Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

The authors observed a decline in autophagy and proteasome activity in the context of Milton knockdown. Through proteomic analysis, they identified an increase in the protein levels of eIF2β, subsequently pinpointing a novel interaction within eIF subunits where eIF2β contributes to the reduction of eIF2α phosphorylation levels. Furthermore, they demonstrated that overexpression of eIF2β suppresses autophagy and leads to diminished motor function. It was also shown that in a heterozygous mutant background of eIF2β, Milton knockdown could be rescued. This work represents a novel and significant contribution to the field, revealing for the first time that the loss of mitochondria from axons can lead to impaired autophagy function via eIF2β, potentially influencing the acceleration of aging. To further support the authors' claims, several improvements are necessary, particularly in the methods of quantification and the points that should be demonstrated quantitatively. It is crucial to investigate the correlation between aging and the proteins eIF2β and eIF2α.

Thank you so much for your review and comments. We included analyses of protein levels of eIF2α, eIF2β, and eIF2γ at 7 days and 21 days (Figure 4D). The manuscript was revised as below;

Lines 242-245 ‘As for the other subunits of eIF2 complex, proteome analysis did not detect a significant difference in the protein levels of eIF2α and eIF2γ between milton knockdown and control flies at 7 and 21 days (Figure 4D).’

Reviewer #2 (Public Review):

In the manuscript, the authors aimed to elucidate the molecular mechanism that explains neurodegeneration caused by the depletion of axonal mitochondria. In Drosophila, starting with siRNA depletion of Milton and Miro, the authors attempted to demonstrate that the depletion of axonal mitochondria induces the defect in autophagy. From proteome analyses, the authors hypothesized that autophagy is impacted by the abundance of eIF2β and the phosphorylation of eIF2α. The authors followed up the proteome analyses by testing the effects of eIF2β overexpression and depletion on autophagy. With the results from those experiments, the authors proposed a novel role of eIF2β in proteostasis that underlies neurodegeneration derived from the depletion of axonal mitochondria.

The manuscript has several weaknesses. The reader should take extra care while reading this manuscript and when acknowledging the findings and the model in this manuscript.

The defect in autophagy by the depletion of axonal mitochondria is one of the main claims in the paper. The authors should work more on describing their results of LC3-II/LC3-I ratio, as there are multiple ways to interpret the LC3 blotting for the autophagy assessment. Lysosomal defects result in the accumulation of LC3-II thus the LC3-II/LC3-I ratio gets higher. On the other hand, the defect in the early steps of autophagosome formation could result in a lower LC3-II/LC3-I ratio. From the results of the actual blotting, the LC3-I abundance is the source of the major difference for all conditions (Milton RNAi and eIF2β overexpression and depletion). In the text, the authors simply state the observation of their LC3 blotting. The manuscript lacks an explanation of how to evaluate the LC3-II/LC3-I ratio. Also, the manuscript lacks an elaboration on what the results of the LC3 blotting indicate about the state of autophagy by the depletion of axonal mitochondria.

Thank you for pointing it out, and we apologize for an insufficient description of the result. We included quantitation of the levels of LC3-I and LC3-II in Figure 2A, 2D, 3D, 6B and 7B. As the reviewer pointed out, changes in the LC3-II/LC3-I ratio do not necessarily indicate autophagy defects. However, since p62 accumulation (Figure 2B, 2E, 3E, 6C, 7C in the original manuscript), these results collectively suggest that autophagy is lowered. We revised the manuscript to include this discussion as below:

Lines 174-186 ‘During autophagy progression, LC3 is conjugated with phosphatidylethanolamine to form LC3-II, which localizes to isolation membranes and autophagosomes. LC3-I accumulation occurs when autophagosome formation is impaired, and LC3-II accumulation is associated with lysosomal defects(31,32). p62 is an autophagy substrate, and its accumulation suggests autophagic defects(31,32). We found that milton knockdown increased LC3-I, and the LC3-II/LC3-I ratio was lower in milton knockdown flies than in control flies at 14-day-old (Figure 2A). We also analyzed p62 levels in head lysates sequentially extracted using detergents with different stringencies (1% Triton X-100 and 2% SDS). Western blotting revealed that p62 levels were increased in the brains of 14-day-old of milton knockdown flies (Figure 2B). The increase in the p62 level was significant in the Triton X-100-soluble fraction but not in the SDS-soluble fraction (Figure 2B), suggesting that depletion of axonal mitochondria impairs the degradation of less-aggregated proteins.’

Line 189-190 : ‘At 30 day-old, LC3-I was still higher, and the LC3-II/LC3-I ratio was lower, in milton knockdown compared to the control (Figure 2D).’

Line 199-201: ‘However, in contrast with milton knockdown, Pfk knockdown did not affect the levels of LC3-I, LC3-II or the LC3-II/LC3-I ratio (Figure 3D).’

Line 275-281: ‘Neuronal overexpression of eIF2β increased LC3-II, while the LC3-II/LC3-I ratio was not significantly different (Figure 6A and B). Overexpression of eIF2β significantly increased the p62 level in the Triton X-100-soluble fraction (Figure 6C, 4-fold vs. control, p < 0.005 (1% Triton X-100)) but not in the SDS-soluble fraction (Figure 6C, 2-fold vs. control, p = 0.062 (2% SDS)), as observed in brains of milton knockdown flies (Figure 2B). These data suggest that neuronal overexpression of eIF2β accumulates autophagic substrates.’

Line 307-315: ‘Neuronal knockdown of milton causes accumulation of autophagic substrate p62 in the Triton X-100-soluble fraction (Figure 2B), and we tested if lowering eIF2β ameliorates it. We found that eIF2β heterozygosity caused a mild increase in LC3-I levels and decreases in LC3-II levels, resulting in a significantly lower LC3-II/LC3-I ratio in milton knockdown flies (Figure 7B). eIF2β heterozygosity decreased the p62 level in the Triton X-100-soluble fraction in the brains of milton knockdown flies (Figure 7C). The p62 level in the SDS-soluble fraction, which is not sensitive to milton knockdown (Figure 2B), was not affected (Figure 7C). These results suggest that suppression of eIF2β ameliorates the impairment of autophagy caused by milton knockdown.’

Another main point of the paper is the up-regulation of eIF2β by depleting the axonal mitochondria leads to the proteostasis crisis. This claim is formed by the findings from the proteome analyses. The authors should have presented their proteomic data with much thorough presentation and explanation. As in the experiment scheme shown in Figure 4A, the author did two proteome analyses: one from the 7-day-old sample and the other from the 21-day-old sample. The manuscript only shows a plot of the result from the 7-day-old sample, but that of the result from the 21-day-old sample. For the 21-day-old sample, the authors only provided data in the supplemental table, in which the abundance ratio of eIF2β from the 21-day-old sample is 0.753, meaning eIF2β is depleted in the 21-day-old sample. The authors should have explained the impact of the eIF2β depletion in the 21-day-old sample, so the reader could fully understand the authors' interpretation of the role of eIF2β on proteostasis.

Thank you for pointing it out. We included plots of the results of 21-day-old proteome as a part of the main figure (Figure 4C). As the reviewer pointed out, eIF2β protein levels are reduced at the 21-day-old. Since a reduction in the eIF2_β_ ameliorated milton knockdown-induced locomotor defects in aged flies (Figure 7D), the reduction in eIF2β observed in the 21-day-old milton knockdown flies is not likely to negatively contribute to milton knockdown-induced defects. We included this discussion in the manuscript as below:

Lines 337-341:‘eIF2β protein levels are reduced at the 21-day-old; however, since a reduction in the eIF2β ameliorated milton knockdown-induced locomotor defects in aged flies (Figure 7), the reduction in eIF2β observed in the 21-day-old is not likely to negatively contribute to milton knockdown-induced defects.’

The manuscript consists of several weaknesses in its data and explanation regarding translation.

(1) The authors are likely misunderstanding the effect of phosphorylation of eIF2α on translation. The P-eIF2α is inhibitory for translation initiation. However, the authors seem to be mistaken that the down-regulation of P-eIF2α inhibits translation.

We are sorry for our insufficient explanation in the previous version. As the reviewer pointed out, it is well known that the phosphorylated form of eIF2α inhibits translation initiation. Neuronal knockdown of milton caused a reduction in p-eIF2α (Figure 4J and K), and it also lowered translation (Figure 5); the relationship between these two events is currently unclear. We do not think that a reduction in the p-eIF2α suppressed translation; rather, we propose that the unbalance of expression levels of the components of eIF2 complexes negatively affects translation. We revised discussion sections to describe our interpretation more in detail as below:

Line 368-378: ‘eIF2β is a component of eIF2, which meditates translational regulation and ISR initiation. When ISR is activated, phosphorylated eIF2α suppresses global translation and induces translation of ATF4, which mediates transcription of autophagy-related genes(39,40). Since ISR can positively regulate autophagy, we suspected that suppression of ISR underlies a reduction in autophagic protein degradation. We found neuronal knockdown of milton reduced phosphorylated eIF2α, suggesting that ISR is reduced (Figure 4). However, we also found that global translation was reduced (Figure 5). It may be possible that increased levels of eIF2β disrupt the eIF2 complex or alter its functions. The stoichiometric mismatch caused by an imbalance of eIF2 components may inhibit ISR induction. Supporting this model, we found that eIF2β upregulation reduced the levels of p-eIF2α (Figure 6).’

We have revised the graphical abstract and removed the eIF2 complex since its role in the loss of proteostasis caused by milton knockdown has not been elucidated yet.

(2) The result of polysome profiling in Figure 4H is implausible. By 10%-25% sucrose density gradient, polysomes are not expected to be observed. The authors should have used a gradient with much denser sucrose, such as 10-50%.

Thank you for pointing it out. It was a mistake of 10-50%, and we apologize for the oversight. It was corrected (Figure 5).

(3) Also on the polysome profiling, as in the method section, the authors seemed to fractionate ultra-centrifuged samples from top to bottom and then measured A260 by a plate reader. In that case, the authors should have provided a line plot with individual data points, not the smoothly connected ones in the manuscript.

Thank you for pointing it out. We revised the graph (Figure 5).

(4) For both the results from polysome profiling and puromycin incorporation (Figure 4H and I), the difference between control siRNA and Milton siRNA are subtle, if not nonexistent. This might arise from the lack of spatial resolution in their experiment as the authors used head lysate for these data but the ratio of Phospho-eIF2α/eIF2α only changes in the axons, based on their results in Figure 4E-G. The authors could have attempted to capture the spatial resolution for the axonal translation to see the difference between control siRNA and Milton siRNA.

Thank you for your comment. We agree that it would be an interesting experiment, but it will take a considerable amount of time to analyze axonal translation with spatial resolution. We will try to include such analyses in the future. For this manuscript, we revised the discussion section to include the reviewer's suggestion as below;

Lines 351-353: ‘Further analyses to dissect the effects of milton knockdown on proteostasis and translation in the cell body and axon by experiments with spatial resolution would be needed.’

Recommendations for the authors:

From the Reviewing Editor:

As the Reviewing Editor, I have read your manuscript and the associated peer reviews. I have concerns about publishing this work in its current form. I think that your manuscript cannot claim to have found a novel function of eIF2beta because of technical uncertainties and conceptual problems that should be addressed.

Thank you so much for your review and comments. We addressed all the concerns raised by the reviewers. Point-by-point responses are listed below.

First, your manuscript is based partly on what appears to be a mistaken understanding of the mechanistic basis of the ISR. Specifically, eIF2 is a heterotrimeric complex of alpha, beta, and gamma subunits. When eIF2a is phosphorylated, the heterotrimer adopts a new conformation. This conformation directly binds and inhibits eIF2B, the decameric GEF that exchanges the GDP bound to the gamma subunit of the eIF2 complex for GTP. Unless I misunderstood your paper, you seem to propose that decreasing levels of phospho-eIF2a will inhibit translation, but this is backward from what we know about the ISR.

Thank you for your insightful comment, and we are sorry for the confusion. We did not mean to propose that decreasing levels of phospho-eIF2_a_ inhibits translation. We apologize for our insufficient explanation, which might have caused a misunderstanding (Lines 312-318 in the original version). We agree with the reviewer that ‘mismatch due to elevated eIF2-beta could change the behavior of the ISR’. We revised the text in the result section as follows:

Lines 259-264 (in the Result section) ‘Phosphorylation of eIF2α induces conformational changes in the eIF2 complex and inhibits global translation(36). To analyze the effects of milton knockdown on translation, we performed polysome gradient centrifugation to examine the level of ribosome binding to mRNA. Since p-eIF2α was downregulated, we hypothesized that milton knockdown would enhance translation. However, unexpectedly, we found that milton knockdown significantly reduced the level of mRNAs associated with polysomes (Figure 5A and B).’

Lines 368-378 (in the Discussion section): ‘eIF2β is a component of eIF2, which meditates translational regulation and ISR initiation. When ISR is activated, phosphorylated eIF2α suppresses global translation and induces translation of ATF4, which mediates transcription of autophagy-related genes(39,40). Since ISR can positively regulate autophagy, we suspected that suppression of ISR underlies a reduction in autophagic protein degradation. We found neuronal knockdown of milton reduced phosphorylated eIF2α, suggesting that ISR is reduced (Figure 4). However, we also found that global translation was reduced (Figure 5). It may be possible that increased levels of eIF2β disrupt the eIF2 complex or alter its functions. The stoichiometric mismatch caused by an imbalance of eIF2 components may inhibit ISR induction. Supporting this model, we found that eIF2β upregulation reduced the levels of p-eIF2α (Figure 6).’

It may be possible that a stoichiometric mismatch due to elevated eIF2-beta could change the behavior of the ISR, but your paper doesn't adequately address the expression levels of all three eIF2 subunits: alpha, beta, and gamma. The proteomic data shown in Fig 4B is unconvincing on its own because the changes in the beta subunit are subtle. The Western blot in Figure 4C suggests that the KD changes the mass or mobility of the beta subunit, and most importantly, there are no Western blots measuring the levels of eIF2a, eIF2a-phospho, or eIF2-gamma.

We appreciate the reviewer’s comment and agree that the stoichiometric mismatch due to elevated eIF2β may interfere with ISR. We found overexpression of eIF2β lowered p-eIF2 alpha (Figure S2 in V1), which supports this model. We included this data in the main figure in the revised manuscript (Figure 6D) and revised the text as below:

Lines 279-281: ‘Since milton knockdown reduced the p-eIF2α level (Figure 4K), we asked whether an increase in eIF2β affects p-eIF2α. Neuronal overexpression of eIF2β did not affect the eIF2α level but significantly decreased the p-eIF2α level (Figure 6D, E).’

Expression data of eIF2α and eIF2γ from proteomic analyses has been extracted from proteome analyses and included as a table (Figure 4D). Western blots of phospho-eIF2a (Figure S1 in V1) in the main figure (Figure 4G). The result section was revised as below;

Lines 242-245: ‘As for the other subunits of eIF2 complex, proteome analysis did not detect a significant difference in the protein levels of eIF2α and eIF2γ between milton knockdown and control flies at 7 and 21 days (Figure 4D).’

Reviewer #1 (Recommendations For The Authors):

L125-128: In this section, while the efficiency of Milton knockdown is referenced from a previous publication, it is necessary to also mention that the Miro knockdown has been similarly reported in the literature. Additionally, the Methods section lacks details on the Miro RNAi line used, and Table 2 does not include the genotype for Miro RNAi. This information should be included for clarity and completeness.

Thank you for pointing it out. Knockdown efficiency with this strain has been reported (Iijima-Ando et al., PLoS Genet, 2012). We revised the text to include citation and knockdown efficiency as follows:

Lines 139-147: ‘There was no significant increase in ubiquitinated proteins in milton knockdown flies at 1-day old, suggesting that the accumulation of ubiquitinated proteins caused by milton knockdown is age-dependent (Figure S1). We also analyzed the effect of the neuronal knockdown of Miro, a partner of milton, on the accumulation of ubiquitin-positive proteins. Since severe knockdown of Miro in neurons causes lethality, we used UAS-Miro RNAi strain with low knockdown efficiency, whose expression driven by elav-GAL4 caused 30% reduction of Miro mRNA in head extract(24). Although there was a tendency for increased ubiquitin-positive puncta in Miro knockdown brains, the difference was not significant (Figure 1B, p>0.05 between control RNAi and Miro RNAi). These data suggest that the depletion of axonal mitochondria induced by milton knockdown leads to the accumulation of ubiquitinated proteins before neurodegeneration occurs.’

L132-L136: The current phrasing in this section suggests an increase in ubiquitinated proteins for both Milton and Miro knockdowns. However, since there is no significant difference noted for Miro, it is incorrect to state an increase in ubiquitin-positive puncta. Furthermore, combining the results of Milton knockdown to claim an increase in ubiquitinated proteins prior to neurodegeneration is misleading. At the very least, the expression here needs to be moderated to accurately reflect the findings.

