paragraph).  

Results Section: At the beginning of the results section, the experimental setup should be described in greater detail to make an easier transition into the results for the reader, rather than relying just on references to Figure 1 captions.

We have clarified the experimental setup (pg. 5).  

There is insufficient comparison between single-cell RNA-seq data from ICI-induced arthritis and previously published single-cell RA datasets. Such a comparison may include DEGs and GSEA, pathway analysis comparison for similar subsets of cells. Ideally, an integration with previous datasets with RA-tissue-derived primary monocytes would allow for a direct comparison of subsets and their transcriptomic features.

We thank the Reviewer for this suggestion, which has increased the impact of our data and analysis. A computationally rigorous representation mapping approach showed that ICI-arthritis myeloid subsets predominantly mapped onto 4 previously defined RA subsets including IL-1β+ cells. This result was corroborated using a complementary data integration approach. Analysis of (TNF + PGE)-induced gene sets (TP signatures) in ICI-arthritis myeloid cells projected onto the RA subsets using the AUCell package showed elevated TP gene expression in similar ICI-arthritis and RA monocytic cell subsets. We also found mutually exclusive expression of TP and IFN signatures in distinct RA and ICI-arthritis myeloid cell subsets, which supports that the opposing cross-regulation between IFN-γ and PGE2 pathways that we identified in vitro also functions similarly in vivo. This analysis is shown in the new Fig. 3, described on pg. 7, and discussed on pp. 13-14.

While it's understandable that arthritis samples are limited in numbers and myeloid cell numbers, it would still be interesting to see the results of PGE2+TNF in vitro stimulation on the primary RA or ICI-RA macrophages. It would be valuable to see RNA-Seq signatures of patient cell reactivation in comparison to primary stimulation of healthy donor-derived monocytes.

We agree that this would be interesting but given limited samples and distribution of samples amongst many studies and investigators this is beyond the scope of the current study.  

Discussion: Prior single-cell studies of RA and RA macrophage subpopulations from 2019, 2020, 2023 publications deserve more discussion. A thorough comparison with these datasets would place the study in a broader scientific context. 

Creating an integrated RA myeloid cell atlas that combines ICI-RA data into the RA landscape would be ideal to add value to the field. 

As one of the next research goals, TNF blockade data in RA and ICI-RA patients would be interesting to add to such an integrated atlas. Combining responders and non-responders to TNF blockade would help to understand patient stratification with the myeloid pathogenic phenotypes. It would be great to read the authors' opinion on this in the Discussion section. 

Please see our response to point 3 above. This point is addressed in Fig. 3, pg. 7, and pp. 13-14, which includes a discussion of responders and nonresponders and patient stratification.  

Conclusion: The authors demonstrated that while PGE2 maintains the inflammatory profile of macrophages, it also induces a distinct phenotype in simultaneous PGE2 and TNF treatment. The study of this specific subset in single-cell data from ICI-RA patients sheds light on the pathogenic mechanisms underlying this condition, however, how it compares with conventional RA is not clear from the manuscript. 

Given the substantial incidence of ICI-induced autoimmune arthritis, understanding the unique macrophage subsets involved for future targeting them therapeutically is an important challenge. The findings are significant for immunologists, cancer researchers, and specialists in autoimmune diseases, making the study relevant to a broad scientific audience. 

Reviewer #2 (Public review): 

Summary/Significance of the findings: 

The authors have done a great job by extensively carrying out transcriptomic and epigenomic analyses in the primary human/mouse monocytes/macrophages to investigate TNF-PGE2 (TP) crosstalk and their regulation by IFN-γ in the Rheumatoid arthritis (RA) synovial macrophages. They proposed that TP induces inflammatory genes via a novel regulatory axis whereby IFN-γ and PGE2 oppose each other to determine the balance between two distinct TNF-induced inflammatory gene expression programs relevant to RA and ICI-arthritis. 

Strengths: 

The authors have done a great job on RT-qPCR analysis of gene expression in primary human monocytes stimulated with TNF and showing the selective agonists of PGE2 receptors EP2 and EP4 22 that signal predominantly via cAMP. They have beautifully shown IFN-γ opposes the effects of PGE2 on TNF-induced gene expression. They found that TP signature genes are activated by cooperation of PGE2-induced AP-1, CEBP, and NR4A with TNF-induced NF-κB activity. On the other hand, they found that IFN-γ suppressed induction of AP-1, CEBP, and NR4A activity to ablate induction of IL-1, Notch, and neutrophil chemokine genes but promoted expression of distinct inflammatory genes such as TNF and T cell chemokines like CXCL10 indicating that TP induces inflammatory genes via IFN-γ in the RA and ICI-arthritis. 

Weaknesses: 

(1) The authors carried out most of the assays in the monocytes/macrophages. How do APCcells like Dendritic cells behave with respect to this TP treatment similar dosing? 

We agree that this is an interesting topic especially as TNF + PGE2 is one of the standard methods of maturing in vitro generated human DCs and promoting antigen-presenting function. As DC maturation is quite different from monocyte activation this would represent a new study and is beyond the scope of the current manuscript. We have instead added a paragraph to the discussion (pg. 12) and cited the literature on DC maturation by TNF + PGE2 including one of our older papers (PMID: 18678606; 2008)  

(2) The authors studied 3h and 24h post-treatment transcriptomic and epigenomic. What happens to TP induce inflammatory genes post-treatment 12h, 36h, 48h, 72h. It is critical to see the upregulated/downregulated genes get normalised or stay the same throughout the innate immune response.

We now clarify that subsets of inducible genes showed distinct kinetics of induction with transient expression at 3 hr versus sustained expression over the 24 hr stimulation period as shown in Supplementary Fig. 1 (pg. 5).

(3) The authors showed IL1-axis in response to the TP-treatment. Do other cytokine axes get modulated? If yes, then how do they cooperate to reduce/induce inflammatory responses along this proposed axis?

This is an interesting question, which we approached using a combination of pathway analysis and targeted inspection of pathways important pathogenesis of RA, which is the inflammatory condition most relevant for this study. In addition to genes in the IL-1-NF-κB core inflammatory pathway, pathway analysis of genes induced by TP co-stimulation showed enrichment of genes related to leukocyte chemotaxis, in particular neutrophil migration. Accordingly, TP costimulation increased expression of CSF3, which plays a key role in mobilizing neutrophils from the bone marrow, and major neutrophil chemokines CXCL1, CXCL2, CXCL3 and CXCL5 that recruit neutrophils to sites of inflammation including in inflammatory arthritis. Analysis of the late response to TNF similarly showed enrichment of genes important in chemotaxis, and suppression of genes in the cholesterol biosynthetic pathway, which we and others have previously linked to IFN responses. Targeted inspection of genes in additional pathways implicated in RA pathogenesis showed increased expression of genes in the Notch pathway. We believe that these pathways work together with the IL-1 pathway to increase immune cell recruitment and activation in inflammatory responses; these results are described on pp. 5-6 and are incorporated into Figures 1, 2 and Supplementary Fig. 2. 

Overall, the data looks good and acceptable but I need to confirm the above-mentioned criticisms. 

Recommendations for the authors: 

Reviewer #1 (Recommendations for the authors):   

The discussion section of the manuscript claims: "In this study, we utilized transcriptomics to demonstrate a 'TNF + PGE2' (TP) signature in RA and ICI-arthritis IL-1β+ synovial macrophages." This statement is misleading, as no new transcriptomic data from RA synovial samples were generated in this study. To support such a claim, the authors would need to compare primary monocytes or macrophages from RA patients using bulk RNA-seq or singlecell RNA-seq. Based on the current data, the comparison is limited to bulk RNA-seq findings from the authors' in vitro model and prior monocyte-fibroblast coculture studies. 

We have modified the abstract and discussion (pg. 10) to reflect that we have compared an in vitro generated TP signature with gene expression in previously identified RA macrophage subsets.

eLife Assessment

The authors study the context of the skeletal remains of three individuals and associated sediment samples to conclude that the hominin species Homo naledi intentionally buried their dead. Demonstration of the earliest known instance of intentional funerary practice – with a relatively small-brained hominin engaging in a highly complex behavior that has otherwise been observed from Homo sapiens and Homo neanderthalensis – would represent a landmark finding. The authors have revised their manuscript extensively in light of the reviews of their initial submission, with improved illustration, context, discussion, and theoretical frameworks, leading to an improved case supporting their conclusion that Homo naledi intentionally buried their dead. One of the reviewers concludes that the findings convincingly demonstrate intentional burial practices, while another considers evidence for such an unambiguous conclusion to be incomplete given a lack of definitive knowledge around how the hominins got into the chamber. We look forward to seeing the continued development and assessment of this hypothesis. It is worth noting that the detailed reviews (both rounds) and comprehensive author response are commendable and consequential parts of the scientific record of this study. The editors note that the authors' response repeatedly invokes precedent from previous publications to help justify the conclusions in this paper. While doing so is helpful, the editors also note that scientific norms and knowledge are constantly evolving, and that any study has to rest on its own scientific merit.

Reviewer #1 (Public review):

Thank you for allowing me to review the paper "Evidence for deliberate burial of the dead by Homo naledi". This remains a very important site for paleoanthropology. I appreciate the work that the crew, especially the junior members of the team, put into this massive project. I appreciate that the authors did revise the paper since that is not a requirement of eLife. Extensive reviews by peer-reviewers have been provided for this paper, as well as professionally published replies (Martinón-Torres et al., 2023; Foecke et al., 2023). The composition, and citations of this version are much improved, though important information, some requested by reviewers, are buried in the supplementary section. It seems important that the authors make these sections more easily accessible to the general reader. The length of the paper is also unnecessary and impedes the readability of the work. Concise clarity is an expectation of most journals. The Netflix documentary was made to appeal to a mass audience, I would hope that the goal of the accompanying publication would be to enable readers to fully comprehend the work behind the claims.

This version of the paper considers at great length many possibilities for how the H. naledi skeletal material came to rest in the cave system with some additional figures and data provided. However, quite a lot is still unclear. In my original review I stated, "The authors have repeatedly described how incredibly challenging it is to get into and out of this cave system and all of its chambers." This was a point emphasized in the Netflix documentary. In this version of the paper the authors have included within the supplementary section a brief discussion of other entrances. The work by Robbins et al. 2021 (a peer-reviewed paper in the impact factor rated journal Chemical Geology) is extremely relevant here. In this revision it is noted in the supplementary section that if the Postbox chamber was used as an opening, it would have reduced the length of the access to the system by 80 m. This fact seems important. This section should be moved out of the supplementary material and expanded because the conclusions published by Robbins et al. (2021) indicate a completely different route by which H. naledi accessed the cave, but this is hardly mentioned in the revision and deserves attention. To quote the Robbins et al.'s (2021) discussion section 6.3:

"We acknowledge that additional data is required in order to confidently assess the relative timing of the Dragon's Back collapse and entry of H. naledi. Nonetheless, the stratigraphic and geochronologic observations presented here, together with those previously published (Dirks et al., 2017) are consistent with the following scenario. Prior to the collapse of the Dragon's Back, sometime before 241 ka (new minimum age for H. naledi from RS68), the cave could be entered by H. naledi via a shaft in the roof of the Postbox Chamber. From there H. naledi could walk along a straight passage that follows a gently descending, SW trending fracture into the Dragon's Back Chamber and, with the Dragon's Back block still attached to the roof, would have only needed to climb over a ~5 m high sill to access the Dinaledi Subsystem behind it. This sill and narrow fracture system behind the Dragon's Back block would have been a major impediment to any flood waters and most other fauna into the Dinaledi Subsystem, but it would have been a more accessible route than that today."

The paper's conclusion continues, "The new dates further constrain the minimum age of H. naledi to 241 ka. Thus, H. naledi entered the subsystem between 241 ka and 335 ka, during a glacial period, when clastic sediment along the access route into the Dinaledi Subsystem experienced erosion. H. naledi would have probably entered the cave in the same way as the clastic sediments did, through an opening in the roof of the Postbox Chamber and may have entered via the Dragon's Back Chamber by climbing a 5 m high sill and passing below the Dragon's Back Block that was then still attached to the roof, to enter the Dinaledi Subsystem. In this context it is important to emphasize that it was not the Dragon's Back Block that prevented high-energy transport of coarse siliciclastic sediment from the Dragon's Back Chamber into the Dinaledi Subsystem, but rather the in situ floor block in the back wall of the Dragon's Back Chamber, against which the Dragon's Back Block slumped after it fell." This conclusion is very different from the complex pathway suggested by Berger et al. Martinón-Torres et al., 2023 also requested elaboration on this point in their reply by stating, "Moreover, recent studies by the Rising Star Cave team also point to a possible different and easier accesses for H. naledi into the fossil-bearing cave chambers than the current restricted access chute used by the research team, making clear that the degree of accessibility remains an open question (Robbins et al., 2021). Based on extensive dating studies of speleothem, this research (Robbins et al., 2021) implies that prior to 241 ka and the collapse of the Dragon's Back block hominins and other species could have more easily entered the cave via the Post Box Chamber and beneath the Dragon's Back Block before it fell. This gives access to a series of rifts that allow easier entry to the Dinaledi and other chambers beyond the present-day chute."

Because this paper introduces very different sets of possibilities, it seems impossible to derive an understanding of the processes that occurred 335-241 ka throughout the cave system without going into detail on these other openings, especially openings that are hypothesized to have been used by the hominins in question.

The world cares deeply about the H. naledi hominins and their story. I hope that in the coming years these issues are addressed, and perhaps other independent teams are allowed to do a full analysis since science is about replication. In any case, the excavation team has contributed important fossils to paleoanthropology.

Literature cited:

• Foecke, Kimberly K., Queffelec, Alain, & Pickering, Robyn. (2023). No Sedimentological Evidence for Deliberate Burial by Homo naledi - A Case Study Highlighting the Need for Best Practices in Geochemical Studies Within Archaeology and Paleoanthropology. PaleoAnthropology, 2024.

• Martinón-Torres, M., Garate, D., Herries, A. I. R., & Petraglia, M. D. (2023). No scientific evidence that Homo naledi buried their dead and produced rock art. Journal of Human Evolution, 103464. https://doi.org/10.1016/j.jhevol.2023.103464

• Robbins, J. L., Dirks, P. H. G. M., Roberts, E. M., Kramers, J. D., Makhubela, T. V., HilbertWolf, H. L., Elliott, M., Wiersma, J. P., Placzek, C. J., Evans, M., & Berger, L. R. (2021). Providing context to the Homo naledi fossils: Constraints from flowstones on the age of sediment deposits in Rising Star Cave, South Africa. Chemical Geology, 567, 120108. https://doi.org/10.1016/j.chemgeo.2021.120108

Reviewer #2 (Public review):

Before providing my review of the revised version of this study by Berger et al., which explores potential deliberate burials of Homo naledi within the Rising Star Cave System, I would like to briefly summarize the key points from my previous review of the earlier version (in 2023). Summarizing my previous review will provide context for assessing how effectively the revised study addresses the concerns I raised previously (in 2023).

In my earlier comments, I highlighted significant methodological and analytical shortcomings that, in my view, undermined the authors' claim of intentional burials by Homo naledi. While the study presented detailed geological and fossil data, I found the evidence for intentional burials unconvincing due to insufficient application of archaeothanatological principles and other methodological gaps.

My key concerns included:

(1) The absence of a comprehensive archaeothanatological analysis, particularly with respect to taphonomic changes, bone articulations, and displacement patterns such as the collapse of sediments and bone remains into voids created by decomposition.

(2) Missing or unclear illustrations of bone arrangements, which are critical for interpreting burial positions and processes.

(3) A lack of detailed discussion on the sequence of decomposition, joint disarticulation, sediment infill, and secondary bone displacement.

To convincingly support claims of deliberate burial, I argued that the study must reconstruct the timeline and processes surrounding death and deposition while clearly distinguishing natural taphonomic changes from intentional human actions. I emphasized the importance of integrating established archaeothanatological frameworks, such as those outlined by Duday et al. or Boulestin et al., to provide the necessary analytical rigor.

I will now explain how the revised version of this study has successfully addressed all the concerns raised in my previous review and why I now think that the authors provide sufficient evidence for the presence of "repeated and patterned" deliberate burials (referred to as "cultural burials" by the authors) by Homo naledi within the Rising Star Cave System.

In their revised manuscript, the authors have implemented substantial improvements in methodology, analytical depth, and overall presentation, which have effectively resolved the critical issues I previously highlighted. These revisions greatly strengthen their argument for intentional funerary practices. Importantly, the authors remain cautious in their interpretation of the evidence, explicitly refraining from inferring "symbolic" behavior or complex cognitive motivations behind these burials. Instead, they focus on presenting clear evidence for deliberate, patterned practices while leaving the broader implications for Homo naledi's cultural and cognitive capacities open for further investigation. This cautious approach adds to the credibility of their conclusions and avoids overextending the interpretation of the data.

The authors' enhanced application of archaeothanatological principles now offers a more comprehensive and convincing interpretation of the burial features. Key gaps in the earlier version, such as the absence of detailed reconstructions of taphonomic processes, bone articulations, and displacement patterns, have been addressed with thorough analyses and clearer illustrations. The study also now includes a well-structured timeline of events surrounding death and deposition, demonstrating an improved ability to differentiate between natural processes and deliberate human actions. These additions lend greater clarity and rigor to the evidence, making the argument for intentional burials both robust and persuasive.

Furthermore, the revised study presents detailed data on skeletal arrangements, decomposition sequences, and spatial patterns. This information is now relatively well illustrated and contextualized, enabling readers to better understand the complex processes involved in these burial practices. Importantly, the authors provide a stronger theoretical framework, integrating established archaeothanatological methodologies and taphonomic studies that situate their findings within broader archaeological and anthropological discussions of funerary behavior.

That being said, there remain relatively minor issues that could be refined further. Addressing these would help ensure the study is as clear and accessible as possible to the reader. Such adjustments would enhance the overall readability and reinforce the study's impact within the scientific community.

A - Suggested changes:

While the revised version of this study marks a significant improvement, successfully addresses my previous major concerns and provides a convincing argument for deliberate burials by Homo naledi, I believe that including both one summary table + one summary figure for each of the three main locations and the-Hill Antechamber, and Dinaledi Chamber (Feature 1 and Puzzle Box)-would further enhance the clarity and accessibility of the findings. Such tables and figures would serve as a valuable reference, allowing readers to more easily follow how the detailed patterns observed at each site fit the criteria for distinguishing intentional from natural processes.

The summary tables should consolidate key information for each location, such as:

(1) Bone articulations: A comprehensive list of articulated skeletal elements, categorized by their anatomical relationships (e.g., labile vs. stable articulations).

(2) Displacement patterns: Documentation of any spatial shifts in bone positions, noting directions and extents of disarticulation.

(3) Sequence of decomposition: Observations regarding the sequence of decomposition, joint disarticulation and associated changes in bone arrangements.

(4) Sediment interaction: Notes on sediment infill and its timing relative to decomposition, including evidence of secondary voids or delayed sediment deposition.

(5) Distinguishing criteria: Clear indications of how each observed pattern supports intentional burial (e.g., structured placement, lack of natural transport mechanisms) versus natural processes (e.g., random dispersal, sediment-driven bone displacement).<br /> Including such tables would not only summarize the complex taphonomic and archaeothanatological data but also allow readers to quickly assess how the evidence supports the authors' conclusions. This approach would bridge the gap between the detailed narrative descriptions and the criteria necessary to differentiate deliberate funerary practices from natural occurrences.

To streamline the main text further, many of the detailed descriptions of individual bones, specific displacement measurements, and other intricate observations could be moved to the supplementary data. This reorganization would maintain the richness of the data for those who wish to explore it in depth, while the summary tables would present the key findings concisely in the main text. This balance between accessibility and detail would ensure that the study appeals to both specialists requiring comprehensive data and readers looking for an overarching understanding of the findings.

In addition to these structural changes, it is crucial to ensure that evidence is consistently illustrated throughout the text.

Importantly the skeletal part representation is provided for Dinaledi Feature 1 in Figure 14, but similar data is not presented for the other burial features, such as those in the Hill Antechamber or Puzzle Box. This inconsistency could make it more challenging for readers to compare the features and fully appreciate the patterns of burial behavior across the different locations. Ensuring that similar types of evidence and analyses are presented uniformly for all features would strengthen the study and make its conclusions more cohesive and compelling.

Adding supplementary figures to represent the skeletal part distribution (as in Figure 14) within each excavated area (i.e., not only for Dinaledi Feature 1 but also for Hill Antechamber and Puzzle Box) would significantly enhance the study's clarity and accessibility. These figures could provide a visual summary of skeletal part representation, allowing readers to easily understand the nature of human remains within each burial context.

Specifically, such figures could:

(1) Illustrate Skeletal Part Representation: By visually mapping the presence and location of various skeletal elements, the figures would make it easier for readers to assess the completeness and arrangement of remains in each feature. This is particularly important for interpreting patterns of bone articulation and disarticulation.<br /> For example, it is quite challenging to determine the exact number and characteristics of the human skeletal remains identified within the Puzzle Box and those recovered through the "subsurface collection" in its surrounding area. The authors state that "at least six individuals" were identified in this area (during "subsurface collection") but provide no further clarification. They simply mention that "most elements" were described previously, without specifying which elements or where this prior description can be found.

(2) Highlight Articulations and Displacements: Figures could indicate which bones are articulated and their relative positions, as well as the spatial distribution of disarticulated elements. This would provide a clear visual context to support interpretations of taphonomic processes.

(3) Facilitate Comparisons Across Locations: By presenting skeletal part representation consistently for each location, the figures would enable readers to directly compare features, reinforcing the argument for "repeated and patterned" behavior.

(4) Simplify Complex Data: Instead of relying solely on textual descriptions, the visual format would allow readers to quickly grasp the key findings, making the study more accessible to a broader audience

By including such figures alongside the proposed summary tables in the main text, the study would achieve a balance between detailed narrative descriptions and concise, visual representation of the data. This approach would strengthen the overall presentation and support the authors' conclusions effectively.

Again, by presenting the data in a structured and comparative format, the new tables + figures could also highlight the differences and similarities between the three locations. This would reinforce the argument for "repeated and patterned" behavior, as the tables would make it easier to observe consistent burial practices across different contexts within the Rising Star Cave System.

Adding these summary tables + figures, ensuring consistent presentation of evidence, and reallocating detailed descriptions to supplementary materials would not require significant new analysis. However, these organizational adjustments would greatly enhance the study's clarity, readability, and overall impact.

B - A few additional changes are needed:

Figure 8: This figure is critical but lacks clarity. Specifically:

Panels 8a-c suffer from low contrast, making details difficult to discern.<br /> Panel 8d (sediment profile) is too small and lacks annotations that would aid interpretation.<br /> Figure S7: While this figure has significantly better contrast than Figures 8a-c, I am unable to identify the "articulated foot ... at right of frame," as mentioned in the caption. Please clarify this by adding annotations directly to the figure.

Page 4, 2nd paragraph: In the sentence "Researchers thus have diverse opinions about how to test whether ...," the word "opinions" should be replaced with a more precise term, such as "approaches."

C - In conclusion, I am impressed by the significant effort and meticulous work that has gone into this revised version of the study. The quality of the new evidence presented is commendable, and the findings now convincingly demonstrate not only clear evidence of intentional burial practices by Homo naledi but also compelling indications of post-depositional reworking. These advancements reflect a major improvement in the study's analytical rigor and the robustness of its conclusions, making it a valuable contribution to the understanding of early hominin funerary behavior.

Author response:

[The following is the authors’ response to the original reviews.]

We extend our sincere thanks to the editor, referees for eLife, and other commentators who have written evaluations of this manuscript, either in whole or in part. Sources of these comments were highly varied, including within the bioRxiv preprint server, social media (including many comments received on X/Twitter and some YouTube presentations and interviews), comments made by colleagues to journalists, and also some reviews of the work published in other academic journals. Some of these are formal and referenced with citations. Others were informal but nonetheless expressed perspectives that helped enable us to revise the manuscript with the inclusion of broader perspectives than the formal review process. It is beyond the scope of this summary to list every one of these, which have often been brought to the attention of different coauthors, but we begin by acknowledging the very wide array of peer and public commentary that have contributed to this work. The reaction speaks to a broad interest in open discussion and review of preprints. 

As we compiled this summary of changes to the manuscript, we recognized that many colleagues made comments about the process of preprint dissemination and evaluation rather than the data or analyses in the manuscript. Addressing such comments is outside the scope of this revised manuscript, but we do feel that a broader discussion of these comments would be valuable in another venue. Many commentators have expressed confusion about the eLife system of evaluation of preprints, which differs from the editorial acceptance or rejection practiced in most academic journals. As authors in many different nations, in varied fields, and in varied career stages, we ourselves are still working to understand how the academic publication landscape is changing, and how best to prepare work for new models of evaluation and dissemination. 

The manuscript and coauthor list reflect an interdisciplinary collaboration. Analyses presented in the manuscript come from a wide range of scientific disciplines. These range from skeletal inventory, morphology, and description, spatial taphonomy, analysis of bone fracture patterns and bone surface modifications, sedimentology, geochemistry, and traditional survey and mapping. The manuscript additionally draws upon a large number of previous studies of the Rising Star cave system and the Dinaledi Subsystem, which have shaped our current work. No analysis within any one area of research stands alone within this body of work: all are interpreted in conjunction with the outcomes of other analyses and data from other areas of research. Any single analysis in isolation might be consistent with many different hypotheses for the formation of sediments and disposition of the skeletal remains. But testing a hypothesis requires considering all data in combination and not leaving out data that do not fit the hypothesis. We highlight this general principle at the outset because a number of the comments from referees and outside specialists have presented alternative hypotheses that may arguably be consistent with one kind of analysis that we have presented, while seeming to overlook other analyses, data, or previous work that exclude these alternatives. In our revision, we have expanded all sections describing results to consider not only the results of each analysis, but how the combination of data from different kinds of analysis relate to hypotheses for the deposition and subsequent history of the Homo naledi remains. We address some specific examples and how we have responded to these in our summary of changes below. 

General organization:

The referee and editor comments are mostly general and not line-by-line questions, and we have compiled them and treated them as a group in this summary of changes, except where specifically noted. 

The editorial comments on the previous version included the suggestion that the manuscript should be reorganized to test “natural” (i.e. noncultural) hypotheses for the situations that we examine. The editorial comment suggested this as a “null hypothesis” testing approach. Some outside comments also viewed noncultural deposition as a null hypothesis to be rejected. We do not concur that noncultural processes should be construed as a null hypothesis, as we discuss further below. However, because of the clear editorial opinion we elected to revise the manuscript to make more explicit how the data and analyses test noncultural depositional hypotheses first, followed by testing of cultural hypotheses. This reorganization means that the revised manuscript now examines each hypothesis separately in turn. 

Taking this approach resulted in a substantial reorganization of the “Results” section of the manuscript. The “Results” section now begins with summaries of analyses and data conducted on material from each excavation area. After the presentation of data and analyses from each area, we then present a separate section for each of several hypotheses for the disposition and sedimentary context of the remains. These hypotheses include deposition of bodies upon a talus (as hypothesized in some previous work), slow sedimentary burial on a cave floor or within a natural depression, rapid burial by gravity-driven slumping, and burial of naturally mummified remains. We then include sections to test the hypothesis of primary cultural burial and secondary cultural burial. This approach adds substantial length to the Results. While some elements may be repeated across sections, we do consider the new version to be easier to take piece by piece for a reader trying to understand how each hypothesis relates to the evidence. 

The Results section includes analyses on several different excavation areas within the Dinaledi Subsystem. Each of these presents somewhat different patterns of data. We conceived of this manuscript combining these distinct areas because each of them provides information about the formation history of the Homo naledi-associated sediments and the deposition of the Homo naledi remains. Together they speak more strongly than separately. In the previous version of the manuscript, two areas of excavation were considered in detail (Dinaledi Feature 1 and the Hill Antechamber Feature), with a third area (the Puzzle Box area) included only in the Discussion and with reference to prior work. We now describe the new work undertaken after the 2013-2014 excavations in more detail. This includes an overview of areas in the Hill Antechamber and Dinaledi Chamber that have not yielded substantial H. naledi remains and that thereby help contextualize the spatial concentration of H. naledi skeletal material. The most substantial change in the data presented is a much expanded reanalysis of the Puzzle Box area. This reanalysis provides greater clarity on how previously published descriptions relate to the new evidence. The reanalysis also provides the data to integrate the detailed information on bone identification fragmentation, and spatial taphonomy from this area with the new excavation results from the other areas. 

In addition to Results, the reorganization also affected the manuscript’s Introduction section. Where the previous version led directly from a brief review of Pleistocene burial into the description of the results, this revised manuscript now includes a review of previous studies of the Rising Star cave system. This review directly addresses referee comments that express some hesitation to accept previous results concerning the structure and formation of sediments, the accessibility of the Dinaledi Subsystem, the geochronological setting of the H. naledi remains, and the relation of the Dinaledi Subsystem to nearby cave areas. Some parts of this overview are further expanded in the Supplementary Information to enable readers to dive more deeply into the previous literature on the site formation and geological configuration of the Rising Star cave system without needing to digest the entirety of the cited sources. 

The Discussion section of the revised manuscript is differentiated from Results and focuses on several areas where the evidence presented in this study may benefit from greater context. One new section addresses hypothesis testing and parsimony for Pleistocene burial evidence, which we address at greater length in this summary below. The majority of the Discussion concerns the criteria for recognizing evidence for burial as applied in other studies. In this research we employ a minimal definition but other researchers have applied varied criteria. We consider whether these other criteria have relevance in light of our observations and whether they are essential to the recognition of burial evidence more broadly. 

Vocabulary:

We introduce the term “cultural burial” in this revised manuscript to refer to the burial of dead bodies as a mortuary practice. “Burial” as an unmodified term may refer to the passive covering of remains by sedimentary processes. Use of the term “intentional burial” would raise the question of interpreting intent, which we do not presume based on the evidence presented in this research. The relevant question in this case is whether the process of burial reflects repeated behavior by a group. As we received input from various colleagues it became clear that burial itself is a highly loaded term. In particular there is a common assumption within the literature and among professionals that burial must by definition be symbolic. We do not take any position on that question in this manuscript, and it is our hope that the term “cultural burial” may focus the conversation around the extent that the behavioral evidence is repeated and patterned. 

Sedimentology and geochemistry of Dinaledi Feature 1:

Reviewer 4 provided detailed comments on the sedimentological and geochemical context that we report in the manuscript. One outside review (Foecke et al. 2024) included some of the points raised by reviewer 4, and additionally addressed the reporting of geochemical and sedimentological data in previous work that we cite. 

To address these comments we have revised the sedimentary context and micromorphology of sediments associated with Dinaledi Feature 1. In the new text we demonstrate the lack of microstratigraphy (supported by grain size analysis) in the unlithified mud clast breccia (UMCB), while such a microstratigraphy is observed in the laminated orange-red mudstones (LORM) that contribute clasts to the UMCB. Thus, we emphasize the presence and importance of a laterally continuous layer of LORM nature occurring at a level that appears to be the maximum depth of fossil occurrence. This layer is severely broken under extensive accumulation of fossils such as Feature 1 and only evidenced by abundant LORM clasts within and around the fossils. 

We have completely reworked the geochemical context associated with Feature 1 following the comments of reviewer 4. We described the variations and trends observed in the major oxides separate from trace and rare-earth elements. We used Harker variations plots to assess relationships between these element groups with CaO and Zn, followed by principal component analysis of all elements analyzed. The new geochemical analysis clearly shows that Feature 1 is associated with localized trace element signatures that exist in the sediments only in association with the fossil bones, which suggests lack of postdepositional mobilization of the fossils and sediments. We additionally have included a fuller description of XRF methods. 

To clarify the relation of all results to the features described in this study, we removed the geochemical and sedimentological samples from other sites within the Dinaledi Subsystem. These localities within the fissure network represent only surface collection of sediment, as no excavation results are available from those sites to allow for comparison in the context of assessing evidence of burial. These were initially included for comparison, but have now been removed to avoid confusion.  

Micromorphology of sediments:

Some referees (1, 3, and 4) and other commentators (including Martinón-Torres et al. 2024) have suggested that the previous manuscript was deficient due to an insufficient inclusion of micromorphological analysis of sediments. Because these commentators have emphasized this kind of evidence as particularly important, we review here what we have included and how our revision has addressed this comment. Previous work in the Dinaledi Chamber (Dirks et al., 2015; 2017) included thin section illustrations and analysis of sediment facies, including sediments in direct association with H. naledi remains within the Puzzle Box area. The previous work by Wiersma and coworkers (2020) used micromorphological analysis as one of several approaches to test the formation history of Unit 3 sediments in the Dinaledi Subsystem, leading to the interpretation of autobrecciation of earlier Unit 1 sediment. In the previous version of this manuscript we provided citations to this earlier work. The previous manuscript also provided new thin section illustrations of Unit 3 sediment near Dinaledi Feature 1 to place the disrupted layer of orange sediment (now designated the laminated orange silty mudstone unit) into context. 

In the new revised manuscript we have added to this information in three ways. First, as noted above in response to reviewer 4, we have revised and added to our discussion of micromorphology within and adjacent to the Dinaledi Feature 1. Second, we have included more discussion in the Supplementary Information of previous descriptions of sediment facies and associated thin section analysis, with illustrations from prior work (CC-BY licensed) brought into this paper as supplementary figures, so that readers can examine these without following the citations. Third, we have included Figure 10 in the manuscript which includes six panels with microtomographic sections from the Hill Antechamber Feature. This figure illustrates the consistency of sub-unit 3b sediment in direct contact with H. naledi skeletal material, including anatomically associated skeletal elements, with previous analyses that demonstrate the angular outlines and chaotic orientations of LORM clasts. It also shows density contrasts of sediment in immediate contact with some skeletal elements, the loose texture of this sediment with air-filled voids, and apparent invertebrate burrowing activity. To our knowledge this is the first application of microtomography to sediment structure in association with a Pleistocene burial feature. 

To forestall possible comments that the revised manuscript does not sufficiently employ micromorphological observations, or that any one particular approach to micromorphology is the standard, we present here some context from related studies of evidence from other research groups working at varied sites in Africa, Europe, and Asia. Hodgkins et al. (2021) noted: “Only a handful of micromorphological studies have been conducted on human burials and even fewer have been conducted on suspected burials from Paleolithic or hunter-gatherer contexts.” In that study, one supplementary figure with four photomicrographs of thin sections of sediments was presented. Interpretation of the evidence for a burial pit by Hodgkins et al. (2021) noted the more open microstructure of sediment but otherwise did not rely upon the thin section data in characterizing the sediments associated with grave fill. Martinón-Torres et al. (2021) included one Extended Data figure illustrating thin sections of sediments and bone, with two panels showing sediments (the remainder showing bone histology). The micromorphological analysis presented in the supplementary information of that paper was restricted to description of two microfacies associated with the proposed “pit” in that study. That study did carry out microCT scanning of the partially-prepared skeletal remains but did not report any sediment analysis from the microtomographic results. Maloney et al. (2022) reported no micromorphological or thin section analysis. Pomeroy et al. (2020a) included one illustration of a thin section; this study may be regarded as a preliminary account rather than a full description of the work undertaken. Goldberg et al. (2017) analyzed the geoarchaeology of the Roc de Marsal deposits in which possible burial-associated sediments had been fully excavated in the 1960s, providing new morphological assessments of sediment facies; the supplementary information to this work included five scans (not microscans) of sediment thin sections and no microphotographs. Fewlass et al. (2023) presented no thin section or micromorphological illustrations or methods. In summary of this research, we note that in one case micromorphological study provided observations that contributed to the evidence for a pit, in other cases micromorphological data did not test this hypothesis, and many researchers do not apply micromorphological techniques in their particular contexts. 