Thank you for pointing it out. We revised the text as above.

L137-L141: Results in Figure 1 indicate that Milton knockdown leads to an increase in ubiquitinated proteins at 14 days, while Miro knockdown shows no difference from the control at either 14 or 30 days. Conversely, both the control and Miro exhibit an increase in ubiquitinated proteins with aging, but this trend does not seem to apply to Milton knockdown. This observation suggests that Milton KD may not affect the changes in protein quality control associated with aging. It implies that Milton's function might be more related to protein homeostasis in younger cells, or that changes due to aging might overshadow the effects of Milton knockdown. These interpretations should be included in the Results or Discussion sections for a more comprehensive analysis.

Thank you for your insightful comment. We revised the text to include those points as follows:

Lines 152-153: ‘These results suggest that depletion of axonal mitochondria may have more impact on proteostasis in young neurons than in old neurons.’

Lines 355-362: ‘The depletion of axonal mitochondria and accumulation of abnormal proteins are both characteristics of aged brains(37,38). Our results suggest that the loss of axonal mitochondria is an event upstream of proteostasis collapse during aging. Neuronal knockdown of milton had more impact on proteostasis in young neurons than the old neurons (Figure 1). Proteome analyses also showed that age-related pathways, such as immune responses, are enhanced in young flies with milton knockdown (Table 2). The reduction in axonal transport of mitochondria may be one of the triggering events of age-related changes and accelerates the onset of aging in the brain.’

L143 : Please remove the erroneously included quotation mark.

Thank you for pointing it out. We corrected it.

L145-L147:

- While it is understood that Milton knockdown results in a reduction of mitochondria in axons, as reported previously and seemingly indicated in Figure 1E, this paper repeatedly refers to axonal depletion of mitochondria. Therefore, it would be beneficial to quantitatively assess the number of mitochondria in the axonal terminals located in the lamina via electron microscopy. Such quantification would robustly reinforce the argument that mitochondrial absence in axons is a consequence of Milton knockdown.

Thank you for pointing it out. We included quantitation of the number of mitochondria in the synaptic terminals (Figure 1E).

The text and figure legend was revised accordingly:

Lines 156-157: ‘As previously reported(24), the number of mitochondria in presynaptic terminals decreased in milton knockdown (Figure 1E).’

- The knockdown of Milton is known to reduce mitochondrial transport from an early stage, but what about swelling? By observing swelling at 1 day and 14 days, it may be possible to confirm the onset of swelling and discuss its correlation with the accumulation of ubiquitinated proteins.

Quantitation of axonal swelling has also been included (Figure 1F).

We appreciate reviewer’s comments on the correlation between the accumulation of ubiquitinated proteins and axonal swelling. Axonal swelling was not observed at 3-days-old (Iijima-Ando et al., PLoS Genetics, 2012), indicating that axonal swelling is an age-dependent event. Dense materials are found in swollen axons more often than in normal axons, suggesting a positive correlation between disruption of proteostasis and axonal damage. It would be interesting to analyze the time course of events further; however, we feel it is beyond the scope of this manuscript. We revised the text as below to include this discussion:

Lines 157-159: ‘The swelling of presynaptic terminals, characterized by the enlargement and roundness, was not reported at 3-day-old(24) but observed at this age with about 4% of total presynaptic terminals (Figure 1F, asterisks).’

Lines 162-167: ‘Dense materials are rarely found in age-matched control neurons, indicating that milton knockdown induces abnormal protein accumulation in the presynaptic terminals (Figure 1G and H). In milton knockdown neurons, dense materials are found in swollen presynaptic terminals more often than in presynaptic terminals without swelling, suggesting a positive correlation between the disruption of proteostasis and axonal damage (Figure 1G).’

Lines 362-365: ‘Disruption of proteostasis is expected to contribute neurodegeneration(38), and it would be interesting to analyze the sequence of protein accumulation and axonal degeneration in milton knockdown ((24,29) and Figure 1) in detail with higher time resolution.’

L147-L151: Though Figures 1F and 1G provide qualitative representations, it is advisable to quantitatively assess whether dense materials significantly accumulate. Such quantitative analysis would be required to verify the accumulation of dense materials in the context of the study.

Thank you for pointing it out. We included quantitation of the number of neurons with dense material (Figure 1G). We revised the manuscript as follows:

Line 161-163: ‘Dense materials are rarely found in age-matched control neurons, indicating that milton knockdown induces abnormal protein accumulation in the presynaptic terminals (Figure 1G and H).’

Regarding Figure 1B, C:

- Even though the count of puncta in the whole brain appears to be fewer than 400, the magnification of the optic lobe suggests a substantial presence of puncta. Please clarify in the Methods section what constitutes a puncta and whether the quantification in the whole brain is based on a 2D or 3D analysis. Detail the methodology used for quantification.

Thank you for your comment. We revised the method section to include more details as below:

Lines 434-437: ‘Quantitative analysis was performed using ImageJ (National Institutes of Health) with maximum projection images derived from Z-stack images acquired with same settings. Puncta was identified with mean intensity and area using ImageJ.’

- What about 1-day-old specimens? Does Milton knockdown already show an increase in ubiquitinated protein accumulation at this early stage? Investigating whether ubiquitin-protein accumulation is involved in aging promotion or is already prevalent during developmental stages is a necessary experiment.

Thank you for your comment. We carried out immunostaining with an anti-ubiquitin antibody in the brains at 1-day-old. No significant difference was detected between the control and milton knockdown. This result has been included as Figure S1 in the revised manuscript. The result section was revised as below:

Line 136-139 ‘There was no significant increase in ubiquitinated proteins in milton knockdown flies at 1-day old, suggesting that the accumulation of ubiquitinated proteins caused by milton knockdown is age-dependent (Figure S1).’

For Figure 1E: In the Electron Microscopy section of the Methods, define how swollen axons were identified and describe the quantification methodology used.

Thank you for your comment. Swollen axons are, unlike normal axons, round in shape and enlarged. We revised the text as below;

Lines 157-160: ‘The swelling of presynaptic terminals, characterized by the enlargement and roundness, was not reported at 3-day-old(24) but observed at this age with about 4% of total presynaptic terminals (Figure 1F, asterisks).’

Lines 683-684, Figure 1 legend: ‘Swollen presynaptic terminals (asterisks in (F)), characterized by the enlargement and higher circularity, were found more frequently in milton knockdown neurons.’

L218-L219: Throughout the text, the expression 'eIF2β is "upregulated" in response to Milton knockdown' is frequently used. However, considering the presented results, it might be more accurate to interpret that under the condition of Milton knockdown, eIF2β is not undergoing degradation but rather remains stable.

Thank you for pointing it out. We replaced ‘upregulated’ with ‘increased’ throughout the text.

L234-L235: On what basis is the conclusion drawn that there is a reduction? Given that three experiments have been conducted, it would be possible and more convincing to quantify the results to determine if there is a significant decrease.

Thank you for pointing it out. We quantified the AUC of polysome fraction and carried out statistical analysis. There is a significant decrease in polysome in milton knockdown, and this result has been included in Figure 5B. We revised the figure and the legend accordingly.

L236: 5H-> 4H

Thank you for pointing it out, and we are sorry for the confusion. We corrected it.

L238-L239: Since there is no significant difference observed, it may not be accurate to interpret a reduction in puromycin incorporation.

Thank you for pointing it out. As described above, quantification of polysome fractions showed that milton knockdown significantly reduce polysome (Figure 5B). We revised the manuscript as below;

Lines 263-264: ‘However, unexpectedly, we found that milton knockdown significantly reduced the level of mRNAs associated with polysomes (Figure 5A and B).’

Figure 5D and Figure 6D: Climbing assays have been conducted, but I believe experiments should also be performed to examine whether overexpression or heterozygous mutants of eIF2β induce or suppress degeneration.

Thank you for pointing it out. We analyzed the eyes with eIF2_β_ overexpression for neurodegeneration. Although there was a tendency of elevated neurodegeneration in the retina with eIF2_β_ overexpression, the difference between control and eIF2_β_ overexpression did not reach statistical significance (Figure S2). This result has been included as Figure S2 in the revised manuscript, and the following sentences have been included in the text:

Lines 288-293: ‘We asked if eIF2β overexpression causes neurodegeneration, as depletion of axonal mitochondria in the photoreceptor neurons causes axon degeneration in an age-dependent manner(24). eIF2β overexpression in photoreceptor neurons tends to increase neurodegeneration in aged flies, while it was not statistically significant (p>0.05, Figure S2).’

L271-L272: The results in Figure 6B are surprising. I anticipated a greater increase compared to the Milton knockdown alone. While p62 appears to be reduced, it is not clear why these results lead to the conclusion that lowering eIF2β rescues autophagic impairment. Please add a discussion section to address this point.

Thank you for pointing it out. We apologize for the unclear description of the result. Milton knockdown flies show p62 accumulation (Figure 2), and deleting one copy of eIF2beta in milton knockdown background reduced p62 accumulation (Figure 7C). We revised the text as below:

Lines 307-315: ‘Neuronal knockdown of milton causes accumulation of autophagic substrate p62 in the Triton X-100-soluble fraction (Figure 2B), and we tested if lowering eIF2β ameliorates it. We found that eIF2β heterozygosity caused a mild increase in LC3-I levels and decreases in LC3-II levels, resulting in a significantly lower LC3-II/LC3-I ratio in milton knockdown flies (Figure 7B). eIF2β heterozygosity decreased the p62 level in the Triton X-100-soluble fraction in the brains of milton knockdown flies (Figure 7C). The p62 level in the SDS-soluble fraction, which is not sensitive to milton knockdown (Figure 2B), was not affected (Figure 7C). These results suggest that suppression of eIF2β ameliorates the impairment of autophagy caused by milton knockdown.’

L369: Please specify the source of the anti-ubiquitin antibody used.

Thank you for pointing it out. We included the antibody information in the method section.

Figure 7: While the relationship between Milton knockdown and the eIF2β and eIF2α proteins has been elucidated through the authors' efforts, I would like to see an investigation into whether eIF2β is upregulated and eIF2α phosphorylation is reduced in simply aged Drosophila. This would help us understand the correlation between aging and eIF2 protein dynamics.

Thank you for your comment. We agree that it is an important question, and we are working on it. However, we feel that it is beyond the scope of the current manuscript.

L645-L646: If the mushroom body is identified using mito-GFP, then include mito-GFP in the genotype listed in Supplementary Table 2.

We are sorry for the oversight. We corrected it in Supplementary Table 2.

Additionally, while it is presumed that the mito-GFP signal decreases in axons with Milton RNAi, how was the lobe tips area accurately selected for analysis? Please include these details along with a comprehensive description of the quantification methodology in the Methods section.

Thank you for your comment. Although the mito-GFP signal in the axon is weak in the milton knockdown neurons, it is sufficient to distinguish the mushroom body structure from the background. We revised the method section to include this information in the method section:

Line 437-438: ‘For eIF2α and p-eIF2α immunostaining, the mushroom body was detected by mitoGFP expression.’

Author response:

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

Point-by-point response to the public review:

General Comment: “Using computational modeling, this manuscript explores the effect of growth feedback on the performance of gene networks capable of adaptation. The authors selected 425 hypothetical synthetic circuits that were shown to achieve nearly perfect adaptation in two earlier computational studies (see Ma et al. 2009, and Shi et al. 2017). They examined the effects of cell growth feedback by introducing additional terms to the ordinary differential equation-based models, and performed numerical simulations to check the retainment and the loss of the adaptation responses of the circuits in the presence of growth feedback. The authors show that growth feedback can disrupt the gene network adaptation dynamics in different ways, and report some exceptional core motifs which allow for robust performance in the presence of growth feedback. They also used a metric to establish a scaling law between a circuit robustness measure and the strength of growth feedback. These results have important implications in the field of synthetic biology, where unforeseen interactions between designed gene circuits and the host often disrupt the desired behavior. The paper’s conclusions are supported by their simulation results, although these are presented in their summary formats and it would be useful for the community if the detailed results for each topology were available as a supplementary file or through the authors’ GitHub repository.”

We are grateful for the referee’s positive evaluation of our work. We have updated our GitHub and OSF repositories with detailed results for each topology. Additionally, we have included other simulation codes, result data, and detailed explanations in these two repositories that may be of interest to our readers.

Strength 1: “This work included a detailed investigation of the reasons for adaptation failure upon introducing cell growth to the systems. The comprehensiveness of the analysis makes the work stand out among studies of functional screening of network topologies of gene regulation.”

We are grateful for the referee’s positive assessment of our work, notably the recognition of the ‘detailed investigation’ we conducted, and the ‘comprehensiveness of the analysis’ we provided.

Strength 2: “The authors’ approaches for assessment of robustness, such as the survival ratio Q, can be useful for a wide range of topologies beyond adaptation. The scaling law obtained with those approaches is interesting.”

We are grateful for the referee’s positive evaluation of our defined factors for assessing circuit robustness. We also appreciate the acknowledgment of the “interesting” nature of the scaling law we discovered using the assessment factor R.

Weaknesses 1: “The title suggests that the work investigates the ’effects of growth feedback on gene circuits’. However, the performance of ’nearly perfect adaptation’ was chosen for the majority of the work, leaving the question of whether the authors’ conclusion regarding the effects of growth feedback is applicable to other functional networks.”

We agree that our present title can be too broad, and we have changed it from “Effects of growth feedback on gene circuits: A dynamical understanding” to “Effects of growth feedback on adaptive gene circuits: A dynamical understanding”. Although we have some brief results and discussions on the gene circuits with bistability, we admit that most of our results and discussions are focused on circuits that have adaptation.

The new title is more specific and should be a more appropriate summary of the paper.

Weaknesses 2: “This work relies extensively on an earlier study, evaluating only a selected set of 425 topologies that were shown to give adaptive responses (Shi et al., 2017). This limited selection has two potential issues. First, as the authors mentioned in the introduction, growth feedback can also induce emerging dynamics even without existing function-enabling gene circuits, as an example of the ”effects of growth feedback on gene circuits”. Limiting the investigation to only successful circuits for adaptation makes it unclear whether growth feedback can turn the circuits that failed to produce adaptation by themselves into adaptation-enabling circuits. Secondly, as the Shi et al. (2017) study also used numerical experiments to achieve their conclusions about successful topologies, it is unclear whether the numerical experiments in the present study are compatible with the earlier work regarding the choice of equation forms and ranges of parameter values. The authors also assumed that all readers have sufficient understanding of the 425 topologies and their derivation before reading this paper.”

We agree with the reviewer that several issues need to be clarified in our new manuscript. We have added new discussions for all of them.

We agree with the reviewer that growth feedback could turn the non-adaptive circuits into adaptationenabling circuits, and this indeed presents a compelling topic for future research. We have added the following discussions to our paper, talking about a relevant matter. We find that in our simulated dataset, there are cases where a higher degree of growth feedback can restore the adaptation that has been lost in a circuit. However, as we discussed in this new paragraph, a comprehensive study in the direction of turning non-adaptive circuits into adaptation-enabling circuits will “require entirely different approaches for sampling circuit parameters and selecting candidate network topologies, demanding significantly high computational costs.” Given that this topic extends beyond the scope of the current paper, we leave this matter to future research.

“Although the primary focus of this paper is on how growth feedback can undermine an originally adaptive circuit and how to design circuits that are robust against such feedback, our simulated dataset reveals instances where growth feedback can benefit the circuit within certain ranges. Specifically, we identified 2,092 circuits across 306 different topologies where adaption, lost at an intermediate level of growth feedback, is restored at higher levels. This is 1.4% of all circuits tested. We anticipate that additional circuits exhibiting this loss-and-recovery behavior exist, as our sampling of six discrete levels of k_g (0,0.2,0.4,0.6,0.8,1.0) might have overlooked numerous cases. This result again suggests the possible advantages of growth feedback in gene circuits (Tan et al., 2009; Nevozhay et al., 2012; Deris et al., 2013; Feng et al., 2014; Melendez-Alvarez and Tian, 2022). A comprehensive study into how growth feedback can endow or enhance adaption in circuits would require entirely different approaches for sampling circuit parameters and selecting candidate network topologies, demanding significantly high computational costs. Given that this topic extends beyond the scope of the current paper, we leave this matter to future research.”

We have added the following discussions about the reasoning behind using the 425 network topologies selected from the study Shi et al. (2017).