Sediment micromorphology is a growing area of research and may have much to provide to the understanding of ancient burial evidence as its standards continue to develop (Pomeroy et al. 2020b). In particular microtomographic analysis of sediments, as we have initiated in this study, may open new horizons that are not possible with more destructive thin-section preparation. In this manuscript, the thin section data reveals valuable evidence about the disruption of sediment structure by features within the Dinaledi Chamber, and microtomographic analysis further documents that the Hill Antechamber Feature reflects similar processes, in addition to possible post-burial diagenesis and invertebrate activity. Following up in detail on these processes will require further analysis outside the scope of this manuscript. 

Access into the Dinaledi Subsystem:

Reviewer 1 emphasizes the difficulty of access into the Dinaledi Subsystem as a reason why the burial hypothesis is not parsimonious. Similar comments have been made by several outside commentators who question whether past accessibility into the Dinaledi Subsystem may at one time have been substantially different from the situation documented in previous work. Several pieces of evidence are relevant to these questions and we have included some discussion of them in the Introduction, and additionally include a section in the Supplementary Information (“Entrances to the cave system”) to provide additional context for these questions. Homo naledi remains are found not only within the Dinaledi Subsystem but also in other parts of the cave system including the Lesedi Chamber, which is similarly difficult for non-expert cavers to access. The body plan, mass, and specific morphology of H. naledi suggest that this species would be vastly more suited to moving and climbing within narrow underground passages than living people. On this basis it is not unparsimonious to suggest that the evidence resulted from H. naledi activity within these spaces. We note that the accessibility of the subsystem is not strictly relevant to the hypothesis of cultural burial, although the location of the remains does inform the overall context which may reflect a selection of a location perceived as special in some way. 

Stuffing bodies down the entry to the subsystem:

Reviewer 3 suggests that one explanation for the emplacement of articulated remains at the top of the sloping floor of the Hill Antechamber is that bodies were “stuffed” into the chute that comprises the entry point of the subsystem and passively buried by additional accumulation of remains. This was one hypothesis presented in earlier work (Dirks et al. 2015) and considered there as a minimal explanation because it did not entail the entry of H. naledi individuals into the subsystem. The further exploration (Elliott et al. 2021) and ongoing survey work, as well as this manuscript, all have resulted in data that rejects this hypothesis. The revised manuscript includes a section in the results “Deposition upon a talus with passive burial” that examines this hypothesis in light of the data. 

Recognition of pits:

Referee 3 and 4 and several additional commentators have emphasized that the recognition of pit features is necessary to the hypothesis of burial, and questioned whether the data presented in the manuscript were sufficient to demonstrate that pits were present. We have revised the manuscript in several ways to clarify how all the different kinds of evidence from the subsystem test the hypothesis that pits were present. This includes the presentation of a minimal definition of burial to include a pit dug by hominins, criteria for recognizing that a pit was present, and an evaluation of the evidence in each case to make clear how the evidence relates to the presence of a pit and subsequent infill. As referee 3 notes, it can be challenging to recognize a pit when sediment is relatively homogeneous. This point was emphasized in the review by Pomeroy and coworkers (2020b), who reflected on the difficulty seeing evidence for shallow pits constructed by hominins, and we have cited this in the main text. As a result, the evidence for pits has been a recurrent topic of debate for most Pleistocene burial sites. However in addition to the sedimentological and contextual evidence in the cases we describe, the current version also reflects upon other possible mechanisms for the accumulation of bones or bodies. The data show that the sedimentary fill associated with the H. naledi remains in the cases we examine could not have passively accumulated slowly and is not indicative of mass movement by slumping or other high-energy flow. To further put these results into context, we added a section to the Discussion that briefly reviews prior work on distinguishing pits in Pleistocene burial contexts, including the substantial number of sites with accepted burial evidence for which no evidence of a pit is present. 

Extent of articulation and anatomical association:

We have added significantly greater detail to the descriptions of articulated remains and orientation of remains in order to describe more specifically the configuration of the skeletal material. We also provide 14 figures in main text (13 of them new) to illustrate the configuration of skeletal remains in our data. For the Puzzle Box area, this now includes substantial evidence on the individuation of skeletal fragments, which enables us to illustrate the spatial configuration of remains associated with the DH7 partial skeleton, as well as the spatial position of fragments refitted as part of the DH1, DH2, DH3, and DH4 crania. For Dinaledi Feature 1 and the Hill Antechamber Feature we now provide figures that key skeletal parts as identified, including material that is unexcavated where possible, and a skeletal part representation figure for elements excavated from Dinaledi Feature 1. 

Archaeothanatology:

Reviewer 2 suggests that a greater focus on the archaeothanatology literature would be helpful to the analysis, with specific reference to the sequence of joint disarticulation, the collapse of sediment and remains into voids created by decomposition, and associated fragmentation of the remains. In the revised manuscript we have provided additional analysis of the Hill Antechamber Feature with this approach in mind. This includes greater detail and illustration of our current hypothesis for individuation of elements. We now discuss a hypothesis of body disposition, describe the persistent joints and articulation of elements, and examine likely decomposition scenarios associated with these remains. Additionally, we expand our description and illustration of the orientation of remains and degree of anatomical association and articulation within Dinaledi Feature 1. For this feature and for the Hill Antechamber Feature we have revised the text to describe how fracturing and crushing patterns are consistent with downward pressure from overlying sediment and material. In these features, postdepositional fracturing occurred subsequent to the decomposition of soft tissue and partial loss of organic integrity of the bone. We also indicate that the loss by postdepositional processes of most long bone epiphyses, vertebral bodies, and other portions of the skeleton less rich in cortical bone, poses a challenge for testing the anatomical associations of the remaining elements. This is a primary reason why we have taken a conservative approach to identification of elements and possible associations. 

A further aspect of the site revealed by our analysis is the selective reworking of sediments within the Puzzle Box area subsequent to the primary deposition of some bodies. The skeletal evidence from this area includes body parts with elements in anatomical association or articulation, juxtaposed closely with bone fragments at varied pitch and orientation. This complexity of events evidenced within this area is a challenge for approaches that have been developed primarily based on comparative data from single-burial situations. In these discussions we deepen our use of references as suggested by the referee.   

Burial positions:

Reviewer 2 further suggests that illustrations of hypothesized burial positions would be valuable. We recognize that a hypothesized burial position may be an appealing illustration, and that some recent studies have created such illustrations in the context of their scientific articles. However such illustrations generally include a great deal of speculation and artist imagination, and tend to have an emotive character. We have added more discussion to the manuscript of possible primary disposition in the case of the Hill Antechamber Feature as discussed above. We have not created new illustrations of hypothesized burial positions for this revision. 

Carnivore involvement:

Referee 1 suggests that the manuscript should provide further consideration of whether carnivore activity may have introduced bones or bodies into the cave system. The reorganized Introduction now includes a review of previous work, and an expanded discussion within the Supplementary Information (“Hypotheses tested in previous work”). This includes a review of literature on the topic of carnivore accumulation and the evidence from the Dinaledi and Lesedi Chamber that rejects this hypothesis. 

Water transport and mud:

The eLife referees broadly accepted previous work showing that water inundation or mass flow of water-saturated sediment did not occur within the history of Unit 2 and 3 sediments, including those associated with H. naledi remains. However several outside commentators did refer specifically to water flow or mud flow as a mechanism for slumping of deposits and possible sedimentary covering of the remains. To address these comments we have added a section to the

Supplementary Information (“Description of the sedimentary deposits of the Dinaledi Subsystem”) that reviews previous work on the sedimentary units and formation processes documented in this area. We also include a subsection specifically discussing the term “mud” as used in the description of the sedimentology within the system, as this term has clearly been confusing for nonspecialists who have read and commented on the work. We appreciate the referees’ attention to the previous work and its terminology.  

Redescription of areas of the cave system:

Reviewer 1 suggests that a detailed reanalysis of all portions of the cave system in and around the Dinaledi Subsystem is warranted to reject the hypothesis that bodies entered the space passively and were scattered from the floor by natural (i.e. noncultural) processes. The referee suggests that National Geographic could help us with these efforts. To address this comment we have made several changes to the manuscript. As noted above, we have added material in Supplementary Information to review the geochronology of the Dinaledi Subsystem and nearby Dragon’s Back Chamber, together with a discussion of the connections between these spaces. 

Most directly in response to this comment we provide additional documentation of the possibility of movement of bodies or body parts by gravity within the subsystem itself. This includes detailed floor maps based on photogrammetry and LIDAR measurement, where these are physically possible, presented in Figures 2 and 3. In some parts of the subsystem the necessary equipment cannot be used due to the extremely confined spaces, and for these areas our maps are based on traditional survey methods. In addition to plan maps we have included a figure showing the elevation of the subsystem floor in a cross-section that includes key excavation areas, showing their relative elevation. All figures that illustrate excavation areas are now keyed to their location with reference to a subsystem plan. These data have been provided in previous publications but the visualization in the revised manuscript should make the relationship of areas clear for readers. The Introduction now includes text that discusses the configuration of the Hill Antechamber, Dinaledi Chamber, and nearby areas, and also discusses the instances in which gravity-driven movement may be possible, at the same time reviewing that gravity-driven movement from the entry point of the subsystem to most of the localities with hominin skeletal remains is not possible. 

Within the Results, we have added a section on the relationship of features to their surroundings in order to assist readers in understanding the context of these bone-bearing areas and the evidence this context brings to the hypothesis in question. We have also included within this new section a discussion of the discrete nature of these features, a question that has been raised by outside commentators. 

Passive sedimentation upon a cave floor or within a natural depression:

Reviewer 3 suggests that the situation in the Dinaledi Subsystem may be similar to a European cave where a cave bear skeleton might remain articulated on a cave floor (or we can add, within a hollow for hibernation), later to be covered in sediment. The reviewer suggests that articulation is therefore no evidence of burial, and suggests that further documentation of disarticulation processes is essential to demonstrating the processes that buried the remains. We concur that articulation by itself is not sufficient evidence of cultural burial. To address this comment we have included a section in the Results that tests the hypothesis that bodies were exposed upon the cave floor or within a natural depression. To a considerable degree, additional data about disarticulation processes subsequent to deposition are provided in our reanalysis of the Puzzle Box area, including evidence for selective reworking of material after burial. 

Postdepositional movement and floor drains:

Reviewer 3 notes that previous work has suggested that subsurface floor drains may have caused some postdepositional movement of skeletal remains. The hypothesis of postdepositional slumping or downslope movement has also been discussed by some external commentators (including Martinón-Torres et al. 2024). We have addressed this question in several places within the revised manuscript. As we now review, previous discussion of floor drains attempted to explain the subvertical orientation of many skeletal elements excavated from the Puzzle Box area. The arrangement of these bones reflects reworking as described in our previous work, and without considering the possibility of reworking by hominins, one mechanism that conceivably might cause reworking was downward movement of sediments into subsurface drains. Further exploration and mapping, combined with additional excavation into the sediments beneath the Puzzle Box area provided more information relevant to this hypothesis. In particular this evidence shows that subsurface drains cannot explain the arrangement of skeletal material observed within the Puzzle Box area. As now discussed in the text, the reworking is selective and initiated from above rather than below. This is best explained by hominin activity subsequent to burial. 

In a new section of the Results we discuss slumping as a hypothesis for the deposition of the remains. This includes discussion of downslope movement within the Hill Antechamber and the idea that floor drains may have been a mechanism for sediment reworking in and around the Puzzle Box area and Dinaledi Feature 1. As described in this section the evidence does not support these hypotheses. 

Hypothesis testing and parsimony:

Referees 1 and 3 and the editorial guidance all suggested that a more appropriate presentation would adopt a null hypothesis and test it. The specific suggestion that the null hypothesis should be a natural sedimentary process of deposition was provided not only by these reviewers but also by some outside commentators. To address this comment, we have edited the manuscript in two ways. The first is the addition of a section to the Discussion that specifically discusses hypothesis testing and parsimony as related to Pleistocene evidence of cultural burial. This includes a brief synopsis of recent disciplinary conversations and citation of work by other groups of authors, none of whom adopted this “null hypothesis” approach in their published work. 

As we now describe in the manuscript, previous work on the Dinaledi evidence never assumed any role for H. naledi in the burial of remains. Reading the reviewer reports caused us to realize that this previous work had followed exactly the “null hypothesis” approach that some suggested we follow. By following this null hypothesis approach, we neglected a valuable avenue of investigation. In retrospect, we see how this approach impeded us from understanding the pattern of evidence within the Puzzle Box area. Thus in the revised manuscript we have mentioned this history within the Discussion and also presented more of the background to our previous work in the Introduction. Hopefully by including this discussion of these issues, the manuscript will broaden conversation about the relation of parsimony to these issues. 

Language and presentation style:

Reviewer 4 criticizes our presentation, suggesting that the text “gives the impression that a hypothesis was formulated before data were collected.” Other outside commentators have mentioned this notion also, including Martinón-Torres et al. (2024) who suggest that the study began from a preferred hypothesis and gathered data to support it. The accurate communication of results and hypotheses in a scientific article is a broader issue than this one study. Preferences about presentation style vary across fields of study as well as across languages. We do not regret using plain language where possible. In any study that combines data and methods from different scientific disciplines, the use of plain language is particularly important to avoid misunderstandings where terms may mean different things in different fields. 

The essential question raised by these comments is whether it is appropriate to present the results of a study in terms of the hypothesis that is best supported. As noted above, we read carefully many recent studies of Pleistocene burial evidence. We note that in each of these studies that concluded that burial is the best hypothesis, the authors framed their results in the same way as our previous manuscript: an introduction that briefly reviews background evidence for treatment of the dead, a presentation of results focused on how each analysis supports the hypothesis of burial for the case, and then in some (but not all) cases discussion of why some alternative hypotheses could be rejected. We do not infer from this that these other studies started from a presupposition and collected data only to confirm it. Rather, this is a simple matter of presentation style. 

The alternative to this approach is to present an exhaustive list of possible hypotheses and to describe how the data relate to each of them, at the end selecting the best. This is the approach that we have followed in the revised manuscript, as described above under the direction of the reviewer and editorial guidance. This approach has the advantage of bringing together evidence in different combinations to show how each data point rejects some hypotheses while supporting others. It has the disadvantage of length and repetition. 

Possible artifact:

We have chosen to keep the description of the possible artifact associated with the Hill Antechamber Feature in the Supplementary Information. We do this while acknowledging that this is against the opinion of reviewer 4, who felt the description should be removed unless the object in question is fully excavated and physically analyzed. The previous version of the manuscript did not rely upon the stone as positive evidence of grave goods or symbolic content, and it noted that the data do not test whether the possible artifact was placed or was intentionally modified. However this did not satisfy reviewer 4, and some outside commentators likewise asserted that the object must be a “geofact” and that it should be removed. 

We have three arguments against this line of thinking. First, we do not omit data from our reporting. Whether Homo naledi shaped the rock or not, used it as a tool or not, whether the rock was placed with the body or not, it is unquestionably there. Omitting this one object from the report would be simply dishonest. Second, the data on this rock are at 16 micron resolution. While physical inspection of its surface may eventually reveal trace evidence and will enable better characterization of the raw material, no mode of surface scanning will produce better evidence about the object’s shape. Third, the position of this possible artifact within the feature provides significant information about the deposition of the skeletal material and associated sediments. The pitch, orientation, and position of the stone is not consistent with slow deposition but are consistent with the hypothesis that the surrounding sediment was rapidly emplaced at the same time as the articulated elements less than 2 cm away. 

In the current version, we have redoubled our efforts to provide information about the position and shape of this stone while not presupposing the intentionality of its shape or placement. We add here that the attitude expressed by referee 4 and other commentators, if followed at other sites, would certainly lead to the loss or underreporting of evidence, which we are trying to avoid.  

Consistency versus variability of behavior:

As described in the revised manuscript, different features within the Dinaledi Subsystem exhibit some shared characteristics. At the same time, they vary in positioning, representation of individuals and extent of commingling. Other localities within the subsystem and broader cave system present different evidence. Some commentators have questioned whether the patterning is consistent with a single common explanation, or whether multiple explanations are necessary. To address this line of questioning, we have added several elements to the manuscript. We created a new section on secondary cultural burial, discussing whether any of the situations may reflect this practice. In the Discussion, we briefly review the ways in which the different features support the involvement of H. naledi without interpreting anything about the intentionality or meaning of the behavior. We further added a section to the Discussion to consider whether variation among the features reflects variation in mortuary practices by H. naledi. One aspect of this section briefly cites variation in the location and treatment of skeletal remains at other sites with evidence of burial. 

Grave goods:

Some commentators have argued that grave goods are a necessary criterion for recognizing evidence of ancient burial. We added a section to the Discussion to review evidence of grave goods at other Pleistocene sites where burial is accepted. 

References:

• Dirks, P. H., Berger, L. R., Roberts, E. M., Kramers, J. D., Hawks, J., Randolph-Quinney, P. S., Elliott, M., Musiba, C. M., Churchill, S. E., de Ruiter, D. J., Schmid, P., Backwell, L. R., Belyanin, G. A., Boshoff, P., Hunter, K. L., Feuerriegel, E. M., Gurtov, A., Harrison, J. du G., Hunter, R., … Tucker, S. (2015). Geological and taphonomic context for the new hominin species Homo naledi from the Dinaledi Chamber, South Africa. eLife, 4, e09561. https://doi.org/10.7554/eLife.09561

• Dirks, P. H., Roberts, E. M., Hilbert-Wolf, H., Kramers, J. D., Hawks, J., Dosseto, A., Duval, M., Elliott, M., Evans, M., Grün, R., Hellstrom, J., Herries, A. I., Joannes-Boyau, R., Makhubela, T. V., Placzek, C. J., Robbins, J., Spandler, C., Wiersma, J., Woodhead, J., & Berger, L. R. (2017). The age of Homo naledi and associated sediments in the Rising Star Cave, South Africa. eLife, 6, e24231. https://doi.org/10.7554/eLife.24231

• Elliott, M., Makhubela, T., Brophy, J., Churchill, S., Peixotto, B., FEUERRIEGEL, E., Morris, H., Van Rooyen, D., Ramalepa, M., Tsikoane, M., Kruger, A., Spandler, C., Kramers, J., Roberts, E., Dirks, P., Hawks, J., & Berger, L. R. (2021). Expanded Explorations of the Dinaledi Subsystem,Rising Star Cave System, South Africa. PaleoAnthropology, 2021(1), 15–22. https://doi.org/10.48738/2021.iss1.68

• Fewlass, H., Zavala, E. I., Fagault, Y., Tuna, T., Bard, E., Hublin, J.-J., Hajdinjak, M., & Wilczyński, J. (2023). Chronological and genetic analysis of an Upper Palaeolithic female infant burial from Borsuka Cave, Poland. iScience, 26(12). https://doi.org/10.1016/j.isci.2023.108283

• Foecke, Kimberly K., Queffelec, Alain, & Pickering, Robyn. (n.d.). No Sedimentological Evidence for Deliberate Burial by Homo naledi – A Case Study Highlighting the Need for Best Practices in Geochemical Studies Within Archaeology and Paleoanthropology. PaleoAnthropology, 2024. https://doi.org/10.48738/202x.issx.xxx

• Goldberg, P., Aldeias, V., Dibble, H., McPherron, S., Sandgathe, D., & Turq, A. (2017). Testing the Roc de Marsal Neandertal “Burial” with Geoarchaeology. Archaeological and Anthropological Sciences, 9(6), 1005–1015. https://doi.org/10.1007/s12520-013-0163-2

• Maloney, T. R., Dilkes-Hall, I. E., Vlok, M., Oktaviana, A. A., Setiawan, P., Priyatno, A. A. D., Ririmasse, M., Geria, I. M., Effendy, M. A. R., Istiawan, B., Atmoko, F. T., Adhityatama, S., Moffat, I., Joannes-Boyau, R., Brumm, A., & Aubert, M. (2022). Surgical amputation of a limb 31,000 years ago in Borneo. Nature, 609(7927), 547–551. https://doi.org/10.1038/s41586-022-05160-8

• Martinón-Torres, M., d’Errico, F., Santos, E., Álvaro Gallo, A., Amano, N., Archer, W., Armitage, S. J., Arsuaga, J. L., Bermúdez de Castro, J. M., Blinkhorn, J., Crowther, A., Douka, K., Dubernet, S., Faulkner, P., Fernández-Colón, P., Kourampas, N., González García, J., Larreina, D., Le Bourdonnec, F.-X., … Petraglia, M. D. (2021). Earliest known human burial in Africa. Nature, 593(7857), Article 7857. https://doi.org/10.1038/s41586021-03457-8

• Martinón-Torres, M., Garate, D., Herries, A. I. R., & Petraglia, M. D. (2023). No scientific evidence that Homo naledi buried their dead and produced rock art. Journal of Human Evolution, 103464. https://doi.org/10.1016/j.jhevol.2023.103464

• Pomeroy, E., Bennett, P., Hunt, C. O., Reynolds, T., Farr, L., Frouin, M., Holman, J., Lane, R., French, C., & Barker, G. (2020a). New Neanderthal remains associated with the ‘flower burial’ at Shanidar Cave. Antiquity, 94(373), 11–26. https://doi.org/10.15184/aqy.2019.207

• Pomeroy, E., Hunt, C. O., Reynolds, T., Abdulmutalb, D., Asouti, E., Bennett, P., Bosch, M., Burke, A., Farr, L., Foley, R., French, C., Frumkin, A., Goldberg, P., Hill, E., Kabukcu, C., Lahr, M. M., Lane, R., Marean, C., Maureille, B., … Barker, G. (2020b). Issues of theory and method in the analysis of Paleolithic mortuary behavior: A view from Shanidar Cave. Evolutionary Anthropology: Issues, News, and Reviews, 29(5), 263–279. https://doi.org/10.1002/evan.21854

• Robbins, J. L., Dirks, P. H. G. M., Roberts, E. M., Kramers, J. D., Makhubela, T. V., HilbertWolf, H. L., Elliott, M., Wiersma, J. P., Placzek, C. J., Evans, M., & Berger, L. R. (2021). Providing context to the Homo naledi fossils: Constraints from flowstones on the age of sediment deposits in Rising Star Cave, South Africa. Chemical Geology, 567, 120108. https://doi.org/10.1016/j.chemgeo.2021.120108

• Wiersma, J. P., Roberts, E. M., & Dirks, P. H. G. M. (2020). Formation of mud clast breccias and the process of sedimentary autobrecciation in the hominin-bearing (Homo naledi) Rising Star Cave system, South Africa. Sedimentology, 67(2), 897–919. https://doi.org/10.1111/sed.12666

eLife Assessment

The songbird vocal motor nucleus HVC contains cells that project to the basal ganglia, the auditory system, or to downstream vocal motor structures. In this fundamental study, the authors conduct optogenetic circuit mapping to clarify how four distinct inputs to HVC act on these distinct HVC cell types. They provide compelling evidence that all long range projections engage inhibitory circuits in HVC and can also exhibit cell-type specific preferences in monosynaptic input strength. Understanding HVC at this microcircuit level is critical for constraining models of song learning and production.

Reviewer #1 (Public review):

Summary:

This work has crated the map of synaptic connectivity between the inputs and outputs of song premotor nucleus, HVC in zebra finches to understand how sensory (auditory) to motor circuit interact to coordinate song production and learning. The authors optimized the optogenetic technique via AAV to manipulate auditory inputs from a specific auditory area one-by-one and recorded synaptic activity from a neuron in HVC with whole-cell recording from slice preparation with identification of projection area by retrograde neuronal tracing. These thorough and detailed analysis provide compelling evidence of synaptic connections between 4 major auditory inputs (3 forebrain and 1 thalamic regions) within three projection neurons in the HVC; all areas give monosynaptic excitatory inputs and polysynaptic inhibitory inputs, but proportions of projection to each projection neuron varied. They also find specific reciprocal connections between mMAN and Av. Taken together the authors provide the map of synaptic connection between intercortical sensory to motor areas which is suggested to be involved in zebra finch song production and learning.

Strengths:

The authors optimized optogenetical tools with eGtACR1 by using AAV which allow them to manipulate synaptic inputs in a projection-specific manner in zebra finches. They also identify HVC cell type based on projection area. With their technical advance and thorough experiments, they provided detailed map synaptic connection and gave insights into the neuronal circuit for auditory guided vocal (motor) learning.

Weaknesses:

As this study is in adult brain slices, there might be a gap to the functions in developmental song learning.

Reviewer #2 (Public review):

Summary:

The manuscript describes synaptic connectivity in Songbird cortex four main classes of sensory neurons afferents onto three known classes of projection neurons of the pre-motor cortical region HVC. HVC is a region associated with the generation of learned bird song. Investigators here use all male zebra finches to examine the functional anatomy of this region using patch clamp methods combined with optogenetic activation of select neuronal groups.

Strengths:

The quality of the recordings is extremely high and the quantity of data is on a very significant scale, this will certainly aid the field.

Weaknesses:

Could make the figures a little easier to navigate by having some atlas drawings.

Comments on revisions:

The authors have addressed the minor concerns and suggestions

Reviewer #3 (Public review):

Nucleus HVC is critical both for song production as well as learning and arguably, sitting at the top of the song control system, is the most critical node in this circuit receiving a multitude of inputs and sending precisely timed commands that determine the temporal structure of song. The complexity of this structure and its underlying organization seem to become more apparent with each experimental manipulation, and yet our understanding of the underlying circuit organization remains relatively poorly understood. In this study, Trusel and Roberts use classic whole-cell patch clamp techniques in brain slices coupled with optogenetic stimulation of select inputs to provide a careful characterization and quantification of synaptic inputs into HVC. By identifying individual projections neurons using retrograde tracer injections combined with pharmacological manipulations, they classify monosynaptic inputs onto each of the three main classes of glutamatergic projection neurons in HVC (RA-, Area X- and Av-projecting neurons). This study is remarkable in the amount of information that it generates, and the tremendous labor involved for each experiment, from the expression of opsins in each of the target inputs (Uva, NIf, mMAN and Av), the retrograde labelling of each type of projection neuron, and ultimately the optical stimulation of infected axons while recording from identified projection neurons. Taken together, this study makes an important contribution to increasing our identification, and ultimately understanding, of the basic synaptic elements that make up the circuit organization of HVC, and how external inputs, which we know to be critical for song production and learning, contribute to the intrinsic computations within this critic circuit.

This study is impressive in its scope, rigorous in its implementation and thoughtful regarding its limitations. The manuscript is well written, and I appreciate the clarity with which the authors use our latest understanding of the evolutionary origins of this circuit to place these studies within a larger context and their relevance to the study of vocal control, including human speech. My comments are minor and primarily about legibility, clarification of certain manipulations and organization of some of the summary figures.

Comments on revisions:

The authors have done a very nice job addressing the reviewers' comments.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

This work tried to map the synaptic connectivity between the inputs and outputs of the song premotor nucleus, HVC in zebra finches to understand how sensory (auditory) to motor circuits interact to coordinate song production and learning. The authors optimized the optogenetic technique via AAV to manipulate auditory inputs from a specific auditory area one-by-one and recorded synaptic activity from a neuron with whole-cell recording from slice preparation with identification of the projection area by retrograde neuronal tracing. This thorough and detailed analysis provides compelling evidence of synaptic connections between 4 major auditory inputs (3 forebrain and 1 thalamic region) within three projection neurons in the HVC; all areas give monosynaptic excitatory inputs and polysynaptic inhibitory inputs, but proportions of projection to each projection neuron varied. They also find specific reciprocal connections between mMAN and Av. Taken together the authors provide the map of the synaptic connection between intercortical sensory to motor areas which is suggested to be involved in zebra finch song production and learning.

Strengths:

The authors optimized optogenetic tools with eGtACR1 by using AAV which allow them to manipulate synaptic inputs in a projection-specific manner in zebra finches. They also identify HVC cell types based on projection area. With their technical advance and thorough experiments, they provided detailed map synaptic connections.

Weaknesses:

As it is the study in brain slice, the functional implication of synaptic connectivity is limited. Especially as all the experiments were done in the adult preparation, there could be a gap in discussing the functions of developmental song learning.

We thank the reviewer for their appreciation of our work. Although we agree that there can be limitations to brain slice preparations, the approaches used here for synaptic connectivity mapping are well-designed to identify long-range synaptic connectivity patterns. Optogenetic stimulation of axon terminals in brain slices does not require intact axons and works well when axons are cut, allowing identification of all inputs expressing optogenetic channels from aXerent regions. Terminal stimulation in slices yields stable post-synaptic responses for hours without rundown, assuring that polysynaptic and monosynaptic connections can be reliably identified in our brain slices.  Additionally, conducting similar types of experiments in vivo can run into important limitations. First, the extent of TTX and 4-AP diXusion, which is necessary for identification of long-range monosynaptic connections, can be diXicult to verify in vivo - potentially confounding identification of monosynaptic connectivity.  Second, conducting whole-cell patch-clamp experiments in vivo, particularly in deeper brain regions, is technically challenging, and would limit the number of cells that can be patched and increase the number of animals needed. 

We agree that there may well be important diXerences between adult connectivity and connectivity patterns in the juvenile brain. Indeed, learning and experience during development almost certainly shape connectivity patterns and these patterns of connectivity may change incrementally and/or dynamically during development. Ultimately, adult connectivity patterns are the result of changes in the brain that accrue over development. Given that this is the first study mapping long-range connectivity of HVC input-output pathways, we reasoned that the adult connectivity would provide a critical reference allowing future studies to map diXerent stages of juvenile connectivity and the changes in connectivity driven by milestones like forming a tutor song memory, sensorimotor learning, and song crystallization.

In this revision we worked to better highlight the points raised above and thank the reviewer for their comments.

Reviewer #2 (Public review):

Summary:

The manuscript describes synaptic connectivity in the Songbird cortex's four main classes of sensory neuron aXerents onto three known classes of projection neurons of the pre-motor cortical region HVC. HVC is a region associated with the generation of learned bird songs. Investigators here use all male zebra finches to examine the functional anatomy of this region using patch clamp methods combined with optogenetic activation of select neuronal groups.

Strengths:

The quality of the recordings is extremely high and the quantity of data is on a very significant scale, this will certainly aid the field.

Weaknesses:

The authors could make the figures a little easier to navigate. Most of the figures use actual anatomical images but it would be nice to have this linked with a zebra finch atlas in more of a cartoon format that accompanied each fluro image. Additionally, for the most part, figures showing the labeling lack scale bar values (in um). These should be added not just shown in the legends.

The authors could make it clear in the abstract that this is all male zebra finches - perhaps this is obvious given the bird song focus, but it should be stated. The number of recordings from each neuron class and the overall number of birds employed should be clearly stated in the methods (this is in the figures, but it should say n=birds or cells as appropriate).

The authors should consider sharing the actual electrophysiology records as data.

We thank the reviewer for their assessment of our research and suggestions. We have implemented many of these suggestions and provide details in our response to their specific Recommendations. Additionally, we are organizing our data and will make it publicly available with the version of record.

Reviewer #3 (Public review):

Nucleus HVC is critical both for song production as well as learning and arguably, sitting at the top of the song control system, is the most critical node in this circuit receiving a multitude of inputs and sending precisely timed commands that determine the temporal structure of song. The complexity of this structure and its underlying organization seem to become more apparent with each experimental manipulation, and yet our understanding of the underlying circuit organization remains relatively poorly understood. In this study, Trusel and Roberts use classic whole-cell patch clamp techniques in brain slices coupled with optogenetic stimulation of select inputs to provide a careful characterization and quantification of synaptic inputs into HVC. By identifying individual projection neurons using retrograde tracer injections combined with pharmacological manipulations, they classify monosynaptic inputs onto each of the three main classes of glutamatergic projection neurons in HVC (RA-, Area X- and Av-projecting neurons). This study is remarkable in the amount of information that it generates, and the tremendous labor involved for each experiment, from the expression of opsins in each of the target inputs (Uva, NIf, mMAN, and Av), the retrograde labelling of each type of projection neuron, and ultimately the optical stimulation of infected axons while recording from identified projection neurons. Taken together, this study makes an important contribution to increasing our identification, and ultimately understanding, of the basic synaptic elements that make up the circuit organization of HVC, and how external inputs, which we know to be critical for song production and learning, contribute to the intrinsic computations within this critic circuit.

This study is impressive in its scope, rigorous in its implementation, and thoughtful regarding its limitations. The manuscript is well-written, and I appreciate the clarity with which the authors use our latest understanding of the evolutionary origins of this circuit to place these studies within a larger context and their relevance to the study of vocal control, including human speech. My comments are minor and primarily about legibility, clarification of certain manipulations, and organization of some of the summary figures.

We thank the reviewer for their thoughtful assessment of our research.

Recommendations for the authors:

The following recommendations were considered by all reviewers to be important to incorporate for improving this paper:

(1) Clarify the site of viral injection and the possibility of labeling other structures a) Show images of viral injection sites.

We provide a representative image of viral expression for each pathway studied in this manuscript. Please see panel A in Figures 2-3 and 5-6 showing our viral expression in Uva, NIf, mMAN, and Av respectively.  

b) Include in discussion caveats that the virus may spread beyond the boundaries of structures (e.g. especially injections into NIF could spread into Field L).

For each HVC aXerent nucleus we have now included a sentence describing the possible spread of viral infection in surrounding structures in the Results. We also now expanded the image from the Av section to include NIf, to showcase lack of viral expression in NIf (see Fig. 6A).

(2) Clarify the logic and precise methods of the TTX and 4-AP experiments

a) Please see the detailed issue raised by Reviewer 3, Major Point 1 below.

The TTX and 4AP application is the gold-standard of opsin-assisted synaptic circuit interrogation, pioneered by the Svoboda lab in 2009 (Petreanu, Mao et al. 2009) and widely used to assess monosynaptic connectivity in multiple brain circuits, as summarized in a recent review(Linders, Supiot et al. 2022). We now better describe the logic of this approach in the second paragraph of the Results section and cite the first description of this method from the Svoboda lab and a recent review weighing this method with other optogenetic methods for tracing synaptic connections in the brain.

(3) Include caveats in discussion

a) Note that there may be other inputs to HVC that were not examined in this study (e.g. CMM, Field L)

In our original manuscript we did state “Although a complete description of HVC circuitry will require the examination of other potential inputs (i.e. RA_HVC PNs, A11 glutamatergic neurons(Roberts, Klein et al. 2008, Ben-Tov, Duarte et al. 2023)) and a characterization of interneuron synaptic connectivity, here we provide a map of the synaptic connections between the 4 best described aPerents to HVC and its 3 populations of projection neurons” in the last paragraph of the Discussion. We have now edited this sentence to include the projection from NCM to HVC and cited Louder et al., 2024.