“We use these 425 network topologies from the study (Shi et al., 2017), avoiding redundancy with established results. Due to the unique focus of our research on the effects of growth feedback and the need to evaluate quantitative ratios of robust circuits among all functional ones, we have chosen to use a 20-fold increase in the number of random parameter sets for each network topology compared to the simulations in (Shi et al., 2017). This approach makes it computationally prohibitive to scan all possible 16,038 three-node circuits. We carefully follow the settings in (Shi et al., 2017), which also analyzed TRNs with the AND logic as in this paper. Detailed descriptions of our simulation experiments are provided in the Methods section. To make our results more convincing, we have adopted a set of adaptation criteria that are stricter than those used in (Shi et al., 2017). Consequently, the ratio of adaptive circuits is somewhat lower in our study, with 4 out of the 425 network topologies not demonstrating adaptation.”

Other than the more strict adaptation criteria and much larger sampling sizes, as we mentioned in this paragraph, we have carefully followed the simulation details of the study Shi et al. (2017). This includes but is not limited to: the dynamical equations (when k_g = 0), the input signals, the scales and ranges of the circuit parameters to be randomly sampled, and the sampling method (Latin hypercube sampling). One of the authors of the current paper was also the first author of the study Shi et al. (2017), who helped us verify the details of simulations (among many other contributions). These identical settings justify our usage of the established results with the 425 network topologies.

To provide more information about these 425 network topologies, We have added the following introduction. It introduces the structural features of the networks, especially the shared core motifs for adaptation. In our GitHub and OSF repositories, we have also provided relevant data about the 425 topologies, including the topology structures and the parameter sets we scanned.

“These topologies can be classified into two families based on the core topology: networks with a negative feedback loop (NFBL) and networks with an incoherent feed-forward loop (IFFL) (Shi et al., 2017). More specifically, there are 206 network topologies in the NFBL family. All of these NFBL topologies have a negative feedback loop for node B. This negative feedback loop can be formed by the loop from node B to A and back to B (such as the circuit shown in Fig. 1 (a)), by node B to C and back to B, or by a longer route, from node B to A and then to C and back to B. There is always a self-activation link from B to B in all these 206 NFBL networks. There are 219 network topologies in the IFFL family. All of them have two feed-forward pathways from the input node A to the output node C. One pathway goes from node A to C directly, while the other involves node B in the middle. One of the pathways is activating while the other one is inhibitory.”

Weaknesses 3: “The authors’ model does not describe the impact of growth via a biological mechanism: they model growth as an additional dilution rate and calculate growth rate based on a phenomenological description with growth rate occurring at a maximum (k_g) scaled by the circuit ’burden’ b(t). Therefore, the authors’ model does not capture potential growth rate changes in parameter values (e.g., synthetic protein production falls with increasing growth rate; see Scott & Hwa, 2023).”

In our paper, we consider dilution due to cell growth as the dominant factor of growth feedback. Here we compared the adaptive circuits under no-growth conditions and their ability to maintain their adaptive behaviors after dilution into a fresh medium, which mediated a significant dilution to the circuits. This is based on our previous work, Zhang, et al. Nature chemical biology 16.6 (2020): 695-701. We agree that an increased growth rate can change synthetic protein production. However, the dynamic roles of the dilution and growthaffected production rate should be analogous, given that they both act as inhibitory factors arising from cell growth as mentioned by the reviewer. Still, we agree that taking the growth effect on the production rate into account would provide a more comprehensive study, but it is beyond the scope of the present work. We have added the following paragraph in the Discussion section of our paper.

“In our paper, we consider dilution due to cell growth as the dominant factor of growth feedback. Here we compared the adaptive circuits under no-growth conditions and their ability to maintain their adaptive behaviors after dilution into a fresh medium, which mediated a significant dilution to the circuits. This is based on our previous work (Zhang et al. (2020)). However, growth feedback is inherently complex (Klumpp et al. (2009)). For instance, an increased growth rate can change protein synthesis rate (Hintsche and Klumpp (2013); Scott and Hwa (2023)), and cell growth rates can affect the distribution of protein expression in cell populations (Gouda et al. (2019)). In our paper, we concentrate on a simplified model with dilution, which we consider to have captured the dominant factor. The dynamic roles of the dilution and growth-affected production rate should be analogous, given that they both act as inhibitory factors arising from cell growth. Incorporating the impact of growth rate on protein synthesis into our model would offer a more comprehensive analysis, a task beyond the scope of this paper but presenting an intriguing opportunity for future research to address the complexities of growth feedback.”

Weaknesses 4: “The authors made several claims about the bifurcations (infinite-period, saddle-node, etc) underlying the abrupt changes leading to failures of adaptations. There is a lack of evidence supporting these claims. Both local and global bifurcations can be demonstrated with semi-analytic approaches such as numerical continuation along with investigations of eigenvalues of the Jacobian matrix. The claims based on ODE solutions alone are not sound.”

After our further simulations and verification, we found that most of the bifurcation-induced failures we mentioned in type-V and type-VI failures should be categorized as bistability or multistability-induced failures. They are still abrupt switching between adaptive and non-adaptive states, as we described in the previous version of the manuscript. However, they are actually still far away from the bifurcation points at the critical k_g. We have corrected all relevant descriptions and figures, including panel Fig. 4 (c) and its captions. We have added the following paragraph in the paper to explain this issue.

“One might expect bifurcations to play an important role in many type-V and type-VI failures. However, in our simulations, failures precisely at the bifurcation point are not observed. This is because the bifurcation points under consideration, such as fold bifurcations, are where one of the attraction basins diminishes to zero. For a failure to occur exactly at the bifurcation point, the initial condition would need to coincide precisely with the infinitesimally small basin just before it vanishes. More realistically, failures almost always largely precede the exact bifurcation point. They happen while the basin is still contracting and the basin boundary crosses the initial condition or O₁. An example is shown in Fig. 4(b), where bistability persists, yet the lighter orange basin with a larger O₁(C) cannot be reached as the boundary shifts away from the initial condition A₀ and B₀. As another example, in Fig. 4 (c) from a different circuit, the higher O₂(C) state disappears at k_g ≈ 0.012 and switches to a lower O₂(C), but this point is not a bifurcation.

It is the point where the stable O₁ continuously crosses the basin boundary of O₂.”

Our further simulations have verified the existence of the oscillation-related bifurcations. We have added a new appendix discussing the phenomena associated with them in more detail.

Weaknesses 5: “The impact of biochemical noise is not evaluated in this work; the author’s analysis is only carried out in a deterministic regime.”

In this paper, we have not taken into account biochemical noise as we focus solely on scenarios where all protein concentrations are high. In these circumstances, the influence of noise is relatively minor. Incorporating biochemical noise, which originates from various sources and possesses diverse characteristics, would significantly complicate the analysis beyond the scope of our current work. However, exploring this aspect could be an intriguing avenue for future research. We have included the following discussions in our paper.

“Our study focuses on scenarios where random noises are ignored. Realistically, gene circuits are subjected to diverse types of noise, which can complicate their predictable behavior and design. These noises can originate externally from a noisy input signal I, or intrinsically, directly affecting the circuit components. Further, these noises can be classified based on various mechanisms that cause them (Colin et al. (2017); Sartori and Tu (2011)) . And with different mechanisms, each type of noise can be characterized by different attributes such as frequency, amplitude, and noise color. These variances can lead to different impacts on the circuits, potentially necessitating unique mechanisms or designs for the attenuation of each category (Sartori and Tu (2011); Qiao et al. (2019) ). Given the extensive complexity and the need for thorough investigation, these noise-related challenges are beyond the scope of this paper and require a series of future studies.”

Point-by-point response to the recommendations for the authors:

Comment 1: - The authors’ github repository, detailed in their code availability statement, is currently unavailable and likely contains some of the answers to the queries here.

We have updated our GitHub and OSF repositories with simulation codes, result data, and detailed explanations. The link to our GitHub repository in the previous version of the manuscript contained a format error, making it inaccessible to the referees. We apologize for this mistake and have corrected it.

Comment 2:   - At present, it is not clear how the 425 topologies are created from the system of equations (Eq. 6-8) or from the circuit diagram in Fig 1a. This could do with being explicitly stated for the reader.

We have added the following paragraph to discuss how the 425 topologies are selected and what the common motifs and connections they share.

“Previous research identified 425 different three-node TRN network topologies that can achieve adaptation in the absence of growth feedback (Shi et al., 2017), providing the base of our computational study. These topologies can be classified into two families based on the core topology: networks with a negative feedback loop (NFBL) and networks with an incoherent feed-forward loop (IFFL) (Shi et al., 2017). More specifically, there are 206 network topologies in the NFBL family. All of these NFBL topologies have a negative feedback loop for node B. This negative feedback loop can be formed by the loop from node B to A and back to B (such as the circuit shown in Fig. 1 (a)), by node B to C and back to B, or by a longer route, from node B to A and then to C and back to B. There is always a self-activation link from B to B in all these 206 NFBL networks. There are 219 network topologies in the IFFL family. All of them have two feed-forward pathways from the input node A to the output node C. One pathway goes from node A to C directly, while the other involves node B in the middle. One of the pathways is activating while the other one is inhibitory. We use these 425 network topologies from the study (Shi et al., 2017), avoiding redundancy with established results. Due to the unique focus of our research on the effects of growth feedback and the need to evaluate quantitative ratios of robust circuits among all functional ones, we have chosen to use a 20-fold increase in the number of random parameter sets for each network topology compared to the simulations in (Shi et al., 2017). This approach makes it computationally prohibitive to scan all possible 16,038 three-node circuits. We carefully follow the settings in (Shi et al., 2017), which also analyzed TRNs with the AND logic as in this paper. Detailed descriptions of our simulation experiments are provided in the Methods section. To make our results more convincing, we have adopted a set of adaptation criteria that are stricter than those used in (Shi et al., 2017). Consequently, the ratio of adaptive circuits is somewhat lower in our study, with 4 out of the 425 network topologies not demonstrating adaptation.”

Comment 3: - In the main text, the authors mentioned that they chose 425 network topologies for this study, whereas the number is 435 in the abstract. Please correct the error.

The number 435 in our previous abstract referred to the 10 four-node circuits that we studied in the appendix, in addition to the 425 three-node network topologies. To avoid confusion and potential misunderstandings among readers, we have revised this expression of “435 distinct topological structures” to “more than four hundred topological structures”.

Comment 4: - Please can the authors include the topologies they have studied in an appendix or as supplementary material. The impact of this work would increase significantly if for each topology the authors could include a pie chart similar to the one shown in Fig 2 so that others can use these results.

We fully acknowledge the potential benefits of providing simulation results for each topology. However, including over four hundred more figures in this paper is not feasible. Moreover, we expect that many readers may also be interested in results not only for individual topologies but also for subsets sharing specific motifs or regulatory connections. Therefore, we have provided all the necessary data and codes in our GitHub repository to make these pie charts. We have included a detailed guide on how to generate these pie charts in the GitHub Readme file. These allow readers to plot the pie chart and extract distributions for any individual topology or use conditions to filter any subset of topologies as required. We believe this approach offers greater flexibility for our readers. We have also added the following explanation in the Methods section.

“The codes implementing these criteria are available in our GitHub repository, with the link provided in the ”Code Availability” section. The failure type results for all circuits tested are available in our OSF repository, with the link provided in the ”Data Availability” section. An additional note is provided in the README file of our GitHub repository for further guidance on generating pie charts similar to Fig. 2 for any network topology or subset of topologies.”

Comment 5: - At present, the authors have not given sufficient detail for their numerical methods (e.g. to identify bistability or oscillations) to enable the work to be repeated. I would appreciate it if the authors could expand their Methods section or provide a description of their method as an appendix. Additionally, the authors must clarify how many parameter sets per topology showed successful adaptation.

In response to this comment, we have reorganized and expanded our Methods section, especially the new “Numerical simulations of circuit dynamics” and “Numerical criteria for functional adaptation and failure types” subsections. We added details on how we define and evaluate a “relatively steady state”, how to determine if there is an oscillation, how to determine the critical k_g value, and how to determine if a failure is continuous or abrupt. Readers can also find the corresponding codes in our GitHub repository, where we provide a README file to help the readers locate the script file they need.

The number of parameter sets per topology showed successful adaptation is precisely our definition of the Q-value. Q-values of most of the circuits we tested are shown in multiple figures in the paper. A complete table of Q-values with different topologies and different k_growth values can be found in our OSF repository.

Comment 6: - Looking at the Model Description, there seem to be multiple issues, as follows. The model should be rewritten and all simulations redone with the model corrected as described below:

(a) The ”strength of growth feedback” is modeled by the maximal growth parameter k_g in Equation (12). However, this rate does not represent growth feedback. In fact, this parameter must be present also for the system without growth feedback, Equations (6 - 8), because those cells grow as well! So Equation (12) with b(t)=0 should also be added to Equations (6 - 8), in addition to the dilution terms in each equation.

(b) The dilution due to growth (dN/dt)*(B/N) is only added to Equations (9 - 11). This is wrong - growthaffects (dilutes) all protein concentrations, even without growth feedback, so similar terms must be added even to equations without growth feedback, i.e., to Equations (6 - 8).

(c) The term representing growth feedback is actually the fraction 1/(1+b(t)). To adjust the strength ofgrowth feedback, some parameters should be introduced into this term. Specifically, the term currently has a Hill form with Hill coefficient = 1 and sensitivity = 1. The term should be converted into a general Hill function, and the parameters of that function should be altered to represent growth feedback. This Hill function is called a cellular (phenotypic) fitness landscape, see Nevozhay et al., 2012.

Equations (6-8) only describe one part of the entire model we are studying. We are having these equations presented solely for the purpose of not overwhelming readers with a large number of parameters that are defined for the first time. They are not actually used in our simulations, but were only for explanations of the meaning of parameters. In our simulations throughout the paper, we only used Eqs. (9-13) (with various topologies). We have revised the texts to make this point clear. We have added the following descriptions in the section Model Description:

“In order not to overwhelm readers with too many terms and parameters, we first describe a partial model (an isolated circuit without growth feedback) before introducing the complete model that we study in this work.”

“Equations. (9) to (13) are the dynamical equations we actually use for simulating the circuit dynamics.”

Additionaly, in the newly added subsection “Numerical simulations of circuit dynamics683” in the Methods, we explicitly mention that:

“The dynamical equations we use are similar to Eqs. (9-13) but with different topologies.”

We consider dilution due to cell growth as the dominant factor of growth feedback. In fact, we study the adaptive circuits without growth and their ability to maintain their adaptive behaviors after dilution into a fresh medium, based on a recent work [Zhang, et al., Nature Chemical Biology 16.6 (2020): 695-701]. The dynamic roles of the dilution and growth-affected production rate should be analogous, given that they both act as inhibitory factors arising from cell growth. The term mentioned in the comment is about how the burden of the circuit affects cell growth. We agree that it can be interesting to have a more comprehensive study on how different degrees of nonlinearity of this term can have different effects on the overall robustness towards the growth feedback problem, but this is not part of our primary focus and is beyond the scope of this paper. In this paper, we are mostly concerned with the variability of the strength of the growth feedback/dilution, controlled by the parameter k_g, instead of the different types of nonlinearity.

Comment 7:  - On the right side of Equation (7), the first term should be inhibitory, right?

This is indeed an error. We accidentally reversed the regulation from A to B and B to A when inputting the formula. We have corrected both terms.

Comment 8: - It seems to me that a better transition from Figs 6 and 7 to Fig 8 can be made. Did the authors choose the three circuits in Fig 8 based on the three distinct groups shown in Fig 6 and 7? The rationale for choosing the three topologies given the clusters identified earlier can be explained more clearly.

We agree more explanation can be provided here. We have added the following descriptions, in the caption of Fig.8:

“The other three curves represent circuits with different robustness levels: high (Circuit No. 98), moderate (Circuit No. 3), and low (Circuit No. 28) values of R, to demonstrate that this scaling behavior is generic. Each of these three circuit topologies is selected from one of the three groups illustrated in Fig. 6 and Fig. 7, and they have the highest Q(k_g = 0) value within their respective groups.”

and in the main text:

“The three other curves represent circuit topologies that have a relatively high, moderate, and low value R among the 425 topologies tested, to demonstrate that this scaling behavior is generic. (These three topologies are the highest Q(k_g = 0) topology in each of the three groups shown in Fig. 6 and Fig. 7.”

Comment 9: - The insights from the neural network model seem to be very limited. It would be interesting to see if the model can predict the performance of network topologies that have not been exposed to the model during training.

Machine learning is not a focus of this paper. For the section the comment was referring to, the main research question is on the relationship between circuit robustness and topology, and the point we are trying to make is that the robustness dependency varies across different connections — some connections are critical, while others are less impactful. The neural-network-based analysis was only used to provide further support to this point by demonstrating that through optimization, neural networks automatically assign different levels of weights to different connections in the circuits.