We have extensively mapped input pathways to HVC, and consistent with Vates (Vates, Broome et al. 1996) we have not found evidence that Field L projects to HVC. Rather that it projects to the shelf region outside of HVC. Consistent with this, we do not see retrogradely labeled neurons in Field L following tracer injections confined to HVC (see Fig. 3G). Additionally, we find that CM projections to HVC arise from the nucleus Avalanche (Roberts, Hisey et al. 2017) which we specifically examine in this study. We do not dispute that there may be other pathways projecting to HVC that will need to be examined in the future, including known projections from neuromodulatory regions and RA, from developmentally restricted pathway(s) like NCM (Louder, Kuroda et al. 2024), and from yet unidentified pathways.

b) Also note that birds in this study were adults and that some inputs to HVC likely to be important for learning may recede during development (e.g. Louder et al, 2024).

In the second to last paragraph of the Discussion we now state: While our opsin-assisted circuit mapping provides us with a new level of insight into HVC synaptic circuitry, there are limitations to this research that should be considered. All circuit mapping in this study was carried out in brain slices from adult male zebra finches. Future studies will be needed to examine how this adult connectivity pattern relates to patterns of connectivity in juveniles during sensory or sensorimotor phases of vocal learning and connectivity patterns in female birds.   

(4) Consider cosmetic changes to figures as suggested by Reviewers 2-3 below.

We thank the reviewers for their suggestions and have implemented the changes as best we can.

(5) Address all minor issues raised below.

Reviewer #1 (Recommendations for the authors):

I see this study is well designed to answer the author's specific question, mapping synaptic auditorymotor connections within HVC. Their experiments with advanced techniques of projection-specific optogenetic manipulation of synaptic inputs and retrograde identification of projection areas revealed input-output combination selective synaptic mapping.

As I found this study advanced our knowledge with the compelling dataset, I have only some minor comments here.

(1) One technical concern is we don't see how much the virus infection was focused on the target area and if we can ignore the eXect of synaptic connectivity from surrounding areas. As the amount of virus they injected is large (1.5ul) and target areas are small, we assume the virus might spread to the surrounding area, such as field L which also projects to HVC when targeting Nif. While I think the majority of the projections were from their target areas, it would be better to mention (also the images with larger view areas) the possibilities of projections of surrounding areas.

We agree with the reviewer about the concern about specificity of viral expression. For this reason, we included sample images of the viral expression in each target area (panel A in Fig. 2,3,5,6). We have now also included a sentence at the beginning of each subsection of our Result to describe how we have ensured interpretability of the results. Uva and mMAN’s surrounding areas are not known to project to HVC. Possible cross-infection is an issue for Av and NIf, and we checked each bird’s injection site to ensure that eGtACR1+ cells were not visible in the unintended HVC-projecting areas.

As mentioned in our response the public comment, consistent with Vates (Vates, Broome et al. 1996) we do not see evidence that Field L projects directly to HVC (see Fig. 3G).

(2) Another concern about the technical issue is the damage to axonal projections. While I understand the authors stimulated axonal terminals axonal projections were assumed to be cut and their ability to release neurotransmitters would be reduced especially after long-term survival or repeated stimulation. Mentioning whether projection pathways were within their 230um-thick slice (probably depends on input sites) or not and the eXect of axonal cut would be helpful.

We agree that slice electrophysiology has limitations. However, we disagree with the claim of reduced reliability or stability of the evoked response. We and others find that electrical and optogenetic repeated terminal stimulation in slices can yield stable post-synaptic responses for tens of minutes and even hours (Bliss and Gardner-Medwin 1973, Bliss and Lomo 1973, Liu, Kurotani et al. 2004, Pastalkova, Serrano et al. 2006, Xu, Yu et al. 2009, Trusel, Cavaccini et al. 2015, Trusel, Nuno-Perez et al. 2019). Indeed, long-term synaptic plasticity experiments in most preparations and across brain areas rely on such stability of the presynaptic machinery for synaptic release, despite axons being severed from their parent soma. Our assumption is the vast majority, if not all, connections between axon terminals and their cell body in the aXerent regions have been cut in our preparations. Nonetheless, the diversity of outcomes we report (currents returning after TTX+4AP or not, depending on the specific combination of input and HVCPN class) is consistent with the robustness of the synaptic interrogation method. 

(3) While I understand this study focused on 4 major input areas and the authors provide good pictures of synaptic HVC connections from those areas, HVC has been reported to receive auditory inputs from other areas as well (CMM, FieldL, etc.). It is worth mentioning that there are other auditory inputs and would be interesting to discuss coordination with the inputs from other areas.

We have extensively mapped input pathways to HVC, and consistent with Vates (Vates, Broome et al. 1996) we have not found evidence that Field L projects to HVC. Rather that it projects to the shelf region outside of HVC. Consistent with this, we do not see retrogradely labeled neurons in Field L following tracer injections confined to HVC (see Fig. 3G). Additionally, we find that CM projections to HVC arise from the nucleus Avalanche (Roberts, Hisey et al. 2017) which we specifically examine in this study. We do not dispute that there may be other pathways projecting to HVC that will need to be examined in the future, including known projections from neuromodulatory regions and RA, from developmentally restricted pathway(s) like NCM (Louder, Kuroda et al. 2024), and from yet unidentified pathways.

(4) The HVC local neuronal connections have been reported to be modified and a recent study revealed the transient auditory inputs into HVC during song learning period. The author discusses the functions of HVC synaptic connections on song learning (also title says synaptic connection for song learning), however, the experiments were done in adults and dp not discuss the possibility of diXerent synaptic connection mapping in juveniles in the song learning period. Mentioning the neuronal activities and connectivity changes during song learning is important. Also, it would be helpful for the readers to discuss the potential diXerences between juveniles/adults if they want to discuss the functions of song learning.

We now mention in the Discussion that this is an important caveat of our research and that future studies will be needed to examine how these adult connectivity patterns relate to connectivity patterns in juveniles during sensory or sensorimotor phases of vocal learning and connectivity patterns in female birds. Nonetheless, the title and abstract cite song learning because it is important for the broader public to understand that at least some of these aXerent brain regions carry an essential role in song learning (Foster and Bottjer 2001, Roberts, Gobes et al. 2012, Roberts, Hisey et al. 2017, Zhao, Garcia-Oscos et al. 2019, Koparkar, Warren et al. 2024).

Reviewer #2 (Recommendations for the authors):

The work is very detailed and will be an important resource to those working in the field. The recordings are of a high quality and lots of information is included such as measures of response kinetics amplitude and pharmacological confirmation of excitatory and inhibitory synaptic responses. In general, I feel the quality is extremely high and the quantity of data is on a very significant exhaustive scale that will certainly aid the field. I have come at this conclusion as a non zebra finch person but I feel the connection information shown will be of benefit given its high quality.

Figure 7 is a nice way of showing the overall organization. Optional suggestion, consider highlighting anything in Figure 7 that results in a new understanding of the song system as compared to previous work on anatomy and function.

We thank the reviewer for the kind comments about our research. We have highlighted our newly found connection between mMAN and Av and all the connections onto the HVC PNs in Panel B are newly identified in this study.

Reviewer #3 (Recommendations for the authors):

Major points

(1) Clarification regarding methods for determining monosynaptic events:

One of the manipulations that I struggled the most with was those describing the use of TTX + 4AP to isolate monosynaptic events. Initially, not being as familiar with the use of optically based photostimulation of axons to release transmitter locally, I was initially confused by statements such as "we found that oEPSC returned after application of TTX+4AP". This might be clear to someone performing these manipulations, but a bit more clarification would be helpful. Should I assume that an existing monosynaptic EPSC would be masked by co-occurring polysynaptic IPSCs which disappear following application of TTX + 4AP, thereby unmasking the monosynaptic EPSC, thereby causing the EPSC to "return"? A word that I am not sure works. Continuing my confusion with these experiments, I am unsure how this cocktail of drugs is added, if it is even added as a cocktail, which is what I initially assumed. The methods and the results are not so clear if they are added in sequence and why and if traces are recorded after the addition of both drugs or if they are recorded for TTX and then again for TTX + 4AP. Finally, looking at the traces in the experimental figures (e.g. Figures 2F, 3F, 5F, and 6F), it is diXicult to see what is being shown, at least for me. First, the authors need to describe better in the results why they stimulate twice in short succession and why they seem to use the response to the second pulse (unless I am mistaken) to measure the monosynaptic event. Second, I was confused by the traces (which are very small) in the presence of TTX. I would have expected to see a response if there was a monosynaptic EPSC but I only seem to see a flat line.  

The confusion that I list above might be due in part to my ignorance, but it is important in these types of papers not to assume too much expertise if you want readers with a less sophisticated understanding of synaptic physiology to understand the data. In other words, a little bit more clarity and hand-holding would be welcome.

We understand the reviewer’s confusion about the methodology.  In Voltage clamp, the amplifier injects current through the electrode maintaining the membrane voltage to -70mV, where the equilibrium potential for Cl- is near equilibrium, and therefore the only synaptic current evoked by light stimulation is due to cation influx, mainly through AMPA receptors (see Fig. 1).  Therefore, cooccurring polysynaptic IPSCs wouldn’t be visible. We examine those holding the membrane voltage at +10mV, see Fig. 1. TTX application suppresses V-dependent Na+ channels and therefore stops all neurotransmission. We show the traces upon TTX to show that currents we were recording prior to TTX application were of synaptic origin, and not due to accidental expression of opsin in the patched cell. Also, this ensures that any current visible after 4AP application is due to monosynaptic transmission and not to a failure of TTX application.

After recording and light stimulation with TTX, we then add 4AP, which is a blocker of presynaptic K+ channels. This prevents the repolarization of the terminals that would occur in response to opsinmediated local depolarization. 4AP application, therefore, allows local opsin-driven depolarizations to reach the threshold for Ca2+-dependent vesicle docking and release. This procedure selectively reveals or unmasks the monosynaptic currents because any non-monosynaptically connected neuron would still need V-dependent Na+ channels to eXectively produce indirect neurotransmission onto the patched cell. The TTX and 4AP application is the gold-standard of opsinassisted synaptic circuit interrogation, pioneered by the Svoboda lab in 2009 and widely used to assess monosynaptic connectivity in multiple brain circuits, as summarized in a recent review (Linders et al., 2022). We now include 2 more sentences near the beginning of the Results to clarify this process and directly point to the Linders review for researchers wanting a deeper explanation of this technique. 

The double stimulation is unrelated to our testing of monosynaptic connections. We originally conducted the experiments by delivering 2 pulses of light separated by 50ms, a common way to examine the pair-pulse ratio (PPR) – a physiological measure which is used to probe synapses for short-term plasticity and release probability. However, through discussions with colleagues we realized that the slow decay time of eGtACR1 may complicate interpretation of the response to the second light pulse. Thus, we elected to not report these results and indicated this in the Methods section:  “We calculated the paired-pulse ratio (PPR) as the amplitude of the second peak divided by the amplitude of the first peak elicited by the twin stimuli, however due to slow kinetics of eGtACR1 the results would be diPicult to interpret, and therefore we are not currently reporting them.” 

(2) Suggestions for improving summary figures:

Summary Figure 1a: The circuit diagram (schematic to the right of 1a) is OK but I initially found it a bit diXicult to interpret. For example, it is not clear why pink RA projecting neurons don't reach as far to the right as X or Av projecting neurons, suggesting that they are not really projection neurons. Also, the big question marks in the intermediate zone are not entirely intuitive. It seems there might be a better way of representing this. It might also be worth stating in the figure legend that the interconnectivity patterns shown in the figure between PNs in HVC are based on specific prior studies.

We thank the reviewer for the constructive criticism. We have modified the figure to extend the RA projection line and mentioned in the figure legend that connectivity between PNs is based on prior studies.

Summary Figure 1a: I am not sure I love this figure. There are a few minor issues. First, there are too many browns [Nif/AV and mMAN] which makes it more challenging to clearly disambiguate the diXerent projections. Second, it is unclear why this figure does not represent projections from RA to HVC. My biggest concern with this figure is that it oversimplifies some of the findings. From the figure, one gets the impression that Uva only projects to RA-PNs and that Av only projects to X-PNs even though the authors show connections to other PNs. With the small sample size in this current study for each projection and each PN type, one really cannot rule out that these "minority" projections are not important. I, therefore, suggest that the authors qualitatively represent the strength/probability of connections by weighting with thickness of aXerent connections.

We assume the reviewer is commenting on our summary figure panel 7B. We agree with the referee that this is a simplified representation of our findings. We had indeed indicated in the legend that this was just a “Schematic of the HVC aXerent connectivity map resulting from the present work” and that “For conceptualization purposes, aXerent connectivity to HVC-PNs is shown only when the rate of monosynaptic connectivity reaches 50% of neurons examined”. We have added a title to highlight that this is but a simplification. We have now adjusted the colors to make the figure easier to follow. Based on the reviewers critique we searched for a better method for summarizing the complex connectivity patterns described in this research. We settled on a Sankey diagram of connectivity. This is now Figure 7C. In this diagram, we are able to show the proportion of connections from each input pathway onto each class of neuron and if these connections are poly or monosynaptic. We find this to a straightforward way of displaying all of the connectivity patterns identified in our figure 2-3 and 4-5 look forward to understanding if the reviewers find this a useful way of illustrating our findings.

Minor points:

(1) Line 50 - typo - song circuits.

Thank you for catching this.

(2) Line 106 - 111 - The findings suggest that 100% of Uva projections onto HVCRA neurons are monosynaptic. However, because the authors only tested 6 neurons their statements that their findings are so diXerent from other studies, should be somewhat tempered since these other studies (e.g. Moll et al.) looked at 251 neurons in HVC and sampling bias could still somewhat explain the diXerence.

We observed oEPSCs in 43 of 51 (84.3%) HVC-RA neurons recorded (mean rise time = 2.4 ms) and monosynaptic connections onto 100% of the HVC-RA neurons tested (n = 6). Moll et al. combined electrical stimulation of Uva with two-photon calcium imaging (GCaMP6s) of putative HVC-RA neurons (n = 251 neurons). We should note that these are putative HVC-RA neurons because they were not visually identified using retrograde tracing or using some other molecular handle. They report that only ~16% of HVC-RA neurons showed reliable calcium responses following Uva stimulation. Although the experiments by Moll et al are technically impressive, calcium imaging is an insensitive technique for measuring post-synaptic responses, particularly subthreshold responses, when compared to whole-cell patch-clamp recordings. This approach cannot identify monosynaptic connections and is likely limited to only be sensitive suprathreshold activity that likely relies on recruitment of other polysynaptic inputs onto the neurons in HVC. Furthermore, as indicated in the Discussion, our opsin-mediated synaptic interrogation recruits any eGtACR1+ Uva terminal in the slice and therefore will have great likelihood of revealing any existing connections. 

A limitation of whole-cell patch-clamp recordings is that it is a laborious low throughput technique. Future experiments using better imaging approaches, like voltage imaging, may be able to weigh in on diXerences between what we report here using whole-cell patch-clamp recordings from visually identified HVC-RA neurons combined with optogenetic manipulations of Uva terminals and the calcium imaging results reported by Moll. Nonetheless, whole-cell patch-clamp recordings combined with optogenetic manipulations is likely to remain the most sensitive method for identifying synaptic connectivity.

(3) Figure 2G - the significance of white circles is not clear.

The figure legend indicates that those highlight and mark the position of “retrogradely labeled HVCprojecting neurons in Uva (cyan, white circles)” to facilitate identification of colocalization with the in-situ markers.

(4) Line 135 - Cardin et al. (J. Neurophys. 2004) is the first to show that song production does not require Nif.

We thank the reviewer pointing this out and we have cited this important study. 

(5) Line 183 - This is a confusing sentence because I initially thought that mMAN-mMANHVC PNs was a category!

We switched the dash with a colon.

(6) Figure 4d could use some arrows to identify what is shown. It is assumed that the box represents mMAN. Should it be assumed that Av is not in the plane of this section? If not, this should be stated in the legend. It is also unclear where the anterograde projections are. Is this the dork highway that goes from the box to the dorsal surface? If yes this should be indicated but it should also be made clear why the projections go both in the dorsal as well as the ventral directions.

The inset, as indicated by the lines around it, is a magnification of the terminal fields in Av. We added an explanation of the inset.

(7) Discussion. In the introduction, the authors mention projections from RA to HVC but never end up studying them in the current manuscript which seems like a missed opportunity and perhaps even a weakness of the study. In the discussion, it would certainly be good for the authors to at least discuss the possible significance of these projections and perhaps why they decided not to study them.

We thank the reviewer for the comment. Unfortunately, we couldn’t reliably evoke interpretable currents from RA, and we elected to publish the current version of the paper with these 4 major inputs. Nonetheless, we have indicated in the Introduction and in the Discussion that more inputs (e.g. RA, A11, NCM) remain to be evaluated. 

(8) Line 622 - Is this reference incomplete?

We thank the reviewer. We have corrected the reference.

• Ben-Tov, M., F. Duarte and R. Mooney (2023). "A neural hub for holistic courtship displays." Curr Biol 33(9): 1640-1653 e1645.

• Bliss, T. V. and A. R. Gardner-Medwin (1973). "Long-lasting potentiation of synaptic transmission in the dentate area of the unanaestetized rabbit following stimulation of the perforant path." J Physiol 232(2): 357-374.

• Bliss, T. V. and T. Lomo (1973). "Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path." J Physiol 232(2): 331-356.

• Foster, E. F. and S. W. Bottjer (2001). "Lesions of a telencephalic nucleus in male zebra finches: Influences on vocal behavior in juveniles and adults." J Neurobiol 46(2): 142-165.

• Koparkar, A., T. L. Warren, J. D. Charlesworth, S. Shin, M. S. Brainard and L. Veit (2024). "Lesions in a songbird vocal circuit increase variability in song syntax." Elife 13.

• Linders, L. E., L. F. Supiot, W. Du, R. D'Angelo, R. A. H. Adan, D. Riga and F. J. Meye (2022). "Studying Synaptic Connectivity and Strength with Optogenetics and Patch-Clamp Electrophysiology." Int J Mol Sci 23(19).

• Liu, H. N., T. Kurotani, M. Ren, K. Yamada, Y. Yoshimura and Y. Komatsu (2004). "Presynaptic activity and Ca2+ entry are required for the maintenance of NMDA receptor-independent LTP at visual cortical excitatory synapses." J Neurophysiol 92(2): 1077-1087.

• Louder, M. I. M., M. Kuroda, D. Taniguchi, J. A. Komorowska-Muller, Y. Morohashi, M. Takahashi, M. Sanchez-Valpuesta, K. Wada, Y. Okada, H. Hioki and Y. Yazaki-Sugiyama (2024). "Transient sensorimotor projections in the developmental song learning period." Cell Rep 43(5): 114196.

• Pastalkova, E., P. Serrano, D. Pinkhasova, E. Wallace, A. A. Fenton and T. C. Sacktor (2006). "Storage of spatial information by the maintenance mechanism of LTP." Science 313(5790): 1141-1144.

• Petreanu, L., T. Mao, S. M. Sternson and K. Svoboda (2009). "The subcellular organization of neocortical excitatory connections." Nature 457(7233): 1142-1145.

• Roberts, T. F., S. M. Gobes, M. Murugan, B. P. Olveczky and R. Mooney (2012). "Motor circuits are required to encode a sensory model for imitative learning." Nat Neurosci 15(10): 1454-1459.

• Roberts, T. F., E. Hisey, M. Tanaka, M. G. Kearney, G. Chattree, C. F. Yang, N. M. Shah and R. Mooney (2017). "Identification of a motor-to-auditory pathway important for vocal learning." Nat Neurosci 20(7): 978-986.

• Roberts, T. F., M. E. Klein, M. F. Kubke, J. M. Wild and R. Mooney (2008). "Telencephalic neurons monosynaptically link brainstem and forebrain premotor networks necessary for song." J Neurosci 28(13): 3479-3489.

• Trusel, M., A. Cavaccini, M. Gritti, B. Greco, P. P. Saintot, C. Nazzaro, M. Cerovic, I. Morella, R. Brambilla and R. Tonini (2015). "Coordinated Regulation of Synaptic Plasticity at Striatopallidal and Striatonigral Neurons Orchestrates Motor Control." Cell Rep 13(7): 1353-1365.

• Trusel, M., A. Nuno-Perez, S. Lecca, H. Harada, A. L. Lalive, M. Congiu, K. Takemoto, T. Takahashi, F. Ferraguti and M. Mameli (2019). "Punishment-Predictive Cues Guide Avoidance through Potentiation of Hypothalamus-to-Habenula Synapses." Neuron 102(1): 120-127.e124.

• Vates, G. E., B. M. Broome, C. V. Mello and F. Nottebohm (1996). "Auditory pathways of caudal telencephalon and their relation to the song system of adult male zebra finches." Journal of Comparative Neurology 366(4): 613-642.

• Xu, T., X. Yu, A. J. Perlik, W. F. Tobin, J. A. Zweig, K. Tennant, T. Jones and Y. Zuo (2009). "Rapid formation and selective stabilization of synapses for enduring motor memories." Nature 462(7275): 915-919.

• Zhao, W., F. Garcia-Oscos, D. Dinh and T. F. Roberts (2019). "Inception of memories that guide vocal learning in the songbird." Science 366: 83 - 89.

eLife Assessment

The study reports valuable findings from a very rich EEG-fMRI dataset, including 107 participants, which was collected during nocturnal naps. Using overall solid methods, the authors link activity in memory-related brain regions (e.g., hippocampus, thalamus, and medial prefrontal cortex), and their functional connectivity to the occurrence of canonical sleep rhythms (spindles and slow oscillations) in non-rapid eye movement sleep. This work will be of broad interest to sleep and memory researchers and beyond.

Reviewer #1 (Public review):

Wang et al., recorded concurrent EEG-fMRI in 107 participants during nocturnal NREM sleep to investigate brain activity and connectivity related to slow oscillations (SO), sleep spindles, and in particular their co-occurrence. The authors found SO-spindle coupling to be correlated with increased thalamic and hippocampal activity, and with increased functional connectivity from the hippocampus to the thalamus and from the thalamus to the neocortex, especially the medial prefrontal cortex (mPFC). They concluded the brain-wide activation pattern to resemble episodic memory processing, but to be dissociated from task-related processing and suggest that the thalamus plays a crucial role in coordinating the hippocampal-cortical dialogue during sleep.

The paper offers an impressively large and highly valuable dataset that provides the opportunity for gaining important new insights into the network substrate involved in SOs, spindles, and their coupling.

Comments on revisions:

While the authors have sufficiently addressed some of my previous comments, I still have severe concerns regarding several key aspects of the methodology, which were even corroborated by the supplementary results presented in response to the last round of reviews. I have the following specific comments (numbers refer to comments raised in the previous review):

Re 1: The revised introduction now cites a couple of papers but discusses them only very superficially, lumping together several studies with very different key results. This is stil not very informative for the reader and does not sufficiently acknowledge previously published work. Here are two examples to illustrate this:<br /> a. "These studies have generally reported that slow oscillations are associated with widespread cortical and subcortical BOLD changes, whereas spindles elicit activation in the thalamus, as well as in several cortical and paralimbic regions."  Several studies even showed e.g., a clear activation of the hippocampus and parahippocampal gyrus associated with spindles, not just the thalamus<br /> b. "Although these findings provide valuable insights into the BOLD correlates of sleep rhythms, they often do not employ sophisticated temporal modeling (Huang et al., 2024) [, ...]." - previous studies have used e.g., spindle event-related regressors with individual spindle amplitudes as parametric modulators, first and second order derivatives of the HRF function, as well as PPI connectivity analyses, which I would consider rather sophisticated temporal modelling.

Re 4+9: The short overall recordings in some subjects on the one hand and the large number of spindles and SOs detected in N1 sleep stages are still highly concerning, in fact even more so, now that the actual numbers have been provided in the Supplementary Tables. Either the sleep staging or the detection of SO and spindle events must be incorrect. I understand that for specific EEG analysis and fMRI modelling purposes sometimes slightly different thresholds are used as compared to clinical sleep staging, but several parameters here are alarmingly off.<br /> a. Given that proper NREM sleep (N2+N3) is the relevant stage for the analyses conducted in this paper, some of the N2+N3 durations are very short (eg 7-8 min) while those subjects' results have the same impact on the group level analyses as those with >100 min of N2+N3. Either subjects with very little relevant data (not overall recording time but N2+N3 time) should be excluded or weighting subject data for the group analyses according to the amount od contributed data should be done.<br /> b. The authors argue that the SO and spindle detection algorithms are valid since widely used and that they were developed for N2+N3 stages, which is why they will also detect events in other stages: "While, because the detection methods for SO and spindle are based on percentiles, this method will always detect a certain number of events when used for other stages (N1 and REM) sleep data, but the differences between these events and those detected in stage N23 remain unclear." I do agree that with very liberal thresholds, also SO and spindle vents may be detected in other stages, but it shouldn't be that many. If the percentiles of amplitude thresholds were defined based on properly scored N2+N3 stages only, very few events should be detected (erroneously!) in N1, as the occurrence of K-complexes (isolated SOs) and spindles per definition makes it N2, and during REM sleep only very few spindles and SOS are allowed to occur, without scoring it NREM instead. For the first subject (just as example, but with similar numbers for the rest of the sample), reveals as many as 60 SOs and 31 spindles within 8 min of N1 sleep (Table S2) as well as 13 SOs and 7 spindles within 2 min of REM sleep (Table S4). These numbers are completely unrealistic and question the correctness of the sleep staging as well as the physiological relevance of the EEG graphoelements identified as SO and spindles. It also completely undermines the interpretability of the respective event regressors for the fMRI analyses.<br /> c. Likely, given the large numbers of coupled SO-spindle events and the apparently very low amplitude criteria for event identification, also the number of SO-spindle couplings is likely severely overestimated.

Re 10: The rationale for using a lateralized frontal electrode (F3) for both SO (should have been at least bilateral or central) and spindle detection (should have been a centro-parietal electrode) is not convincing. Other EEG-fMRI spindle or SO papers have used a number of frontal (SO) or centro-parietal (spindles) electrodes averaged or even approaches including all EEG electrodes. Searching events with low thresholds at suboptimal recording sites does not dot this highly valuable dataset justice.

Re 7: It is not clear to me why/how larger voxels would reduce susceptibility-related distortions and partial volume effects. Usually, the opposite is true. This should be elaborated.

Reviewer #2 (Public review):

In this study, Wang and colleagues aimed to explore brain-wide activation patterns associated with NREM sleep oscillations, including slow oscillations (SOs), spindles, and SO-spindle coupling events. Their findings reveal that SO-spindle events corresponded with increased activation in both the thalamus and hippocampus. Additionally, they observed that SO-spindle coupling was linked to heightened functional connectivity from the hippocampus to the thalamus, and from the thalamus to the medial prefrontal cortex-three key regions involved in memory consolidation and episodic memory processes.

This study's findings are timely and highly relevant to the field. The authors' extensive data collection, involving 107 participants sleeping in an fMRI while undergoing simultaneous EEG recording, deserves special recognition. If shared, this unique dataset could lead to further valuable insights.

Comments on revisions:

The authors' efforts in revising the manuscript and addressing the reviewers' comments are certainly commendable. However, I remain concerned about potential issues in detecting sleep-related oscillations (SOs, spindles, and consequently coupled SO-spindle events), which may arise due to suboptimal parameter selection or inaccurate sleep staging, potentially impacting all subsequent analyses.

A review of Supplementary Tables 1-4 reveals an unusually high number of detected SOs and spindles during sleep stage N1 and REM sleep. While the authors correctly note that a percentile-based detection approach will always identify a certain number of events across sleep stages, the particularly high counts in N1 and REM are concerning. To mitigate the limitations of this method, the authors could have performed event detection independently of sleep stages (i.e., across the entire dataset for each participant) and subsequently assigned the detected events to the corresponding sleep stages. If the event counts in N1 and REM remained disproportionately high, this would indicate a fundamental issue with the detection procedure.

Reviewer #3 (Public review):

Summary:

Wang et al., examined the brain activity patterns during sleep, especially when locked to those canonical sleep rhythms such as SO, spindle, and their coupling. Analyzing data from a large sample, the authors found significant coupling between spindles and SOs, particularly during the up-state of the SO. Moreover, the authors examined the patterns of whole-brain activity locked to these sleep rhythms. The authors next investigated the functional connectivity analyses, and found enhanced connectivity between the hippocampus and the thalamus and the medial PFC. These results reinforced the theoretical model of sleep-dependent memory consolidation, such that SO-spindle coupling is conducive for systems-level memory reactivation and consolidation.

Strengths:

There are obvious strengths in this work, including the large sample size, state-of-the-art neuroimaging and neural oscillation analyses, and the richness of results. The results now inform hemodynamic neural activity that coincided with SO-spindle couplings.

Weaknesses:

My earlier comments were about the inability to make inferences on memory given the lack of memory tasks, and the weakness in using the open-ended cognitive state decoding.

The current revision has addressed these major concerns. The authors expanded discussions regarding the theoretical implications of the work in a more nuanced manner.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Wang et al., recorded concurrent EEG-fMRI in 107 participants during nocturnal NREM sleep to investigate brain activity and connectivity related to slow oscillations (SO), sleep spindles, and in particular their co-occurrence. The authors found SO-spindle coupling to be correlated with increased thalamic and hippocampal activity, and with increased functional connectivity from the hippocampus to the thalamus and from the thalamus to the neocortex, especially the medial prefrontal cortex (mPFC). They concluded the brain-wide activation pattern to resemble episodic memory processing, but to be dissociated from task-related processing and suggest that the thalamus plays a crucial role in coordinating the hippocampal-cortical dialogue during sleep.

The paper offers an impressively large and highly valuable dataset that provides the opportunity for gaining important new insights into the network substrate involved in SOs, spindles, and their coupling. However, the paper does unfortunately not exploit the full potential of this dataset with the analyses currently provided, and the interpretation of the results is often not backed up by the results presented. I have the following specific comments.

Thank you for your thoughtful and constructive feedback. We greatly appreciate your recognition of the strengths of our dataset and findings Below, we address your specific comments and provide responses to each point you raised to ensure our methods and results are as transparent and comprehensible as possible. We hope these revisions address your comments and further strengthen our manuscript. Thank you again for the constructive feedback.

(1) The introduction is lacking sufficient review of the already existing literature on EEG-fMRI during sleep and the BOLD-correlates of slow oscillations and spindles in particular (Laufs et al., 2007; Schabus et al., 2007; Horovitz et al., 2008; Laufs, 2008; Czisch et al., 2009; Picchioni et al., 2010; Spoormaker et al., 2010; Caporro et al., 2011; Bergmann et al., 2012; Hale et al., 2016; Fogel et al., 2017; Moehlman et al., 2018; Ilhan-Bayrakci et al., 2022). The few studies mentioned are not discussed in terms of the methods used or insights gained.

We acknowledge the need for a more comprehensive review of prior EEG-fMRI studies investigating BOLD correlates of slow oscillations and spindles. However, these articles are not all related to sleep SO or spindle. Articles (Hale et al., 2016; Horovitz et al., 2008; Laufs, 2008; Laufs, Walker, & Lund, 2007; Spoormaker et al., 2010) mainly focus on methodology for EEG-fMRI, sleep stages, or brain networks, which are not the focus of our study. Thank you again for your attention to the comprehensiveness of our literature review, and we will expand the introduction to include a more detailed discussion of the existing literature, ensuring that the contributions of previous EEG-fMRI sleep studies are adequately acknowledged.  

Introduction, Page 4 Lines 62-76

“Investigating these sleep-related neural processes in humans is challenging because it requires tracking transient sleep rhythms while simultaneously assessing their widespread brain activation. Recent advances in simultaneous EEG-fMRI techniques provide a unique opportunity to explore these processes. EEG allows for precise event-based detection of neural signal, while fMRI provides insight into the broader spatial patterns of brain activation and functional connectivity (Horovitz et al., 2008; Huang et al., 2024; Laufs, 2008; Laufs, Walker, & Lund, 2007; Schabus et al., 2007; Spoormaker et al., 2010). Previous EEG-fMRI studies on sleep have focused on classifying sleep stages or examining the neural correlates of specific waves (Bergmann et al., 2012; Caporro et al., 2012; Czisch et al., 2009; Fogel et al., 2017; Hale et al., 2016; Ilhan-Bayrakcı et al., 2022; Moehlman et al., 2019; Picchioni et al., 2011). These studies have generally reported that slow oscillations are associated with widespread cortical and subcortical BOLD changes, whereas spindles elicit activation in the thalamus, as well as in several cortical and paralimbic regions. Although these findings provide valuable insights into the BOLD correlates of sleep rhythms, they often do not employ sophisticated temporal modeling (Huang et al., 2024), to capture the dynamic interactions between different oscillatory events, e.g., the coupling between SOs and spindles.”

(2) The paper falls short in discussing the specific insights gained into the neurobiological substrate of the investigated slow oscillations, spindles, and their interactions. The validity of the inverse inference approach ("Open ended cognitive state decoding"), assuming certain cognitive functions to be related to these oscillations because of the brain regions/networks activated in temporal association with these events, is debatable at best. It is also unclear why eventually only episodic memory processing-like brain-wide activation is discussed further, despite the activity of 16 of 50 feature terms from the NeuroSynth v3 dataset were significant (episodic memory, declarative memory, working memory, task representation, language, learning, faces, visuospatial processing, category recognition, cognitive control, reading, cued attention, inhibition, and action).

Thank you for pointing this out, particularly regarding the use of inverse inference approaches such as “open-ended cognitive state decoding.” Given the concerns about the indirectness of this approach, we decided to remove its related content and results from Figure 3 in the main text and include it in Supplementary Figure 7. We will refocus the main text on direct neurobiological insights gained from our EEG-fMRI analyses, particularly emphasizing the hippocampal-thalamocortical network dynamics underlying SO-spindle coupling, and we will acknowledge the exploratory nature of these findings and highlight their limitations.

Discussion, Page 17-18 Lines 323-332

“To explore functional relevance, we employed an open-ended cognitive state decoding approach using meta-analytic data (NeuroSynth: Yarkoni et al. (2011)). Although this method usefully generates hypotheses about potential cognitive processes, particularly in the absence of a pre- and post-sleep memory task, it is inherently indirect. Many cognitive terms showed significant associations (16 of 50), such as “episodic memory,” “declarative memory,” and “working memory.” We focused on episodic/declarative memory given the known link with hippocampal reactivation (Diekelmann & Born, 2010; Staresina et al., 2015; Staresina et al., 2023). Nonetheless, these inferences regarding memory reactivation should be interpreted cautiously without direct behavioral measures. Future research incorporating explicit tasks before and after sleep would more rigorously validate these potential functional claims.”

(3) Hippocampal activation during SO-spindles is stated as a main hypothesis of the paper - for good reasons - however, other regions (e.g., several cortical as well as thalamic) would be equally expected given the known origin of both oscillations and the existing sleep-EEG-fMRI literature. However, this focus on the hippocampus contrasts with the focus on investigating the key role of the thalamus instead in the Results section.

We appreciate your insight regarding the relative emphasis on hippocampal and thalamic activation in our study. We recognize that the manuscript may currently present an inconsistency between our initial hypothesis and the main focus of the results. To address this concern, we will ensure that our Introduction and Discussion section explicitly discusses both regions, highlighting the complementary roles of the hippocampus (memory processing and reactivation) and the thalamus (spindle generation and cortico-hippocampal coordination) in SO-spindle dynamics.

Introduction, Page 5 Lines 87-103

“To address this gap, our study investigates brain-wide activation and functional connectivity patterns associated with SO-spindle coupling, and employs a cognitive state decoding approach (Margulies et al., 2016; Yarkoni et al., 2011)—albeit indirectly—to infer potential cognitive functions. In the current study, we used simultaneous EEG-fMRI recordings during nocturnal naps (detailed sleep staging results are provided in the Methods and Table S1) in 107 participants. Although directly detecting hippocampal ripples using scalp EEG or fMRI is challenging, we expected that hippocampal activation in fMRI would coincide with SO-spindle coupling detected by EEG, given that SOs, spindles, and ripples frequently co-occur during NREM sleep. We also anticipated a critical role of the thalamus, particularly thalamic spindles, in coordinating hippocampal-cortical communication.

We found significant coupling between SOs and spindles during NREM sleep (N2/3), with spindle peaks occurring slightly before the SO peak. This coupling was associated with increased activation in both the thalamus and hippocampus, with functional connectivity patterns suggesting thalamic coordination of hippocampal-cortical communication. These findings highlight the key role of the thalamus in coordinating hippocampal-cortical interactions during human sleep and provide new insights into the neural mechanisms underlying sleep-dependent brain communication. A deeper understanding of these mechanisms may contribute to future neuromodulation approaches aimed at enhancing sleep-dependent cognitive function and treating sleep-related disorders.”

Discussion, Page 16-17 Lines 292-307

“When modeling the timing of these sleep rhythms in the fMRI, we observed hippocampal activation selectively during SO-spindle events. This suggests the possibility of triple coupling (SOs–spindles–ripples), even though our scalp EEG was not sufficiently sensitive to detect hippocampal ripples—key markers of memory replay (Buzsáki, 2015). Recent iEEG evidence indicates that ripples often co-occur with both spindles (Ngo, Fell, & Staresina, 2020) and SOs (Staresina et al., 2015; Staresina et al., 2023). Therefore, the hippocampal involvement during SO-spindle events in our study may reflect memory replay from the hippocampus, propagated via thalamic spindles to distributed cortical regions.

The thalamus, known to generate spindles (Halassa et al., 2011), plays a key role in producing and coordinating sleep rhythms (Coulon, Budde, & Pape, 2012; Crunelli et al., 2018), while the hippocampus is found essential for memory consolidation (Buzsáki, 2015; Diba & Buzsá ki, 2007; Singh, Norman, & Schapiro, 2022). The increased hippocampal and thalamic activity, along with strengthened connectivity between these regions and the mPFC during SO-spindle events, underscores a hippocampal-thalamic-neocortical information flow. This aligns with recent findings suggesting the thalamus orchestrates neocortical oscillations during sleep (Schreiner et al., 2022). The thalamus and hippocampus thus appear central to memory consolidation during sleep, guiding information transfer to the neocortex, e.g., mPFC.”

(4) The study included an impressive number of 107 subjects. It is surprising though that only 31 subjects had to be excluded under these difficult recording conditions, especially since no adaptation night was performed. Since only subjects were excluded who slept less than 10 min (or had excessive head movements) there are likely several datasets included with comparably short durations and only a small number of SOs and spindles and even less combined SO-spindle events. A comprehensive table should be provided (supplement) including for each subject (included and excluded) the duration of included NREM sleep, number of SOs, spindles, and SO+spindle events. Also, some descriptive statistics (mean/SD/range) would be helpful.

We appreciate your recognition of our sample size and the challenges associated with simultaneous EEG-fMRI sleep recordings. We acknowledge the importance of transparently reporting individual subject data, particularly regarding sleep duration and the number of detected SOs, spindles, and SO-spindle events. To address this, we will provide comprehensive tables in the supplementary materials, contains descriptive information about sleep-related characteristics (Table S1), as well as detailed information about sleep waves at each sleep stage for all 107 subjects(Table S2-S4), listing for each subject:(1)Different sleep stage duration; (2)Number of detected SOs; (3)Number of detected spindles; (4)Number of detected SO-spindle coupling events; (5)Density of detected SOs; (6)Density of detected spindles; (7)Density of detected SO-spindle coupling events.

However, most of the excluded participants were unable to fall asleep or had too short a sleep duration, so they basically had no NREM sleep period, so it was impossible to count the NREM sleep duration, SO, spindle, and coupling numbers.

Supplementary Materials, Page 42-54, Table S1-S4

(5) Was the 20-channel head coil dedicated for EEG-fMRI measurements? How were the electrode cables guided through/out of the head coil? Usually, the 64-channel head coil is used for EEG-fMRI measurements in a Siemens PRISMA 3T scanner, which has a cable duct at the back that allows to guide the cables straight out of the head coil (to minimize MR-related artifacts). The choice for the 20-channel head coil should be motivated. Photos of the recording setup would also be helpful.

Thank you for your comment regarding our choice of the 20-channel head coil for EEG-fMRI measurements. We acknowledge that the 64-channel head coil is commonly used in Siemens PRISMA 3T scanners; however, the 20-channel coil was selected due to specific practical and technical considerations in our study. In particular, the 20-channel head coil was compatible with our EEG system and ensured sufficient signal-to-noise ratio (SNR) for both EEG and fMRI acquisition. The EEG electrode cables were guided through the lateral and posterior openings of the head coil, secured with foam padding to reduce motion and minimize MR-related artifacts. Moreover, given the extended nature of nocturnal sleep recordings, the 20-channel coil allowed us to maintain participant comfort while still achieving high-quality simultaneous EEG-fMRI data.

We have made this clearer in the revised manuscript. 

Methods, Page 20 Lines 385-392

“All MRI data were acquired using a 20-channel head coil on a research-dedicated 3-Tesla Siemens Magnetom Prisma MRI scanner. Earplugs and cushions were provided for noise protection and head motion restriction. We chose the 20-channel head coil because it was compatible with our EEG system and ensured sufficient signal-to-noise ratio (SNR) for both EEG and fMRI acquisition. The EEG electrode cables were guided through the lateral and posterior openings of the head coil, secured with foam padding to reduce motion and minimize MR-related artifacts. Moreover, given the extended nature of nocturnal sleep recordings, the 20-channel coil helped maintain participant comfort while still achieving high-quality simultaneous EEG-fMRI data.”

(6) Was the EEG sampling synchronized to the MR scanner (gradient system) clock (the 10 MHz signal; not referring to the volume TTL triggers here)? This is a requirement for stable gradient artifact shape over time and thus accurate gradient noise removal.

Thank you for raising this important point. We confirm that the EEG sampling was synchronized to the MR scanner’s 10 MHz gradient system clock, ensuring a stable gradient artifact shape over time and enabling accurate artifact removal. This synchronization was achieved using the standard clock synchronization interface of the EEG amplifier, minimizing timing jitter and drift. As a result, the gradient artifact waveform remained stable across volumes, allowing for more effective artifact correction during preprocessing. We appreciate your attention to this critical aspect of EEG-fMRI data acquisition.

We have made this clearer in the revised manuscript. 

Methods, Page 19-20 Lines 371-383

“EEG was recorded simultaneously with fMRI data using an MR-compatible EEG amplifier system (BrainAmps MR-Plus, Brain Products, Germany), along with a specialized electrode cap. The recording was done using 64 channels in the international 10/20 system, with the reference channel positioned at FCz. In order to adhere to polysomnography (PSG) recording standards, six electrodes were removed from the EEG cap: one for electrocardiogram (ECG) recording, two for electrooculogram (EOG) recording, and three for electromyogram (EMG) recording. EEG data was recorded at a sample rate of 5000 Hz, the resistance of the reference and ground channels was kept below 10 kΩ, and the resistance of the other channels was kept below 20 kΩ. To synchronize the EEG and fMRI recordings, the BrainVision recording software (BrainProducts, Germany) was utilized to capture triggers from the MRI scanner. The EEG sampling was synchronized to the MR scanner’s 10 MHz gradient system clock, ensuring a stable gradient artifact shape over time and enabling accurate artifact removal. This was achieved via the standard clock synchronization interface of the EEG amplifier, minimizing timing jitter and drift.”

(7) The TR is quite long and the voxel size is quite large in comparison to state-of-the-art EPI sequences. What was the rationale behind choosing a sequence with relatively low temporal and spatial resolution?

We acknowledge that our chosen TR and voxel size are relatively long and large compared to state-of-the-art EPI sequences. This decision was made to optimize the signal-to-noise ratio (SNR) and reduce susceptibility-related distortions, which are particularly critical in EEG-fMRI sleep studies where head motion and physiological noise can be substantial. A longer TR allowed us to sample whole-brain activity with sufficient coverage, while a larger voxel size helped enhance BOLD sensitivity and minimize partial volume effects in deep brain structures such as the thalamus and hippocampus, which are key regions of interest in our study. We appreciate your concern and hope this clarification provides sufficient rationale for our sequence parameters.

We have made this clearer in the revised manuscript. 

Methods, Page 20-21 Lines 398-408

“Then, the “sleep” session began after the participants were instructed to try and fall asleep. For the functional scans, whole-brain images were acquired using k-space and steady-state T2*-weighted gradient echo-planar imaging (EPI) sequence that is sensitive to the BOLD contrast. This measures local magnetic changes caused by changes in blood oxygenation that accompany neural activity (sequence specification: 33 slices in interleaved ascending order, TR = 2000 ms, TE = 30 ms, voxel size = 3.5 × 3.5 × 4.2 mm3, FA = 90°, matrix = 64 × 64, gap = 0.7 mm). A relatively long TR and larger voxel size were chosen to optimize SNR and reduce susceptibility-related distortions, which are critical in EEG-fMRI sleep studies where head motion and physiological noise can be substantial. The longer TR allowed whole-brain coverage with sufficient temporal resolution, while the larger voxel size helped enhance BOLD sensitivity and minimize partial volume effects in deep brain structures (e.g., the thalamus and hippocampus), which are key regions of interest in this study.”

(8) The anatomically defined ROIs are quite large. It should be elaborated on how this might reduce sensitivity to sleep rhythm-specific activity within sub-regions, especially for the thalamus, which has distinct nuclei involved in sleep functions.

We appreciate your insight regarding the use of anatomically defined ROIs and their potential limitations in detecting sleep rhythm-specific activity within sub-regions, particularly in the thalamus. Given the distinct functional roles of thalamic nuclei in sleep processes, we acknowledge that using a single, large thalamic ROI may reduce sensitivity to localized activity patterns. To address this, we will discuss this limitation in the revised manuscript, acknowledging that our approach prioritizes whole-structure effects but may not fully capture nucleus-specific contributions.

Discussion, Page 18 Lines 333-341

“Despite providing new insights, our study has several limitations. First, our scalp EEG did not directly capture hippocampal ripples, preventing us from conclusively demonstrating triple coupling. Second, the combination of EEG-fMRI and the lack of a memory task limit our ability to parse fine-grained BOLD responses at the DOWN- vs. UP-states of SOs and link observed activations to behavioral outcomes. Third, the use of large anatomical ROIs may mask subregional contributions of specific thalamic nuclei or hippocampal subfields. Finally, without a memory task, we cannot establish a direct behavioral link between sleep-rhythm-locked activation and memory consolidation. Future studies combining techniques such as ultra-high-field fMRI or iEEG with cognitive tasks may refine our understanding of subregional network dynamics and functional significance during sleep.”

(9) The study reports SO & spindle amplitudes & densities, as well as SO+spindle coupling, to be larger during N2/3 sleep compared to N1 and REM sleep, which is trivial but can be seen as a sanity check of the data. However, the amount of SOs and spindles reported for N1 and REM sleep is concerning, as per definition there should be hardly any (if SOs or spindles occur in N1 it becomes by definition N2, and the interval between spindles has to be considerably large in REM to still be scored as such). Thus, on the one hand, the report of these comparisons takes too much space in the main manuscript as it is trivial, but on the other hand, it raises concerns about the validity of the scoring.

We appreciate your concern regarding the reported presence of SOs and spindles in N1 and REM sleep and the potential implications. Our detection method for detecting SO, spindle, and coupling were originally designed only for N2&N3 sleep data based on the characteristics of the data itself, and this method is widely recognized and used in the sleep research (Hahn et al., 2020; Helfrich et al., 2019; Helfrich et al., 2018; Ngo, Fell, & Staresina, 2020; Schreiner et al., 2022; Schreiner et al., 2021; Staresina et al., 2015; Staresina et al., 2023). While, because the detection methods for SO and spindle are based on percentiles, this method will always detect a certain number of events when used for other stages (N1 and REM) sleep data, but the differences between these events and those detected in stage N23 remain unclear. We will acknowledge the reasons for these results in the Methods section and emphasize that they are used only for sanity checks.

Methods, Page 25 Lines 515-524

“We note that the above methods for detecting SOs, spindles, and their couplings were originally developed for N2 and N3 sleep data, based on the specific characteristics of these stages. These methods are widely recognized in sleep research (Hahn et al., 2020; Helfrich et al., 2019; Helfrich et al., 2018; Ngo, Fell, & Staresina, 2020; Schreiner et al., 2022; Schreiner et al., 2021; Staresina et al., 2015; Staresina et al., 2023). However, because this percentile-based detection approach will inherently identify a certain number of events if applied to other stages (e.g., N1 and REM), the nature of these events in those stages remains unclear compared to N2/N3. We nevertheless identified and reported the detailed descriptive statistics of these sleep rhythms in all sleep stages, under the same operational definitions, both for completeness and as a sanity check. Within the same subject, there should be more SOs, spindles, and their couplings in N2/N3 than in N1 or REM (see also Figure S2-S4, Table S1-S4).”

(10) Why was electrode F3 used to quantify the occurrence of SOs and spindles? Why not a midline frontal electrode like Fz (or a number of frontal electrodes for SOs) and Cz (or a number of centroparietal electrodes) for spindles to be closer to their maximum topography?

We appreciate your suggestion regarding electrode selection for SO and spindle quantification. Our choice of F3 was primarily based on previous studies (Massimini et al., 2004; Molle et al., 2011), where bilateral frontal electrodes are commonly used for detecting SOs and spindles. Additionally, we considered the impact of MRI-related noise and, after a comprehensive evaluation, determined that F3 provided an optimal balance between signal quality and artifact minimization. We also acknowledge that alternative electrode choices, such as Fz for SOs and Cz for spindles, could provide additional insights into their topographical distributions.

(11) Functional connectivity (hippocampus -> thalamus -> cortex (mPFC)) is reported to be increased during SO-spindle coupling and interpreted as evidence for coordination of hippocampo-neocortical communication likely by thalamic spindles. However, functional connectivity was only analysed during coupled SO+spindle events, not during isolated SOs or isolated spindles. Without the direct comparison of the connectivity patterns between these three events, it remains unclear whether this is specific for coupled SO+spindle events or rather associated with one or both of the other isolated events. The PPIs need to be conducted for those isolated events as well and compared statistically to the coupled events.

We appreciate your critical perspective on our functional connectivity analysis and the interpretation of hippocampus-thalamus-cortex (mPFC) interactions during SO-spindle coupling. We acknowledge that, in the current analysis, functional connectivity was only examined during coupled SO-spindle events, without direct comparison to isolated SOs or isolated spindles. To address this concern, we have conducted PPI analyses for all three ROIs(Hippocampus, Thalamus, mPFC) and all three event types (SO-spindle couplings, isolated SOs, and isolated spindles). Our results indicate that neither isolated SOs nor isolated Spindles yielded significant connectivity changes in all three ROIs, as all failed to survive multiple comparison corrections. This suggests that the observed connectivity increase is specific to SO-spindle coupling, rather than being independently driven by either SOs or spindles alone.

Results, Page 14 Lines 248-255

“Crucially, the interaction between FC and SO-spindle coupling revealed that only the functional connectivity of hippocampus -> thalamus (ROI analysis, t(106) = 1.86, p = 0.0328) and thalamus -> mPFC (ROI analysis, t(106) = 1.98, p = 0.0251) significantly increased during SO-spindle coupling, with no significant changes in all other pathways (Fig. 4e). We also conducted PPI analyses for the other two events (SOs and spindles), and neither yielded significant connectivity changes in the three ROIs, as all failed to survive whole-brain FWE correction at the cluster level (p < 0.05). Together, these findings suggest that the thalamus, likely via spindles, coordinates hippocampal-cortical communication selectively during SO-spindle coupling, but not isolated SOs or spindle events alone.”

(12) The limited temporal resolution of fMRI does indeed not allow for easily distinguishing between fMRI activation patterns related to SO-up- vs. SO-down-states. For this, one could try to extract the amplitudes of SO-up- and SO-down-states separately for each SO event and model them as two separate parametric modulators (with the risk of collinearity as they are likely correlated).

We appreciate your insightful comment regarding the challenge of distinguishing fMRI activation patterns related to SO-up vs. SO-down states due to the limited temporal resolution of fMRI. While our current analysis does not differentiate between these two phases, we acknowledge that separately modeling SO-up and SO-down states using parametric modulators could provide a more refined understanding of their distinct neural correlates. However, as you notes, this approach carries the risk of collinearity, and there is indeed a high correlation between the two amplitudes across all subjects in our results (r=0.98). Future studies could explore more on leveraging high-temporal-resolution techniques. While implementing this in the current study is beyond our scope, we will acknowledge this limitation in the Discussion section.

Discussion, Page 17 Lines 308-322

“An intriguing aspect of our findings is the reduced DMN activity during SOs when modeled at the SO trough (DOWN-state). This reduced DMN activity may reflect large-scale neural inhibition characteristic of the SO trough. The DMN is typically active during internally oriented cognition (e.g., self-referential processing or mind-wandering) and is suppressed during external stimuli processing (Yeshurun, Nguyen, & Hasson, 2021). It is unlikely, however, that this suppression of DMN during SO events is related to a shift from internal cognition to external responses given it is during deep sleep time. Instead, it could be driven by the inherent rhythmic pattern of SOs, which makes it difficult to separate UP- from DOWN-states (the two temporal regressors were highly correlated, and similar brain activation during SOs events was obtained if modelled at the SO peak instead, Fig. S5). Since the amplitude at the SO trough is consistently larger than that at the SO peak, the neural activation we detected may primarily capture the large-scale inhibition from DOWN-state. Interestingly, no such DMN reduction was found during SO-spindle coupling, implying that coupling may involve distinct neural dynamics that partially re-engage DMN-related processes, possibly reflecting memory-related reactivation. Future research using high-temporal-resolution techniques like iEEG could clarify these possibilities.”

Discussion, Page 18 Lines 333-341

“Despite providing new insights, our study has several limitations. First, our scalp EEG did not directly capture hippocampal ripples, preventing us from conclusively demonstrating triple coupling. Second, the combination of EEG-fMRI and the lack of a memory task limit our ability to parse fine-grained BOLD responses at the DOWN- vs. UP-states of SOs and link observed activations to behavioral outcomes. Third, the use of large anatomical ROIs may mask subregional contributions of specific thalamic nuclei or hippocampal subfields. Finally, without a memory task, we cannot establish a direct behavioral link between sleep-rhythm-locked activation and memory consolidation. Future studies combining techniques such as ultra-high-field fMRI or iEEG with cognitive tasks may refine our understanding of subregional network dynamics and functional significance during sleep.”

(13) L327: "It is likely that our findings of diminished DMN activity reflect brain activity during the SO DOWN-state, as this state consistently shows higher amplitude compared to the UP-state within subjects, which is why we modelled the SO trough as its onset in the fMRI analysis." This conclusion is not justified as the fact that SO down-states are larger in amplitude does not mean their impact on the BOLD response is larger.

We appreciate your concern regarding our interpretation of diminished DMN activity reflecting the SO down-state. We acknowledge that the current expression is somewhat misleading, and our interpretation of it is: it could be driven by the inherent rhythmic pattern of SOs, which makes it difficult to separate UP- from DOWN-states (the two temporal regressors were highly correlated, and similar brain activation during SOs events was obtained if modelled at the SO peak instead). Since the amplitude at the SO trough is consistently larger than that at the SO peak, the neural activation we detected may primarily capture the large-scale inhibition from DOWN-state. And we will make this clear in the Discussion section.

Discussion, Page 17 Lines 308-322

“An intriguing aspect of our findings is the reduced DMN activity during SOs when modeled at the SO trough (DOWN-state). This reduced DMN activity may reflect large-scale neural inhibition characteristic of the SO trough. The DMN is typically active during internally oriented cognition (e.g., self-referential processing or mind-wandering) and is suppressed during external stimuli processing (Yeshurun, Nguyen, & Hasson, 2021). It is unlikely, however, that this suppression of DMN during SO events is related to a shift from internal cognition to external responses given it is during deep sleep time. Instead, it could be driven by the inherent rhythmic pattern of SOs, which makes it difficult to separate UP- from DOWN-states (the two temporal regressors were highly correlated, and similar brain activation during SOs events was obtained if modelled at the SO peak instead, Fig. S5). Since the amplitude at the SO trough is consistently larger than that at the SO peak, the neural activation we detected may primarily capture the large-scale inhibition from DOWN-state. Interestingly, no such DMN reduction was found during SO-spindle coupling, implying that coupling may involve distinct neural dynamics that partially re-engage DMN-related processes, possibly reflecting memory-related reactivation. Future research using high-temporal-resolution techniques like iEEG could clarify these possibilities.”

(14) Line 77: "In the current study, while directly capturing hippocampal ripples with scalp EEG or fMRI is difficult, we expect to observe hippocampal activation in fMRI whenever SOs-spindles coupling is detected by EEG, if SOs- spindles-ripples triple coupling occurs during human NREM sleep". Not all SO-spindle events are associated with ripples (Staresina et al., 2015), but hippocampal activation may also be expected based on the occurrence of spindles alone (Bergmann et al., 2012).

We appreciate your clarification regarding the relationship between SO-spindle coupling and hippocampal ripples. We acknowledge that not all SO-spindle events are necessarily accompanied by ripples (Staresina et al., 2015). However, based on previous research, we found that hippocampal ripples are significantly more likely to occur during SO-spindle coupling events. This suggests that while ripple occurrence is not guaranteed, SO-spindle coupling creates a favorable network state for ripple generation and potential hippocampal activation. To ensure accuracy, we will revise the manuscript to delete this misleading sentence in the Introduction section and acknowledge in the Discussion that our results cannot conclusively directly observe the triple coupling of SO, spindle, and hippocampal ripples.

Discussion, Page 18 Lines 333-341

“Despite providing new insights, our study has several limitations. First, our scalp EEG did not directly capture hippocampal ripples, preventing us from conclusively demonstrating triple coupling. Second, the combination of EEG-fMRI and the lack of a memory task limit our ability to parse fine-grained BOLD responses at the DOWN- vs. UP-states of SOs and link observed activations to behavioral outcomes. Third, the use of large anatomical ROIs may mask subregional contributions of specific thalamic nuclei or hippocampal subfields. Finally, without a memory task, we cannot establish a direct behavioral link between sleep-rhythm-locked activation and memory consolidation. Future studies combining techniques such as ultra-high-field fMRI or iEEG with cognitive tasks may refine our understanding of subregional network dynamics and functional significance during sleep.”

Reviewer #2 (Public review):

In this study, Wang and colleagues aimed to explore brain-wide activation patterns associated with NREM sleep oscillations, including slow oscillations (SOs), spindles, and SO-spindle coupling events. Their findings reveal that SO-spindle events corresponded with increased activation in both the thalamus and hippocampus. Additionally, they observed that SO-spindle coupling was linked to heightened functional connectivity from the hippocampus to the thalamus, and from the thalamus to the medial prefrontal cortex-three key regions involved in memory consolidation and episodic memory processes.

This study's findings are timely and highly relevant to the field. The authors' extensive data collection, involving 107 participants sleeping in an fMRI while undergoing simultaneous EEG recording, deserves special recognition. If shared, this unique dataset could lead to further valuable insights. While the conclusions of the data seem overall well supported by the data, some aspects with regard to the detection of sleep oscillations need clarification.

The authors report that coupled SO-spindle events were most frequent during NREM sleep (2.46 [plus minus] 0.06 events/min), but they also observed a surprisingly high occurrence of these events during N1 and REM sleep (2.23 [plus minus] 0.09 and 2.32 [plus minus] 0.09 events/min, respectively), where SO-spindle coupling would not typically be expected. Combined with the relatively modest SO amplitudes reported (~25 µV, whereas >75 µV would be expected when using mastoids as reference electrodes), this raises the possibility that the parameters used for event detection may not have been conservative enough - or that sleep staging was inaccurately performed. This issue could present a significant challenge, as the fMRI findings are largely dependent on the reliability of these detected events.

Thank you very much for your thorough and encouraging review. We appreciate your recognition of the significance and relevance of our study and dataset, particularly in highlighting how simultaneous EEG-fMRI recordings can provide complementary insights into the temporal dynamics of neural oscillations and their associated spatial activation patterns during sleep. In the sections that follow, we address each of your comments in detail. We have revised the text and conducted additional analyses wherever possible to strengthen our argument, clarify our methodological choices. We believe these revisions improve the clarity and rigor of our work, and we thank you for helping us refine it.

We appreciate your insightful comments regarding the detection of sleep oscillations. Our methods for detecting SOs, spindles, and their couplings were originally developed for N2 and N3 sleep data, based on the specific characteristics of these stages. These methods are widely recognized in sleep research (Hahn et al., 2020; Helfrich et al., 2019; Helfrich et al., 2018; Ngo, Fell, & Staresina, 2020; Schreiner et al., 2022; Schreiner et al., 2021; Staresina et al., 2015; Staresina et al., 2023). However, because this percentile-based detection approach will inherently identify a certain number of events if applied to other stages (e.g., N1 and REM), the nature of these events in those stages remains unclear compared to N2/N3. We nevertheless identified and reported the detailed descriptive statistics of these sleep rhythms in all sleep stages, under the same operational definitions, both for completeness and as a sanity check. Within the same subject, there should be more SOs, spindles, and their couplings in N2/N3 than in N1 or REM. We will acknowledge the reasons for these results in the Methods section and emphasize that they are used only for sanity checks.

Regarding the reported SO amplitudes (~25 µV), during preprocessing, we applied the Signal Space Projection (SSP) method to more effectively remove MRI gradient artifacts and cardiac pulse noise. While this approach enhances data quality, it also reduces overall signal power, leading to systematically lower reported amplitudes. Despite this, our SO detection in NREM sleep (especially N2/N3) remain physiologically meaningful and are consistent with previous fMRI studies using similar artifact removal techniques. We appreciate your careful evaluation and valuable suggestions.

In addition, we will provide comprehensive tables in the supplementary materials, contains descriptive information about sleep-related characteristics (Table S1), as well as detailed information about sleep waves at each sleep stage for all 107 subjects(Table S2-S4), listing for each subject:(1)Different sleep stage duration; (2)Number of detected SOs; (3)Number of detected spindles; (4)Number of detected SO-spindle coupling events; (2)Density of detected SOs; (3)Density of detected spindles; (4)Density of detected SO-spindle coupling events.

Methods, Page 25 Lines 515-524

“We note that the above methods for detecting SOs, spindles, and their couplings were originally developed for N2 and N3 sleep data, based on the specific characteristics of these stages. These methods are widely recognized in sleep research (Hahn et al., 2020; Helfrich et al., 2019; Helfrich et al., 2018; Ngo, Fell, & Staresina, 2020; Schreiner et al., 2022; Schreiner et al., 2021; Staresina et al., 2015; Staresina et al., 2023). However, because this percentile-based detection approach will inherently identify a certain number of events if applied to other stages (e.g., N1 and REM), the nature of these events in those stages remains unclear compared to N2/N3. We nevertheless identified and reported the detailed descriptive statistics of these sleep rhythms in all sleep stages, under the same operational definitions, both for completeness and as a sanity check. Within the same subject, there should be more SOs, spindles, and their couplings in N2/N3 than in N1 or REM (see also Figure S2-S4, Table S1-S4).”

Supplementary Materials, Page 42-54, Table S1-S4

Reviewer #3 (Public review):

Summary:

Wang et al., examined the brain activity patterns during sleep, especially when locked to those canonical sleep rhythms such as SO, spindle, and their coupling. Analyzing data from a large sample, the authors found significant coupling between spindles and SOs, particularly during the upstate of the SO. Moreover, the authors examined the patterns of whole-brain activity locked to these sleep rhythms. To understand the functional significance of these brain activities, the authors further conducted open-ended cognitive state decoding and found a variety of cognitive processing may be involved during SO-spindle coupling and during other sleep events. The authors next investigated the functional connectivity analyses and found enhanced connectivity between the hippocampus, the thalamus, and the medial PFC. These results reinforced the theoretical model of sleep-dependent memory consolidation, such that SO-spindle coupling is conducive to systems-level memory reactivation and consolidation.

Strengths:

There are obvious strengths in this work, including the large sample size, state-of-the-art neuroimaging and neural oscillation analyses, and the richness of results.

Weaknesses:

Despite these strengths and the insights gained, there are weaknesses in the design, the analyses, and inferences.

Thank you for your detailed and thoughtful review of our manuscript. We are delighted that you recognize our advanced analysis methods and rich results of neuroimaging and neural oscillations as well as the large sample size data. In the following sections, we provide detailed responses to each of your comments. And we have revised the text and conducted additional analyses to strengthen our arguments and clarify our methodological choices. We believe these revisions enhance the clarity and rigor of our work, and we sincerely appreciate your thoughtful feedback in helping us refine the manuscript.

(1) A repeating statement in the manuscript is that brain activity could indicate memory reactivation and thus consolidation. This is indeed a highly relevant question that could be informed by the current data/results. However, an inherent weakness of the design is that there is no memory task before and after sleep. Thus, it is difficult (if not impossible) to make a strong argument linking SO/spindle/coupling-locked brain activity with memory reactivation or consolidation.

We appreciate your suggestion regarding the lack of a pre- and post-sleep memory task in our study design. We acknowledge that, in the absence of behavioral measures, it is hard to directly link SO-spindle coupling to memory consolidation in an outcome-driven manner. Our interpretation is instead based on the well-established role of these oscillations in memory processes, as demonstrated in previous studies. We sincerely appreciate this feedback and will adjust our Discussion accordingly to reflect a more precise interpretation of our findings.

Discussion, Page 18 Lines 333-341

“Despite providing new insights, our study has several limitations. First, our scalp EEG did not directly capture hippocampal ripples, preventing us from conclusively demonstrating triple coupling. Second, the combination of EEG-fMRI and the lack of a memory task limit our ability to parse fine-grained BOLD responses at the DOWN- vs. UP-states of SOs and link observed activations to behavioral outcomes. Third, the use of large anatomical ROIs may mask subregional contributions of specific thalamic nuclei or hippocampal subfields. Finally, without a memory task, we cannot establish a direct behavioral link between sleep-rhythm-locked activation and memory consolidation. Future studies combining techniques such as ultra-high-field fMRI or iEEG with cognitive tasks may refine our understanding of subregional network dynamics and functional significance during sleep.”

(2) Relatedly, to understand the functional implications of the sleep rhythm-locked brain activity, the authors employed the "open-ended cognitive state decoding" method. While this method is interesting, it is rather indirect given that there were no behavioral indices in the manuscript. Thus, discussions based on these analyses are speculative at best. Please either tone down the language or find additional evidence to support these claims.

Moreover, the results from this method are difficult to understand. Figure 3e showed that for all three types of sleep events (SO, spindle, SO-spindle), the same mental states (e.g., working memory, episodic memory, declarative memory) showed opposite directions of activation (left and right panels showed negative and positive activation, respectively). How to interpret these conflicting results? This ambiguity is also reflected by the term used: declarative memory and episodic memories are both indexed in the results. Yet these two processes can be largely overlapped. So which specific memory processes do these brain activity patterns reflect? The Discussion shall discuss these results and the limitations of this method.

We appreciate your critical assessment of the open-ended cognitive state decoding method and its interpretational challenges. Given the concerns about the indirectness of this approach, we decided to remove its related content and results from Figure 3 in the main text and include it in Supplementary Figure 7. 

Due to the complexity of memory-related processes, we acknowledge that distinguishing between episodic and declarative memory based solely on this approach is not straightforward. We will revise the Supplementary Materials to explicitly discuss these limitations and clarify that our findings do not isolate specific cognitive processes but rather suggest general associations with memory-related networks.

Discussion, Page 17-18 Lines 323-332

“To explore functional relevance, we employed an open-ended cognitive state decoding approach using meta-analytic data (NeuroSynth: Yarkoni et al. (2011)). Although this method usefully generates hypotheses about potential cognitive processes, particularly in the absence of a pre- and post-sleep memory task, it is inherently indirect. Many cognitive terms showed significant associations (16 of 50), such as “episodic memory,” “declarative memory,” and “working memory.” We focused on episodic/declarative memory given the known link with hippocampal reactivation (Diekelmann & Born, 2010; Staresina et al., 2015; Staresina et al., 2023). Nonetheless, these inferences regarding memory reactivation should be interpreted cautiously without direct behavioral measures. Future research incorporating explicit tasks before and after sleep would more rigorously validate these potenial functional claims.”

(3) The coupling strength is somehow inconsistent with prior results (Hahn et al., 2020, eLife, Helfrich et al., 2018, Neuron). Specifically, Helfrich et al. showed that among young adults, the spindle is coupled to the peak of the SO. Here, the authors reported that the spindles were coupled to down-to-up transitions of SO and before the SO peak. It is possible that participants' age may influence the coupling (see Helfrich et al., 2018). Please discuss the findings in the context of previous research on SO-spindle coupling.

We appreciate your concern regarding the temporal characteristics of SO-spindle coupling. We acknowledge that the SO-spindle coupling phase results in our study are not identical to those reported by Hahn et al. (2020); Helfrich et al. (2018). However, these differences may arise due to slight variations in event detection parameters, which can influence the precise phase estimation of coupling. Notably, Hahn et al. (2020) also reported slight discrepancies in their group-level coupling phase results, highlighting that methodological differences can contribute to variability across studies. Furthermore, our findings are consistent with those of Schreiner et al. (2021), further supporting the robustness of our observations.  

That said, we acknowledge that our original description of SO-spindle coupling as occurring at the "transition from the lower state to the upper state" was not entirely precise. The -π/2 phase represents the true transition point, while our observed coupling phase is actually closer to the SO peak rather than strictly at the transition. We will revise this statement in the manuscript to ensure clarity and accuracy in describing the coupling phase.  

Discussion, Page 16 Lines 283-291

“Our data provide insights into the neurobiological underpinnings of these sleep rhythms. SOs, originating mainly in neocortical areas such as the mPFC, alternate between DOWN- and UP-states. The thalamus generates sleep spindles, which in turn couple with SOs. Our finding that spindle peaks consistently occurred slightly before the UP-state peak of SOs (in 83 out of 107 participants), concurs with prior studies, including Schreiner et al. (2021). Yet it differs from some results suggesting spindles might peak right at the SO UP-state (Hahn et al., 2020; Helfrich et al., 2018). Such discrepancies could arise from differences in detection algorithms, participant age (Helfrich et al., 2018), or subtle variations in cortical-thalamic timing. Nonetheless, these results underscore the importance of coordinated SO-spindle interplay in supporting sleep-dependent processes.”