We agree that it can be an interesting topic to study how machine learning can be used to help us design functional and robust circuits, as discussed in the final paragraph of the Discussion section. However, such an investigation would require a series of more comprehensive and carefully designed simulation experiments to validate if “neural networks can predict the performance of network topologies that have not been exposed to the model during training”. One point one should take extra care of is that many network topologies we study are very similar to many others, with shared motifs and links. These considerations extend beyond the scope of this paper.

Other potential improvements or future work

Comment 10: - The growth feedback examined in this paper comes from the effect of protein levels on the cell division rate (growth rate). However, the opposite effect can also occur; cell growth rates can affect the distribution of protein expression in cell populations. A good reference is Kheir Gouda et al., which is already on the list of references. These opposite effects should be described and discussed.

We agree that growth feedback is inherently complex and has many biological effects, and in our paper, we are using a simplified model to study the dominant factor of growth feedback. We have added the following paragraph in the Discussion section, which involves the opposite effect mentioned in the comment.

“In our paper, we consider dilution due to cell growth as the dominant factor of growth feedback. Here we compared the adaptive circuits under no-growth conditions and their ability to maintain their adaptive behaviors after dilution into a fresh medium, which mediated a significant dilution to the circuits. This is based on our previous work (Zhang et al. (2020)). However, growth feedback is inherently complex (Klumpp et al. (2009)). For instance, an increased growth rate can change protein synthesis rate (Hintsche and Klumpp (2013); Scott and Hwa (2023)), and cell growth rates can affect the distribution of protein expression in cell populations (Gouda et al. (2019)). In our paper, we concentrate on a simplified model with dilution, which we consider to have captured the dominant factor. The dynamic roles of the dilution and growth-affected production rate should be analogous, given that they both act as inhibitory factors arising from cell growth. Incorporating the impact of growth rate on protein synthesis into our model would offer a more comprehensive analysis, a task beyond the scope of this paper but presenting an intriguing opportunity for future research to address the complexities of growth feedback.”

Comment11: - It may be worth mentioning that growth feedback can lead to persistence, see PMID:27010473.

We have included this research as a citation.

Comment 12: - While some other networks (two-node) are discussed, it would be worth doing this analysis for all one- and two-node networks, perhaps controlled by small molecules added externally. If not here, then as a future plan.

We agree that this is an interesting idea for future studies.

Comment 13: - The manuscript analyzes the deterministic dynamics of a set of gene networks. However, gene expression is always stochastic, and gene circuits have been designed to control stochastic gene expression. For example, gene expression distributions can be reshaped, or even new peaks can appear, which would be worth mentioning, PMID: 30341217. The effect of growth feedback on stochastic gene expression and future perspectives of systematically studying this should be discussed.

We have added the following paragraph in the Discussion section to discuss the effects of noises and stochasticity. The research mentioned in the comment is also included.

“Our study focuses on scenarios where random noises are ignored. Realistically, gene circuits are subjected to diverse types of noise, which can complicate their predictable behavior and design. These noises can originate externally from a noisy input signal I, or intrinsically, directly affecting the circuit components. Further, these noises can be classified based on various mechanisms that cause them (Colin et al. (2017); Sartori and Tu (2011)). And with different mechanisms, each type of noise can be characterized by different attributes such as frequency, amplitude, and noise color. These variances can lead to different impacts on the circuits, potentially necessitating unique mechanisms or designs for the attenuation of each category (Sartori and Tu (2011); Qiao et al. (2019)). Given the extensive complexity and the need for thorough investigation, these noise-related challenges are beyond the scope of this paper and require a series of future studies.”

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this work, the authors present a cornucopia of data generated using deep mutational scanning (DMS) of variants in MET kinase, a protein target implicated in many different forms of cancer. The authors conducted a heroic amount of deep mutational scanning, using computational structural models to augment the interpretation of their DMS findings.

Strengths:

This powerful combination of computational models, experimental structures in the literature, dose-response curves, and DMS enables them to identify resistance and sensitizing mutations in the MET kinase domain, as well as consider inhibitors in the context of the clinically relevant exon-14 deletion. They then try to use the existing language model ESM1b augmented by an XGBoost regressor to identify key biophysical drivers of fitness. The authors provide an incredible study that has a treasure trove of data on a clinically relevant target that will appeal to many.

We thank Reviewer 1 for their generous assessment of our manuscript!

Weaknesses:

However, the authors do not equally consider alternative possible mechanisms of resistance or sensitivity beyond the impact of mutation on binding, even though the measure used to discuss resistance and sensitivity is ultimately a resistance score derived from the increase or decrease of the presence of a variant during cell growth.

For this resistance screen, Ba/F3 was a carefully chosen cellular selection system due to its addiction to exogenously provided IL-3, undetected expression of endogenous RTKs (including MET), and dependence on kinase transgenes to promote signaling and growth under IL-3 withdrawal. Together this allows for the readout of variants that alter kinase-driven proliferation without the caveat of bypass resistance. In our previous phenotypic screen (Estevam et al., 2024, eLife), we also carefully examined the impact of all possible MET kinase domain mutations both in the presence and absence of IL-3 withdrawal, but no inhibitors. There, we identified a small group of mutations that were associated with gain-of-function behavior located at conserved regulatory motifs outside of the catalytic site, yet these mutations were largely sensitive to inhibitors within this screen.

Here, the majority of resistance mutations were located at or near the ATP-binding pocket, suggesting an impact on resistance through direct drug interactions. However, there was also a small population of distal mutations that met our statistical definitions of resistance. Within the crizotinib selection, sites such as T1293, L1272, T1261, amongst others, demonstrated resistance profiles but were located in C-lobe away from the catalytic site. While we did not experimentally validate these specific mutations, it is possible that non-direct drug binders instead promote resistance through allosteric or conformational mechanisms which preserve kinase activity and signaling. Indeed, our ML framework explicitly included conformational and stability effects as significant in improving predictions.

We would be happy to further discuss any specific alternative resistance mechanisms Reviewer 1 has in mind! Thank you for highlighting this!

There are also points of discussion and interpretation that rely heavily on docked models of kinase-inhibitor pairs without considering alternative binding modes or providing any validation of the docked pose. Lastly, the use of ESM1b is powerful but constrained heavily by the limited structural training data provided, which can lead to misleading interpretations without considering alternative conformations or poses.

The majority of our interpretations are grounded in the X-ray structures of WT MET bound to the inhibitors studied (or close analogs). The use of docked models (note - to mutant structures predicted by UMol, not ESM, that can have conformational changes) is primarily in the ML part of the manuscript. Indeed, in our models, conformational and binding mode changes are taken into account as features (see Ligand RMSD, Residue RMSD). There are certainly improved methods (AF3 variants) emerging that might have even more power to model these changes, but they come with greater computational costs and are something we will be evaluating in the future.

We added to the results section: “While our features can account for some changes in MET-mutant conformation and altered inhibitor binding pose, the prediction of these aspects can likely be improved with new methods.”

Reviewer #2 (Public review):

Summary:

This manuscript provides a comprehensive overview of potential resistance mutations within MET Receptor Tyrosine Kinase and defines how specific mutations affect different inhibitors and modes of target engagement. The goal is to identify inhibitor combinations with the lowest overlap in their sensitivity to resistant mutations and determine if certain resistance mutations/mechanisms are more prevalent for specific modes of ATP-binding site engagement. To achieve this, the authors measured the ability of ~6000 single mutants of MET's kinase domain (in the context of a cytosolic TPR fusion) to drive IL-3-independent proliferation (used as a proxy for activity) of Ba/F3 cells (deep mutational profiling) in the presence of 11 different inhibitors. The authors then used co-crystal and docked structures of inhibitor-bound MET complexes to define the mechanistic basis of resistance and applied a protein language model to develop a predictive model of inhibitor sensitivity/resistance.

Strengths:

The major strengths of this manuscript are the comprehensive nature of the study and the rigorous methods used to measure the sensitivity of ~6000 MET mutants in a pooled format. The dataset generated will be a valuable resource for researchers interested in understanding kinase inhibitor sensitivity and, more broadly, small molecule ligand/protein interactions. The structural analyses are systematic and comprehensive, providing interesting insights into resistance mechanisms. Furthermore, the use of machine learning to define inhibitor-specific fitness landscapes is a valuable addition to the narrative. Although the ESM1b protein language model is only moderately successful in identifying the underlying mechanistic basis of resistance, the authors' attempt to integrate systematic sequence/function datasets with machine learning serves as a foundation for future efforts.

We thank Reviewer 2 for their thoughtful assessment of our manuscript!

Weaknesses:

The main limitation of this study is that the authors' efforts to define general mechanisms between inhibitor classes were only moderately successful due to the challenge of uncoupling inhibitor-specific interaction effects from more general mechanisms related to the mode of ATP-binding site engagement. However, this is a minor limitation that only minimally detracts from the impressive overall scope of the study.

We agree. We have added to the discussion: “A full landscape of mutational effects can help to predict drug response and guide small molecule design to counteract acquired resistance. The ability to define molecular mechanisms towards that goal will likely require more purposefully chosen chemical inhibitors and combinatorial mutational libraries to be maximally informative.”

Reviewer #3 (Public review):

Summary:

In the manuscript 'Mapping kinase domain resistance mechanisms for the MET receptor tyrosine kinase via deep mutational scanning' by Estevam et al, deep mutational scanning is used to assess the impact of ~5,764 mutants in the MET kinase domain on the binding of 11 inhibitors. Analyses were divided by individual inhibitor and kinase inhibitor subtypes (I, II, I 1/2, and III). While a number of mutants were consistent with previous clinical reports, novel potential resistance mutants were also described. This study has implications for the development of combination therapies, namely which combination of inhibitors to avoid based on overlapping resistance mutant profiles. While one suggested pair of inhibitors with the least overlapping resistance mutation profiles was suggested, this manuscript presents a proof of concept toward a more systematic approach for improved selection of combination therapeutics. Furthermore, in a final part of this manuscript the data was used to train a machine learning model, the ESM-1b protein language model augmented with an XG Boost Regressor framework, and found that they could improve predictions of resistance mutations above the initial ESM-1b model.

Strengths:

Overall this paper is a tour-de-force of data collection and analysis to establish a more systematic approach for the design of combination therapies, especially in targeting MET and other kinases, a family of proteins significant to therapeutic intervention for a variety of diseases. The presentation of the work is mostly concise and clear with thousands of data points presented neatly and clearly. The discovery of novel resistance mutants for individual MET inhibitors, kinase inhibitor subtypes within the context of MET, and all resistance mutants across inhibitor subtypes for MET has clinical relevance. However, probably the most promising outcome of this paper is the proposal of the inhibitor combination of Crizotinib and Cabozantib as Type I and Type II inhibitors, respectively, with the least overlapping resistance mutation profiles and therefore potentially the most successful combination therapy for MET. While this specific combination is not necessarily the point, it illustrates a compelling systematic approach for deciding how to proceed in developing combination therapy schedules for kinases. In an insightful final section of this paper, the authors approach using their data to train a machine learning model, perhaps understanding that performing these experiments for every kinase for every inhibitor could be prohibitive to applying this method in practice.

We thank Reviewer 3 for their assessment of our manuscript (we are very happy to have it described as a tour-de-force!)

Weaknesses:

This paper presents a clear set of experiments with a compelling justification. The content of the paper is overall of high quality. Below are mostly regarding clarifications in presentation.

Two places could use more computational experiments and analysis, however. Both are presented as suggestions, but at least a discussion of these topics would improve the overall relevance of this work. In the first case it seems that while the analyses conducted on this dataset were chosen with care to be the most relevant to human health, further analyses of these results and their implications of our understanding of allosteric interactions and their effects on inhibitor binding would be a relevant addition. For example, for any given residue type found to be a resistance mutant are there consistent amino acid mutations to which a large or small or effect is found. For example is a mutation from alanine to phenylalanine always deleterious, though one can assume the exact location of a residue matters significantly. Some of this analysis is done in dividing resistance mutants by those that are near the inhibitor binding site and those that aren't, but more of these types of analyses could help the reader understand the large amount of data presented here. A mention at least of the existing literature in this area and the lack or presence of trends would be worthwhile. For example, is there any correlation with a simpler metric like the Grantham score to predict effects of mutations (in a way the ESM-1b model is a better version of this, so this is somewhat implicitly discussed).

Indeed we experimented with including these types of features in the XGBoost scheme (particularly residue volume change and distance) to augment the predictive power of the ESM model - see Figure 8 - figure supplement 1; however, we didn’t find them as significant. Therefore, the signal is likely very small and/or incorporated into the baseline ESM model.

Indeed, this discussion relates to the second point this manuscript could improve upon: the machine learning section. The main actionable item here is that this results section seems the least polished and could do a better job describing what was done. In the figure it looks like results for certain inhibitors were held out as test data - was this all mutants for a single inhibitor, or some other scheme? Overall I think the implications of this section could be fleshed out, potentially with more experiments.

Figure 8A and the methods section contain a very detailed explanation of test data. We have thought about it and do not have any easy path to improve the description, which we reproduce here:

“Experimental fitness scores of MET variants in the presence of DMSO and AMG458 were ignored in model training and testing since having just one set of data for a type I ½ inhibitor and DMSO leads to learning by simply memorizing the inhibitor type, without generalizability. The remaining dataset was split into training and test sets to further avoid overfitting (Figure 8A). The following data points were held out for testing - (a) all mutations in the presence of one type I (crizotinib) and one type II (glesatinib analog) inhibitor, (b) 20% of randomly chosen positions (columns) and (c) all mutations in two randomly selected amino acids (rows) (e.g. all mutations to Phe, Ser). After splitting the dataset into train and test sets, the train set was used for XGBoost hyperparameter tuning and cross-validation. For tuning the hyperparameters of each of the XGBoost models, we held out 20% of randomly sampled data points in the training set and used the remaining 80% data for Bayesian hyperparameter optimization of the models with Optuna (Akiba et al., 2019), with an objective to minimize the mean squared error between the fitness predictions on 20% held out split and the corresponding experimental fitness scores. The following hyperparameters were sampled and tuned: type of booster (booster - gbtree or dart), maximum tree depth (max_depth), number of trees (n_estimators), learning rate (eta), minimum leaf split loss (gamma), subsample ratio of columns when constructing each tree (colsample_bytree), L1 and L2 regularization terms (alpha and beta) and tree growth policy (grow_policy - depthwise or lossguide). After identifying the best combination of hyperparameters for each of the models, we performed 10-fold cross validation (with re-sampling) of the models on the full training set. The training set consists of data points corresponding to 230 positions and 18 amino acids. We split these into 10 parts such that each part corresponds to data from 23 positions and 2 amino acids. Then, at each of 10 iterations of cross-validation, models were trained on 9 of 10 parts (207 positions and 16 amino acids) and evaluated on the 1 held out part (23 positions and 2 amino acids). Through this protocol we ensure that we evaluate performance of the models with different subsets of positions and amino acids. The average Pearson correlation and mean squared error of the models from these 10 iterations were calculated and the best performing model out of 8192 models was chosen as the one with the highest cross-validation correlation. The final XGBoost models were obtained by training on the full training set and also used to obtain the fitness score predictions for the validation and test sets. These predictions were used to calculate the inhibitor-wise correlations shown in Figure 8B.“

As mentioned in the 'Strengths' section, one of the appealing aspects of this paper is indeed its potential wide applicability across kinases -- could you use this ML model to predict resistance mutants for an entirely different kinase? This doesn't seem far-fetched, and would be an extremely compelling addition to this paper to prove the value of this approach.

This is exactly where we want to go next! But as we see here, it is going to be hard and require more purposeful selection of chemicals and likely combinatorial mutations to be maximally informative (see also reviewer 2 response where we have added text)

Another area in which this paper could improve its clarity is in the description of caveats of the assay. The exact math used to define resistance mutants and its dependence on the DMSO control is interesting, it is worth discussing where the failure modes of this procedure might be. Could it be that the resistance mutants identified in this assay would differ significantly from those found in patients? That results here are consistent with those seen in the clinic is promising, but discrepancies could remain.

Thank you for pointing this out. The greatest trade-off of probing the intracellular MET kinase (juxtamembrane, kinase domain, c-tail) in the constitutively active TPR system is that while we gain cytoplasmic expression, constitutive oligomerization, and HGF-independent activation, other features like membrane-proximal effects are lost and translatability of some mutations in non-proliferative conditions may also be limited. Nevertheless, Ba/F3 allows IL-3 withdrawal to serve as an effective variant readout of transgenic kinase variant effects due to its undetectable expression of endogenous RTKs and addiction to exogenous interleukin-3 (IL-3).

In our previous study, we were also interested in comparing the phenotypic results to available patient populations in cBioPortal. We observed that our DMS captured known oncogenic MET kinase variants, in addition to a population of gain-of-function variants within clinical residue positions that have not been clinically reported. Interestingly, the population of possible novel gain-of-function mutant codons were more distant in genetic space (2-3 Hamming distance) from wild type than the clinically reported variant codon (1-2 Hamming distance).