(4) The discussion is rather superficial with only two pages, without delving into many important arguments regarding the possible functional significance of these results. For example, the author wrote, "This internal processing contrasts with the brain patterns associated with external tasks, such as working memory." Without any references to working memory, and without delineating why WM is considered as an external task even working memory operations can be internal. Similarly, for the interesting results on SO and reduced DMN activity, the authors wrote "The DMN is typically active during wakeful rest and is associated with self-referential processes like mind-wandering, daydreaming, and task representation (Yeshurun, Nguyen, & Hasson, 2021). Its reduced activity during SOs may signal a shift towards endogenous processes such as memory consolidation." This argument is flawed. DMN is active during self-referential processing and mind-wandering, i.e., when the brain shifts from external stimuli processing to internal mental processing. During sleep, endogenous memory reactivation and consolidation are also part of the internal mental processing given the lack of external environmental stimulation. So why during SO or during memory consolidation, the DMN activity would be reduced? Were there differences in DMN activity between SO and SO-spindle coupling events?

We appreciate your concerns regarding the brevity of the discussion and the need for clearer theoretical arguments. We will expand this section to provide more in-depth interpretations of our findings in the context of prior literature. Regarding working memory (WM), we acknowledge that our phrasing was ambiguous. We will modify this statement in the Discussion section.

For the SO-related reduction in DMN activity, we recognize the need for a more precise explanation. This reduced DMN activity may reflect large-scale neural inhibition characteristic of the SO trough. The DMN is typically active during internally oriented cognition (e.g., self-referential processing or mind-wandering) and is suppressed during external stimuli processing (Yeshurun, Nguyen, & Hasson, 2021). It is unlikely, however, that this suppression of DMN during SO events is related to a shift from internal cognition to external responses given it is during deep sleep time. Instead, it could be driven by the inherent rhythmic pattern of SOs, which makes it difficult to separate UP- from DOWN-states (the two temporal regressors were highly correlated, and similar brain activation during SOs events was obtained if modelled at the SO peak instead). Since the amplitude at the SO trough is consistently larger than that at the SO peak, the neural activation we detected may primarily capture the large-scale inhibition from DOWN-state.

To address your final question, we have conducted the additional post hoc comparison of DMN activity between isolated SOs and SO-spindle coupling events. Our results indicate that

DMN activation during SOs was significantly lower than during SO-spindle coupling (t(106) = -4.17, p < 1e-4). This suggests that SO-spindle coupling may involve distinct neural dynamics that partially re-engage DMN-related processes, possibly reflecting memory-related reactivation. We appreciate your constructive feedback and will integrate these expanded analyses and discussions into our revised manuscript.

Results, Page 11 Lines 199-208

“Spindles were correlated with positive activation in the thalamus (ROI analysis, t(106) = 15.39, p < 1e-4), the anterior cingulate cortex (ACC), and the putamen, alongside deactivation in the DMN (Fig. 3c). Notably, SO-spindle coupling was linked to significant activation in both the thalamus (ROI analysis, t(106) \= 3.38, p = 0.0005) and the hippocampus (ROI analysis, t(106) \= 2.50, p = 0.0070, Fig. 3d). However, no decrease in DMN activity was found during SO-spindle coupling, and DMN activity during SO was significantly lower than during coupling (ROI analysis, t(106) \= -4.17, p < 1e-4). For more detailed activation patterns, see Table S5-S7. We also varied the threshold used to detect SO events to assess its effect on hippocampal activation during SO-spindle coupling and observed that hippocampal activation remained significant when the percentile thresholds for SO detection ranged between 71% and 80% (see Fig. S6).”

Discussion, Page 17-18 Lines 308-332

“An intriguing aspect of our findings is the reduced DMN activity during SOs when modeled at the SO trough (DOWN-state). This reduced DMN activity may reflect large-scale neural inhibition characteristic of the SO trough. The DMN is typically active during internally oriented cognition (e.g., self-referential processing or mind-wandering) and is suppressed during external stimuli processing (Yeshurun, Nguyen, & Hasson, 2021). It is unlikely, however, that this suppression of DMN during SO events is related to a shift from internal cognition to external responses given it is during deep sleep time. Instead, it could be driven by the inherent rhythmic pattern of SOs, which makes it difficult to separate UP- from DOWN-states (the two temporal regressors were highly correlated, and similar brain activation during SOs events was obtained if modelled at the SO peak instead, Fig. S5). Since the amplitude at the SO trough is consistently larger than that at the SO peak, the neural activation we detected may primarily capture the large-scale inhibition from DOWN-state. Interestingly, no such DMN reduction was found during SO-spindle coupling, implying that coupling may involve distinct neural dynamics that partially re-engage DMN-related processes, possibly reflecting memory-related reactivation. Future research using high-temporal-resolution techniques like iEEG could clarify these possibilities.

To explore functional relevance, we employed an open-ended cognitive state decoding approach using meta-analytic data (NeuroSynth: Yarkoni et al. (2011)). Although this method usefully generates hypotheses about potential cognitive processes, particularly in the absence of a pre- and post-sleep memory task, it is inherently indirect. Many cognitive terms showed significant associations (16 of 50), such as “episodic memory,” “declarative memory,” and “working memory.” We focused on episodic/declarative memory given the known link with hippocampal reactivation (Diekelmann & Born, 2010; Staresina et al., 2015; Staresina et al., 2023). Nonetheless, these inferences regarding memory reactivation should be interpreted cautiously without direct behavioral measures. Future research incorporating explicit tasks before and after sleep would more rigorously validate these potential functional claims.”

Recommendations for the authors:

Reviewing Editor Comment:

The reviewers think that you are working on a relevant and important topic. They are praising the large sample size used in the study. The reviewers are not all in line regarding the overall significance of the findings, but they all agree the paper would strongly benefit from some extra work, as all reviewers raise various critical points that need serious consideration.

We appreciate your recognition of the relevance and importance of our study, as well as your acknowledgment of the large sample size as a strength of our work. We understand that there are differing perspectives regarding the overall significance of our findings, and we value the constructive critiques provided. We are committed to addressing the key concerns raised by all reviewers, including refining our analyses, clarifying our interpretations, and incorporating additional discussions to strengthen the manuscript. Below, we address your specific recommendations and provide responses to each point you raised to ensure our methods and results are as transparent and comprehensible as possible. We believe that these revisions will significantly enhance the rigor and impact of our study, and we sincerely appreciate your thoughtful feedback in helping us improve our work.

Reviewer #1 (Recommendations for the authors):

(1) The phrase "overnight sleep" suggests an entire night, while these were rather "nocturnal naps". Please rephrase.

Response: Thank you for pointing this out. We have revised the phrasing in our manuscript to "nocturnal naps" instead of "overnight sleep" to more accurately reflect the duration of the sleep recordings.

(2) Sleep staging results (macroscopic sleep architecture) should be provided in more detail (at least min and % of the different sleep stages, sleep onset latency, total sleep duration, total recording duration), at least mean/SD/range.

Thank you for this suggestion. We will provide comprehensive tables in the supplementary materials, contains descriptive information about sleep-related characteristics. This information will help provide a clearer overview of the macroscopic sleep architecture in our dataset.

Reviewer #2 (Recommendations for the authors):

In order to allow for a better estimation of the reliability of the detected sleep events, please:

(1) Provide densities and absolute numbers of all detected SOs and spindles (N1, NREM, and REM sleep).

Thank you for pointing this out. We will provide comprehensive tables in the supplementary materials, contains detailed information about sleep waves at each sleep stage for all 107 subjects (Table S2-S4), listing for each subject:1) Different sleep stage duration; 2) Number of detected SOs; 3) Number of detected spindles; 4) Number of detected SO-spindle coupling events; 5) Density of detected SOs; 6) Density of detected spindles; 7) Density of detected SO-spindle coupling events.

Supplementary Materials, Page 43-54, Table S2-S4

(2) Show ERPs for all detected SOs and spindles (per sleep stage).

Thank you for the suggestion. We will provide ERPs for all detected SOs and spindles, separated by sleep stage (N1, N2&N3, and REM) in supplementary Fig. S2-S4. These ERP waveforms will help illustrate the characteristic temporal profiles of SOs and spindles across different sleep stages.

Methods, Page 25, Line 525-532

“Event-related potentials (ERP) analysis. After completing the detection of each sleep rhythm event, we performed ERP analyses for SOs, spindles, and coupling events in different sleep stages. Specifically, for SO events, we took the trough of the DOWN-state of each SO as the zero-time point, then extracted data in a [-2 s to 2 s] window from the broadband (0.1–30 Hz) EEG and used [-2 s to -0.5 s] for baseline correction; the results were then averaged across 107 subjects (see Fig. S2a). For spindle events, we used the peak of each spindle as the zero-time point and applied the same data extraction window and baseline correction before averaging across 107 subjects (see Fig. S2b). Finally, for SO-spindle coupling events, we followed the same procedure used for SO events (see Fig. 2a, Figs. S3–S4).”

(3) Provide detailed info concerning sleep characteristics (time spent in each sleep stage etc.).

Thank you for this suggestion. Same as the response above, we will provide comprehensive tables in the supplementary materials, contains descriptive information about sleep-related characteristics.

Supplementary Materials, Page 42, Table S1 (same as above)

(4) What would happen if more stringent parameters were used for event detection? Would the authors still observe a significant number of SO spindles during N1 and REM? Would this affect the fMRI-related results?

Thank you for this suggestion. Our methods for detecting SOs, spindles, and their couplings were originally developed for N2 and N3 sleep data, based on the specific characteristics of these stages. These methods are widely recognized in sleep research (Hahn et al., 2020; Helfrich et al., 2019; Helfrich et al., 2018; Ngo, Fell, & Staresina, 2020; Schreiner et al., 2022; Schreiner et al., 2021; Staresina et al., 2015; Staresina et al., 2023). However, because this percentile-based detection approach will inherently identify a certain number of events if applied to other stages (e.g., N1 and REM), the nature of these events in those stages remains unclear compared to N2/N3. We nevertheless identified and reported the detailed descriptive statistics of these sleep rhythms in all sleep stages, under the same operational definitions, both for completeness and as a sanity check. Within the same subject, there should be more SOs, spindles, and their couplings in N2/N3 than in N1 or REM (see also Figure S2-S4, Table S1-S4).

Furthermore, in order to explore the impact of this on our fMRI results, we conducted an additional sensitivity analysis by applying different detection parameters for SOs. Specifically, we adjusted amplitude percentile thresholds for SO detection (the parameter that has the greatest impact on the results). We used the hippocampal activation value during N2&N3 stage SO-spindle coupling as an anchor value and found that when the parameters gradually became stricter, the results were similar to or even better than the current results. However, when we continued to increase the threshold, the results began to gradually decrease until the threshold was increased to 80%, and the results were no longer significant. This indicates that our results are robust within a specific range of parameters, but as the threshold increases, the number of trials decreases, ultimately weakening the statistical power of the fMRI analysis.

Thank you again for your suggestions on sleep rhythm event detection. We will add the results in Supplementary and revise our manuscript accordingly.

Results, Page 11, Line 199-208

“Spindles were correlated with positive activation in the thalamus (ROI analysis, t(106) = 15.39, p < 1e-4), the anterior cingulate cortex (ACC), and the putamen, alongside deactivation in the DMN (Fig. 3c). Notably, SO-spindle coupling was linked to significant activation in both the thalamus (ROI analysis, t(106) \= 3.38, p = 0.0005) and the hippocampus (ROI analysis, t(106) \= 2.50, p = 0.0070, Fig. 3d). However, no decrease in DMN activity was found during SO-spindle coupling, and DMN activity during SO was significantly lower than during coupling (ROI analysis, t(106) \= -4.17, p < 1e-4). For more detailed activation patterns, see Table S5-S7. We also varied the threshold used to detect SO events to assess its effect on hippocampal activation during SO-spindle coupling and observed that hippocampal activation remained significant when the percentile thresholds for SO detection ranged between 71% and 80% (see Fig. S6).”

Finally, we sincerely thank all again for your thoughtful and constructive feedback. Your insights have been invaluable in refining our analyses, strengthening our interpretations, and improving the clarity and rigor of our manuscript. We appreciate the time and effort you have dedicated to reviewing our work, and we are grateful for the opportunity to enhance our study based on your recommendations.  

References:

Bergmann, T. O., Mölle, M., Diedrichs, J., Born, J., & Siebner, H. R. (2012). Sleep spindle-related reactivation of category-specific cortical regions after learning face-scene associations. NeuroImage, 59(3), 2733-2742. 

Buzsáki, G. (2015). Hippocampal sharp wave‐ripple: A cognitive biomarker for episodic memory and planning. Hippocampus, 25(10), 1073-1188. 

Caporro, M., Haneef, Z., Yeh, H. J., Lenartowicz, A., Buttinelli, C., Parvizi, J., & Stern, J. M. (2012). Functional MRI of sleep spindles and K-complexes. Clinical neurophysiology, 123(2), 303-309. 

Coulon, P., Budde, T., & Pape, H.-C. (2012). The sleep relay—the role of the thalamus in central and decentral sleep regulation. Pflügers Archiv-European Journal of Physiology, 463, 53-71. 

Crunelli, V., Lőrincz, M. L., Connelly, W. M., David, F., Hughes, S. W., Lambert, R. C., Leresche, N., & Errington, A. C. (2018). Dual function of thalamic low-vigilance state oscillations: rhythm-regulation and plasticity. Nature Reviews Neuroscience, 19(2), 107-118. 

Czisch, M., Wehrle, R., Stiegler, A., Peters, H., Andrade, K., Holsboer, F., & Sämann, P. G. (2009). Acoustic oddball during NREM sleep: a combined EEG/fMRI study. PloS one, 4(8), e6749. 

Diba, K., & Buzsáki, G. (2007). Forward and reverse hippocampal place-cell sequences during ripples. Nature Neuroscience, 10(10), 1241. 

Diekelmann, S., & Born, J. (2010). The memory function of sleep. Nature Reviews Neuroscience, 11(2), 114-126. 

Fogel, S., Albouy, G., King, B. R., Lungu, O., Vien, C., Bore, A., Pinsard, B., Benali, H., Carrier, J., & Doyon, J. (2017). Reactivation or transformation? Motor memory consolidation associated with cerebral activation time-locked to sleep spindles. PloS one, 12(4), e0174755. 

Hahn, M. A., Heib, D., Schabus, M., Hoedlmoser, K., & Helfrich, R. F. (2020). Slow oscillation-spindle coupling predicts enhanced memory formation from childhood to adolescence. Elife, 9, e53730. 

Halassa, M. M., Siegle, J. H., Ritt, J. T., Ting, J. T., Feng, G., & Moore, C. I. (2011). Selective optical drive of thalamic reticular nucleus generates thalamic bursts and cortical spindles. Nature Neuroscience, 14(9), 1118-1120. 

Hale, J. R., White, T. P., Mayhew, S. D., Wilson, R. S., Rollings, D. T., Khalsa, S., Arvanitis, T. N., & Bagshaw, A. P. (2016). Altered thalamocortical and intra-thalamic functional connectivity during light sleep compared with wake. NeuroImage, 125, 657-667. 

Helfrich, R. F., Lendner, J. D., Mander, B. A., Guillen, H., Paff, M., Mnatsakanyan, L., Vadera, S., Walker, M. P., Lin, J. J., & Knight, R. T. (2019). Bidirectional prefrontal-hippocampal dynamics organize information transfer during sleep in humans. Nature Communications, 10(1), 3572. 

Helfrich, R. F., Mander, B. A., Jagust, W. J., Knight, R. T., & Walker, M. P. (2018). Old brains come uncoupled in sleep: slow wave-spindle synchrony, brain atrophy, and forgetting. Neuron, 97(1), 221-230. e224. 

Horovitz, S. G., Fukunaga, M., de Zwart, J. A., van Gelderen, P., Fulton, S. C., Balkin, T. J., & Duyn, J. H. (2008). Low frequency BOLD fluctuations during resting wakefulness and light sleep: A simultaneous EEG‐fMRI study. Human brain mapping, 29(6), 671-682. 

Huang, Q., Xiao, Z., Yu, Q., Luo, Y., Xu, J., Qu, Y., Dolan, R., Behrens, T., & Liu, Y. (2024). Replay-triggered brain-wide activation in humans. Nature Communications, 15(1), 7185. 

Ilhan-Bayrakcı, M., Cabral-Calderin, Y., Bergmann, T. O., Tüscher, O., & Stroh, A. (2022). Individual slow wave events give rise to macroscopic fMRI signatures and drive the strength of the BOLD signal in human resting-state EEG-fMRI recordings. Cerebral Cortex, 32(21), 4782-4796. 

Laufs, H. (2008). Endogenous brain oscillations and related networks detected by surface EEG‐combined fMRI. Human brain mapping, 29(7), 762-769. 

Laufs, H., Walker, M. C., & Lund, T. E. (2007). ‘Brain activation and hypothalamic functional connectivity during human non-rapid eye movement sleep: an EEG/fMRI study’—its limitations and an alternative approach. Brain, 130(7), e75. 

Margulies, D. S., Ghosh, S. S., Goulas, A., Falkiewicz, M., Huntenburg, J. M., Langs, G., Bezgin, G., Eickhoff, S. B., Castellanos, F. X., & Petrides, M. (2016). Situating the default-mode network along a principal gradient of macroscale cortical organization. Proceedings of the National Academy of Sciences, 113(44), 12574-12579. 

Massimini, M., Huber, R., Ferrarelli, F., Hill, S., & Tononi, G. (2004). The sleep slow oscillation as a traveling wave. Journal of Neuroscience, 24(31), 6862-6870. 

Moehlman, T. M., de Zwart, J. A., Chappel-Farley, M. G., Liu, X., McClain, I. B., Chang, C., Mandelkow, H., Özbay, P. S., Johnson, N. L., & Bieber, R. E. (2019). All-night functional magnetic resonance imaging sleep studies. Journal of neuroscience methods, 316, 83-98. 

Molle, M., Bergmann, T. O., Marshall, L., & Born, J. (2011). Fast and slow spindles during the sleep slow oscillation: disparate coalescence and engagement in memory processing. Sleep, 34(10), 1411-1421. 

Ngo, H.-V., Fell, J., & Staresina, B. (2020). Sleep spindles mediate hippocampal-neocortical coupling during long-duration ripples. Elife, 9, e57011. 

Picchioni, D., Horovitz, S. G., Fukunaga, M., Carr, W. S., Meltzer, J. A., Balkin, T. J., Duyn, J. H., & Braun, A. R. (2011). Infraslow EEG oscillations organize large-scale cortical– subcortical interactions during sleep: a combined EEG/fMRI study. Brain research, 1374, 63-72. 

Schabus, M., Dang-Vu, T. T., Albouy, G., Balteau, E., Boly, M., Carrier, J., Darsaud, A., Degueldre, C., Desseilles, M., & Gais, S. (2007). Hemodynamic cerebral correlates of sleep spindles during human non-rapid eye movement sleep. Proceedings of the National Academy of Sciences, 104(32), 13164-13169. 

Schreiner, T., Kaufmann, E., Noachtar, S., Mehrkens, J.-H., & Staudigl, T. (2022). The human thalamus orchestrates neocortical oscillations during NREM sleep. Nature communications, 13(1), 5231. 

Schreiner, T., Petzka, M., Staudigl, T., & Staresina, B. P. (2021). Endogenous memory reactivation during sleep in humans is clocked by slow oscillation-spindle complexes. Nature Communications, 12(1), 3112. 

Singh, D., Norman, K. A., & Schapiro, A. C. (2022). A model of autonomous interactions between hippocampus and neocortex driving sleep-dependent memory consolidation. Proceedings of the National Academy of Sciences, 119(44), e2123432119. 

Spoormaker, V. I., Schröter, M. S., Gleiser, P. M., Andrade, K. C., Dresler, M., Wehrle, R., Sämann, P. G., & Czisch, M. (2010). Development of a large-scale functional brain network during human non-rapid eye movement sleep. Journal of Neuroscience, 30(34), 11379-11387. 

Staresina, B. P., Bergmann, T. O., Bonnefond, M., van der Meij, R., Jensen, O., Deuker, L., Elger, C. E., Axmacher, N., & Fell, J. (2015). Hierarchical nesting of slow oscillations, spindles and ripples in the human hippocampus during sleep. Nature Neuroscience, 18(11), 1679-1686. 

Staresina, B. P., Niediek, J., Borger, V., Surges, R., & Mormann, F. (2023). How coupled slow oscillations, spindles and ripples coordinate neuronal processing and communication during human sleep. Nature Neuroscience, 1-9. 

Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. Nature methods, 8(8), 665-670. 

Yeshurun, Y., Nguyen, M., & Hasson, U. (2021). The default mode network: where the idiosyncratic self meets the shared social world. Nature Reviews Neuroscience, 1-12.

eLife Assessment

The findings of this study are valuable, offering insights into the neural representation of reversal probability in decision-making tasks, with potential implications for understanding flexible behavior in changing environments. The study contains interesting comparisons between neural data and models, including evidence for partial consistency with line attractor models in this probabilistic reversal learning task. However, it remains incomplete due to issues related to how the RNN training and the analysis of its dynamics, which renders the evidence as not complete.

Reviewer #1 (Public review):

The authors aimed to investigate how the probability of a reversal in a decision-making task is computed in cortical neurons. They analyzed neural activity in the prefrontal cortex of monkeys and units in recurrent neural networks (RNNs) trained on a similar task. Their goal was to understand how the dynamical systems that implement computation perform a probabilistic reversal learning task in RNNs and nonhuman primates.

Major strengths and weaknesses:

Strengths:

(1) Integrative Approach: The study exemplifies a modern approach by combining empirical data from monkey experiments with computational modeling using RNNs. This integration allows for a more comprehensive understanding of the dynamical systems that implement computation in both biological and artificial neural networks.<br /> (2) The focus on using perturbations to identify causal relationships in dynamical systems is a good goal. This approach aims to go beyond correlational observations.<br /> (3) The revised manuscript provides a more nuanced interpretation of the dynamics, reconciling the observations with aspects of line attractor models.

Weaknesses:

(1) The use of targeted dimensionality reduction (TDR) to identify the axis determining reversal probability may not necessarily isolate the dimension along which the RNN computes reversal probability. This should be computed from the RNN update itself rather than through a readout of network variance. Depending on how this is formulated, it could be something like the Jacobian of the state update with respect to inputs at input onset and with respect to the state during relaxation dynamics. This is worth thinking through further. It's important to try to take advantage of access afforded by using RNNs rather than solely relying on analyses available to us in neural data.

Appraisal of aims and conclusions:

The authors have substantially revised their interpretation of the results to reconcile their findings with line attractor models. They now acknowledge that their observation of reward integration explaining reversal probability activity (x_rev) is compatible with line attractor models, which addresses one of my main concerns.

Their expanded analysis now differentiates between two activity modes: (1) substantial non-stationary dynamics during a trial (incompatible with line attractors) and (2) stationary and stable dynamics at trial start (compatible with point attractors and line attractor models). This dual characterization provides a more complete picture of the dynamical system and highlights the composability of dynamical features.

Likely impact and utility:

This work makes a stronger contribution to our understanding of how probabilistic information is represented in neural circuits with intervening behaviors. The augmented model that combines elements of attractor dynamics with non-stationary trajectories offers a more comprehensive framework for understanding neural computations in decision-making tasks.

The data and methods could be useful to the community. While the authors have improved their analysis of network dynamics, additional reverse engineering that takes full advantage of access to the RNN's update equations could further strengthen the work.

Reviewer #2 (Public review):

Summary:

In this work the authors trained RNN to perform a reversal task also performed by animals while PFC activity is recorded. The authors devised a new method to train RNN on this type of reversal task, which in principle ensures that the behavior of the RNN matches the behavior of the animal. They then performed some analysis of neural activity, both RNN and PFC recording, focusing on the neural representation of the reversal probability and its evolution across trials. Given the analysis presented, it has been difficult for me to asses at which point RNN can reasonably be compared to PFC recordings.

Strengths:

Focusing on a reversal task, the authors address a challenge in RNN training, as they do not use a standard supervised learning procedure where the desired output is available for each trial. They propose a new way of doing that.

They attempt to confront RNN and neural recordings in behaving animals.

Weaknesses:

It would be nice to better articulate the analysis results of the two training set-ups (with and without 0 response during fixation). The dynamical system analysis is confusing, the notions of stationary and non-stationary dynamics and its relationship with attractors are puzzling. Is there a line attractor in one case (with inputs orthogonal to the integration direction being called back to the attractor, and reward input aligned with the stable direction)? In the other case, do we have a cylindrical attracting manifold on which activity circles around and is pushed along the axis of the cylinder by reward inputs? Which case is closest to the PFC recordings?

Reviewer #3 (Public review):

Summary:

Kim et al. present a study of the neural dynamics underlying reversal learning in monkey PFC and neural networks. Their main finding is that neural activity during fixation resembles a line attractor storing the current belief of the reversal state of the task. This is followed by richer dynamics unfolding throughout the remainder of the trial, which eventually converge to a new point on the line attractor by the start of the next trial. The idea of studying neural dynamics throughout the task (including intervening behaviour) is interesting, and the data provides some insights into the neural dynamics driving reversal learning. The modelling seems to support the analyses, but both the modelling and analyses also leave several open questions.

Strengths:

The paper addresses an interesting topic of the neural dynamics underlying reversal learning in PFC, using a combination of biological and simulated data. Reversal learning has been studied extensively in neuroscience, but this paper takes a step further by analysing neural dynamics throughout the trials instead of focusing on just the evidence integration epoch.

The authors show some close parallels between the experimental data and RNN simulations, both in terms of behaviour and neural dynamics. The analyses of how rewarded and unrewarded trials differentially affect dynamics throughout the trials in RNNs and PFC were particularly interesting. This work has the potential to provide new insights into the neural underpinnings of reversal learning.

Weaknesses:

Data analyses:

While the analyses seem mostly sound, one shortcoming is that they are all aligned to the inferred reversal trial rather than the true experimental reversal trial. For example, the analyses showing that 'x_rev' decays strongly after the reversal trial, irrespective of the reward outcome, seem like they are true essentially by design. The choice to align to the inferred reversal trial also makes this trial seem 'special' (e.g. in Fig 2 & Fig 6A), but it is unclear whether this is a real feature of the data or an artifact of effectively conditioning on a change in behaviour. It would be useful to investigate whether any of these analyses differ when aligned to the true reversal trial. It is also unsurprising that x_rev increases before the reversal and decreases after the reversal (it is hard to imagine a system where this is not the case), yet all of Fig 6 and several other analyses are devoted to this point.

Most of the analyses focus on the dynamics specifically in the x_rev subspace, but a major point of the paper is to say that biological (and artificial) networks may also have to do other things at different times in the trial. If that is the case, it would be interesting to also ask what happens in other subspaces of neural activity, which are not specifically related to evidence integration or choice - are there other subspaces that explain substantial variance? Do they relate to any meaningful features of the experiment?

This is especially important when considering analyses trying to establish the presence (or absence) of attractor dynamics in the circuit. In particular, activity in the x_rev subspace both affects and depends on other subspaces of neural activity, so it is not as meaningful to analyse the dynamics of this subspace in isolation. It would e.g. have been preferable to analyse the early-trial dynamics in the full state space and then possibly projecting onto x_rev, rather than first projecting activity onto x_rev and then fitting a linear autoregressive model.

Modelling:

There are a number of surprising and non-standard modelling choices made in this paper. For example, the choice to only use inhibitory neurons is non-conventional and it is not clear whether and how this impacts the results. The inputs are also provided without any learnable input weights, which makes it harder to interpret the input-driven dynamics during the different phases of a trial.

It is surprising that the RNN is "trained to flip its preferred choice a few trials after the inferred scheduled reversal trial", with the reversal trial inferred by an ideal Bayesian observer. A more natural approach would be to directly train the RNN to solve the task (by predicting the optimal choice) and then investigating the emergent behaviour & dynamics. If the authors prefer their imitation learning approach, it is also surprising that the network is trained to predict the reversal trial inferred using Bayesian smoothing instead of Bayesian filtering.

Finally, it was surprising that the network is trained and tested with different block lengths (24 & 36 trials, respectively), and it is not mentioned whether or how this affects behaviour.

Author response:

The following is the authors’ response to the original reviews

Main revision made to the manuscript

The main revision made to the manuscript is to reconcile our findings with the line attractor model. The revision is based on Reviewer 1’s comment on reinterpreting our results as a superposition of an attractor model with fast timescale dynamics. We expanded our analysis regime to the start of a trial and characterized the overall within-trial dynamics to reinterpret our findings.

We first acknolwedge that our results are not in contradiction with evidence integration on a line attractor. As pointed out by the reviewers, our finding that the integration of reward outcome explains the reversal probability activity x_rev (Figure 3) is compatible with the line attractor model. However, the reward integration equation is an algebraic relation and does not characterize the dynamics of reversal probability activity. So a closer analysis on the neural dynamics is needed to assess the feasibility of line attractor.

In the revised manuscript, we show that x_rev exhibits two different activity modes (Figure 4). First, x_rev has substantial non-stationary dynamics during a trial, and this non-stationary activity is incompatible with the line attractor model, as claimed in the original manuscript. Second, we present new results showing that x_rev is stationary (i.e., constant in time) and stable (i.e., contracting) at the start of a trial. These two properties of x_rev support that it is a point attractor at the start of a trial and is compatible with the line attractor model. 

We further analyze how the two activity modes are linked (Figure 4, Support vector regression). We show that the non-stationary activity is predictable from the stationary activity if the underlying dynamics can be inferred. In other words, the non-stationary activity during a trial is generated by an underlying dynamics with the initial condition provided by the stationary state at the start of trial.

These results suggest an extension of the line attractor model where an attractor state at the start of a trial provides an initial condition from which non-stationary activity is generated during a trial by an underlying dynamics associated with task-related behavior (Figure 4, Augmented model). 

The separability of non-stationary trajectories (Figure 5 and 6) is a property of the non-stationary dynamics that allows separable points in the initial stationary state to remain separable during a trial, thus making it possible to represent distinct probabilistic values in non-stationary activity.

This revised interpretation of our results (1) retains our original claim that the non-stationary dynamics during a trial is incompatible with the line attractor model and (2) introduces attractor state at the start of a trial which is compatible with the line attractor model. Our anlaysis shows that the two activity modes are linked by an underlying dynamics, and the attractor state serves as initial state to launch the non-stationary activity.

Responses to the Public Reviews:

Reviewer # 1:

(1) To provide better explanation of the reversal learning task and network training method, we added detailed description of RNN and monkey task structure (Result Section 1), included a schematic of target outputs (Figure1B), explained the rationale behind using inhibitory network model (Method Section 1) and explained the supervised RNN training scheme (Result Section 1). This information can also be found in the Methods.

(2) Our understanding is that the augmented model discussed in the previous page is aligned with the model suggested by Reviewer 1: “a curved line attractor, with faster timescale dynamics superimposed on this structure”. It is likely that the “fast” non-stationary activity observed during the trial is driven by task-related behavior, thus is transient. For instance, we do not observe such non-stationary activity in the inter-trial-interval when the task-related behavior is absent. For this reason, the non-stationary trajectories were not considered to be part of the attractor. Instead, they are transient activity generated by the underlying neural dynamics associated with task-related behavior. We believe such characterization of faster timescale dynamics is consistent with Reviewer 1’s view and wanted to clarify that there are two different activity modes.

(3) We appreciate the reviewers (Reviewer 1 and Reviewer 2) comment that TDR may be limited in isolating the neural subspace of interest. Our study presents what could be learned from TDR but is by no means the only way to interpret the neural data. It would be of future work to apply other methods for isolating task-related neural activities.

We would appreciate it if the reviewers could share thoughts on what other alternative methods could better isolate the reversal probability activity.

Reviewer # 2:

(1) (i) We respectfully disagree with Reviewer 2’s comment that “no action is required to be performed by neurons in the RNN”. In our network setup, the output of RNN learns to choose a sign (+ or -), as Reviewer 2 pointed out, to make a choice. This is how the RNN takes an action. It is unclear to us what Reviewer 2 has intended by “action” and how reaching a target value (not just taking a sign) would make a significant difference in how the network performs the task. 

(ii)  From Reviewer 2’s comment that “no intervening behavior is thus performed by neurons”, we noticed that the term “intervening behavior” has caused confusion. It refers to task-related behavior, such as making choices or receiving reward, that the subject must perform across trials before reversing its preferred choice. These are the behaviors that intervene the reversal of preferred choice. To clarify its meaning, in the revised manuscript, we changed the term to “task-related behavior” and put them in context. For example, in the Introduction we state that “However, during a trial, task-related behavior, such as making decisions or receiving feedback, produced …”

(iii) As pointed out by Reviewer 2, the lack of fixation period in the RNN could make differences in the neural dynamics of RNN and PFC, especially at the start of a trial. We demonstrate this issue in Result Section 4 where we analyze the stationary activity at the start of a trial. We find that fixating the choice output to zero before making a choice promotes stationary activity and makes the RNN activity more similar to the PFC activity.

Reviewer #3:

(1) (i) In the previous study (Figure 1 in [Bartolo and Averbeck ‘20]), it was shown that neural activity can predict the behavioral reversal trial. This is the reason we examined the neural activity in the trials centered at the behavioral reversal trial. We explained in Result Section 2 that we followed this line of analysis in our study.

(ii) We would like to emphasize that the main point of Figures 4 and 5 is to show the separability of neural trajectories: the entire trajectory shifts without overlapping. It is not obvious that high-dimensional neural population activity from two trials should remain separated when their activities are compressed into a one-dimensional subspace. The onedimensional activities can easily collide since their activities are compressed into a lowdimensional space. We revised the manuscript to bring out these points. We added an opening paragraph that discusses separability of trajectories and revised the main text to bring out the findings on separability. 

(iii) We agree with Reviewer 3 that it would be interesting to look at what happens in other subspace of neural activity that are not related to reversal probability and characterize how different neural subspace interact with each. However, the focus of this paper was the reversal probability activity, and we’d consider these questions out of the scope of current paper. We point out that, using the same dataset, neural activity related to other experimental variables were analyzed in other papers [Bartolo and Averbeck ’20; Tang, Bartolo and Averbeck ‘21] 

(2) (i) In the revised manuscript, we added explanation on the rational behind choosing inhibitory network as a simplified model for the balanced state. In brief, strong inhibitory recurrent connections with strong excitatory external input operates in the balanced state, as in the standard excitatory-inhibitory network. We included references that studied this inhibitory network. We also explained the technical reason (GPU memory) for choosing the inhibitory model.

(ii) We thank the reviewer for pointing out that the original manuscript did not mention how the feedback and cue were initialized. They were random vectors sample from Gaussian distribution. We added this information in the revised manuscript. In our opinion, it is common to use random external inputs for training RNNs, as it is a priori unclear how to choose them. In fact, it is possible to analyze the effects of random feedback on one-dimensional x_rev dynamics by projecting the random feedback vector to the reversal probability vector. This is shown in Figure 4F.

(iii) We agree that it would be more natural to train the RNN to solve the task without using the Bayesian model. We point out this issue in the Discussion in the revised manuscript.

Recommendations for the authors:

Reviewer #1:

(1) My understanding of network training was that a Bayesian ideal observer signaled target output based on previous reward outcomes. However, the authors never mention that networks are trained by supervised learning in the main text until the last paragraph of the discussion. There is no mention that there was an offset in the target based on the behavior of the monkeys in the main text. These are really important things to consider in the context of the network solution after training. I couldn't actually find any figure that presents the target output for the network. Did I miss something key here?

In Result Section 1, we added a paragraph that describes in detail how the RNN is trained. We explained that the network is first simulated and then the choice outputs and reward outcomes are fed into the Bayesian model to infer the scheduled reversal trial. A few trials are added to the inferred reversal trial to obtain the behavioral reversal trial, as found in a previous study [Bartolo and Averbeck ‘20]. Then the network weights are updated by backpropagation-through-time via supervised learning. 

In the original manuscript, the target output for the network was described in Methods Section 2.5, Step 4. To make this information readily accessible, we added a schematic in Figure 1B that shows the scheduled, inferred and behavioral reversal trials. It also shows how the target choice ouputs are defined. They switch abruptly at the behavioral reversal trial.

(2) The role of block structure in the task is an important consideration. What are the statistics of block switches? The authors say on average the reversals are every 36 trials, but also say there are random block switches. The reviewer's notes suggest that both the networks and monkeys may be learning about the typical duration of blocks, which could influence their expectations of reversals. This aspect of the task design should be explained more thoroughly and considered in the context of Figure 1E and 5 results.

We provided more detailed description of the reversal learning task in Result Section 1. We clarified that (1) a task is completed by executing a block of fixed number of trials and (2) reversal of reward schedule occurrs at a random trial around the mid-trial in a block. The differences in the number of trials in a block that the RNNs (36) and the monkeys (80) perform are also explained. We also pointed out the differences in how the reversal trial is randomly sampled.

However, it is unclear what Reviewer 1 meant by random block switches. Our reversal learning task is completed when a block of fixed number of trials is executed. Reversal of reward schedule occurs only once on a randomly selected trial in the block, and the reversed reward schedule is maintained until the end of a block. It is different from other versions of reveral learning where the reward schedule switches multiple times across trials. We clarified this point in Result Section 1.

(3) The relationship between the supervised learning approach used in the RNNs and reinforcement learning was confused in the discussion. "Although RNNs in our study were trained via supervised learning, animals learn a reversal-learning task from reward feedback, making it into a reinforcement learning (RL) problem." This is fundamentally not true. In the case of this work, the outcome of the previous trial updates the target output, rather than the trial and error type learning as is typical in reinforcement learning. Networks are not learning by reinforcement learning and this statement is confusing.

We agree with Reviewer 1’s comment that the statement in the original manuscript is confusing. Our intention was to point out that our study used supervised learning, and this is different from animals learn by reinforcement learning in rea life. We revised the sentence in Discussion as follows:

“The RNNs in our study were trained via supervised learning. However, in real life, animals learn a reversal learning task via reinforcement learning (RL), i.e., learn the task from reward outcomes.”

(4) The distinction between line attractors and the dynamic trajectories described by the authors deserves further investigation. A significant concern arises from the authors' use of targeted dimensionality reduction (TDR), a form of regression, to identify the axis determining reversal probability. While this approach can reveal interesting patterns in the data, it may not necessarily isolate the dimension along which the RNN computes reversal probability. This limitation could lead to misinterpretation of the underlying neural dynamics.

a) This manuscript cites work described in "Prefrontal cortex as a meta-reinforcement learning system," which examined a similar task. In that study, the authors identified a v-shaped curve in the principal component space of network states, representing the probability of choosing left or right.

Importantly, this curve is topologically equivalent to a line and likely represents a line attractor. However, regressing against reversal probability in such a case would show that a single principal component (PC2) directly correlates with reversal probability.

b) The dynamics observed in the current study bear a striking resemblance to this structure, with the addition of intervening loops in the network state corresponding to within-trial state evolution. Crucially, these observations do not preclude the existence of a line attractor. Instead, they may reflect the network's need to produce fast timescale dynamics within each trial, superimposed on the slower dynamics of the line attractor.

c) This alternative interpretation suggests that reward signals could function as inputs that shift the network state along the line attractor, with information being maintained across trials. The fast "intervening behaviors" observed by the authors could represent faster timescale dynamics occurring on top of the underlying line attractor dynamics, without erasing the accumulated evidence for reversals.

d) Given these considerations, the authors' conclusion that their results are better described by separable dynamic trajectories rather than fixed points on a line attractor may be premature. The observed dynamics could potentially be reconciled with a more nuanced understanding of line attractor models, where the attractor itself may be curved and coexist with faster timescale dynamics.

We appreciate the insightful comments on (1) the similarity of the work by Wang et al ’18 with our findings and (2) an alternative interpretation that augments the line attractor with fast timescale dynamics. 

(1) We added a discussion of the work by Wang et al ’18 in Result Section 2 to point out the similarity of their findings in the principal component space with ours in the x_rev and x_choice space. We commented that such network dynamics could emerge when learning to perform the reversal learning the task, regardless of the training schemes. 

We also mention that the RL approach in Wang et al ’18 does not consider within-trial dynamics, therefore lacks the non-stationary activity observed during the trial in the PFC of monkeys and our trained RNNs.

(2) We revised our original manuscript substantially to reconcile the line attractor model with the nonstationary activity observed during a trial. 

Here are the highlights of the revised interpretation of the PFC and the RNN network activity

- The dynamics of x_rev consists of two activity modes, i.e., stationary activity at the start of a trial and non-stationary activity during the trial. Schematic of the augmented model that reconciles two activity modes is shown in Figure 4A. Analysis of the time derivative (dx_reverse / dt) and contractivity of the stationary state are shown in Figure 4B,C to demonstrate two activity modes.

- We discuss in Result Section 4 main text that the stationary activity is consistent with the line attractor model, but the non-stationary activity deviates from the model. 

- The two activity modes are linked dynamically. There is an underlying dynamics that can map the stationary state to the non-stationary trajectory. This is shown by predicting the nonstationary trajectory with the stationary state using a support vector regression model. The prediction results are shown in Figure 4D,E,F.

- We discuss in Result Section 4 an extension of the standard line attractor model: points on the line attractor can serve as initial states that launch non-stationary activity associated with taskrelated behavior.

- The separability of neural trajectories presented in Result Section 5 is framed as a property of the non-stationary dynamics associated with task-related behavior.

To strengthen their claims, the authors should:

(1) Provide a more detailed description of their RNN training paradigm and task structure, including clear illustrations of target outputs.

(2) Discuss how their findings relate to and potentially extend previous work on similar tasks, particularly addressing the similarities and differences with the v-shaped state organization observed in reinforcement learning contexts. (https://www.nature.com/articles/s41593-018-0147-8 Figure1).

(3) Explore whether their results could be consistent with a curved line attractor model, rather than treating line attractors and dynamic trajectories as mutually exclusive alternatives.

Our response to these three comments is described above.

Addressing these points would significantly enhance the impact of the study and provide a more nuanced understanding of how reversal probabilities are represented in neural circuits.

In conclusion, while this study provides interesting insights into the neural representation of reversal probability, there are several areas where the methodology and interpretations could be refined.

Additional Minor Concerns:

(1) Network Training and Reversal Timing: The authors mention that the network was trained to switch after a reversal to match animal behavior, stating "Maximum a Posterior (MAP) of the reversal probability converges a few trials past the MAP estimate." More explanation of how this training strategy relates to actual animal behavior would enhance the reader's understanding of the meaning of the model's similarity to animal behavior in Figure 1.

In Method Section 2.5, we described how our observation that the running estimate of MAP converges a few trials after the actual MAP is analogous to the animal’s reversal behavior.

“This observation can be interpreted as follows. If a subject performing the reversal learning task employs the ideal observer model to detect the trial at which reward schedule is reversed, the subject can infer the reversal of reward schedule a few trials past the actual reversal and then switch its preferred choice. This delay in behavioral reversal, relative to the reversal of reward schedule, is analogous to the monkeys switching their preferred choice a few trials after the reversal of reward schedule.”

In Step 4, we also mentioned that the target choice outputs are defined based on our observation in Step 3.

“We used the observation from Step 3 to define target choice outputs that switch abruptly a few trials after the reversal of reward schedule, denoted as $t^*$ in the following. An example of target outputs are shown in Fig.\,\ref{fig_behavior}B.”

(2) How is the network simulated in step 1 of training? Is it just randomly initialized? What defines this network structure?

The initial state at the start of a block was random. We think the initial state is less relevant as the external inputs (i.e., cue and feedback) are strong and drive the network dynamics. We mentioned these setup and observation in Step 1 of training.

“Step 1. Simulate the network starting from a random initial state, apply the external inputs, i.e., cue and feedback inputs, at each trial and store the network choices and reward outcomes at all the trials in a block. The network dynamics is driven by the external inputs applied periodically over the trials.”

(3) Clarification on Learning Approach: More description of the approach in the main text would be beneficial. The statement "Here, we trained RNNs that learned from a Bayesian inference model to mimic the behavioral strategies of monkeys performing the reversal learning task [2, 4]" is somewhat confusing, as the model isn't directly fit to monkey data. A more detailed explanation of how the Bayesian inference model relates to monkey behavior and how it's used in RNN training would improve clarity.

We described the learning approach in more detail, but also tried to be concise without going into technical details.

We revised the sentence in Introduction as follows:

“We sought to train RNNs to mimic the behavioral strategies of monkeys performing the reversal learning task. Previous studies \cite{costa2015reversal, bartolo2020prefrontal} have shown that a Bayesian inference model can capture a key aspect of the monkey's behavioral strategy, i.e., adhere to the preferred choice until the reversal of reward is detected and then switch abruptly. We trained the RNNs to replicate this behavioral strategy by training them on target behaviors generated from the Bayesian model.”

We also added a paragraph in Result Section 1 that explains in detail how the training approach works.

(4) In Figure 1B, it would be helpful to show the target output.

We added a figure in Fig1B that shows a schematic of how the target output is generated.

(5) An important point to consider is that a line attractor can be curved while still being topologically equivalent to a line. This nuance makes Figure 4A somewhat difficult to interpret. It might be helpful to discuss how the observed dynamics relate to potentially curved line attractors, which could provide a more nuanced understanding of the neural representations.

As discussed above, we interpret the “curved” activity during the trial as non-stationary activity. We do not think this non-stationary activity would be characterized as attractor. Attractor is (1) a minimal set of states that is (2) invariant under the dynamics and (3) attracting when perturbed into its neighborhood [Strogatz, Nonlinear dynamics and chaos]. If we consider the autonomous system without the behavior-related external input as the base system, then the non-stationary states could satisfy (2) and (3) but not (1), so they are not part of the attractor. If we include the behavior-related external input to the autonomous dynamics, then it may be possible that the non-stationary trajectories are part of the attractor. We adopted the former interpretation as the behavior-related inputs are external and transient.

(6) The results of the perturbation experiments seem to follow necessarily from the way x_rev was defined. It would be valuable to clarify if there's more to these results than what appears to be a direct consequence of the definition, or if there are subtleties in the experimental design or analysis that aren't immediately apparent.

The neural activity x_rev is correlated to the reversal probability, but it is unclear if the activity in this neural subspace is causally linked to behavioral variables, such as choice output. We added this explanation at the beginning of Results Section 7 to clarify the reason for performing the perturbation experiments.

“The neural activity $x_{rev}$ is obtained by identifying a neural subspace correlated to reversal probability. However, it remains to be shown if activity within this neural subspace is causally linked to behavioral variables, such as choice output.”

Reviewer #2:

Below is a list of things I have found difficult to understand, and been puzzled/concerned about while reading the manuscript:

(1) It would be nice to say a bit more about the dataset that has been used for PFC analysis, e.g. number of neurons used and in what conditions is Figure 2A obtained (one has to go to supplementary to get the reference).

We added information about the PFC dataset in the opening paragraph of Result Section 2 to provide an overview of what type of neural data we’ve analyzed. It includes information about the number of recorded neurons, recording method and spike binning process.

(2) It would be nice to give more detail about the monkey task and better explain its trial structure.

In Result Section 1 we added a description of the overall task structure (and its difference with other versions of revesal learning task), the RNN / monkey trial structure and differences in RNN and monkey tasks.

(3) In the introduction it is mentioned that during the hold period, the probability of reversal is represented. Where does this statement come from?

The fact that neural activity during a hold period, i.e., fixation period before presenting the target images, encodes the probability of reversal was demonstrated in a previous study (Bartolo and Averbeck ’20). 

We realize that our intention was to state that, during the hold period, the reversal probability activity is stationary as in the line attractor model, instead of focusing on that the probability of reversal is represented during this period. We revised the sentence to convey this message. In addition, we revised the entire paragraph to reinterpret our findings: there are two activity modes where the stationary activity is consistent with the line attractor model but the non-stationary activity deviates from it.

(4) "Around the behavioral reversal trial, reversal probabilities were represented by a family of rankordered trajectories that shifted monotonically". This sentence is confusing and hard to understand.

Thank you for point this out. We rewrote the paragraph to reflect our revised interpretation. This sentence was removed, as it can be considered as part of the result on separable trajectories.

(5) For clarity, in the first section, when it is written that "The reversal behavior of trained RNNs was similar to the monkey's behavior on the same task" it would be nice to be more precise, that this is to be expected given the strategy used to train the network.

We removed this sentence as it makes a blanket statement. Instead, we compared the behavioral outputs of the RNNs and the monkeys one by one.

We added a sentence in Result Section 1 that the RNN’s abrupt behavioral reversal is expected as they are trained to mimic the target choice outputs of the Bayesian model.

“Such abrupt reversal behavior was expected as the RNNs were trained to mimic the target outputs of the Bayesian inference model.”

(6) What is the value of tau used in eq (1), and how does it compare to trial duration?

We described the value of time constant tau in Eq (1) and also discussed in Result Section 1 that tau=20ms is much faster than trial duration 500ms, thus the persistent behavior seen in trained RNNs is due to learning.

(7) It would be nice to expand around the notion of « temporally flexible representation » to help readers grasp what this means.

Instead of stating that the separable dynamic trajectories have “temporally flexible representation”, we break down in what sense it is temporally flexible: separable dynamic trajectories can accommodate the effects that task-related behavior have on generating non-stationary neural dynamics.

“In sum, our results show that, in a probabilistic reversal learning task, recurrent neural networks encode reversal probability by adopting, not only stationary states as in a line attractor, but also separable dynamic trajectories that can represent distinct probabilistic values while accommodating non-stationary dynamics associated with task-related behavior.”

Reviewer #3:

(1) Data:

It would be useful to describe the experimental task, recording setup, and analyses in much more detail - both in the text and in the methods. What part of PFC are the recordings from? How many neurons were recorded over how many sessions? Which other papers have they been used in? All of these things are important for the reader to know, but are not listed anywhere. There are also some inconsistencies, with the main text e.g. listing the 'typical block length' as 36 trials, and the methods listing the block length as 24 trials (if this is a difference between the biological data and RNN, that should be more explicit and motivated).

We provided more detailed description of the monkey experimental task and PFC recordings in Result Section 1. We also added a new section in Methods 2.1 to describe the monkey experiment.

The experimental analyses should be explained in more detail in the methods. There is e.g. no detailed description of the analysis in Figure 6F.

We added a new section in Methods 6 to describe how the residual PFC activity is computed. It also describes the RNN perturbation experiments.

Finally, it would be useful for more analyses of monkey behaviour and performance, either in the main text or supplementary figures.

We did not pursue this comment as it is unclear how additional behavioral analyses would improve the manuscript.

(2) Model:

When fitting the network, 'step 1' of training in 2.3 seems superfluous. The posterior update from getting a reward at A is the same as that from not getting a reward at B (and vice versa), and it is therefore completely independent of the network choice. The reversal trial can therefore be inferred without ever simulating the network, simply by generating a sample of which trials have the 'good' option being rewarded and which trials have the 'bad' option being rewarded.

We respectfully disagree with Reviewer 3’s comment that the reversal trial can be inferred without ever simulating the network. The only way for the network to know about the underlying reward schedule is to perform the task by itself. By simulating the network, it can sample the options and the reward outcomes. 

Our understanding is that Review 3 described a strategy that a human would use to perform this task. Our goal was to train the RNN to perform the task.

Do the blocks always start with choice A being optimal? Is everything similar if the network is trained with a variable initial rewarded option? E.g. in Fig 6, would you see the appropriate swap in the effect of the perturbation on choice probability if choice B was initially optimal?

Thank you for pointing out that the initial high-value option can be random. When setting up the reward schedule, the initial high-value option was chosen randomly from two choice outputs and, at the scheduled reversal, it was switched to the other option. We did not describe this in the original manuscript.

We added a descrption in Training Scheme Step 4 that the the initial high-value option is selected randomly. This is also explained in Result Section 1 when we give an overview of the RNN training procedure.

(3) Content:

It is rarely explained what the error bars represent (e.g. Figures 3B, 4C, ...) - this should be clear in all figures.

We added that the error bars represent the standard error of mean.

Figure 2A: this colour scheme is not great. There are abrupt colour changes both before and after the 'reversal' trial, and both of the extremes are hard to see.

We changed the color scheme to contrast pre- and post-reversal trials without the abrupt color change.

Figure 3E/F: how is prediction accuracy defined?

We added that the prediction accuracy is based on Pearson correlation.

Figure 4B: why focus on the derivative of the dynamics? The subsequent plots looking at the actual trajectories are much easier to understand. Also - what is 'relative trial' relative to?

The derivative was analyzed to demonstrate stationarity or non-stationarity of the neural activity. We think it will be clearer in the revised manuscript that the derivative allows us to characterize those two activity modes.

Relative trial number indicate the trial position relative to the behavioral reversal trial. We added this description to the figures when “relative trial” is used.

Figure 4C: what do these analyses look like if you match the trial numbers for the shift in trajectories? As it is now, there will presumably be more rewarded trials early and late in each block, and more unrewarded trials around the reversal point. Does this introduce biases in the analysis? A related question is (i) why the black lines are different in the top and bottom plots, and (ii) why the ends of the black lines are discontinuous with the beginnings of the red/blue lines.

We could not understand what Reviewer 3 was asking in this comment. It’d help if Review 3 could clarify the following question:

“Figure 4C: what do these analyses look like if you match the trial numbers for the shift in trajectories?”

Question (i): We wanted to look at how the trajectory shifts in the subsequent trial if a reward is or is not received in the current trial. The top panel analyzed all the trials in which the subsquent trial did not receive a reward. The bottom panel analyzed all the trials in which the subsequent trial received a reward. So, the trials analyzed in the top and bottom panels are different, and the black lines (x_rev of “current” trial) in the top and bottom panels are different.

Question (ii): Black line is from the preceding trial of the red/blue lines, so if trials are designed to be continuous with the inter-trial-interval, then black and red/blue should be continuous. However, in the monkey experiment, the inter-trial-intervals were variable, so the end of current trial does not match with the start of next trial. The neural trajectories presented in the manuscript did not include the activity in this inter-trial-interval.

Figure 6C: are the individual dots different RNNs? Claiming that there is a decrease in Delta x_choice for a v_+ stimulation is very misleading.

Yes individual dots are different RNN perturbations. We added explanation about the dots in Figure7C caption. 

We agree with the comment that \Delta x_choice did not decrease. This sentence was removed. Instead, we revised the manuscript to state that x_choice for v_+ stimulation was smaller than the x_choice for v_- stimulation. We performed KS-test to confirm statistical significance.

Discussion: "...exhibited behaviour consistent with an ideal Bayesian observer, as found in our study". The RNN was explicitly trained to reproduce an ideal Bayesian observer, so this can only really be considered an assumption (not a result) in the present study.

We agree that the statement in the original manuscript is inaccurate. It was revised to reflect that, in the other study, behavior outputs similar to a Bayesian observer emerged by simply learning to do the task, intead of directly mimicking the outputs of Bayesian observer as done in our study.

“Authors showed that trained RNNs exhibited behavior outputs consistent with an ideal Bayesian observer without explicitly learning from the Bayesian observer. This finding shows that the behavioral strategies of monkeys could emerge by simply learning to do the task, instead of directly mimicking the outputs of Bayesian observer as done in our study.”

Methods: Would the results differ if your Bayesian observer model used the true prior (i.e. the reversal happens in the middle 10 trials) rather than a uniform prior? Given the extensive literature on prior effects on animal behaviour, it is reasonable to expect that monkeys incorporate some non-uniform prior over the reversal point.

Thank you for pointing out the non-uniform prior. We haven’t conducted this analysis, but would guess that the convergence to the posterior distribution would be faster. We’d have to perform further analysis, which is out of the scope of this paper, to investigate whether the posteior distribution would be different from what we obtained from uniform prior.

Making the code available would make the work more transparent and useful to the community.

The code is available in the following Github repository: https://github.com/chrismkkim/LearnToReverse

eLife Assessment

This valuable paper uses a quantitative modeling approach to explore a well-studied transition in motor behavior in the nematode C. elegans. The authors provide solid evidence that this transition, which has been considered by many in the field to be a two-state behavior, can instead be described as a process whose parameters are smoothly modulated within a single state. This finding provides insight into the relationships between latent internal states and observable behavioral states, and suggests that relatively simple neuronal mechanisms can drive behavioral sequences that appear more complex.

Reviewer #1 (Public review):

Summary of what the authors were trying to achieve

This paper concerns mechanisms of foraging behavior in C. elegans. Upon removal from food, C. elegans first executes a stereotypical local search behavior in which it explores a small area by executing many random, undirected reversals and turns called "reorientations." If the worm fails to find food, it transitions to a global search in which it explores larger areas by suppressing reorientations and executing long forward runs (Hills et al., 2004). At the population level, reorientation rate declines gradually. Nevertheless, about 50% of individual worms appear to exhibit an abrupt transition between local and global search, which is evident as a discrete transition from high to low reorientation rate (Lopez-Cruz et al., 2019). This observation has given rise to the hypothesis that local and global search correspond to separate internal states with the possibility of sudden transitions between them (Calhoun et al., 2014). The objective of the paper is to demonstrate that is not necessary to posit distinct internal states to account for discrete transitions from high to low reorientation rate. On the contrary, discrete transitions can occur simply because of the stochastic nature of the reorientation behavior itself.

Major strengths and weaknesses of the methods and results

• The model was not explicitly designed to match the sudden, stable changes in reorientation rates observed in the experimental data from individual worms. Kinetic parameters were simply chosen to match the average population behavior. Nevertheless, many sudden stable changes in reorientation rates occurred. This is a strong argument that apparent state changes can arise as an epiphenomenon of stochastic processes.

• The new stochastic model is more parsimonious than reorientation-state change model because it posits one state rather than two.

• A prominent feature of the empirical data is that 50% of the worms exhibit a single (apparent) state change and the rest show either no state changes or multiple state changes. Does the model reproduce these proportions? This obvious question was not addressed.

• There is no obvious candidate for the neuronal basis of the decaying factor M. The authors speculate that decreasing sensory neuron activity might be the correlate of M but then provide contradictory evidence that seems to undermine that hypothesis. The absence of a plausible neuronal correlate of M weakens the case for the model.

Appraisal of whether the authors achieved their aims, and whether the results support their conclusions

The authors have made a solid case that is not necessary to posit distinct internal states to account for discrete transitions from high to low reorientation rate. On the contrary, discrete transitions can occur simply because of the stochastic nature of the reorientation behavior itself.

Impact of the work on the field, and the utility of the methods and data to the community

Posting hidden internal states to explain behavioral sequences is gaining acceptance in behavioral neuroscience. The likely impact of the paper is to establish a compelling example of how statistical reasoning can reduce the number of hidden states to achieve more parsimonious models.

Reviewer #2 (Public review):

Summary:

In this study, the authors build a statistical model that stochastically samples from a time-interval distribution of reorientation rates. The form of the distribution is extracted from a large array of behavioral data, is then used to describe not only the dynamics of individual worms (including the inter-individual variability in behavior), but also the aggregate population behavior. The authors note that the model does not require an assumption about behavioral state transitions, or evidence accumulation, as has been done previously, but rather that the stochastic nature of behavior is "simply the product of stochastic sampling from an exponential function".

Strengths:

This model provides a strong juxtaposition to other foraging models in the worm. Rather than evoking a behavioral transition function (that might arise from a change in internal state or the activity of a cell type in the network), or evidence accumulation (which again maps onto a cell type, or the activity of a network) - this model explains behavior via the stochastic sampling of a function of an exponential decay. The underlying model and the dynamics being simulated, as well as the process of stochastic sampling are well described and the model fits the exponential function (equation 1) to data on a large array of worms exhibiting diverse behaviors (1600+ worms from Lopez-Cruz et al). The work of this study is able to explain or describe the inter-individual diversity of worm behavior across a large population. The model is also able to capture two aspects of the reorientations, including the dynamics (to switch or not to switch) and the kinetics (slow vs fast reorientations). The authors also work to compare their model to a few others including the Levy walk (whose construction arises from a Markov process) to a simple exponential distribution, all of which have been used to study foraging and search behaviors.

Weaknesses:

This manuscript has two weaknesses that dampen the enthusiasm for the results. First, in all of the examples the authors cite where a Gillespie algorithm is used to sample from a distribution, be it the kinetics associated with chemical dynamics, or a Lotka-Volterra Competition Model, there are underlying processes that govern the evolution of the dynamics, and thus the sampling from distributions. In one of their references for instance, the stochasticity arises from the birth and death rates, thereby influencing the genetic drift in the model. In these examples, the process governing the dynamics (and thus generating the distributions from which one samples) are distinct from the behavior being studied. In this manuscript, the distribution being sampled from is the exponential decay function of the reorientation rate (lines 100-102). This appears to be tautological - a decay function fitted to the reorientation data is then sampled to generate the distributions of the reorientation data. That the model performs well, and matches the data is commendable, but it is unclear how that could not be the case if the underlying function generating the distribution was fit to the data.

The second weakness is somewhat related to the first, in that absent an underlying mechanism or framework, one is left wondering what insight the model provides. Stochastic sampling a function generated by fitting the data to produce stochastic behavior is where one ends up in this framework, and the authors indeed point this out: "simple stochastic models should be sufficient to explain observably stochastic behaviors." (Line 233-234). But if that is the case, what do we learn about how the foraging is happening. The authors suggest that the decay parameter M can be considered a memory timescale; which offers some suggestion, but then go on to say that the "physical basis of M can come from multiple sources". Here is where one is left for want: The mechanisms suggested, including loss of sensory stimuli, alternations in motor integration, ionotropic glutamate signaling, dopamine, and neuropeptides are all suggested: this is basically all of the possible biological sources that can govern behavior, and one is left not knowing what insight the model provides. The array of biological processes listed are so variable in dynamics and meaning, that their explanation of what govern M is at best unsatisfying. Molecular dynamics models that generate distributions can point to certain properties of the model, such as the binding kinetics (on and off rates, etc.) as explanations for the mechanisms generating the distributions, and therefore point to how a change in the biology affects the stochasticity of the process. It is unclear how this model provides such a connection, especially taken in aggregate with the previous weakness.

Providing a roadmap of how to think about the processes generating M, the meaning of those processes in search, and potential frameworks that are more constrained and with more precise biological underpinning (beyond the array of possibilities described) would go a long way to assuaging the weaknesses.

Comments on revised version:

The authors have addressed the main concerns of the manuscript.

eLife Assessment

This study provides valuable insights into the evolutionary conservation of sex determination mechanisms in ants by identifying a candidate sex-determining region in a parthenogenetic species. The strength of evidence is solid, using genomic analyses to identify differences in heterozygosity between females and diploid males, though the conclusions are limited by the lack of functional analysis.

Reviewer #1 (Public review):

This study investigates the sex determination mechanism in the clonal ant Ooceraea biroi, focusing on a candidate complementary sex determination (CSD) locus-one of the key mechanisms supporting haplodiploid sex determination in hymenopteran insects. Using whole genome sequencing, the authors analyze diploid females and the rarely occurring diploid males of O. biroi, identifying a 46 kb candidate region that is consistently heterozygous in females and predominantly homozygous in diploid males. This region shows elevated genetic diversity, as expected under balancing selection. The study also reports the presence of an lncRNA near this heterozygous region, which, though only distantly related in sequence, resembles the ANTSR lncRNA involved in female development in the Argentine ant, Linepithema humile (Pan et al. 2024). Together, these findings suggest a potentially conserved sex determination mechanism across ant species. However, while the analyses are well conducted and the paper is clearly written, the insights are largely incremental. The central conclusion - that the sex determination locus is conserved in ants - was already proposed and experimentally supported by Pan et al. (2024), who included O. biroi among the studied species and validated the locus's functional role in the Argentine ant. The present study thus largely reiterates existing findings without providing novel conceptual or experimental advances.

Other comments:

The mapping is based on a very small sample size: 19 females and 16 diploid males, and these all derive from a single clonal line. This implies a rather high probability for false-positive inference. In combination with the fact that only 11 out of the 16 genotyped males are actually homozygous at the candidate locus, I think a more careful interpretation regarding the role of the mapped region in sex determination would be appropriate. The main argument supporting the role of the candidate region in sex determination is based on the putative homology with the lncRNA involved in sex determination in the Argentine ant, but this argument was made in a previous study (as mentioned above).<br /> In the abstract, it is stated that CSD loci have been mapped in honeybees and two ant species, but we know little about their evolutionary history. But CSD candidate loci were also mapped in a wasp with multi-locus CSD (study cited in the introduction). This wasp is also parthenogenetic via central fusion automixis and produces diploid males. This is a very similar situation to the present study and should be referenced and discussed accordingly, particularly since the authors make the interesting suggestion that their ant also has multi-locus CSD and neither the wasp nor the ant has tra homologs in the CSD candidate regions. Also, is there any homology to the CSD candidate regions in the wasp species and the studied ant?

The authors used different clonal lines of O. biroi to investigate whether heterozygosity at the mapped CSD locus is required for female development in all clonal lines of O. biroi (L187-196). However, given the described parthenogenesis mechanism in this species conserves heterozygosity, additional females that are heterozygous are not very informative here. Indeed, one would need diploid males in these other clonal lines as well (but such males have not yet been found) to make any inference regarding this locus in other lines.

Reviewer #2 (Public review):

The manuscript by Lacy et al. is well written, with a clear and compelling introduction that effectively conveys the significance of the study. The methods are appropriate and well-executed, and the results, both in the main text and supplementary materials, are presented in a clear and detailed manner. The authors interpret their findings with appropriate caution.

This work makes a valuable contribution to our understanding of the evolution of complementary sex determination (CSD) in ants. In particular, it provides important evidence for the ancient origin of a non-coding locus implicated in sex determination, and shows that, remarkably, this sex locus is conserved even in an ant species with a non-canonical reproductive system that typically does not produce males. I found this to be an excellent and well-rounded study, carefully analyzed and well contextualized.

That said, I do have a few minor comments, primarily concerning the discussion of the potential 'ghost' CSD locus. While the authors acknowledge (line 367) that they currently have no data to distinguish among the alternative hypotheses, I found the evidence for an additional CSD locus presented in the results (lines 261-302) somewhat limited and at times a bit difficult to follow. I wonder whether further clarification or supporting evidence could already be extracted from the existing data. Specifically:

(1) Line 268: I doubt the relevance of comparing the proportion of diploid males among all males between lines A and B to infer the presence of additional CSD loci. Since the mechanisms producing these two types of males differ, it might be more appropriate to compare the proportion of diploid males among all diploid offspring. This ratio has been used in previous studies on CSD in Hymenoptera to estimate the number of sex loci (see, for example, Cook 1993, de Boer et al. 2008, 2012, Ma et al. 2013, and Chen et al., 2021). The exact method might not be applicable to clonal raider ants, but I think comparing the percentage of diploid males among the total number of (diploid) offspring produced between the two lineages might be a better argument for a difference in CSD loci number.

(2) If line B indeed carries an additional CSD locus, one would expect that some females could be homozygous at the ANTSR locus but still viable, being heterozygous only at the other locus. Do the authors detect any females in line B that are homozygous at the ANTSR locus? If so, this would support the existence of an additional, functionally independent CSD locus.

(3) Line 281: The description of the two tra-containing CSD loci as "conserved" between Vollenhovia and the honey bee may be misleading. It suggests shared ancestry, whereas the honey bee csd gene is known to have arisen via a relatively recent gene duplication from fem/tra (10.1038/nature07052). It would be more accurate to refer to this similarity as a case of convergent evolution rather than conservation.

(4) Finally, since the authors successfully identified multiple alleles of the first CSD locus using previously sequenced haploid males, I wonder whether they also observed comparable allelic diversity at the candidate second CSD locus. This would provide useful supporting evidence for its functional relevance.

Overall, these are relatively minor points in the context of a strong manuscript, but I believe addressing them would improve the clarity and robustness of the authors' conclusions.

Reviewer #3 (Public review):

Summary:

The sex determination mechanism governed by the complementary sex determination (CSD) locus is one of the mechanisms that support the haplodiploid sex determination system evolved in hymenopteran insects. While many ant species are believed to possess a CSD locus, it has only been specifically identified in two species. The authors analyzed diploid females and the rarely occurring diploid males of the clonal ant Ooceraea biroi and identified a 46 kb CSD candidate region that is consistently heterozygous in females and predominantly homozygous in males. This region was found to be homologous to the CSD locus reported in distantly related ants. In the Argentine ant, Linepithema humile, the CSD locus overlaps with an lncRNA (ANTSR) that is essential for female development and is associated with the heterozygous region (Pan et al. 2024). Similarly, an lncRNA is encoded near the heterozygous region within the CSD candidate region of O. biroi. Although this lncRNA shares low sequence similarity with ANTSR, its potential functional involvement in sex determination is suggested. Based on these findings, the authors propose that the heterozygous region and the adjacent lncRNA in O. biroi may trigger female development via a mechanism similar to that of L. humile. They further suggest that the molecular mechanisms of sex determination involving the CSD locus in ants have been highly conserved for approximately 112 million years. This study is one of the few to identify a CSD candidate region in ants and is particularly noteworthy as the first to do so in a parthenogenetic species.

Strengths:

(1) The CSD candidate region was found to be homologous to the CSD locus reported in distantly related ant species, enhancing the significance of the findings.

(2) Identifying the CSD candidate region in a parthenogenetic species like O. biroi is a notable achievement and adds novelty to the research.

Weaknesses

(1) Functional validation of the lncRNA's role is lacking, and further investigation through knockout or knockdown experiments is necessary to confirm its involvement in sex determination.

(2) The claim that the lncRNA is essential for female development appears to reiterate findings already proposed by Pan et al. (2024), which may reduce the novelty of the study.

Author response:

Reviewer #1 (Public review):

This study investigates the sex determination mechanism in the clonal ant Ooceraea biroi, focusing on a candidate complementary sex determination (CSD) locus-one of the key mechanisms supporting haplodiploid sex determination in hymenopteran insects. Using whole genome sequencing, the authors analyze diploid females and the rarely occurring diploid males of O. biroi, identifying a 46 kb candidate region that is consistently heterozygous in females and predominantly homozygous in diploid males. This region shows elevated genetic diversity, as expected under balancing selection. The study also reports the presence of an lncRNA near this heterozygous region, which, though only distantly related in sequence, resembles the ANTSR lncRNA involved in female development in the Argentine ant, Linepithema humile (Pan et al. 2024). Together, these findings suggest a potentially conserved sex determination mechanism across ant species. However, while the analyses are well conducted and the paper is clearly written, the insights are largely incremental. The central conclusion - that the sex determination locus is conserved in ants - was already proposed and experimentally supported by Pan et al. (2024), who included O. biroi among the studied species and validated the locus's functional role in the Argentine ant. The present study thus largely reiterates existing findings without providing novel conceptual or experimental advances.