For this inhibitor screen, we also carefully compared previously reported and validated resistance mutations across referenced publications to that of our inhibitor screen, and observed large agreement as noted in-text. While discrepancies could definitely remain, there is precedence for consistency.

Furthermore a more in depth discussion of the MetdelEx14 results is warranted. For example, why is the DMSO signature in Figure 1 - supplement 4 so different from that of Figure 1?

In our previous study (Estevam et al., 2024), we more directly compared MET and METΔExon14, and while observed several differences, especially at conserved regulatory motifs, the TPR expression system did not provide a robust differential. Therefore, we hypothesize that a membrane-bound context is likely necessary to obtain a differential that captures juxtamembrane regulatory effects for these two isoforms. For that reason, we did not place heavy emphasis on the differences between MET and METΔExon14 in this study. Nevertheless, we performed parallel analysis of the METΔExon14 inhibitor DMS and provided all source and analyzed data in our GitHub repository (https://github.com/fraser-lab/MET_kinase_Inhibitor_DMS).

In our analysis of resistance, we used Rosace to score and compare DMSO and inhibitor landscapes. We present the full distribution of raw scores in Figure 1 for each condition. However, to visually highlight resistance mutations as a heatmap, we subtracted the scores of each variant in each inhibitor condition from the raw DMSO score, making the heatmaps in Figure 1 - supplement 4 appear more “blue.”

And finally, there is a lot of emphasis put on the unexpected results of this assay for the tivantinib "type III" inhibitor - could this in fact be because the molecule "is highly selective for the inactive or unphosphorylated form of c-Met" according to Eathiraj et al JBC 2011?

The work presented by Eathiraj et al JBC 2011 is a key study we reference and is foundational to tivantinib. While the point brought up about tivantinib’s selective preference for an inactive conformation is valid, this is also true for type II kinase inhibitors. In our study, regardless of inhibitor conformational preference, tivantinib was the only one with a nearly identical landscape to DMSO and exhibited selection even in the absence of Ba/F3 MET-addiction (Figure 1E). This result is in closer agreement with MET agnostic behavior reported by Basilico et al., 2013 and Katayama et al., 2013.

While this paper is crisply written with beautiful figures, the complexity of the data warrants a bit more clarity in how the results are visualized. Namely, clearly highlighting mutants that have previously reported and those identified by this study across all figures could help significantly in understanding the more novel findings of the work.

To better compare and contrast novel mutation identified in this study to others, we compiled a list of reported resistance mutations from recent clinical and experimental studies (Pecci et al 2024; Yao et al., 2023; Bahcall et al., 2022; Recondo et al., 2020; Rotow et al ., 2020; Fujino et al., 2019), since a direct database with resistance annotations does not exist for MET, to the best of our knowledge. In total, this amounted to 31 annotated resistance mutations across crizotinib, capmatinib, tepotinib, savolitinib, cabozantinib, merestinib, and glesatinib, which we have now tabulated in a new figure (Figure 4) and commentary in the main text:

To assess the agreement between our DMS and previously annotated resistance mutations, we compiled a list of reported resistance mutations from recent clinical and experimental studies (Pecci et al 2024; Yao et al., 2023; Bahcall et al., 2022; Recondo et al., 2020; Rotow et al ., 2020; Fujino et al., 2019) (Figure 4A,B). Overall, previously discovered mutations are strongly shifted to a GOF distribution for the drugs where resistance is reported from treatment or experiment; in contrast, the distribution is centered around neutral for those sites for other drugs not reported in the literature (Figure 4C). However, even in cases such as L1195V, we observe GOF DMS scores indicative of resistance to previously reported inhibitors. Given this overall strong concordance with prior literature and clinical results, we can also provide hypotheses to clarify the role of mutations that are observed in combination with others. For example, H1094Y is a reported driver mutation that has been linked to resistance in METΔEx14 for glesatinib with either the secondary L1195V mutation or in isolation (Recodo et al., 2020). However, in our assay H1094Y demonstrated slight sensitivity to gelesatinib, suggesting that either resistance is linked to the exon14 deletion isoform, the L1195V mutation, or a cellular factor not modeled well by the BaF3 system.

Finally, the potential impacts and follow-ups of this excellent study could be communicated better - it is recommended that they advertise better this paper as a resource for the community both as a dataset and as a proof of concept. In this realm I would encourage the authors to emphasize the multiple potential uses of this dataset by others to provide answers and insights on a variety of problems.

Please see below

Related to this, the decision to include the MetdelEx14 results, but not discuss them at all is interesting, do the authors expect future analyses to lead to useful insights? Is it surprising that trends are broadly the same to the data discussed?

Our previous paper suggests that Ba/F3 isn’t a great model for measuring the differences between MET and METΔEx14, so we haven’t emphasized other than to point to our previous paper. We include the full analysis here nonetheless as a resource. Potentially where the greatest differences between resistance mutant behaviors would be observed is in the full-length, membrane-bound MET and METΔEx14 receptor isoforms. While outside of the scope of this study, there is great potential to use the resistance mutations identified in this study as a filtered group to test and map differential inhibitor sensitivities between receptor isoforms.

And finally it could be valuable to have a small addition of introspection from the authors on how this approach could be altered and/or improved in the future to facilitate the general application of this approach for combination therapies for other targets.

See also reviewer 2 response where we have added text.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Major points of revision:

(1) It seems like much of the structural interpretation of the inhibitor binding mode, outside of crizotinib binding, appears to come from docked models of the inhibitor to the MET kinase domain. Given the potential variability of the docked structure to the kinase domain, it would be useful for the authors to consider alternative possible binding modes that their docking pipeline may have suggested. It could also be useful to provide some degree of validation or contextualization of their docking models.

All individual figures are very carefully inspected based on either existing crystal structures of the inhibitor or closely related inhibitors (ATP, 3DKC; crizotinib, 2WGJ; tepotinib, 4R1V; tivantinib, 3RHK; AMG-458, 5T3Q; NVP-BVU972, 3QTI; merestinib, 4EEV; savolitinib, 6SDE). In total, four structural interpretations were the result of docking onto reference experimental structures (capmatinib, cabozantinib, glumetinib, glesatinib). As we wrote above, different conformations and binding modes are possible in predicted mutant structures (as we did here at scale) and included in the ML analysis already.

(2) In the first section, the authors classify an inhibitor as Type Ia on docking models, but mention the conflicting literature describing it as type Ib - it would be helpful to provide a contextualization of why this distinction between Ia and Ib matters, and what difference it might make. It would also be useful to know if their docking score only suggested poses compatible with Ia or if other poses were provided as well. Validation using other method might be beneficial, especially since they acknowledge the conflicting literature for classification. Or at least recontextualization that more evidence would be needed.

Kinase inhibitors have several canonical structural definitions we use to base the classifications in this study. Specifically, type I inhibitors are classified in MET by interactions with Y1230, D1228, K1110 in addition to its conformation in the ATP-binding site. Type I inhibitors are further subdivided into type 1a in MET if it leverages interactions with the solvent front and residue G1163. In prior literature referenced, tepotinib was classified as type 1b, which would imply it does not have solvent front interactions, like savolitinib (PDB 6SDE) or NVP-BVU972 (PDB 3QTI). However, in the tepotinib experimental structure (PDB 4R1V), we observed a greater structural resemblance to other type 1a inhibitors opposed to type 1b (Figure 1 - figure supplement 1b).

(3) The measure used to discuss resistance and sensitivity is ultimately a resistance score derived from the increase or decrease of the presence of a variant during cell growth. This is not a measure of direct binding. It would be helpful if the authors discussed alternative mechanisms through which these variants may impact resistance and/or sensitivity, such as stability, protonation effects, or kinase activity. The score itself may be convolving over all these potential mechanisms to drive GOF and LOF observed behavior.

See the response to the public review. Indeed, our ML framework explicitly included conformational and stability effects as significant in improving predictions.

(4) While it is promising to try and improve the predictive properties of ESM1b, it is not exactly clear why the authors considered their structural data of 11 inhibitors a sufficient dataset with which to augment the model. It would be useful for the authors to provide some additional context for why they wished to augment ESM1b in particular with their dataset, and provide any metrics indicating that their training data of 11 inhibitors provided an adequate statistical sample.

We don’t understand what this means. Sorry!

(5) The authors use ESM-1b to predict the fitness impact of each mutation and augment it using protein structural data of drug-target interactions. However, using an XGBoost regressor on a single set of 11 kinase-inhibitor interaction pairs is an incredibly sparse dataset to train upon. It would be useful for the authors to consider the limitations of their model, as well as its extensibility in the context of alternate binding poses, alternate conformations, or changes in protonation states of ligand or inhibitor.

On the contrary - this is 11 chemicals across 3000 mutations. We have discussed alternative interpretations above.

Minor points:

(1) It would also be useful for the authors to provide more context around their choice of regressor. XGBoost is a powerful regressor but can easily overfit high dimensional data when paired with language models such as ESM-1b. This would be particularly useful since some of the features to train on were also generated using existing models such as ThermoMPNN.

Yes - we are quite concerned about overfitting and have tried to assess overfitting by careful design of test and validation sets.

(2) The authors also mention excluding their DMSO and AMG458 scores in the model training and testing due to overfitting issues - it would be useful to have an SI figure pointing to this data.

No - we exclude the DMSO because that is the reference (baseline) and AMG because it has a different binding mode. This isn’t related to overfitting.

(3) The authors mention in their docking pipeline that 5 binding modes were used for each ligand docking, but it appears that only one binding mode is considered in the main figures. It would be useful for the authors to provide additional details about what were the other binding modes used for, how different were each binding mode, and how was the "primary" mode selected (and how much better was its score than the others).

The reviewer misinterprets the difference between poses shown in figures, based on mostly crystal structures or carefully selected templates, and the use of docked models in feature engineering for the ML part of the study. Where existing crystal structures do not exist, we performed docking for capmatinib, cabozantinib, glumetinib, glesatinib onto reference structures bound to type I (2WGJ) and type II (4EEV) inhibitors. We selected one representative binding mode based on the reference inhibitor, and while not exact, at a minimum these models provide a basis for structural interpretation.

Reviewer #2 (Recommendations for the authors):

My main suggestion is for the authors to add a few sentences (in non-technical language) to the results section, specifically before the results shown in Figure 3, defining gain-of-function, loss-of-function, resistance, and sensitivity. While these definitions are present in the materials and methods section, explicitly discussing them prior to the relevant results would significantly improve the overall readability of the manuscript.

We defined “gain-of-function” and “loss-of-function” mutations as those with fitness scores statistically greater or lower than wild-type. Within the DMSO condition, gain-of-function and loss-of -function labels describe mutational perturbation to protein function, whereas within inhibitor conditions, the labels describe the difference in fitness introduced by an inhibitor.

We have also clarified these definitions where the terms are first introduced: “As expected, the DMSO control population displayed a bimodal distribution with mutations exhibiting wild-type fitness centered around 0, with a wider distribution of mutations that exhibited loss- or gain-of-function effects, as defined by fitness scores with statistically significant lower or greater scores than wild-type, respectively.”

Figure 7D. Please add a bit more detail to the legend on how fold change (y-axis) was calculated.

Here, fold change represents the number of viable cells at each inhibitor concentration relative to the TKI control, measured with the CellTiter-Glo® Luminescent Cell Viability Assay (Promega) as an end point readout. We have updated the legend of Figure 7D with calculation details: “Dose-response for each inhibitor concentration is represented as the fraction of viable cells relative to the TKI free control.”

I must admit, I did not understand what "Specific inhibitor fitness landscapes also aid in identifying mutations with potential drug sensitivity, such as R1086 and C1091 in the MET P-loop" means. These are positions where most mutations lead to greater sensitivity to crizotinib. Is the idea that there are potentially clinically-relevant MET mutations that can be targeted over wild type with crizotinib?

Thank you for highlighting this! The P-loop (phosphate-binding loop) is a glycine-rich structural motif conserved in kinase domains. This motif is located in the N-lobe, where its primary role is to gate ATP entry into the active site and stabilize the phosphate groups of ATP when bound. Therefore, the P-loop is a common target region for ATP-competitive inhibitor design, but also a site where resistance can emerge (Roumiantsev et al., 2002). The idea we’d like to convey is that identifying residues that offer the potential for drug stabilization with the added benefit of having lower risk resistance, is an attractive consideration for novel inhibitor design.

We have added to the text: “Individual inhibitor resistance landscapes also aid in identifying target residues for novel drug design by providing insights into mutability and known resistance cases. This enables the selection of vectors for chemical elaboration with potential lower risk of resistance development. Sites with mutational profiles such as R1086 and C1091, located in the common drug target P-loop of MET, could be likely candidates for crizotinib.”

Reviewer #3 (Recommendations for the authors):

(1) Suggested Improvements to the Figures:

a)  Figure 4A - T1261 seems to be mislabeled

b)  In Figure 3A it's suggested to highlight mutants determined to be resistance mutants by this scheme.

c)  In Figure 3D it would be informative to highlight which of these resistance mutants have already been previously reported and which are novel to this study

d)  Throughout figures 3A, 3D, and 4G the graphical choices on how to highlight synonymous mutations and mutations not performed in the assay needs improvement.

The Green vs Grey 'TRUE' vs 'FALSE' boxes are confusing. Just a green box indicating synonymous mutations would be sufficient. Additionally these green boxes are hard to see, and often edges of this green box are currently missing making it even more difficult to see and interpret.

* In Figure 4A mutants do not seem to be indicated by a line or plus sign, but this is not explained in the legend or the caption. Please add.

* In 3D and 4G it is not clear if the mutants not performed are indicated at all - perhaps they are indicated in white, making them indistinguishable from scores with 0. Please clarify.

T1261 and G1242 are now correctly labeled.

In text we have also highlighted reported resistance mutations for crizotinib, which are inclusive of clinical reports and in vitro characterization: “These sites, and many of the individual mutations, have been noted in prior reports, such as: D1228N/H/V/Y, Y1230C/H/N/S, G1163R.”

We have adjusted the heatmaps to improve visual clarity. Mutations with score 0 are white, as indicated by the scale bar, and mutations uncaptured by the screen are now in light yellow. The green outline distinguishing WT synonymous mutations have also been adjusted so edges are no longer cut off. In our representations, we only distinguished mutations by the score color scale bar and WT outline. What looked like a “plus” or “line” in the original figure was only the heatmap background, which now should be resolved in the updated figure and legends for Figure 3 and Figure 4.

(2) Some Minor Suggested Improvements to the Text:

a)  The abbreviation CBL for 'CBL docking site' is used without being defined.

b)  Figure 3G is referenced, but it does not exist.

c)  In the sentence 'Beyond these well characterized sites, regions with sensitivity occurred throughout the kinase, primarily in loop-regions which have the greatest mutational tolerance in DMSO, but do not provide a growth advantage in the presence of an inhibitor (Figure 1 - Figure Supplement 1; Figure 1 - Figure Supplement 2).'. It is not clear why these supplemental figures are being referenced.

d)  In the supplement section 'Enrich2 Scoring' has what seem like placeholders for citations in [brackets]

Cbl is a E3 ubiquitin ligase that plays a role in MET regulation through engagement with exon 14, specifically at Y1003 when phosphorylated. This mode of regulation was more highlighted in our previous study. However, since Cbl was only mentioned briefly in this study, we have removed reference to it to simplify the text.

In addition, we have removed the figure 3G reference and corrected the in-text range. We have also removed references to figure supplements where unnecessary and edited the “Enrich2 scoring” method section to now reference missing citations.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary: 

In organisms with open mitosis, nuclear envelope breakdown at mitotic entry and re‐assembly of the nuclear envelope at the end of mitosis are important, highly regulated processes. One key regulator of nuclear envelope re‐assembly is the BAF (Barrier‐to‐Autointegration) protein, which contributes to cross‐linking of chromosomes to the nuclear envelope. Crucially, BAF has to be in a dephosphorylated form to carry out this function, and PP2A has been shown to be the phosphatase that dephosphorylates BAF. The Ankle2/LEM4 protein has previously been identified as an important regulator of PP2A in the dephosphorylation of BAF but its precise function is not fully understood, and Li and colleagues set out to investigate the function of Ankle2/LEM4 in both Drosophila flies and Drosophila cell lines.

Strengths: 

The authors use a combination of biochemical and imaging techniques to understand the biology of Ankle2/LEM4. On the whole, the experiments are well conducted and the results look convincing. A particular strength of this manuscript is that the authors are able to study both cellular phenotypes and organismal effects of their mutants by studying both Drosophila D‐mel cells and whole flies.