Although it is true that Pan et al., 2024 demonstrated (in Figure 4 of their paper) that the synteny of the region flanking ANTSR is conserved across aculeate Hymenoptera (including O. biroi), Reviewer 1’s claim that that paper provides experimental support for the hypothesis that the sex determination locus is conserved in ants is inaccurate. Pan et al., 2024 only performed experimental work in a single ant species (Linepithema humile) and merely compared reference genomes of multiple species to show synteny of the region, rather than functionally mapping or characterizing these regions.

Other comments:

The mapping is based on a very small sample size: 19 females and 16 diploid males, and these all derive from a single clonal line. This implies a rather high probability for false-positive inference. In combination with the fact that only 11 out of the 16 genotyped males are actually homozygous at the candidate locus, I think a more careful interpretation regarding the role of the mapped region in sex determination would be appropriate. The main argument supporting the role of the candidate region in sex determination is based on the putative homology with the lncRNA involved in sex determination in the Argentine ant, but this argument was made in a previous study (as mentioned above).

Our main argument supporting the role of the candidate region in sex determination is not based on putative homology with the lncRNA in L. humile. Instead, our main argument comes from our genetic mapping (in Fig. 2), and the elevated nucleotide diversity within the identified region (Fig. 4). Additionally, we highlight that multiple genes within our mapped region are homologous to those in mapped sex determining regions in both L. humile and Vollenhovia emeryi, possibly including the lncRNA.

In response to the Reviewer’s assertion that the mapping is based on a small sample size from a single clonal line, we want to highlight that we used all diploid males available to us. Although the primary shortcoming of a small sample size is to increase the probability of a false negative, small sample sizes can also produce false positives. We used two approaches to explore the statistical robustness of our conclusions. First, we generated a null distribution by randomly shuffling sex labels within colonies and calculating the probability of observing our CSD index values by chance (shown in Fig. 2). Second, we directly tested the association between homozygosity and sex using Fisher’s Exact Test (shown in Supplementary Fig. S2). In both cases, the association of the candidate locus with sex was statistically significant after multiple-testing correction using the Benjamini-Hochberg False Discovery Rate. These approaches are clearly described in the “CSD Index Mapping” section of the Methods.

We also note that, because complementary sex determination loci are expected to evolve under balancing selection, our finding that the mapped region exhibits a peak of nucleotide diversity lends orthogonal support to the notion that the mapped locus is indeed a complementary sex determination locus.

The fourth paragraph of the results and the sixth paragraph of the discussion are devoted to explaining the possible reasons why only 11/16 genotyped males are homozygous in the mapped region. The revised manuscript will include an additional sentence (in what will be lines 384-388) in this paragraph that includes the possible explanation that this locus is, in fact, a false positive, while also emphasizing that we find this possibility to be unlikely given our multiple lines of evidence.

In response to Reviewer 1’s suggestion that we carefully interpret the role of the mapped region in sex determination, we highlight our careful wording choices, nearly always referring to the mapped locus as a “candidate sex determination locus” in the title and throughout the manuscript. For consistency, the revised manuscript version will change the second results subheading from “The O. biroi CSD locus is homologous to another ant sex determination locus but not to honeybee csd” to “O. biroi’s candidate CSD locus is homologous to another ant sex determination locus but not to honeybee csd,” and will add the word “candidate” in what will be line 320 at the beginning of the Discussion, and will change “putative” to “candidate” in what will be line 426 at the end of the Discussion.

In the abstract, it is stated that CSD loci have been mapped in honeybees and two ant species, but we know little about their evolutionary history. But CSD candidate loci were also mapped in a wasp with multi-locus CSD (study cited in the introduction). This wasp is also parthenogenetic via central fusion automixis and produces diploid males. This is a very similar situation to the present study and should be referenced and discussed accordingly, particularly since the authors make the interesting suggestion that their ant also has multi-locus CSD and neither the wasp nor the ant has tra homologs in the CSD candidate regions. Also, is there any homology to the CSD candidate regions in the wasp species and the studied ant?

In response to Reviewer 1’s suggestion that we reference the (Matthey-Doret et al. 2019) study in the context of diploid males being produced via losses of heterozygosity during asexual reproduction, the revised manuscript will include the following sentence: “Therefore, if O. biroi uses CSD, diploid males might result from losses of heterozygosity at sex determination loci (Fig. 1C), similar to what is thought to occur in other asexual Hymenoptera that produce diploid males (Rabeling and Kronauer 2012; Matthey-Doret et al. 2019).”

We note, however, that in their 2019 study, Matthey-Doret et al. did not directly test the hypothesis that diploid males result from losses of heterozygosity at CSD loci during asexual reproduction, because the diploid males they used for their mapping study came from inbred crosses in a sexual population of that species.

We address this further below, but we want to emphasize that we do not intend to argue that O. biroi has multiple CSD loci. Instead, we suggest that additional, undetected CSD loci is one possible explanation for the absence of diploid males from any clonal line other than clonal line A. In response to Reviewer 1’s suggestion that we reference the (Matthey-Doret et al. 2019) study in the context of multilocus CSD, the revised manuscript version will include the following additional sentence in the fifth paragraph of the discussion: “Multi-locus CSD has been suggested to limit the extent of diploid male production in asexual species under some circumstances (Vorburger 2013; Matthey-Doret et al. 2019).”

Regarding Reviewer 2’s question about homology between the putative CSD loci from the (Matthey-Doret et al. 2019) study and O. biroi, we note that there is no homology. The revised manuscript version will have an additional Supplementary Table (which will be the new Supplementary Table S3) that will report the results of this homology search. The revised manuscript will also include the following additional sentence in the Results: “We found no homology between the genes within the O. biroi CSD index peak and any of the genes within the putative L. fabarum CSD loci (Supplementary Table S3).”

The authors used different clonal lines of O. biroi to investigate whether heterozygosity at the mapped CSD locus is required for female development in all clonal lines of O. biroi (L187-196). However, given the described parthenogenesis mechanism in this species conserves heterozygosity, additional females that are heterozygous are not very informative here. Indeed, one would need diploid males in these other clonal lines as well (but such males have not yet been found) to make any inference regarding this locus in other lines.

We agree that a full mapping study including diploid males from all clonal lines would be preferable, but as stated earlier in that same paragraph, we have only found diploid males from clonal line A. We stand behind our modest claim that “Females from all six clonal lines were heterozygous at the CSD index peak, consistent with its putative role as a CSD locus in all O. biroi.” In the revised manuscript version, this sentence (in what will be lines 199-201) will be changed slightly in response to a reviewer comment below: “All females from all six clonal lines (including 26 diploid females from clonal line B) were heterozygous at the CSD index peak, consistent with its putative role as a CSD locus in all O. biroi.”

Reviewer #2 (Public review):

The manuscript by Lacy et al. is well written, with a clear and compelling introduction that effectively conveys the significance of the study. The methods are appropriate and well-executed, and the results, both in the main text and supplementary materials, are presented in a clear and detailed manner. The authors interpret their findings with appropriate caution.

This work makes a valuable contribution to our understanding of the evolution of complementary sex determination (CSD) in ants. In particular, it provides important evidence for the ancient origin of a non-coding locus implicated in sex determination, and shows that, remarkably, this sex locus is conserved even in an ant species with a non-canonical reproductive system that typically does not produce males. I found this to be an excellent and well-rounded study, carefully analyzed and well contextualized.

That said, I do have a few minor comments, primarily concerning the discussion of the potential 'ghost' CSD locus. While the authors acknowledge (line 367) that they currently have no data to distinguish among the alternative hypotheses, I found the evidence for an additional CSD locus presented in the results (lines 261-302) somewhat limited and at times a bit difficult to follow. I wonder whether further clarification or supporting evidence could already be extracted from the existing data. Specifically:

We agree with Reviewer 2 that the evidence for a second CSD locus is limited. In fact, we do not intend to advocate for there being a second locus, but we suggest that a second CSD locus is one possible explanation for the absence of diploid males outside of clonal line A. In our initial version, we intentionally conveyed this ambiguity by titling this section “O. biroi may have one or multiple sex determination loci.” However, we now see that this leads to undue emphasis on the possibility of a second locus. In the revised manuscript, we will split this into two separate sections: “Diploid male production differs across O. biroi clonal lines” and “O. biroi lacks a tra-containing CSD locus.”

(1) Line 268: I doubt the relevance of comparing the proportion of diploid males among all males between lines A and B to infer the presence of additional CSD loci. Since the mechanisms producing these two types of males differ, it might be more appropriate to compare the proportion of diploid males among all diploid offspring. This ratio has been used in previous studies on CSD in Hymenoptera to estimate the number of sex loci (see, for example, Cook 1993, de Boer et al. 2008, 2012, Ma et al. 2013, and Chen et al., 2021). The exact method might not be applicable to clonal raider ants, but I think comparing the percentage of diploid males among the total number of (diploid) offspring produced between the two lineages might be a better argument for a difference in CSD loci number.

We want to re-emphasize here that we do not wish to advocate for there being two CSD loci in O. biroi. Rather, we want to explain that this is one possible explanation for the apparent absence of diploid males outside of clonal line A. We hope that the modifications to the manuscript described in the previous response help to clarify this.

Reviewer 2 is correct that comparing the number of diploid males to diploid females does not apply to clonal raider ants. This is because males are vanishingly rare among the vast numbers of females produced. We do not count how many females are produced in laboratory stock colonies, and males are sampled opportunistically. Therefore, we cannot report exact numbers. However, we will add the following sentence to the revised manuscript: “Despite the fact that we maintain more colonies of clonal line B than of clonal line A in the lab, all the diploid males we detected came from clonal line A.”

(2) If line B indeed carries an additional CSD locus, one would expect that some females could be homozygous at the ANTSR locus but still viable, being heterozygous only at the other locus. Do the authors detect any females in line B that are homozygous at the ANTSR locus? If so, this would support the existence of an additional, functionally independent CSD locus.

We thank the reviewer for this suggestion, and again we emphasize that we do not want to argue in favor of multiple CSD loci. We just want to introduce it as one possible explanation for the absence of diploid males outside of clonal line A.

The 26 sequenced diploid females from clonal line B are all heterozygous at the mapped locus, and the revised manuscript will clarify this in what will be lines 199-201. Previously, only six of those diploid females were included in Supplementary Table S2, and that will be modified accordingly.

(3) Line 281: The description of the two tra-containing CSD loci as "conserved" between Vollenhovia and the honey bee may be misleading. It suggests shared ancestry, whereas the honey bee csd gene is known to have arisen via a relatively recent gene duplication from fem/tra (10.1038/nature07052). It would be more accurate to refer to this similarity as a case of convergent evolution rather than conservation.

In the sentence that Reviewer 2 refers to, we are representing the assertion made in the (Miyakawa and Mikheyev 2015) paper in which, regarding their mapping of a candidate CSD locus that contains two linked tra homologs, they write in the abstract: “these data support the prediction that the same CSD mechanism has indeed been conserved for over 100 million years.” In that same paper, Miyakawa and Mikheyev write in the discussion section: “As ants and bees diverged more than 100 million years ago, sex determination in honey bees and V. emeryi is probably homologous and has been conserved for at least this long.”

As noted by Reviewer 2, this appears to conflict with a previously advanced hypothesis: that because fem and csd were found in Apis mellifera, Apis cerana, and Apis dorsata, but only fem was found in Mellipona compressipes, Bombus terrestris, and Nasonia vitripennis, that the csd gene evolved after the honeybee (Apis) lineage diverged from other bees (Hasselmann et al. 2008). However, it remains possible that the csd gene evolved after ants and bees diverged from N. vitripennis, but before the divergence of ants and bees, and then was subsequently lost in B. terrestris and M. compressipes. This view was previously put forward based on bioinformatic identification of putative orthologs of csd and fem in bumblebees and in ants [(Schmieder et al. 2012), see also (Privman et al. 2013)]. However, subsequent work disagreed and argued that the duplications of tra found in ants and in bumblebees represented convergent evolution rather than homology (Koch et al. 2014). Distinguishing between these possibilities will be aided by additional sex determination locus mapping studies and functional dissection of the underlying molecular mechanisms in diverse Aculeata.

Distinguishing between these competing hypotheses is beyond the scope of our paper, but the revised manuscript will include additional text to incorporate some of this nuance. We will include these modified lines below:

“A second QTL region identified in V. emeryi (V.emeryiCsdQTL1) contains two closely linked tra homologs, similar to the closely linked honeybee tra homologs, csd and fem (Miyakawa and Mikheyev 2015). This, along with the discovery of duplicated tra homologs that undergo concerted evolution in bumblebees and ants (Schmieder et al. 2012; Privman et al. 2013) has led to the hypothesis that the function of tra homologs as CSD loci is conserved with the csd-containing region of honeybees (Schmieder et al. 2012; Miyakawa and Mikheyev 2015). However, other work has suggested that tra duplications occurred independently in honeybees, bumblebees, and ants (Hasselmann et al. 2008; Koch et al. 2014), and it remains to be demonstrated that either of these tra homologs acts as a primary CSD signal in V. emeryi.”

(4) Finally, since the authors successfully identified multiple alleles of the first CSD locus using previously sequenced haploid males, I wonder whether they also observed comparable allelic diversity at the candidate second CSD locus. This would provide useful supporting evidence for its functional relevance.

As is already addressed in the final paragraph of the results and in Supplementary Fig. S4, there is no peak of nucleotide diversity in any of the regions homologous to V.emeryiQTL1, which is the tra-containing candidate sex determination locus (Miyakawa and Mikheyev 2015). In the revised manuscript, the relevant lines will be 307-310. We want to restate that we do not propose that there is a second candidate CSD locus in O. biroi, but we simply raise the possibility that multi-locus CSD *might* explain the absence of diploid males from clonal lines other than clonal line A (as one of several alternative possibilities).

Overall, these are relatively minor points in the context of a strong manuscript, but I believe addressing them would improve the clarity and robustness of the authors' conclusions.

Reviewer #3 (Public review):

Summary:

The sex determination mechanism governed by the complementary sex determination (CSD) locus is one of the mechanisms that support the haplodiploid sex determination system evolved in hymenopteran insects. While many ant species are believed to possess a CSD locus, it has only been specifically identified in two species. The authors analyzed diploid females and the rarely occurring diploid males of the clonal ant Ooceraea biroi and identified a 46 kb CSD candidate region that is consistently heterozygous in females and predominantly homozygous in males. This region was found to be homologous to the CSD locus reported in distantly related ants. In the Argentine ant, Linepithema humile, the CSD locus overlaps with an lncRNA (ANTSR) that is essential for female development and is associated with the heterozygous region (Pan et al. 2024). Similarly, an lncRNA is encoded near the heterozygous region within the CSD candidate region of O. biroi. Although this lncRNA shares low sequence similarity with ANTSR, its potential functional involvement in sex determination is suggested. Based on these findings, the authors propose that the heterozygous region and the adjacent lncRNA in O. biroi may trigger female development via a mechanism similar to that of L. humile. They further suggest that the molecular mechanisms of sex determination involving the CSD locus in ants have been highly conserved for approximately 112 million years. This study is one of the few to identify a CSD candidate region in ants and is particularly noteworthy as the first to do so in a parthenogenetic species.

Strengths:

(1) The CSD candidate region was found to be homologous to the CSD locus reported in distantly related ant species, enhancing the significance of the findings.

(2) Identifying the CSD candidate region in a parthenogenetic species like O. biroi is a notable achievement and adds novelty to the research.

Weaknesses

(1) Functional validation of the lncRNA's role is lacking, and further investigation through knockout or knockdown experiments is necessary to confirm its involvement in sex determination.

See response below.

(2) The claim that the lncRNA is essential for female development appears to reiterate findings already proposed by Pan et al. (2024), which may reduce the novelty of the study.

We do not claim that the lncRNA is essential for female development in O. biroi, but simply mention the possibility that, as in L. humile, it is somehow involved in sex determination. We do not have any functional evidence for this, so this is purely based on its genomic position immediately adjacent to our mapped candidate region. We agree with the reviewer that the study by Pan et al. (2024) decreases the novelty of our findings. Another way of looking at this is that our study supports and bolsters previous findings by partially replicating the results in a different species.

eLife Assessment

This manuscript presents an interesting new framework (VARX) for simultaneously quantifying effective connectivity in brain activity during sensory stimulation and how that brain activity is being driven by that sensory stimulation. The reviewers thought the model was original and its conclusion that intrinsic connectivity is reduced (rather than increased) during sensory stimulation is very interesting, but that for ideal performance, one must specify all sensory features in the model, which is not possible. Overall, however, this work is important with convincing evidence for its conclusions - it will be of interest to neuroscientists working on brain connectivity and dynamics.

Reviewer #1 (Public review):

This manuscript presents an interesting new framework (VARX) for simultaneously quantifying effective connectivity in brain activity during sensory stimulation and how that brain activity is being driven by that sensory stimulation. The core idea is to combine the Vector Autoregressive model that is often used to infer Granger-causal connectivity in brain data with an encoding model that maps the features of a sensory stimulus to that brain data. The authors do a nice job of explaining the framework. And then they demonstrate its utility through some simulations and some analysis of real intracranial EEG data recorded from subjects as they watched movies. They infer from their analyses that the functional connectivity in these brain recordings is essentially unaltered during movie watching, that accounting for the driving movie stimulus can protect one against misidentifying brain responses to the stimulus as functional connectivity, and that recurrent brain activity enhances and prolongs the putative neural responses to a stimulus.

This manuscript presents an interesting new framework (VARX) for simultaneously quantifying effective connectivity in brain activity during sensory stimulation and how that brain activity is being driven by that sensory stimulation. Overall, I thought this was an interesting manuscript with some rich and intriguing ideas.

Comments on revisions:'

The responses to the previous comments are very helpful. I think the manuscript does a nice job now of presenting its interesting findings in a convincing and measured manner.

I had only one small remaining suggestion - to maybe link the finding of reduced intrinsic connectivity during stimulation to previous work on that topic. I thought of Nauhaus et al., Nature Neurosci, 2009.

Reviewer #2 (Public review):

Summary:

The authors apply the recently developed VARX model, which explicitly models intrinsic dynamics and the effect of extrinsic inputs, to simulated data and intracranial EEG recordings. This method provides a directed method of 'intrinsic connectivity'. They argue this model is better suited to the analysis of task neuroimaging data because it separates the intrinsic and extrinsic activity. They show: that intrinsic connectivity is largely unaltered during a movie-watching task compared to eyes open rest; intrinsic noise is reduced in the task; and there is intrinsic directed connectivity from sensory to higher-order brain areas.

Strengths:

(1) The paper tackles an important issue with an appropriate method.

(2) The authors validated their method on data simulated with a neural mass model.

(3) They use intracranial EEG, which provides a direct measure of neuronal activity.

(4) Code is made publicly available and the paper is written well.

Comments on revisions:'

The authors have addressed my comments.

eLife Assessment

This valuable work presents an interpretable protein-DNA Energy Associative (IDEA) model for predicting binding sites and affinities of DNA-binding proteins. While the method is convincing, it requires some adaptation for application to different proteins. The IDEA method is available and can be potentially used for predicting genome-wide protein-DNA binding sites.

Reviewer #2 (Public review):

Summary:

Zhang et al. present a methodology to model protein-DNA interactions via learning an optimizable energy model, taking into account a represetative bound structure for the system and binding data. The methodology is sound and interesting. They apply this model for predicting binding affinity data and binding sites in vivo.

Strengths:

The manuscript is well organized with good visualizations and is easy to follow. The methodology is discussed in detail. The IDEA energy model seems like an interesting way to study a protein-DNA system in the context of a given structure and binding data. The authors show that an IDEA model trained on one system can be transferred to other structurally similar systems. The authors show good performance in discriminating between binding-vs-decoy sequences for various systems, and binding affinity prediction. The authors also show evidence of the ability to predict genome-wide binding sites.

Weaknesses:

An energy-based model which needs to be optimized for specific systems is inherently an uncomfortable idea. Prediction of binding affinity is a well-studied domain and many competitors exist, some of which are well used. The usefulness of this method will be a test of time. The methodology is interpretable in a limited sense. The model is dependent on preserved interface geometry which might lead to suboptimal results for novel folds. The model predicts different output for reverse complement sequence (which in reality are the same as far as double helix is concerned). This is unintuitive.

Comments on revisions:

The authors have addressed my points regarding comparisons with existing methods, clarifying discussion terminologies and proper discussion of the existing literature. This resulted in a stronger manuscript with a clearer understanding of applicability.

Reviewer #3 (Public review):

Summary:

Protein-DNA interactions and sequence readout represent a challenging and rapidly evolving field of study. Recognizing the complexity of this task, the authors have developed a compact and elegant model. They applied well-established approaches to address a difficult problem, effectively enhancing the information extracted from sparse contact maps by integrating an artificial decoy sequence set and available experimental data. This has resulted in a practical tool that can be adapted for use with other proteins.

Strengths:

The authors integrate sparse information with available experimental data to construct a model whose utility extends beyond the limited set of structures used for training.

A comprehensive methods section is included, ensuring reproducibility.

The authors provide a well-represented performance comparison between their model and other existing models.

Additionally, the authors have shared their model as a GitHub project, reflecting their commitment to research transparency.

Weaknesses:

The coarse-graining procedure is quite convoluted, but the authors provide reasoning for the proposed scheme. The authors acknowledge discrepancies between data-driven and simulation models.

eLife Assessment

In this important paper, the authors use single-cell RNA sequencing to understand post-mitotic cone and rod developmental states and identify cone-specific features that contribute to retinoblastoma genesis. The authors report findings that have practical implications for retinal development, gene expression, and cell fate specification. The evidence is compelling as the experimental design and analysis are exceptionally rigorous.

Reviewer #1 (Public review):

Summary:

The authors have used full length single cell sequencing on a sorted population of human fetal retina to delineate expression patterns associated with the progression of progenitors to rod and cone photoreceptors. They find that rod.cone precursors contain a mix of rod/cone determinants, with a bias in both amounts and isoform balance likely deciding the ultimate cell fate. Markers of early rod/cone hybrids are clarified, and a gradient of lncRNAs is uncovered in maturing cones. Comparison of early rods and cones exposes an enriched MYCN regulon, as well as expression of SYK, which may contribute to tumor initiation in RB1 deficient cone precursors.

Strengths:

The insight into how cone and rod transcripts are mixed together at first is important and clarifies a long-standing notion in the field.

The discovery of distinct active vs inactive mRNA isoforms for rod and cone determinants is crucial to understand how cells make the decision to form one or the other cell type. This is only really possible with full length scRNAseq analysis.

New markers of subpopulations are also uncovered, such as CHRNA1 in rod/cone hybrids that seem to give rise to either rods or cones.

Regulon analyses provide insight into key transcription factor programs linked to rod or cone fates.

The gradient of lncRNAs in maturing cones is novel, and while the functional significance is unclear, it opens up a new line of questioning around photoreceptor maturation.

The finding that SYK mRNA is naturally expressed in cone precursors is novel, as previously it was assumed that SYK expression required epigenetic rewiring in tumors.

Weaknesses:

Functional data on many new hypothesis regarding potential players in cone genesis are not performed, but these are beyond the scope of the current work.

Validation of the SYK inhibitor data e.g. by genetic means, is not included, but the authors acknowledge this caveat throughout.

Reviewer #2 (Public review):

Summary:

The authors used deep full-length single-cell sequencing to study the human photoreceptor development, with a particular emphasis on the characteristics of photoreceptors that may contribute to retinoblastoma.

Strengths:

This single-cell study captures gene regulation in photoreceptors across different developmental stages, defining post-mitotic cone and rod populations by highlighting their unique gene expression profiles through analyses such as RNA velocity and SCENIC. By leveraging full-length sequencing data, the study identifies differentially expressed isoforms of NRL and THRB in L/M cone and rod precursors, illustrating the dynamic gene regulation involved in photoreceptor fate commitment. Additionally, the authors performed high-resolution clustering to explore markers defining developing photoreceptors across the fovea and peripheral retina, particularly characterizing SYK's role in the proliferative response of cones in the RB loss background. The study provides an in-depth analysis of developing human photoreceptors, with the authors conducting thorough analyses using full-length single-cell RNA sequencing. The strength of the study lies in its design, which integrates single-cell full-length RNA-seq, long-read RNA-seq, and follow-up histological and functional experiments to provide compelling evidence supporting their conclusions. The model of cell type-dependent splicing for NRL and THRB is particularly intriguing. Moreover, the potential involvement of the SYK and MYC pathways with RB in cone progenitor cells aligns with previous literature, offering additional insights into RB development.

Weaknesses:

The manuscript feels somewhat unfocused, with a lack of a strong connection between the analysis of developing photoreceptors, which constitutes the bulk of the manuscript, and the discussion on retinoblastoma. Additionally, given the recent publication of several single-cell studies on developing human retina, it is important for the authors to cross-validate their findings and adjust their statements where appropriate.

Comments on revisions:

The authors have done quite thorough work addressing concerns raised by myself and other reviewers. The identification of unresolved developing state of rod/cone precursor cell is interesting and intriguing. I do not have much more to add.

Reviewer #3 (Public review):

Summary:

The authors use high-depth, full-length scRNA-Seq analysis of fetal human retina to identify novel regulators of photoreceptor specification and retinoblastoma progression.

Strengths:

The use of high-depth, full-length scRNA-Seq to identify functionally important alternatively spliced variants of transcription factors controlling photoreceptor subtype specification, and identification of SYK as a potential mediator of RB1-dependent cell cycle reentry in immature cone photoreceptors.

Weaknesses:

Relatively minor. This is a technically strong and thorough study that is broadly useful to investigators studying retinal development and retinoblastoma.

Comments on revisions:

The authors have addressed all points raised in the review and considerably strengthened the manuscript. No additional changes are required.

Author response:

The following is the authors’ response to the original reviews

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors have used full-length single-cell sequencing on a sorted population of human fetal retina to delineate expression patterns associated with the progression of progenitors to rod and cone photoreceptors. They find that rod and cone precursors contain a mix of rod/cone determinants, with a bias in both amounts and isoform balance likely deciding the ultimate cell fate. Markers of early rod/cone hybrids are clarified, and a gradient of lncRNAs is uncovered in maturing cones. Comparison of early rods and cones exposes an enriched MYCN regulon, as well as expression of SYK, which may contribute to tumor initiation in RB1 deficient cone precursors.

Strengths:

(1) The insight into how cone and rod transcripts are mixed together at first is important and clarifies a long-standing notion in the field.

(2) The discovery of distinct active vs inactive mRNA isoforms for rod and cone determinants is crucial to understanding how cells make the decision to form one or the other cell type. This is only really possible with full-length scRNAseq analysis.

(3) New markers of subpopulations are also uncovered, such as CHRNA1 in rod/cone hybrids that seem to give rise to either rods or cones.

(4) Regulon analyses provide insight into key transcription factor programs linked to rod or cone fates.

(5) The gradient of lncRNAs in maturing cones is novel, and while the functional significance is unclear, it opens up a new line of questioning around photoreceptor maturation.

(6) The finding that SYK mRNA is naturally expressed in cone precursors is novel, as previously it was assumed that SYK expression required epigenetic rewiring in tumors.

We thank the reviewer for describing the study’s strengths, reflecting the major conclusions of the initially submitted manuscript.  However, based on new analyses – including the requested analyses of other scRNA-seq datasets, our revision clarifies that:

-  related to point (1), cone and rod transcripts do not appear to be mixed together at first (i.e., in immediately post-mitotic immature cone and rod precursors) but appear to be coexpressed in subsequent cone and rod precursor stages; and 

- related to point (3), CHRNA1 appears to mark immature cone precursors that are distinct from the maturing cone and rod precursors that co-express cone- and rod-related RNAs (despite the similar UMAP positions of the two populations in our dataset). 

Weaknesses:

(1) The writing is very difficult to follow. The nomenclature is confusing and there are contradictory statements that need to be clarified.

(2) The drug data is not enough to conclude that SYK inhibition is sufficient to prevent the division of RB1 null cone precursors. Drugs are never completely specific so validation is critical to make the conclusion drawn in the paper.

We thank the reviewer for noting these important issues. Accordingly, in the revised manuscript:

(1) We improve the writing and clarify the nomenclature and contradictory statements, particularly those noted in the Reviewer’s Recommendations for Authors. 

(2) We scale back claims related to the role of SYK in the cone precursor response to RB1 loss, with wording changes in the Abstract, Results, and Discussion, which now recognize that the inhibitor studies only support the possibility that cone-intrinsic SYK expression contributes to retinoblastoma initiation, as detailed in our responses to Reviewer’s Recommendations for Authors. We agree and now mention that genetic perturbation of SYK is required to prove its role.  

Reviewer #2 (Public review):

Summary:

The authors used deep full-length single-cell sequencing to study human photoreceptor development, with a particular emphasis on the characteristics of photoreceptors that may contribute to retinoblastoma.

Strengths:

This single-cell study captures gene regulation in photoreceptors across different developmental stages, defining post-mitotic cone and rod populations by highlighting their unique gene expression profiles through analyses such as RNA velocity and SCENIC. By leveraging fulllength sequencing data, the study identifies differentially expressed isoforms of NRL and THRB in L/M cone and rod precursors, illustrating the dynamic gene regulation involved in photoreceptor fate commitment. Additionally, the authors performed high-resolution clustering to explore markers defining developing photoreceptors across the fovea and peripheral retina, particularly characterizing SYK's role in the proliferative response of cones in the RB loss background. The study provides an in-depth analysis of developing human photoreceptors, with the authors conducting thorough analyses using full-length single-cell RNA sequencing. The strength of the study lies in its design, which integrates single-cell full-length RNA-seq, longread RNA-seq, and follow-up histological and functional experiments to provide compelling evidence supporting their conclusions. The model of cell type-dependent splicing for NRL and THRB is particularly intriguing. Moreover, the potential involvement of the SYK and MYC pathways with RB in cone progenitor cells aligns with previous literature, offering additional insights into RB development.

We thank the reviewer for summarizing the main findings and noting the compelling support for the conclusions, the intriguing cell type-dependent splicing of rod and cone lineage factors, and the insights into retinoblastoma development.  

Weaknesses:

The manuscript feels somewhat unfocused, with a lack of a strong connection between the analysis of developing photoreceptors, which constitutes the bulk of the manuscript, and the discussion on retinoblastoma. Additionally, given the recent publication of several single-cell studies on the developing human retina, it is important for the authors to cross-validate their findings and adjust their statements where appropriate.

We agree that the manuscript covers a range of topics resulting from the full-length scRNAseq analyses and concur that some studies of developing photoreceptors were not well connected to retinoblastoma. However, we also note that the connection to retinoblastoma is emphasized in several places in the Introduction and throughout the manuscript and was a significant motivation for pursuing the analyses. We suggest that it was valuable to highlight how deep, fulllength scRNA-seq of developing retina provides insights into retinoblastoma, including i) the similar biased expression of NRL transcript isoforms in cone precursors and RB tumors, ii) the cone precursors’ co-expression of rod- and cone-related genes such as NR2E3 and GNAT2, which may explain similar co-expression in RB cells, and iii) the expression of  SYK in early cones and RB cells.  While the earlier version had mainly highlighted point (iii), the revised Discussion further refers to points (i) and (ii) as described further in the response to the Reviewer’s Recommendations for Authors. 

We address the Reviewer’s request to cross-validate our findings with those of other single-cell studies of developing human retina by relating the different photoreceptor-related cell populations identified in our study to those characterized by Zuo et al (PMID 39117640), which was specifically highlighted by the reviewer and is especially useful for such cross-validation given the extraordinarily large ~ 220,000 cell dataset covering a wide range of retinal ages (pcw 8–23) and spatiotemporally stratified by macular or peripheral retina location. Relevant analyses of the Zuo et al dataset are shown in Supplementary Figures S3G-H, S10B, S11A-F, and S13A,B. 

Reviewer #3 (Public review):

Summary:

The authors use high-depth, full-length scRNA-Seq analysis of fetal human retina to identify novel regulators of photoreceptor specification and retinoblastoma progression.

Strengths:

The use of high-depth, full-length scRNA-Seq to identify functionally important alternatively spliced variants of transcription factors controlling photoreceptor subtype specification, and identification of SYK as a potential mediator of RB1-dependent cell cycle reentry in immature cone photoreceptors.

Human developing fetal retinal tissue samples were collected between 13-19 gestational weeks and this provides a substantially higher depth of sequencing coverage, thereby identifying both rare transcripts and alternative splice forms, and thereby representing an important advance over previous droplet-based scRNA-Seq studies of human retinal development.

Weaknesses:

The weaknesses identified are relatively minor. This is a technically strong and thorough study, that is broadly useful to investigators studying retinal development and retinoblastoma.

We thank the reviewer for describing the strengths of the study. Our revision addresses the concerns raised separately in the Reviewer’s Recommendations for Authors, as detailed in the responses below.  

Recommendations for the authors:

Reviewing Editor Comments:

The reviewers have completed their reviews. Generally, they note that your work is important and that the evidence is generally convincing. The reviewers are in general agreement that the paper adds to the field. The findings of rod/cone fate determination at a very early stage are intriguing. Generally, the paper would benefit from clarifications in the writing and figures. Experimentally, the paper would benefit from validation of the drug data, for example using RNAi or another assay. Alternatively, the authors could note the caveats of the drug experiments and describe how they could be improved. In terms of analysis, the paper would be improved by additional comparisons of the authors' data to previously published datasets.

We thank the reviewing editor for this summary. As described in the individual reviewer responses, we clarify the writing and figures and provide comparisons to previously published datasets (in particular, the large snRNA-seq dataset of Zuo et al., 2024 (PMID 39117640).  With regard to the drug (i.e., SYK inhibitor) studies, we opted to provide caveats and describe the need for genetic approaches to validate the role of SYK, owing to the infeasibility of completing genetic perturbation experiments in the appropriate timeframe.  We are grateful for the opportunity to present our findings with appropriate caveats. 

Reviewer #1 (Recommendations for the authors):

Shayler cell sort human progenitor/rod/cone populations then full-length single cell RNAseq to expose features that distinguish paths towards rods or cones. They initially distinguish progenitors (RPCs), immature photoreceptor precursors (iPRPs), long/medium wavelength (LM) cones, late-LM cones, short wavelength (S) cones, early rods (ER) and late rods (LR), which exhibit distinct transcription factor regulons (Figures 1, 2). These data expose expected and novel enriched genes, and support the notion that S cones are a default state lacking expression of rod (NRL) or cone (THRB) determinants but retaining expression of generic photoreceptor drivers (CRX/OTX2/NEUROD1 regulons). They identify changes in regulon activity, such as increasing NRL activity from iPRP to ER to LR, but decreasing from iPRP to cones, or increasing RAX/ISL2/THRB regulon activity from iPRP to LM cones, but decreasing from iPRP to S cones or rods.

They report co-expression of rod/cone determinants in LM and ER clusters, and the ratios are in the expected directions (NRLTHRB or RXRG in ER). A novel insight from the FL seq is that there are differing variants generated in each cell population. Full-length NRL (FL-NRL) predominates in the rod path, whereas truncated NRL (Tr-NRL) does so in the cone path, then similar (but opposite) findings are presented for THRB (Fig 3, 4), whereas isoforms are not a feature of RXRG expression, just the higher expression in cones.