The work presented in this manuscript significantly enhances our understanding of how Ankle2/LEM4 supports BAF dephosphorylation at the end of mitosis. Particularly interesting is the finding that Ankle2/LEM4 appears to be a bona fide PP2A regulatory protein in Drosophila, as well as the localisation of Ankle2/LEM4 and how this is influenced by the interaction between Ankle2 and the ER protein Vap33. It would be interesting to see, though, whether these insights are conserved in mammalian cells, e.g. does mammalian Vap33 also interact with LEM4? Is LEM4 also a part of the PP2A holoenzyme complex in mammalian cells? 

We feel that conducting experiments to test the level of conservation of our findings in mammalian cells is outside the scope of our study, and we will leave it for other labs to investigate.

Weaknesses: 

This work is certainly impactful but more discussion and comparison of the Drosophila versus mammalian cell system would be helpful. Also, to attract the largest possible readership, the Ankle2 protein should be referred to as Ankle2/LEM4 throughout the paper to make it clear that this is the same molecule. 

We have reinforced our presentation and discussion of similarities and differences between Ankle2 from Drosophila vs humans where relevant throughout the Introduction and Discussion sections. Additionally, we have added the mention that Ankle2 is also called LEM4 in humans in the Abstract and Introduction. However, when referring to Drosophila Ankle2, we do not use LEM4 because it is not listed as an alternate name for this gene/protein in FlyBase.

A schematic model at the end of the final figure would be very useful to summarise the findings.

We have already provided a schematic model in Figure S3, where we think it is better placed.

Reviewer #2 (Public review):

The authors first identify Ankle2 as a regulatory subunit and direct interactor of PP2A, showing they interact both in vitro and in vivo to promote BAF dephosphorylation. The Ankyrin domain of Ankle2 is important for the interaction with PP2A. They then show Ankle2 also interacts with the ER protein Vap33 through FFAT motifs and they particularly co‐localize during mitosis. The recruitment of Ankle2 to Vap33 is essential to ER and nuclear envelop membrane in telophase while earlier in mitosis, it relies on the C terminus but not the FFAT motifs for recruitments to the nuclear membrane and spindle envelop in early mitosis. The molecular determinants and receptors are currently not known. The authors check the function of the PP2A recruitment to Ankle2/Vap33 in the context of embryos and show this recruitment pathway is functionally important. While the Ankle2/Vap33 interaction is dispensable in adult flies ‐looking at wing development, the PP2A/Ankle2 interaction is essential for correct wing and fly development. Overall, this is a very complete paper that reveals the molecular mechanism of PP2A recruitment to Ankle2 and studies both the cellular and the physiological effect of this interaction in the context of fly development.

Strengths: 

The paper is well written and the narrative is well‐developed. The figures are of high quality, wellcontrolled, clearly labelled, and easy to understand. They support the claims made by the authors. 

Weaknesses: 

The study would benefit from being discussed in the context of what is already known on Ankle2 biology in C.elegans and human cells. It is important to highlight the structures shown in the paper are alphafold models, rather than validated structures. 

We have enhanced our presentation of what is known about LEM‐4L/Ankle2 in C. elegans and humans in the Introduction, and further developed comparisons of our findings regarding Drosophila Ankle2 with these orthologs in the Results and Discussion sections. We have also specified in all sections and figure legends that the structures shown are AlphaFold3 models.

Reviewer #3 (Public review): 

Summary: 

The authors were interested in how Ankle2 regulates nuclear envelope reformation after cell division. Other published manuscripts, including those from the authors, show without a doubt that Ankle2 plays a role in this critical process. However, the mechanism by which Ankle2 functions was unclear. Previous work using worms and humans (Asencio et al., 2012) established that human ANKLE2 could bind endogenous PP2A subunits. The binding was direct and was mediated through a region before and including the first ankyrin repeat in human ANKLE2. In addition to its interaction with PP2A, Asencio et al., 2012 also show that ANKLE2 regulates VRK1 kinase activity. Together PP2A and VRK1 regulate BAF phosphorylation for proper nuclear envelope reformation. Here, the authors provide more evidence for interaction with PP2A by also mapping the domain of interaction to the ankyrin repeat in Drosophila. In addition, the ankyrin repeat is essential for nuclear envelope reformation after division. They show that Ankle2 can bind in a PP2A complex without other known regulatory subunits of PP2A. The authors also identify a novel interaction with ER protein Vap33, but functional relevance for this interaction in nuclear envelope reformation is not provided in the manuscript, which the authors explicitly state. This manuscript does not comment on the activity of Ballchen/VRK1 in relation to Ankle2 loss and BAF phosphorylation or nuclear envelope reformation, even though links were previously shown by multiple studies (Asencio et al., Link et al., Apridita Sebastian et al.,). Nuclear envelope defects were rescued by the reduction of VRK1 in two of these manuscripts. It is possible that BAF phosphorylation phenotypes can be contributed by both PP2A inactivity and VRK1 overactivity due to the loss of Ankle2.

Strengths: 

This manuscript is a useful finding linking Ankle2 function during nuclear envelope reformation to the PP2A complex. The authors present solid data showing that Ankle2 can form a complex with PP2A‐29B and Mts and generate a phosphoproteomic resource that is fundamentally important to understanding Ankle2 biology. 

Weaknesses: 

However, the main findings/conclusions about subcellular localization might be incomplete since they are drawn from overexpression experiments. In addition, throughout the text, some conclusions are overstated or are not supported by data. 

It is true that all experiments studying subcellular localization were done with tagged proteins overexpressed in flies and cell culture. Nevertheless, we show that Ankle2‐GFP is functional since it rescues phenotypes resulting from the loss of endogenous Ankle2 in both flies and cultured cells. The antibodies we generated against Ankle2 were unable to reliably detect the endogenous protein by immunofluorescence. We have now stated this caveat in our manuscript. Regarding the validity of our conclusions in relation to our data, we address each point raised by the reviewer under the Recommendations for the authors. In some cases, we have adjusted our conclusions and in other cases, we have provided additional clarification or justification. 

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors): 

There are a few experimental issues that should be addressed, specific comments are listed below: 

(1) Figure 1F: In this experiment, the authors immunoprecipitate GFP‐PP2A‐29B or PP2A‐B29BGFP and Western blot for Ankle2 and Mts to demonstrate that both are co‐immunoprecipitated. To demonstrate that these interactions are specific, the authors should also blot for a protein that is expected to definitely NOT co‐immunoprecipitate with PP2A‐B29; e.g. tubulin. 

Our conclusion that GFP‐PP2A‐29B and PP2A‐29B‐GFP specifically interact with Ankle2 and Mts is also based on mass spectrometry analysis of the purification products from embryos and cells in culture, comparing with products of purification of GFP alone (Fig 1E‐F, S1C‐D and Tables S2, S3). The lists of identified proteins reveal that most proteins (including tubulins) are not enriched with GFP‐PP2A‐29B or PP2A‐29B‐GFP like Ankle2 and Mts are.

(2) Figure 2A: The colour coding of the dots is not explained in the figure legend. 

We have now added the explanation.

(3) Figure 2B: The competition experiment is a good idea. Do the authors get the same results when they conduct the experiment the other way round, i.e. keep the concentration of Tws the same but increase the concentration of Ankle2? 

We have tried this reverse experiment but saw little effect. The failure to observe displacement of Tws by Ankle2 in this context could be due to a higher affinity of Tws than Ankle2 in the PP2A complex, or to lower expression levels achieved for Ankle2 (a larger protein) relative to Tws.

(4) Figure 5D: The hyperphosphorylation of BAF is very difficult to see, and it is impossible to tell whether the hyperphosphorylation has been rescued or not by the different Ankle2 constructs. Can the phosphorylated and the hyperphosphorylated bands be separated better? This panel needs significant improvements to support the claims in the text.

In our opinion, the hyperphosphorylated (upper band) and unphosphorylated (lower band) forms of BAF are well resolved and readily distinguishable. The fainter band in the middle could correspond to a partially phosphorylated form of BAF but we do not venture to speculate on its precise identity nor do we need it to draw our conclusions. The important information from this blot is that the level of unphosphorylated BAF after Ankle2 RNAi increases when Ankle2WT‐GFP and Ankle2Fm+FL1‐GFP are expressed but not when Flag‐GFP or Ankle2ANK‐GFP are expressed. In these experiments, the rescue of unphosphorylated BAF is incomplete because not all cells express the GFP‐tagged protein in our non‐clonal stable cell lines.

Reviewer #2 (Recommendations for the authors):

(1) The alphafold models need to be labelled as such better on the figures, to distinguish them from X‐ray crystallography structures. Alphafold will always propose a solution but it is not necessarily correct. 

We have added the note “MODEL” directly in Figures 2C, 2D, 4F and S3B, in addition to the information already provided in the text and figure legends specifying that these are models generated by AlphaFold3.

(2) Figure 4 F. Annotate the Ankle2 FL1 peptide. 

We have indicated the amino acid residues in the figure.

(3) Problems with the statistical tests. T‐tests cannot be used for comparing multiple groups, as this favors error propagation. 

All of our t‐tests compare only two groups at a time, as indicated. In this regard, our labeling in Fig 5C may have been misleading. We have now changed it.

(4) Close‐ups of ring canal in Figure S2. In Figure S2, there seem to be lots of GFP‐Ankle2 vesicles in the cytoplasm of the oocyte. 

We agree that the image showing Ankle2‐GFP alone in the RNAi Vap33 condition suggested a cytoplasmic granular localization of unknown nature. However, upon examination, we realized that this image did not correspond to the same z‐step as the matching merged image (which also

included DNA staining). We have now replaced the image with the correct one.

Reviewer #3 (Recommendations for the authors): 

Be more accurate about what conclusions can be made from reported data, particularly from overexpression and deletion studies. 

(1) The domain analysis for physical interaction is quite thorough. However, localization information is taken from overexpressed constructs. While these data show what could happen, the authors are not using endogenous levels of Ankle2 in cells or tissues that are known to require Ankle2. As a result, it is difficult to determine whether localization results are biologically meaningful. 

We have added the following text at the end of the third Results section:

“We were unable to examine the localization of endogenous Ankle2 because the antibodies that we generated gave inconclusive results in immunofluorescence. For the remainder of our study, we relied on the overexpression of Ankle2‐GFP, which may not perfectly reflect the localization and function of endogenous Ankle2. However, Ankle2‐GFP is functional as it can rescue phenotypes observed when endogenous Ankle2 is depleted (see below).”

(2) The data showing that Ankle2 is a regulator unit of the PP2A complex also relies on in vitro binding assays in an over‐expression context. Data certainly show Ankle2 can bind proteins in the PP2A complex when overexpressed. However, the authors could not isolate enough of the complex from the animal to test function, so Ankle2 acting as a regulatory subunit isn't functionally shown. There are other possibilities, such as Ankle2 acts as a scaffold for complex assembly.  

The competition experiments shown in Fig 2 are based on complexes assembling in cells and are not in vitro binding assays. We show 4 lines of evidence supporting the idea that Ankle2 functions as a regulatory subunit of PP2A: 1) Ankle2 interacts with the structural (PP2A‐29B) and catalytic (Mts) subunits of PP2A without any known regulatory subunit of PP2A. 2) Depletion of Ankle2 leads to the hyperphosphorylation of the known PP2A substrate BAF. 3) The PP2A regulatory subunit Tws/B55 competes with Ankle2 for formation of a complex with PP2A. 4) AlphaFold3 predicts that Ankle2 engages in a complex with PP2A at a position similar to that of known regulatory subunits of PP2A including Tws/B55, and consistent with their mutually exclusive presence in PP2A complexes. If Ankle2 acted as a scaffold for the formation of a PP2A complex containing other regulatory subunits, we would expect to detect Ankle2 and another regulatory subunit in the same complex.

(3) Throughout the text, some conclusions are overstated or are not supported by data. Examples are below: 

a. Page 1: "we show for the first time that Ankle2 is a regulatory subunit of PP2A"  The authors show binding and changes in BAF phosphorylation levels, but changes in PP2A activity with modulation of Ankle2 weren't shown. 

We have replaced this phrase with this one:

“…we provide several lines of evidence that suggest that Ankle2 is a regulatory subunit of PP2A…”

b. Page 3: "The requirement for Ankle2 in the development of the central nervous system was initially discovered through its targeting by the microcephaly‐causing Zika virus (Shah et al.,

2018)." 

This is not the first paper showing ANKLE2 plays a role in the development of the CNS. Yamamoto et al., 2014 identified mutants in Ankle2 with defects in CNS development in flies and humans, establishing it as a human microcephaly‐causing gene. 

We are sorry for this oversight. We have now cited this important work.

c. Page 6: "Moreover, BAF appears to be the only obligatory substrate of Ankle2‐dependent dephosphorylation for cell proliferation as lowering the dose of the BAF kinase NHK‐1/Ballchen rescues wing development defects caused by the partial depletion of Ankle2 (Li et al., 2024)."  It is unclear why the authors conclude this since Ballchen/VRK1 can phosphorylate many things besides BAF. 

Although the conclusion cannot be drawn categorically, it seems to be by far the most likely scenario. However, we agree that in principle, other mechanisms could also account for these genetic observations, such as the dephosphorylation of another, still unidentified obligatory substrate of PP2A‐Ankle2 that would also be phosphorylated by NHK‐1/Ballchen. However, we have also shown that expression of an unphosphorylatable mutant form of BAF rescues phenotypes observed upon loss of Ankle2 function (Li et al, 2024). We have changed our sentence as follows:

"Moreover, BAF could be the only obligatory substrate of Ankle2‐dependent dephosphorylation for cell proliferation as lowering the dose of the BAF kinase NHK‐1/Ballchen or expression of an unphosphorylatable mutant form of BAF rescues wing development defects caused by the partial depletion of Ankle2 (Li et al., 2024).”

d. Page 10: "These results suggest that a Vap33‐Ankle2‐PP2A complex can mediate the recruitment of a pool of PP2A at the NE."

There is insufficient evidence to indicate that Vap33‐Ankle2‐PP2A exists in a stable state in the cell and that this complex mediates recruitment of PP2A at the NE. The images do not include Vap33, showing no evidence it is present when PP2A is at the NE and the complex could only be detected with overexpression. 

We agree with this caveat and recognize the need to be cautious when proposing our model. In this regard, we feel that our wording is reasonable and appropriate, using “suggest” rather than “prove”, “show” or “indicate”.

e. Page 11: These results suggest that the interaction of Ankle2 with PP2A is essential for its function in BAF dephosphorylation and nuclear reassembly." Page 14: "these results indicate that the interaction of Ankle2 with PP2A is essential during embryo". Page 14: "These results indicate that the interaction of Ankle2 with PP2A but not with Vap33 is essential for its function during cell proliferation in imaginal wing disc development." 

These experiments show that the ankyrin repeat in Ankle2 is necessary for these processes. It does not say PP2A interaction with Ankle2 is necessary because other things could bind the domain. 

We have revised the segments of the text mentioned, taking the reviewer’s legitimate concerns into consideration. We have also added the following sentence to the Discussion:

“However, it remains formally possible that the deletion of Ankyrin repeats used to disrupt the Ankle2‐PP2A interaction abrogated another, unknown aspect of Ankle2 function.”

f. Page 12: "Overall, we conclude that in addition to its N‐terminal PP2A‐interacting Ankyrin domain, Ankle2 requires the integrity of its C‐terminal portion for its essential function in nuclear reassembly." 

No data was shown for differences in nuclear reassembly, only the ability for ANKLE2 truncation mutants to localize to the nuclear envelope. It isn't clear whether the nuclear envelope reformation is normal in Figure S6 which the authors refer to. Lamin staining could help determine and conclude the C‐terminal region is important for nuclear envelope reformation. 

Our conclusion is drawn from the results shown in Figures S4 and S5 (described in the same section), where a rescue assay in cells was performed to assess the functionality of different variants of Ankle2‐GFP when endogenous Ankle2 was depleted. In this assay, Lamin and DNA staining were used to examine nuclear reassembly (as in Figure 5). Figure S6 shows the localizations of the different variants of Ankle2‐GFP, but endogenous Ankle2 is not depleted in these cells.

g. Page 13: "We conclude that the ability of Ankle2 to interact with PP2A is required for the timely recruitment of BAF at reassembling nuclei and ensuing NE reassembly."

It's possible the Ankyrin domain in ANKLE2 is interacting with proteins other than PP2A to recruit BAF at reassembling nuclei, especially since ANKLE2 is found to regulate VRK1 (Link 2019) which has been found to phosphorylate BAF during the cell cycle (Molitor 2014). Additionally, the images in Figure 6A appear to show fully reassembled nuclear envelopes in all mutants by 180s. 

This point relates to point e, raised above by this reviewer. We have re‐written the sentence as follows:

“We conclude that the Ankyrin domain, required for the ability of Ankle2 to interact with PP2A, is necessary for the timely recruitment of BAF at reassembling nuclei and ensuing NE reassembly.”

Please note that in this paragraph, we discuss a delay in RFP‐BAF recruitment, rather than the complete elimination of this recruitment. 

h. Page 16: "Our unbiased phosphoproteomic analysis confirmed that BAF dephosphorylation depends on Ankle2, despite the absence of a detectable interaction between Drosophila Ankle2 and BAF, which may be due to the lack of a LEM domain in the former (Fishburn et al., 2024). Moreover, while Ankle2 was shown to bind and inhibit the BAF counteracting kinase VRK1 in humans (Asencio et al., 2012), we detected no interaction between Ankle2 and NHK‐1/Ballchen (VRK1 ortholog) in Drosophila. This suggests that the loss of Ankle2 causes BAF hyperphosphorylation by preventing PP2A‐dependent dephosphorylation rather than by preventing inhibition of NHK‐1"

There could be transient binding between Ankle2 and Ballchen/VRK1/NHK‐1 or activity can be indirect, but that doesn't mean there is not a contribution of BAF phosphorylation by Ballchen/VRK1/NHK‐1. Genetic evidence from three model systems, including Drosophila, indicates there is a strong genetic interaction between Ankle2 and Ballchen/VRK1/NHK‐1 that includes rescue of lethality.

We agree and we have re‐written in this way:

“While a putative interaction between Ankle2 and NHK‐1 in Drosophila could occur transiently, thereby escaping detection, the simplest interpretation of our results is that the loss of Ankle2 causes BAF hyperphosphorylation by preventing PP2A‐dependent dephosphorylation rather than by preventing inhibition of NHK‐1.”

We do not question the fact that Ballchen/VRK1/NHK‐1 phosphorylates BAF and genetically interacts with Ankle2. The antagonistic relationship between Ballchen/VRK1/NHK‐1 and Ankle2 observed genetically can be explained by the fact that the kinase phosphorylates BAF while PP2AAnkle2 dephosphorylates it, without the need to invoke an additional inhibition of the kinase by Ankle2.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

The hypothesis is based on the idea that inversions capture genetic variants that have antagonistic effects on male sexual success (via some display traits) and survival of females (or both sexes) until reproduction. Furthermore, a sufficiently skewed distribution of male sexual success will tend to generate synergistic epistasis for male fitness even if the individual loci contribute to sexually selected traits in an additive way. This should favor inversions that keep these male-beneficial alleles at different loci together at a cis-LD. A series of simulations are presented and show that the scenario works at least under some conditions. While a polymorphism at a single locus with large antagonistic effects can be maintained for a certain range of parameters, a second such variant with somewhat smaller effects tends to be lost unless closely linked. It becomes much more likely for genomically distant variants that add to the antagonism to spread if they get trapped in an inversion; the model predicts this should drive accumulation of sexually antagonistic variants on the inversion versus standard haplotype, leading to the evolution of haplotypes with very strong cumulative antagonistic pleiotropic effects. This idea has some analogies with one of predominant hypotheses for the evolution of sex chromosomes, and the authors discuss these similarities. The model is quite specific, but the basic idea is intuitive and thus should be robust to the details of model assumption. It makes perfect sense in the context of the geographic pattern of inversion frequencies. One prediction of the models (notably that leads to the evolution of nearly homozygously lethal haplotypes) does not seem to reflect the reality of chromosomal inversions in Drosophila, as the authors carefully discuss, but it is the case of some other "supergenes", notably in ants. So the theoretical part is a strong novel contribution.

We appreciate the detailed and accurate summary of our main theoretic results.

To provide empirical support for this idea, the authors study the dynamics of inversions in population cages over one generation, tracking their frequencies through amplicon sequencing at three time points: (young adults), embryos and very old adult offspring of either sex (>2 months from adult emergence). Out of four inversions included in the experiment, two show patterns consistent with antagonistic effects on male sexual success (competitive paternity) and the survival of offspring, especially females, until an old age, which the authors interpret as consistent with their theory.

As I have argued in my comments on previous versions, the experiment only addresses one of the elements of the theoretical hypothesis, namely antagonistic effects of inversions on male reproductive success and other fitness components, in particular of females. Furthermore, the design of this experiment is not ideal from the viewpoint of the biological hypothesis it is aiming to test. This is in part because, rather than testing for the effects of inversion on male reproductive success versus the key fitness components of survival to maturity and female reproductive output, it looks at the effects on male reproductive success versus survival to a rather old age of 2 months. The relevance of survival until old age to fitness under natural conditions is unclear, as the authors now acknowledge. Furthermore, up to 15% of males that may have contributed to the next generation did not survive until genotyping, and thus the difference between these males' inversion frequency and that in their offspring may be confounded by this potential survival-based sampling bias. The experiment does not test for two other key elements of the proposed theory: the assumption of frequency-dependence of selection on male sexual success, and the prediction of synergistic epistasis for male fitness among genetic variants in the inversion. To be fair, particularly testing for synergistic epistasis would be exceedingly difficult, and the authors have now included a discussion of the above caveats and limitations, making their conclusions more tentative. This is good but of course does not make these limitations of the experiment go away. These limitations mean that the paper is stronger as a theoretical than as an empirical contribution.

We discuss the choice to focus on exploring the potential antagonistic effects of the inversion karyotype on male reproductive success and survival in our general response above. Primarily, this prediction seemed to be the most specific to the proposed model as compared to other alternate models. Still, further studies are clearly needed to elucidate the potential frequency dependence and genetic architecture of the inversions.

Regarding the choice of age at collection, it is unknown to what degree our selected collection age of 10 weeks correlates with survival in the wild, but we feel confident that there will be some positive correlation.

We now further clarify that across our experiments, a minimum of 5% and a mean of 9% of the males used in the parental generation died before collection. These proportions do not appear sufficient to explain the differences between paternal and embryo inversion frequencies shown in Figure 9.

Reviewer #2 (Public review):

Summary:

In their manuscript the authors address the question whether the inversion polymorphism in D. melanogaster can be explained by sexually antagonistic selection. They designed a new simulation tool to perform computer simulations, which confirmed their hypothesis. They also show a tradeoff between male reproduction and survival. Furthermore, some inversions display sex-specific survival.

Strengths:

It is an interesting idea on how chromosomal inversions may be maintained

Weaknesses:

The authors motivate their study by the observation that inversions are maintained in D. melanogaster and because inversions are more frequent closer to the equator, the authors conclude that it is unlikely that the inversion contributes to adaptation in more stressful environments. Rather the inversion seems to be more common in habitats that are closer to the native environment of ancestral Drosophila populations.

While I do agree with the authors that this observation is interesting, I do not think that it rules out a role in local adaptation. After all, the inversion is common in Africa, so it is perfectly conceivable that the non-inverted chromosome may have acquired a mutation contributing to the novel environment.

Based on their hypothesis, the authors propose an alternative strategy, which could maintain the inversion in a population. They perform some computer simulations, which are in line with the predicted behavior. Finally, the authors perform experiments and interpret the results as empirical evidence for their hypothesis. While the reviewer is not fully convinced about the empirical support, the key problem is that the proposed model does not explain the patterns of clinal variation observed for inversions in D. melanogaster. According to the proposed model, the inversions should have a similar frequency along latitudinal clines. So in essence, the authors develop a complicated theory because they felt that the current models do not explain the patterns of clinal variation, but this model also fails to explain the pattern of clinal variation.

To the contrary – in the Discussion paragraph beginning on Line 671, we explain why we would predict that a tradeoff between survival and reproduction should lead to clinal inversion frequencies. We suggest that a karyotype associated with a survival penalty should be increasingly disadvantageous in more challenging environments (such as high altitudes and latitudes for this species). Furthermore, an advantage in male reproductive competition conferred by that same haplotype may be reduced by the lower population densities that we would expect in more challenging environments (meaning that each female should encounter fewer males). Individually or jointly, these two factors predict that the equilibrium frequency of a balanced inversion frequency polymorphism should depend on a local population’s environmental harshness and population density, with the ensuing prediction that inversion frequency should correlate with certain environmental variables.

Reviewer #3 (Public review):

Summary:

In this study, McAllester and Pool develop a new model to explain the maintenance of balanced inversion polymorphism, based on (sexually) antagonistic alleles and a trade-off between male reproduction and survival (in females or both sexes). Simulations of this model support the plausibility of this mechanism. In addition, the authors use experiments on four naturally occurring inversion polymorphisms in D. melanogaster and find tentative evidence for one aspect of their theoretical model, namely the existence of the above-mentioned trade-off in two out of the four inversions.

Strengths:

(1) The study develops and analyzes a new (Drosophila melanogaster-inspired) model for the maintenance of balanced inversion polymorphism, combining elements of (sexually) antagonistically (pleiotropic) alleles, negative frequency-dependent selection and synergistic epistasis. Simulations of the model suggest that the hypothesized mechanism might be plausible.

(2) The above-mentioned model assumes, as a specific example, a trade-off between male reproductive display and survival; in the second part of their study, the authors perform laboratory experiments on four common D. melanogaster inversions to study whether these polymorphisms may be subject to such a trade-off. The authors observe that two of the four inversions show suggestive evidence that is consistent with a trade-off between male reproduction and survival.

Open issues:

(1) A gap in the current modeling is that, while a diploid situation is being studied, the model does not investigate the effects of varying degrees of dominance. It would thus be important and interesting, as the authors mention, to fill this gap in future work.

(2) It will also be important to further explore and corroborate the potential importance and generality of trade-offs between different fitness components in maintaining inversion polymorphisms in future work.

We appreciate the work put in to evaluating, improving, and summarizing our study. We agree that further work studying the effects of dominance and of the fitness components of the inversions is important.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

l. 354 : I don't understand what the authors mean by "an antagonistic and non-antagonistic allele". If there is a antagonistic polymorphism at a locus, then both alleles have antagonistic effects; i.e., allele B increases trait 1 and reduced trait 2 relative to allele A and vice versa.

Edited, agreed that the terminology used here was sub-optimal.

Reviewer #2 (Recommendations for the authors):

The motivation for their model is their claim that the clinal inversion frequencies are not compatible with local adaptation. The reviewer doubts this strong statement. Furthermore, the proposed model also fails to explain the inversion frequencies in natural populations.

Hence, rather than building a straw man, it would be better if the authors first show their experiments and then present their model as an explanation for the empirical results. Nevertheless, it is also clear that the empirical data are not very strong and cannot be fully explained by the proposed model.

This claim that we reject any role of local adaptation in clinal variation and selection upon inversion polymorphism does not hold up in a reading of our manuscript. We even suggest that locally varying selective pressures must be playing some role, although that does not imply that local adaptation is the ultimate driver of inversion frequencies. Indeed, we suggest that local adaptation alone is an insufficient explanation for inversion frequency clines in D. melanogaster, including because (1) these frequency clines do not approach the alternate fixed genotypes predicted by local directional selection, (2) these derived inversions tend to be more frequent in more ancestral environments (l.113-158).

In our public review response above, and in the Discussion section of our paper, we explain why our model can predict both the clinal frequencies of many Drosophila inversions and their intermediate maximal frequencies. Of course, we do not predict that most inversions in this species should follow the specific tradeoff investigated here. In fact, we were surprised to find even two inversions that experimentally supported our predicted tradeoff. Still, it remains possible that other inversions in this species are subject to other balanced tradeoffs not investigated here, which could help explain why they rarely reach high local frequencies.

Reviewer #3 (Recommendations for the authors):

My previous comments have been adequately addressed.

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

Reviewer #1 (Public Review):

[…]

To provide empirical support for this idea, the authors study the dynamics of inversions in population cages over one generation, tracking their frequencies through amplicon sequencing at three time points: (young adults), embryos and very old adult offspring of either sex (>2 months from adult emergence). Out of four inversions included in the experiment, two show patterns consistent with antagonistic effects on male sexual success (competitive paternity) and the survival of offspring, especially females, until an old age, which the authors interpret as consistent with their theory.

There are several reasons why the support from these data for the proposed theory is not waterproof.

(1) As I have already pointed out in my previous review, survival until 2 months (in fact, it is 10 weeks and so 2.3 months) of age is of little direct relevance to fitness, whether under natural conditions or under typical lab conditions.

The authors argue this objection away with two arguments

First, citing Pool (2015) they claim that the average generation time (i.e. the average age at which flies reproduce) in nature is 24 days. That paper made an estimate of 14.7 generations per year under the North Carolina climate. As also stated in Pool (2015), the conditions in that locality for Drosophila reproduction and development are not suitable during three months of the year. This yields an average generation length of about 19.5 days during the 9 months during which the flies can reproduce. On the highly nutritional food used in the lab and at the optimal temperature of 25 C, Drosophila need about 11-12 days to develop from egg to adult. Even assuming these perfect conditions, the average age (counted from adult eclosion) would be about 8 days. In practice, larval development in nature is likely longer for nutritional and temperature reasons, and thus the genomic data analyzed by Pool imply that the average adult age of reproducing flies in nature would be about 5 days, and not 24 days, and even less 10 weeks. This corresponds neatly to the 2-6 days median life expectancy of Drosophila adults in the field based on capture-recapture (e.g., Rosewell and Shorrocks 1987).

Second, the authors also claim that survival over a period of 2 month is highly relevant because flies have to survive long periods where reproduction is not possible. However, to survive the winter flies enter a reproductive diapause, which involves profound physiological changes that indeed allow them to survive for months, remaining mostly inactive, stress resistant and hidden from predators. Flies in the authors' experiment were not diapausing, given that they were given plentiful food and kept warm. It is still possible that survival to the ripe old age of 10 weeks under these conditions still correlates well with surviving diapause under harsh conditions, but if so, the authors should cite relevant data. Even then, I do not think this allows the authors to conclude that longevity is "the main selective pressure" on Drosophila (l. 936).

This is overall a thoughtfully presented critique and we have endeavored to improve our discussion of Pool (2015) and to clarify some of the language used about survival elsewhere. While we agree that challenges other than survival to 10 weeks are very relevant to Drosophila melanogaster, collection at 10 weeks does encompass some of these other challenges. Egg to adult viability still contributes to the frequencies of the inversions at collection and is not separable from longevity in this data. Collection at longevity was chosen in part to encompass all lifetime fitness challenges that might influence the inversion frequency at collection, albeit still within permissive laboratory conditions. Future experiments exploring specific stressors independently and beyond permissive lab conditions would generate a clearer picture.

In addition to general edits, the specific phrase mentioned at 1. 936 [now line 1003] has been revised from “In many such cases females are in reproductive diapause, and so longevity is the main selective pressure.” to “While longevity is a key selective pressure underlying overwintering, the relationship between longevity in permissive lab conditions without diapause and in natural conditions under diapause is unclear (Schmidt et al. 2005; Flatt 2020), and our experiment represents just one of many possible ways to examine tradeoffs involving survival.”

(2) It appears that the "parental" (in fact, paternal) inversion frequency was estimated by sequencing sires that survived until the end of the two-week mating period. No information is provided on male mortality during the mating period, but substantial mortality is likely given constant courtship and mating opportunities. If so, the difference between the parental and embryo inversion frequency could reflect the differential survival of males until the point of sampling rather than / in addition to sexual selection.

We have further clarified that when referenced as parental frequency, the frequency presented is ½ the paternal frequency as the mothers were homokaryotypic for the standard arrangement. We chose to present both due to considerations in representing the frequency change from paternal to embryo frequencies, where a hypothetical change from 0.20 frequency in fathers to 0.15 frequency in embryos represents a selective benefit (a frequency increase in the population), despite the reality that this is a decrease in allele frequency between paternal and embryo cohorts.

We mentioned a maximum 15% paternal mortality at line 827 [now l.1056], but have now added complete data on the counts of flies in the experiment as a supplemental table (Table S1) and have added or corrected further references to this in the results and methods [lines 555, 638, 975]. It is true that this may influence the observed frequency changes to some degree, and while we adjusted our sampling method to account for the effects of this mortality on statistical power [l.1056ff], we have now edited the manuscript to better highlight potential effects of this phenomenon on the recorded frequency changes.

It is also worth noting that, if mortality among fathers over the mating period is codirectional with mortality among aged offspring, this would bias the results against detecting an opposing antagonistic selective effect of the inversions on paternity share. This is now also mentioned in the manuscript, l.639ff.

(3) Finally, irrespective of the above caveats, the experimental data only address one of the elements of the theoretical hypothesis, namely antagonistic effects of inversions on reproduction and survival, notably that of females. It does not test for two other key elements of the proposed theory: the assumption of frequency-dependence of selection on male sexual success, and the prediction of synergistic epistasis for male fitness among genetic variants in the inversion. To be fair, particularly testing the latter prediction would be exceedingly difficult. Nonetheless, these limitations of the experiment mean that the paper is much stronger theoretical than empirical contribution.