The authors then further subcluster and perform RNA velocity to uncover decision points in the tree (Figure 5). They identify two photoreceptor precursor streams, the Transitional Rods (TRs) that provide one source for rod maturation and (reusing the name from the initial clustering) iPRPs that form cones, but also provide a second route to rods. TR cells closest to RPCs (immediately post-mitotic) have higher levels of the rod determinant NR2E3 and NRL, whereas the higher resolution iPRPs near RPCs lack NR2E3 and have higher levels of ONECUT1, THRB, and GNAT2, a cone bias. These distinct rod-biased TR and cone-biased high-resolution iPRPs were not evident in published scRNAseq with 3′ end-counting (i.e. not FL seq). Regulon analysis confirmed higher NRL activity in TR cells, with higher THRB activity in highresolution iPRP cells.

Many of the more mature high-resolution iPRPs show combinations of rod (GNAT1, NR2E3) and cone (GNAT2, THRB) paths as well as both NRL and THRB regulons, but with a bias towards cone-ness (Figure 6). Combined FISH/immunofluorescence in fetal retina uncovers cone-biased RXRG-protein-high/NR2E3-protein-absent cone-fated cells that nevertheless expressed NR2E3 mRNA. Thus early cone-biased iPRP cells express rod gene mRNA, implying a rod-cone hybrid in early photoreceptor development. The authors refer to these as "bridge region iPRP cells".

In Figure 7, they identify CHRNA1 as the most specific marker of these bridge cells (overlapping with ATOH7 and DLL3, previously linked to cone-biased precursors), and FISH shows it is expressed in rod-biased NRL protein-positive and cone-biased RXRG proteinpositive cones at fetal week 12.

Figure 8 outlines the graded expression of various lncRNAs during cone maturation, a novel pattern.

Finally (Figure 9), the authors identify differential genes expressed in early rods (ER cluster from Figure 1) vs early cones (LM cluster, excluding the most mature opsin+ cells), revealing high levels of MYCN targets in cones. They also find SYK expression in cones. SYK was previously linked to retinoblastoma, so intrinsic expression may predispose cone precursors to transformation upon RB loss. They finish by showing that a SYK inhibitor blocks the proliferation of dividing RB1 knockdown cone precursors in the human fetal retina.

Overall, the authors have uncovered interesting patterns of biased expression in cone/rod developmental paths, especially relating to the isoform differences for NRL and THRB which add a new layer to our understanding of this fate choice. The analyses also imply that very soon after RPCs exit the cell cycle, they generate post-mitotic precursors biased towards a rod or cone fate, that carry varying proportions of mixed rod/cone determinants and other rod/cone marker genes. They also introduce new markers that may tag key populations of cells that precede the final rod/cone choice (e.g. CHRNA1), catalogue a new lncRNA gradient in cone maturation, and provide insight into potential genes that may contribute to retinoblastoma initiation, like SYK, due to intrinsic expression in cone precursors. However, as detailed below, the text needs to be improved considerably, and overinterpretations need to be moderated, removed, or tested more rigorously with extra data.

Major Comments

The manuscript is very difficult to follow. The nomenclature is at times torturous, and the description of hybrid rod/cone hybrid cells is confusing in many aspects.

(1) A single term, iPRP, is used to refer to an initial low-resolution cluster, and then to a subset of that cluster later in the paper.

We agree that using immature photoreceptor precursor (iPRP) for both high-resolution and lowresolution clusters was confusing. We kept this name for the low-resolution cluster (which includes both immature cone and immature rod precursors), renamed the high-resolution iPRP cluster immature cone precursors (iCPs). and renamed their transitional rod (TR) counterparts immature rod precursors (iRPs). These designations are based on 

- the biased expression of THRB, ONECUT1, and the THRB regulon in iCPs (Fig. 5D,E);

- the biased expression of NRL, NR2E3, and NRL regulon iRPs (Fig. 5D,E);

- the partially distinct iCP and iRP UMAP positions (Figure 5C); and 

- the evidence of similar immature cone versus rod precursor populations in the Zuo et al 3’ snRNA-seq dataset, as noted below and described in two new paragraphs starting at the bottom of p. 12.

(2) To complicate matters further, the reader needs to understand the subset within the iPRP referred to as bridge cells, and we are told at one point that the earliest iPRPs lack NR2E3, then that they later co-express NR2E3, and while the authors may be referring to protein and RNA, it serves to further confuse an already difficult to follow distinction. I had to read and re-read the iPRP data many times, but it never really became totally clear.

We agree that the description of the high-resolution iPRP (now “iCP”) subsets was unclear, although our further analyses of a large 3’ snRNA-seq dataset in Figure S11 support the impression given in the original manuscript that the earliest iCPs lack NR2E3 and then later coexpress NR2E3 while the earliest iRPs lack THRB and then later express THRB. As described in new text in the Two post-mitotic immature photoreceptor precursor populations section (starting on line 7 of p. 13): 

When considering only the main cone and rod precursor UMAP regions, early (pcw 8 – 13) cone precursors expressed THRB and lacked NR2E3 (Figure S11D,E, blue arrows), while early (pcw 10 – 15) rod precursors expressed NR2E3 and lacked THRB (Figure S11D,E, red arrows), similar to RPC-localized iCPs and iRPs in our study (Figure 5D).

Next, as summarized in new text in the Early cone and rod precursors with rod- and conerelated RNA co-expression section (new paragraph at top of p. 16): 

Thus, a 3’ snRNA-seq analysis confirmed the initial production of immature photoreceptor precursors with either L/M cone-precursor-specific THRB or rod-precursor-specific NR2E3 expression, followed by lower-level co-expression of their counterparts, NR2E3 in cone precursors and THRB in rod precursors. However, in the Zuo et al. analyses, the co-expression was first observed in well-separated UMAP regions, as opposed to a region that bridges the early cone and early rod populations in our UMAP plots. These findings are consistent with the notion that cone- and rod-related RNA co-expression begins in already fate-determined cone and rod precursors, and that such precursors aberrantly intermixed in our UMAP bridge region due to their insufficient representation in our dataset.  

Importantly, and as noted in our ‘Public response’ to Reviewer 1, “CHRNA1 appears to mark immature cone precursors that are distinct from the maturing cone and rod precursors that coexpress cone- and rod-related RNAs (despite the similar UMAP positions of the two populations in our dataset).” In support of this notion, the immature cone precursors expressing CHRNA1  and other  populations did not overlap in UMAP space in the Zuo et al dataset. We hope the new text cited above along with other changes will significantly clarify the observations.

(3) The term "cone/rod precursor" shows up late in the paper (page 12), but it was clear (was it not?) much earlier in this manuscript that cone and rod genes are co-expressed because of the coexpressed NRL and THRB isoforms in Figures 3/4.

We thank the reviewer for noting that the differential NRL and THRB isoform expression already implies that cone and rod genes are co-expressed. However, as we now state, the co-expression of RNAs encoding an additional cone marker (GNAT2) and rod markers (GNAT1, NR2E3) was 

“suggestive of a proposed hybrid cone/rod precursor state more extensive than implied by the coexpression of different THRB and NRL isoforms” (first paragraph of “Early cone and rod …” section on p. 14; new text underlined). 

(4) The (incorrect) impression given later in the manuscript is that the rod/cone transcript mixture applies to just a subset of the iPRP cells, or maybe just the bridge cells (writing is not clear), but actually, neither of those is correct as the more abundant and more mature LM and ER populations analyzed earlier coexpress NRL and THRB mRNAs (Figures 2, 3). Overall, the authors need to vastly improve the writing, simplify/clarify the nomenclature, and better label figures to match the text and help the reader follow more easily and clearly. As it stands, it is, at best, obtuse, and at worst, totally confusing.

We thank the reviewer for bringing the extent of the confusing terminology and wording to our attention. We revised the terminology (as in our response to point 1) and extensively revised the text.  We also performed similar analyses of the Zuo et al. data (as described in more detail in our response to Reviewer 2), which clarifies the distinct status of cells with the “rod/cone transcript mixture” and cells co-expressing early cone and rod precursor markers.  

To more clearly describe data related to cells with rod- and cone-related RNA co-expression, we divided the former Figure 6 into two figures, with Figure 6 now showing the cone- and rodrelated RNA co-expression inferred from scRNA-seq and Figure 7 showing GNAT2 and NR2E3 co-expression in FISH analyses of human retina plus a new schematic in the new panel 7E.

To separate the conceptually distinct analyses of cone and rod related RNA co-expression and the expression of early photoreceptor precursor markers (which were both found in the so-called bridge region – yet now recognized to be different subpopulations), we separated the analyses of the early photoreceptor precursor markers to form a new section, “Developmental expression of photoreceptor precursor markers and fate determinants,” starting on p. 16. 

Additionally, we further review the findings and their implications in four revised Discussion paragraphs starting at the bottom of p. 23).

(5) The data showing that overexpressing Tr-NRL in murine NIH3T3 fibroblasts blocks FL-NRL function is presented at the end of page 7 and in Figure 3G. Subsequent analysis two paragraphs and two figures later (end page 8, Figure 5C + supp figs) reveal that Tr-NRL protein is not detectable in retinoblastoma cells which derive from cone precursors cells and express Tr-NRL mRNA, and the protein is also not detected upon lentiviral expression of Tr-NRL in human fetal retinal explants, suggesting it is unstable or not translated. It would be preferable to have the 3T3 data and retinoblastoma/explant data juxtaposed. E.g. they could present the latter, then show the 3T3 that even if it were expressed (e.g. briefly) it would interfere with FL-NRL. The current order and spacing are somewhat confusing.

We thank the reviewer for this suggestion and moved the description of the luciferase assays to follow the retinoblastoma and explant data and switched the order of Figure panels 3G and 3H.  

(6) On page 15, regarding early rod vs early cone gene expression, the authors state: "although MYCN mRNA was not detected....", yet on the volcano plot in Figure S14A MYCN is one of the marked genes that is higher in cones than rods, meaning it was detected, and a couple of sentences later: "Concordantly, the LM cluster had increased MYCN RNA". The text is thus confusing.

With respect, we note that the original text read, “although MYC RNA was not detected,” which related to a statement in the previous sentence that the gene ontology analysis identified “MYC targets.” However, given that this distinction is subtle and may be difficult for readers to recognize, we revised the text (now on p. 19) to more clearly describe expression of MYCN (but not MYC) as follows:

“The upregulation of MYC target genes was of interest given that many MYC target genes are also targets of MYCN, that MYCN protein is highly expressed in maturing (ARR3+) cone precursors but not in NRL+ rods (Figure 10A), and that MYCN is critical to the cone precursor proliferative response to pRB loss8–10.  Indeed, whereas MYC RNA was not detected, the LM cone cluster had increased MYCN RNA …”

(7) The authors state that the SYK drug is "highly specific". They provide no evidence, but no drug is 100% specific, and it is possible that off-target hits are important for the drug phenotype. This data should be removed or validated by co-targeting the SYK gene along with RB1.

We agree that our data only show the potential for SYK to contribute to the cone proliferative response; however, we believe the inhibitor study retains value in that a negative result (no effect of the SYK inhibitor) would disprove its potential involvement. To reflect this, we changed wording related to this experiment as follows:

In the Abstract, we changed:

(1) “SYK, which contributed to the early cone precursors’ proliferative response to RB1 loss” To: “SYK, which was implicated in the early cone precursors’ proliferative response to RB1 loss.”  

(2) “These findings reveal … and a role for early cone-precursor-intrinsic SYK expression.” To:  “These findings reveal … and suggest a role for early cone-precursor-intrinsic SYK expression.”

In the last paragraph of the Results, we changed:

(1) “To determine if SYK contributes…” To:  “To determine if SYK might contribute…”

(2) “the highly specific SYK inhibitor” To:  “the selective SYK inhibitor”  

(3)  “indicating that cone precursor intrinsic SYK activity is critical to the proliferative response” To: “consistent with the notion that cone precursor intrinsic SYK activity contributes to the proliferative response.”

In the Results, we added a final sentence: 

“However, given potential SYK inhibitor off-target effects, validation of the role of SYK in retinoblastoma initiation will require genetic ablation studies.”

In the Discussion (2nd-to-last paragraph), we changed: 

“SYK inhibition impaired pRB-depleted cone precursor cell cycle entry, implying that native SYK expression rather than de novo induction contributes to the cone precursors’ initial proliferation.” To: “…the pRB-depleted cone precursors’ sensitivity to a SYK inhibitor suggests that native SYK expression rather than de novo induction contributes to the cone precursors’ initial proliferation, although genetic ablation of SYK is needed to confirm this notion.” In the Discussion last sentence, we changed:

“enabled the identification of developmental stage-specific cone precursor features that underlie retinoblastoma predisposition.” To: “enabled the identification of developmental stage-specific cone precursor features that are associated with the cone precursors’ predisposition to form retinoblastoma tumors.”

Minor/Typos

Figure 7 legend, H should be D.

We corrected the figure legend (now related to Figure 8).

Reviewer #2 (Recommendations for the authors):

(1) The author should take advantage of recently published human fetal retina data, such as PMID:39117640, which includes a larger dataset of cells that could help validate the findings. Consequently, statements like "To our knowledge, this is the first indication of two immediately post-mitotic photoreceptor precursor populations with cone versus rod-biased gene expression" may need to be revised.

We thank the reviewer for noting the evidence of distinct immediately post-mitotic rod and cone populations published by others after we submitted our manuscript. In response, we omitted the sentence mentioned and extensively cross-checked our results including:

- comparison of our early versus late cone and rod maturation states to the cone and rod precursor versus cone and rod states identified by Zuo et al (new paragraph on the top half of p. 6 and new figure panels S3G,H);

- detection of distinct immediately post-mitotic versus later cone and rod precursor populations (two new paragraphs on pp. 12-13 and new Figures S10B and S11A-E); 

- identification of cone and rod precursor populations that co-express cone and rod marker genes (two new paragraphs starting at the bottom of p. 15 and new Figures S11D-F);

- comparison of expression patterns of immature cone precursor (iCP) marker genes in our and the Zuo et al dataset (new paragraph on top half of p. 17 and new Figure S13).

We also compare the cell states discerned in our study and the Zuo et al. study in a new Discussion paragraph (bottom of p. 23) and new Figure S17.

(2) The data generated comes from dissociated cells, which inherently lack spatial context. Additionally, it is unclear whether the dataset represents a pool of retinas from multiple developmental stages, and if so, whether the developmental stage is known for each cell profiled. If this information is available, the authors should examine the distribution of developmental stages on the UMAP and trajectory analysis as part of the quality control process. 

We thank the reviewer for highlighting the importance of spatial context and developmental stage. 

Related to whether the dataset represents a pool of retinae from multiple developmental stages, the different cell numbers examined at each time point are indicated in Figure S1A. To draw the readers’ attention to this detail, Figure S1A is now cited in the first sentence of the Results. 

Related to the age-related cell distributions in UMAP plots, the distribution of cells from each retina and age was (and is) shown in Fig. S1F. In addition, we now highlight the age distributions by segregating the FW13, FW15-17, and FW17-18-19 UMAP positions in the new Figure 1C. We describe the rod temporal changes in a new sentence at the top of  p. 5:

“Few rods were detected at FW13, whereas both early and late rods were detected from FW15-19 (Figure 1C), corroborating prior reports [15,20].”  

We describe the cone temporal changes and note the likely greater discrimination of cell state changes that would be afforded by separately analyzing macula versus peripheral retina at each age in a new sentence at the bottom of p. 5:

“L/M cone precursors from different age retinae occupied different UMAP regions, suggesting age-related differences in L/M cone precursor maturation (Figure 1C).”

Moreover, they should assess whether different developmental stages impact gene expression and isoform ratios. It is well established that cone and rod progenitors typically emerge at different developmental times and in distinct regions of the retina, with minimal physical overlap. Grouping progenitor cells based solely on their UMAP positioning may lead to an oversimplified interpretation of the data.

(2a) We agree that different developmental stages may impact gene expression and isoform ratios, and evaluated stages primarily based on established Louvain clustering rather than UMAP position. However, we also used UMAP position to segregate so-called RPC-localized and nonRPC-localized iCPs and iRPs, as well as to characterize the bridge region iCP sub-populations. In the revision, we examine whether cell groups defined by UMAP positions helped to identify transcriptomically distinct populations and further examine the spatiotemporal gene expression patterns of the same genes in the Zuo et al. 3’ snRNA-seq dataset. 

(2b) Related to analyses of immediately post-mitotic iRPs and iCPs, the new Figure S10A expanded the violin plots first shown in Figure 5D to compare gene expression in RPC-localized versus non-RPC-localized iCPs and iRPs and subsequent cone and rod precursor clusters (also presented in response to Reviewer 3). The new Figure S10C, shows a similar analysis of UMAP region-specific regulon activities. These figures support the idea that there are only subtle UMAP region-related differences in the expression of the selected gene and regulons. 

To further evaluate early cone and rod precursors, we compared expression patterns in our cluster- and UMAP-defined cell groups to those of the spatiotemporally defined cell groups in the Zuo et al. 3’ snRNA-seq study. The results revealed similar expression timing of the genes examined, although the cluster assignments of a subset of cells were brought into question, especially the assigned rod precursors at pcw 10 and 13, as shown in new Figures S10B (grey columns) and S11, and as described in two new paragraphs starting near the bottom of p.12. 

(2c) Related to analyses of iCPs in the so-called bridge region, our analyses of the Zuo et al dataset helped distinguish early cone and rod precursor populations (expressing early markers such as ATOH7 and CHRNA1) from the later stages exhibiting rod- and cone-related gene coexpression, which had intermixed in the UMAP bridge region in our dataset. Further parsing of early cone precursor marker spatiotemporal expression revealed intriguing differences as now described in the second half of a new paragraph at the top of p. 17, as follows:

“Also, different iCP markers had different spatiotemporal expression: CHRNA1 and ATOH7 were most prominent in peripheral retina with ATOH7 strongest at pcw 10 and CHRNA1 strongest at pcw 13; CTC-378H22.2 was prominently expressed from pcw 10-13 in both the macula and the periphery; and DLL3 and ONECUT1 showed the earliest, strongest, and broadest expression (Figure S13B). The distinct patterns suggest spatiotemporally distinct roles for these factors in cone precursor differentiation.”

(3) I would commend the authors for performing a validation experiment via RNA in situ to validate some of the findings. However, drawing conclusions from analyzing a small number of cells can still be dangerous. Furthermore, it is not entirely clear how the subclustering is done. Some cells change cell type identities in the high-resolution plot. For example, some iPRP cells from the low-resolution plots in Figure 1 are assigned as TR in high-resolution plots in Figure 5.

The authors should provide justification on the identifies of RPC localized iPRP and TR.

Comparison of their data with other publicly available data should strengthen their annotation

We agree that drawing conclusions from scRNA-seq or in situ hybridization analysis of a small number of cells can be dangerous and have followed the reviewer’s suggestion to compare our data with other publicly available data, focusing on the 3’ snRNA-seq of Zuo et al. given its large size and extensive annotation. Our analysis of  the Zuo et al. dataset helped clarify cell identities by segregating cone and rod precursors with similar gene expression properties in distinct UMAP regions. However, we noted that the clustering of early cone and rod precursors likely gave numerous mis-assigned cells (as noted in response 2b above and shown in the new Figure S11). It would appear that insights may be derived from the combination of relatively shallow sequencing of a high number of cells and deep sequencing of substantially fewer cells. 

Related to how subclustering was done, the Methods state, “A nearest-neighbors graph was constructed from the PCA embedding and clusters were identified using a Louvain algorithm at low and high resolutions (0.4 and 1.6)[70],” citing the Blondel et al reference for the Louvain clustering algorithm used in the Seurat package.  To clarify this, the results text was revised such that it now indicates the levels used to cluster at low resolution (0.4, p. 4, 2nd paragraph) and at high resolution (1.6, top of p. 11) .

Related to the assignment of some iPRP cells from the low-resolution plots in Figure 1 to the TR cluster (now called the ‘iRP’ ‘cluster) in the high-resolution plots in Figure 5, we suggest that this is consistent with Louvain clustering, which does not follow a single dendrogram hierarchy. 

The justification for referring to these groups as RPC-localized iCPs and iRPs relates to their biased gene and regulon expression in Fig. 5D and 5E, as stated on p. 12: 

“In the RPC-localized region, iCPs had higher ONECUT1, THRB, and GNAT2, whereas iRPs trended towards higher NRL and NR2E3 (p= 0.19, p=0.054, respectively).”

(4) Late-stage LM5 cluster Figure 9 is not defined anywhere in previous figures, in which LM clusters only range from 1 to 4. The inconsistency in cluster identification should be addressed.

We revised the text related to this as follows: 

“Indeed, our scRNA-seq analyses revealed that SYK RNA expression increased from the iCP stage through cluster LM4, in contrast to its minimal expression in rods (Figure 10E).  Moreover, SYK expression was abolished in the five-cell group with properties of late maturing cones (characterized in Figure 1E), here displayed separately from the other LM4 cells and designated LM5 (Figure 10E).”  (p. 19-20)

(5) Syk inhibitor has been shown to be involved in RB cell survival in previous studies. The manuscript seems to abruptly make the connection between the single-cell data to RB in the last figure. The title and abstract should not distract from the bulk of the manuscript focusing on the rod and cone development, or the manuscript should make more connection to retinoblastoma.

We appreciate the reviewer’s concern that the title may seem to over-emphasize the connection to retinoblastoma based solely on the SYK inhibitor studies. However, we suggest the title also emphasizes the identification and characterization of early human photoreceptor states, per se, and that there are a number of important connections beyond the SYK studies that could warrant the mention of cell-state-specific retinoblastoma-related features in the title.

Most importantly, a prior concern with the cone cell-of-origin theory was that retinoblastoma cells express RNAs thought to mark retinal cell types other than cones, especially rods. The evidence presented here, that cone precursors also express the rod-related genes helps resolve this issue. The issue is noted numerous times in the manuscript, as follows:  

In the Introduction, we write:

“However, retinoblastoma cells also express rod lineage factor NRL RNAs, which – along with other evidence – suggested a heretofore unexplained connection between rod gene expression and retinoblastoma development[12,13]. Improved discrimination of early photoreceptor states is needed to determine if co-expression of rod- and cone-related genes is adopted during tumorigenesis or reflects the co-expression of such genes in the retinoblastoma cell of origin.” (bottom, p. 2) And: 

“In this study, we sought to further define the transcriptomic underpinnings of human  photoreceptor development and their relationship to retinoblastoma tumorigenesis.” (last paragraph, p. 3)

The Discussion also alluded to this issue and in the revised Discussion, we aimed to make the connection clearer.  We previously ended the 3rd-to-last paragraph with,  

“iPRP [now iCP] and early LM cone precursors’ expression of NR2E3 and NRL RNAs suggest that their presence in retinoblastomas[12,13] reflects their normal expression in the L/M cone precursor cells of origin.” 

We now separate and elaborate on this point in a new paragraph as follows: 

“Our characterization of cone and rod-related RNA co-expression may help resolve questions about the retinoblastoma cell of origin. Past studies suggested that retinoblastoma cells co-express RNAs associated with rods, cones, or other retinal cells due to a loss of lineage fidelity[12]. However, the early L/M cone precursors’ expression of NR2E3 and NRL RNAs suggest that their presence in retinoblastomas[12,13] reflects their normal expression in the L/M cone precursor cells of origin. This idea is further supported by the retinoblastoma cells’ preferential expression of cone-enriched NRL transcript isoforms (Figure S5B).” (middle of p. 24) Based on the above, we elected to retain the title.  

Minor comments:

(1) It is difficult to see the orange and magenta colors in the Fig 3E RNA-FISH image. The colors should be changed, or the contrast threshold needs to be adjusted to make the puncta stand out more.

We re-assigned colors, with red for FL-NRL puncta and green for Tr-NRL puncta. 

(2) Figure 5C on page 8 should be corrected to Supplementary Figure 5C.

We thank the reviewer for noting this error and changed the figure citation.

Reviewer #3 (Recommendations for the authors):

(1) Minor concerns

a. Abbreviation of some words needs to be included, example: FW. 

We now provide abbreviation definitions for FW and others throughout the manuscript.  

b. Cat # does not matches with the 'key resource table' for many reagents/kits. Some examples are: CD133-PE mentioned on Page # 22 on # 71, SMART-Seq V4 Ultra Low Input RNA Kit and SMARTer Ultra Low RNA Kit for the Fluidigm C1 Sytem on Page # 22 on # 77, Nextera XT DNA Library preparation kit on Page # 23 on # 77.

We thank the reviewer for noting these discrepancies. We have now checked all catalog numbers and made corrections as needed.

c. Cat # and brand name of few reagents & kits is missing and not mentioned either in methods or in key resource table or both. Eg: FBS, Insulin, Glutamine, Penicillin, Streptomycin, HBSS, Quant-iT PicoGreen dsDNA assay, Nextera XT DNA LibraryPreparation Kit, 5' PCR Primer II A with CloneAmp HiFi PCR Premix. 

Catalog numbers and brand names are now provided for the tissue culture and related reagents within the methods text and for kits in the Key Resources Table. Additional descriptions of the primers used for re-amplification and RACE were added to the Methods (p. 28-29).

d. Spell and grammar check is needed throughout the manuscript is needed. Example. In Page # 46 RXRγlo is misspelled as RXRlo.

Spelling and grammar checks were reviewed.

(2) Methods & Key Resource table.

a. In Page # 21, IRB# needs to be stated.      

The IRB protocols have been added, now at top of p. 26.

b. In Page # 21, Did the authors dissociate retinae in ice-cold phosphate-buffered saline or papain?   

The relevant sentence was corrected to “dissected while submerged in ice-cold phosphatebuffered saline (PBS) and dissociated as described10.” ( p. 26)

c. In Page # 21, How did the authors count or enumerate the cell count? Provide the details.

We now state, “… a 10 µl volume was combined with 10 µl trypan blue and counted using a hemocytometer” (top of p. 27)

d. Why did the authors choose to specifically use only 8 cells for cDNA preparation in Page # 22? State the reason and provide the details.

The reasons for using 8 cells (to prevent evaporation and to manually transfer one slide-worth of droplets to one strip of PCR tubes) and additional single cell collection details are now provided as follows (new text underlined): 

“Single cells were sorted on a BD FACSAria I at 4°C using 100 µm nozzle in single-cell mode into each of eight 1.2 µl lysis buffer droplets on parafilm-covered glass slides, with droplets positioned over pre-defined marks … .  Upon collection of eight cells per slide, droplets were transferred to individual low-retention PCR tubes (eight tubes per strip) (Bioplastics K69901, B57801) pre-cooled on ice to minimize evaporation. The process was repeated with a fresh piece of parafilm for up to 12 rounds to collect 96 cells). (p. 27, new text underlined)

e. Key resource table does not include several resources used in this study. Example - NR2E3 antibody.

We added the NR2E3 antibody and checked for other omissions.

(3) Results & Figures & Figure Legends

a. Regulon-defined RPC and photoreceptor precursor states

i. On page # 4, 1 paragraph - Clarify the sentence 'Exclusion of all cells with <100,000 cells read and 18 cells.........Emsembl transcripts inferred'. Did the authors use 18 cells or 18FW retinae? 

The sentence was changed to:

“After sequencing, we excluded all cells with <100,000 read counts and 18 cells expressing one or more markers of retinal ganglion, amacrine, and/or horizontal cells (POU4F1, POU4F2, POU4F3, TFAP2A, TFAP2B, ISL1) and concurrently lacking photoreceptor lineage marker OTX2. This yielded 794 single cells with averages of 3,750,417 uniquely aligned reads, 8,278 genes detected, and 20,343 Ensembl transcripts inferred (Figure S1A-C).” (p. 4, new words underlined)

To clarify that 18 retinae were used, the first sentence of the Results was revised as follows:

“To interrogate transcriptomic changes during human photoreceptor development, dissociated RPCs and photoreceptor precursors were FACS-enriched from 18 retinae, ages FW13-19 …” (p. 4).

Why did the authors 'exclude cells lacking photoreceptor lineage marker OTX2' from analysis especially when the purpose here was to choose photoreceptor precursor states & further results in the next paragraph clearly state that 5 clusters were comprised of cells with OTX2 and CRX expression. This is confusing.

We apologize for the imprecise diction. We divided the evidently confusing sentence into two sentences to more clearly indicate that we removed cells that did not express OTX2, as in the first response to the previous question.

ii. In Page # 5, the authors reported the number of cell populations (363 large and 5 distal) identified in the THRB+ L/M-cone cluster. What were the # of cell populations identified in the remaining 5 clusters of the UMAP space?

We added the cell numbers in each group to Fig. 1B. We corrected the large LM group to 366 cells (p. 5) and note 371 LM cells , which includes the five distal cells, in Figure 1B.

b. Differential expression of NRL and THRB isoforms in rod and cone precursors

i. In Figure 3B, the authors compare and show the presence of 5 different NRL isoforms for all the 6 clusters that were defined in 3A. However, in the results, the ENST# of just 2 highly assigned transcript isoforms is given. What are the annotated names of the three other isoforms which are shown in 3B? Please explain in the Results.

As requested, we now annotate the remaining isoforms as encoding full-length or truncated NRL in Fig. 3B and show isoform structures in new Supplementary Figure S4B.  We also refer to each transcript isoform in the Results (p. 7, last paragraph) and similarly evaluate all isoforms in RB31 cells (Fig. S5B).

ii. What does the Mean FPM in the y-axis of Fig 3C refer to?

Mean FPM represents mean read counts (fragments per million, FPM) for each position across Ensembl NRL exons for each cluster, as now stated in the 6th line of the Fig. 3 legend.

iii. A clear explanation of the results for Figures 3E-3F is missing.

We revised the text to more clearly describe the experiment as follows:

“The cone cells’ higher proportional expression of Tr-NRL first exon sequences was validated by RNA fluorescence in situ hybridization (FISH) of FW16 fetal retina in which NRL immunofluorescence was used to identify rod precursors, RXRg immunofluorescence was used to identify cone precursors, and FISH probes specific to truncated Tr-NRL exon 1T or FL-NRL exons 1 and 2 were used to assess Tr-NRL and FL-NRL expression (Figure 3E,F).” (p. 8, new text underlined).

c. Two post-mitotic photoreceptor precursor populations

i. Although deep-sequencing and SCENIC analysis clarified the identities of four RPC-localized clusters as MG, RPC, and iPRP indicative of cone-bias and TR indicative of rod-bias. It would be interesting to see the discriminating determinant between the TR and ER by SCENIC and deep-sequencing gene expression violin/box plots.

We agree it is of interest to see the discriminating determinant between the TR [now termed iRP] and ER clusters by SCENIC and deep-sequencing gene expression violin/box plots. We now provide this information for selected genes and regulons of interest in the new Supplementary Figures S10A and S10C, along with a similar comparison between the prior high-resolution iPRP (now termed iCP) cluster and the first high-resolution LM cluster, LM1, as described for gene expression on p. 12:

“Notably, THRB and GNAT2 expression did not significantly change while ONECUT1 declined in the subsequent non-RPC-localized iCP and LM1 stages, whereas NR2E3 and NRL dramatically increased on transitioning to the ER state (Figure S10A).”

And as described for regulon activities on pp. 13-14:

“Finally, activities of the cone-specific THRB and ISL2 regulons, the rod-specific NRL regulon, and the pan-photoreceptor LHX3, OTX2, CRX, and NEUROD1 regulons increased to varying extents on transitioning from the immature iCP or iRP states to the early-maturing LM1 or ER states (Figure 10C).”

We also show expression of the same genes for spatiotemporally grouped cells from the Zuo et al. dataset in the new Figure S10B, which displays a similar pattern (apart from the possibly mixed pcw 10 and pcw13 designated rod precursors).

d. Early cone precursors with cone- and rod-related RNA expression

i. On page #12, the last paragraph where the authors explain the multiplex RNA FISH results of RXRγ and NR2E3 by citing Figure S8E. However, in Fig S8E, the authors used NRL to identify the rods. Please clarify which one of the rod markers was used to perform RNA FISH?

Figure S8E (where NRL was used as a rod marker) was cited to remind readers that RXRg has low expression in rods and high expression in cones, rather than to describe the results of this multiplex FISH section. To avoid confusion on this point, Figure S8E is now cited using “(as earlier shown in Figure S8E).” With this issue clarified, we expect the markers used in the FISH + IF analysis will be clear from the revised explanation, 

“… we examined GNAT2 and NR2E3 RNA co-expression in RXRg+ cone precursors in the outermost NBL and in RXRg+ rod precursors in the middle NBL … .” (p. 14-15).

To provide further clarity, we provide a diagram of the FISH probes, protein markers, and expression patterns in the new Figure 7E.

ii. The Y-axis of Fig 6G-6H needs to be labelled.

The axes have been re-labeled from “Nb of cells” to “Number of RXRg+ outermost NBL cells in each region” (original Fig. 6G, now Fig. 7C) and “Number of RXRg+ middle NBL cells in each region” (original Fig. 6H, now Fig. 7D).

iii. The legends of Figures 6G and 6H are unclear. In the Figure 6G legend, the authors indicate 'all cells are NR2E3 protein-'. Does that imply the yellow and green bars alone? Similarly, clarify the Figure 6H legend, what does the dark and light magenta refer to? What does the light magenta color referring to NR2E3+/ NR2E3- and the dark magenta color referring to NR2E3+/ NR2E3+ indicate? 

We regret the insufficient clarity. We revised the Fig. 6G (now Fig. 7C) key, which now reads

“All outermost NBL cells are NR2E3 protein-negative.”  We added to the figure legend for panel 7C,D “(n.b., italics are used for RNAs, non-italics for proteins).”  The new scheme in Figure 7E shows the RNAs in italics proteins in non-italics. We hope these changes will clarify when RNA or protein are represented in each histogram category.

Overall, the results (on page # 13) reflecting Figures 6E-6H & Figure S11 are confusing and difficult to understand. Clear descriptions and explanations are needed.

We revised this results section described in the paragraph now spanning p. 14:

-  We now refer to the bar colors in Figures 7C and 7D that support each statement. 

-  We provide an illustration of the findings in Figure 7E.

iv. Previously published literature has shown that cells of the inner NBL are RXRγ+ ganglion cells. So, how were these RXRγ+ ganglion cells in the inner NBL discriminated during multiplex RNA FISH (in Fig 6E-6H and in Fig S11)?

We thank the reviewer for requesting this clarification. We agree that “inner NBL” is the incorrect term for the region in which we examined RXRg+ photoreceptor precursors, as this could include RXRγ+ nascent RGCs. We now clarify that 

“we examined GNAT2 and NR2E3 RNA co-expression in RXRg+ cone precursors in the outermost NBL and in RXRg+ rod precursors in the middle NBL … .”  (p. 14-15) We further state, 

“Limiting our analysis to the outer and middle NBL allowed us to disregard RXRγ+ retinal ganglion cells in the retinal ganglion cell layer or inner NBL (top of p. 15)”

Figure 7E is provided to further aid the reader in understanding the positions examined, and the legend states “RXRg+ retinal ganglion cells in the inner NBL and ganglion cell layer not shown. 

v. In Figure 6E, what marker does each color cell correspond to?