This is a fair criticism of the limitations of our results, and we now summarize such caveats more directly in the discussion summary, lines 876ff.

Reviewer #2 (Public Review): 

[…]

Comments on the latest version:

I would like to give an example of the confusing terminology of the authors:

"Additionally, fitness conveyed by an allele favoring display quality is also frequency-dependent: since mating success depends on the display qualities of other males, the relative advantage of a display trait will be diminished as more males carry it..."

I do not understand the difference to an advantageous allele, as it increases in frequency the frequency increase of this allele decreases, but this has nothing to do with frequency dependent selection. In my opinion, the authors re-define frequency dependent selection, as for frequency dependent selection needs to change with frequency, but from their verbal description this is not clear.

We have edited this text for greater clarity, now line 232ff. We did not seek to redefine frequency dependence, and did mean by “the relative advantage of a display trait will be diminished” that an equivalent s would diminish with frequency. We have now remedied terminological issues introduced in the prior revision with regard to frequency dependent selection.

One example of how challenging the style of the manuscript is comes from their description of the DNA extraction procedure. In principle a straightforward method, but even here the authors provide a convoluted uninformative description of the procedure.

We have edited for clarity the text on lines 1016-1020. Citing a published protocol and mentioning our modifications seems an appropriate trade-off between representing what was done accurately, citing the sources we relied on in doing it, and limiting the volume of information in the main text for such a straightforward and common method. 

It is not apparent to the reviewer why the authors have not invested more effort to make their manuscript digestible.

We have invested a great deal of effort in making this manuscript as clear as we are able to.  We regret that our writing has not been to this reviewer’s liking. We believe we have been highly responsive to all specific criticisms, including revising all passages cited as unclear. In this round, we have again scrutinized the entire manuscript for any opportunity to clarify it, and we have made further changes throughout.  Although our subject matter is conceptually nuanced, we nevertheless remain optimistic that a careful, fresh reading of our revised manuscript would yield a more favorable impression.

Reviewer #3 (Public Review):

[…]

Weaknesses:

A gap in the current modeling is that, while a diploid situation is being studied, the model does not investigate the effects of varying degrees of dominance. It would be important and interesting to fill this gap in future work.

Agreed, and now reinforced at lines 892ff.

Comments on the latest version:

Most of the comments which I have made in my public review have been adequately addressed.

Some of the writing still seems somewhat verbose and perhaps not yet maximally succinct; some additional line-by-line polishing might still be helpful at this stage in terms of further improving clarity and flow (for the authors to consider and decide).

We have made further changes and some polishing in this draft, and greatly appreciate the guidance provided in improving the draft so far. 

Reviewer #1 (Recommendations For The Authors):

(1) While the model results are convincing, some of the verbal interpretation is confusing. In particular, the authors state that in their model the allele favoring male display quality shows a negative frequency dependence whereas the alternative allele has a positive frequency dependence. This does not make sense to me in the context of population genetics theory. For a one-locus, two-allele model the change of allele frequency under selection depends on the fitness of the genotypes concerned relative to each other. Thus, at least under no dominance assumed in this model, if the relative fitness of AA decreases with the frequency of allele A, the relative fitness of aa must decrease with the frequency of allele a. I.e., if selection is negatively frequency dependent, then it is so for both alleles.

This phrasing was wrong, and we have edited the relevant section.

(2) I am still not entirely sure that the synergistic epistasis assumed in the verbal model is actually generated in the simulations; this would be easy enough to check by extracting the mating success of males with different genotypes from the simulation output should be reported, e.g., as a figure supplement.

Our new Figure S2, which depicts haplotype frequencies for a set of the simulations presented in Figure 4, should demonstrate a necessary presence of synergistic epistasis. These results further clarify that the weaker allele B is only kept when linked to A. The same fitness classes of genotype are present in the simulations with and without the inversion, so the only mechanical difference is the rate of recombination, and the only way this might change selection on the alleles is if a variant has a different fitness in one haplotype background than another – i.e. epistasis. The maintenance of haplotypes AB and ab to the exclusion of Ab and aB relies on the lesser relative fitness of Ab and aB. And since survival values are multiplicative, this additional contribution must come from the mate success of AB being disproportionately larger than Ab or aB, indicating the emergent synergistic epistasis posited by our model. We have clarified this point in the text at line 363ff.

(3) l. 318ff: What was this set number of males? I could not find this information anywhere. Also, this model of the mating system is commonly referred to as "best of N", so the authors may want to include this label in the description.

We indicate this detail just after the referenced line, now reworded and on l. 338-340 as “For each female’s mating competition, 100 males were sampled, though see Figure S1 for plots with varying encounter number.”  Among these edits, “one hundred” has been changed to a numeral for easier skimming, and Figure S1 is now referenced here earlier in the text. Several edits have also been made in the caption of Figures 2 and 3, and in the relevant methods section to clarify the number of encountered males simulated, mention best of N terminology, and clarify how the quality score is used in the mate competition.

(4) The description of the experiment is still confusing. The number of individuals of each sex entered in each mating cage is missing from the Methods (l. 914); although I did finally find it in the Results. These flies were laying over 2 weeks - does this mean that offspring from the entire period were used to obtain the embryo and aged offspring frequencies, or only from a particular egg collection? If the former, does this mean that the offspring obtained from different egg batches were aged separately? Were the offspring aged in cages or bottles, at what density? Given that only those males that survived until the end of the two-week mating period were sequenced, it is important to know what % of the initial number of males these survivors were. A substantial mortality of the parental males could bias the estimate of parental frequencies. How many parental males, embryos and aged offspring were sequenced? Were all individuals of a given cage and stage extracted and sequenced as a single pool or were there multiple pools? The description could also be structured better. For example, the food and grape agar recipes and cage construction are inserted at random points of the description of the crossing design, which does not help.

We have now reorganized and edited these portions of the Methods text. Portions of this comment overlap with edits responding to (2) of the Public Review and below for l. 921 in Details. Offspring from different laying periods were aged in different bottles, further separated by the time at which they eclosed. They were then pooled for DNA extraction and library preparation by sex and a binary early or late eclosion time. This data was present in the “D. mel. Sample Size” column of supplemental tables S6 and S7 (now S7 and S8), but we have added and referenced a new table to specifically collate the sample sizes of different experimental stages, table S1. Now referenced at lines 555, 638, 975, 1057.

(5) The caption of figure 9 and the discussion of its results should be clear and explicit about the fact that "adult offspring" in Fig 9A and "female" and "male" refers to adults surviving to old age (whereas "parental" in Fig 9A refers to young adults in their reproductive prime. This has consequences for the interpretation of the difference between "parental" and "adult offspring", as it combines one generation of usual selection as it occurs under the conditions of the lab culture (young adult at generation t -> young adult in generation t+1) with an additional step of selection for longevity. Thus, a marked change in allele frequency does not imply that the "parental" frequency does not represent an equilibrium frequency of the inversions under the lab culture conditions. Furthermore, it would be useful to state explicitly that Figure 9B represents the same results as figure 9A, but with the aged offspring split by sex.

Figure caption edited to provide further clarity on the age of cohorts and presented data, along with the relevant results section (2.3) referencing this figure.

We avoid making any statements about the equilibrium frequencies of inversions under lab conditions, and whether or not any step of our experiment reflects such equilibria, because our investigation does not rely upon or test for such conditions. Instead, our analysis focuses on whether inversions have contrasting effects (as indicated by frequency changes that are incompatible with neutral sampling) between different life history components.  Under our model, such frequency reversals might be detectable both at equilibrium balanced inversion frequencies and also at frequencies some distance away from equilibria. We have now clarified this point at l. 970-972.

Details:

l. 211: this should be modified as male-only costs are now included.

Edited. “survival likelihood (of either or both sexes).”

l. 343: misplaced period

Edited.

l. 814: "We confirmed model predictions...": This sounds like it refers to an empirical confirmation of a theory prediction, but I think the authors just want to say that their simulations predicted antagonistic variants can be maintained at an intermediate equilibrium frequency. So the wording should be changed to avoid ambiguity.

Edited. Now line 869.

l. 853: How can a genome be "empty"? Do the authors mean an absence of any polymorphism?

Edited to: “In SAIsim, a population is instantiated as a python object, and populated with individuals which are also represented by python objects. These individuals may be instantiated using genomes specified by the user, or by default carry no genomic variation.” Lines 913ff.

l. 853: I do not see this diagramed in Figure 5

Apologies, fixed to Fig. 2

l. 864: is crossing-over in the model limited to female gametogenesis (reflecting the Drosophila case) or does it occur in both sexes?

There is a variable in the simulator to make crossover female-specific. All simulations were performed with female-only crossover. Edited for clarity. “While the simulator can allow recombination in both sexes, all simulations presented only generate crossovers and gene conversion events for female gametes, in accordance with the biology of D. melanogaster.” Lines 928-929.

l. 906: "F2" is ambiguous; does this mean that the mix of lines was allowed to breed for two generations? Also, in other places in the manuscript these flies appear to be referred to are "parental". So do not use F2.

Edited, F2 language removed and replaced with being allowed to breed for two generations. Now lines 967ff.

l. 910: this is incorrect/imprecise; what can be inferred is the frequency of the inversions in male gametes that contributed to fertilization. This would correspond to the frequency in successful males only if each successful male genotype had the same paternity share.

Edited, now “Since no inversions could be inherited through the mothers, inversion frequencies among successful male gametes could be inferred from their pooled offspring.” Now line 994.

l. 912: "without a controlled day/night cycle" meaning what? Constant light? Constant darkness? Daylight falling through the windows?

Edited to “Unless otherwise noted, all flies were kept in a lab space of 23°C with around a degree of temperature fluctuation and without a controlled day/night cycle. Light exposure was dependent on the varying use of the space by laboratory workers but amounted to near constant exposure to at least a minimal level of lighting, with some variable light due to indirect lighting from adjacent rooms with exterior windows.” Now lines 1007-1010.

l. 921: I cannot parse this sentence. Were the offspring isolated as virgins?

No, the logistics of collecting virgins would have been prohibitive, and it did not seem essential for our experiment. Hopefully the edits to this section are clearer, now lines 978ff.

Author response:

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

Reviewer #1 (Public review): 

Summary: 

This manuscript reports the substrate-bound structure of SiaQM from F. nucleatum, which is the membrane component of a Neu5Ac-specific Tripartite ATP-dependent Periplasmic (TRAP) transporter. Until recently, there was no experimentally derived structural information regarding the membrane components of TRAP transporter, limiting our understanding of the transport mechanism. Since 2022, there have been 3 different studies reporting the structures of the membrane components of Neu5Ac-specific TRAP transporters. While it was possible to narrow down the binding site location by comparing the structures to proteins of the same fold, a structure with substrate bound has been missing. In this work, the authors report the Na+-bound state and the Na+ plus Neu5Ac state of FnSiaQM, revealing information regarding substrate coordination. In previous studies, 2 Na+ ion sites were identified. Here, the authors also tentatively assign a 3rd Na+ site. The authors reconstitute the transporter to assess the effects of mutating the binding site residues they identified in their structures. Of the 2 positions tested, only one of them appears to be critical to substrate binding.

Strengths: 

The main strength of this work is the capture of the substrate bound state of SiaQM, which provides insight into an important part of the transport cycle.

Weaknesses: 

The main weakness is the lack of experimental validation of the structural findings. The authors identified the Neu5Ac binding site, but only test 2 residues for their involvement in substrate interactions, which is quite limited. However, comparison with previous mutagenesis studies on homologues supports the location of the Neu5Ac binding site. The authors tentatively identified a 3rd Na+ binding site, which if true would be an impactful finding, but this site was not sufficiently experimentally tested for its contribution to Na+ dependent transport. This lack of experimental validation prevents the authors from unequivocally assigning this site as a Na+ binding site. However, the reporting of these new data is important as it will facilitate follow up studies by the authors or other researchers. 

Comments on revisions: 

Overall, the authors have done a good job of addressing the reviewers' comments. It's good to know that the authors are working on the characterisation of the potential metal binding site mutants - characterizing just a few of these will provide much-needed experimental support for this potential Na+ site. 

The new MD simulations provide additional support for the new Na+ site and could be included.

However, as the authors know, direct experimental characterisation of mutants is the ideal evidence of the Na+ site.

Aside from the characterisation of mutants, which seems to be held up by technical issues, the only remaining issue is the comparison of the Na+- and Na+/Neu5Ac-bound states with ASCT2. It still does not make sense to me why the authors are not directly comparing their Na+ only and Na+/Neu5Ac states with the structures of VcINDY in the Na+-only and Na+/succinate bound states. These VcINDY structures also revealed no conformational changes in the HP loops upon binding succinate, as the authors see for SiaQM. Therefore, this comparison is very supportive. It is understood that the similarity to the DASS structure is mentioned on p.17, but it is also interesting and useful to note that TRAP and DASS transporters also share a lack of substrateinduced local conformational changes, to the extent these things have been measured.

We acknowledge the summary weakness that experimental data to support the third Na binding site is critical.

Based on the reviewer’s suggestion, we added the following in the main text and a supplementary figure comparing the Na ion binding sites between VcINDY and SiaQM. Page 13.

“These two sodium ion binding sites are also conserved in the structure of VcINDY (Supplementary Figure 7) (Sauer et al., 2022). In both cases, the sodium ions are bound at the helix-loop-helix ends of HP1 and HP2. The binding sites utilize both side chains and main chain carbonyl groups. The number of main chain carbonyl interactions suggests that they are critical, and using main chain rather than side chain interactions minimizes the likelihood of point mutations affecting the binding.”

Reviewer #3 (Public review): 

The manuscript by Goyal et al report substrate-bound and substrate-free structures of a tripartite ATP independent periplasmic (TRAP) transporter from a previously uncharacterized homolog, F. nucleatum. This is one of most mechanistically fascinating transporter families, by means of its QM domain (the domain reported in his manuscript) operating as a monomeric 'elevator', and its P domain functioning as a substrate-binding 'operator' that is required to deliver the substrate to the QM domain; together, this is termed an 'elevator with an operator' mechanism.

Remarkably, previous structures had not demonstrated the substrate Neu5Ac bound. In addition, they confirm the previously reported Na+ binding sites, and report a new metal binding site in the transporter, which seems to be mechanistically relevant. Finally, they mutate the substrate binding site and use proteoliposomal uptake assays to show the mechanistic relevance of the proposed substrate binding residues.

Strengths: 

The structures are of good quality, the presentation of the structural data has improved, the functional data is robust, the text is well-written, and the authors are appropriately careful with their interpretations. Determination of a substrate bound structure is an important achievement and fills an important gap in the 'elevator with an operator' mechanism.

Weaknesses: 

Although the possibility of the third metal site is compelling, I do not feel it is appropriate to model in a publicly deposited PDB structure without directly confirming experimentally. The authors do not extensively test the binding sites due to technical limitations of producing relevant mutants; however, their model is consistent with genetic assays of previously characterized orthologs, which will be of benefit to the field. Finally, some clarifications of EM processing would be useful to readers, and it would be nice to have a figure visualizing the unmodeled lipid densities - this would be important to contextualize to their proposed mechanism.

Reviewer #3 (Recommendations for the authors): 

I appreciate the authors' responses to our critiques; the revised manuscript is much improved and has addressed most of my concerns. I look forward to seeing their follow up experiments testing mutational e=ects. I think MD simulations of ion-binding sites on their own are supportive but by themselves not su=icient to prove the existence of a functional Na+-binding site. Some clarifications in the methods/supplements would satisfy my concerns about data processing and analysis.

- Unliganded map: were the 141,272 particles used for one class of ab initio? This is unusual, usually multiple ab initio classes are used to further eliminate junk particles. The authors themselves use 6 classes for the substrate-bound dataset.

We classified the particles into multiple 3-D classes.  There was no improvement in statistics or maps on splitting these further.  Hence, we did not pursue that further. 

- Substrate-bound map: how did the four 'identical' classes independently refine? Are similar Na+/substate densities found in each separate class?

The other classes refined to worse than 4.5 Å resolution. We stopped characterizing them past that point.  We were hoping to see multiple conformations that are diLerent – and hopefully a class where only two sodium ions could be bound.  However, any interpretation at 4.5 Å would be unreliable.

- Both maps: all ab initio classes prior to final refinement should be displayed in the supplementary workflow, this is common for EM processing diagrams.

We agree it is common – however, unless there is a good reason to discuss the other classes, we are not convinced of the value of crowding the figures.

- What specific refinement package and version of Phenix are the authors using? It seems unusual that it is not possible to refine without a metal in Phenix real-space refinement, I have seen many structures where there is no issue refining without critical ions/waters. The authors should double check that they are using the appropriate scattering table for cryo-EM, which should be "electron".

Sorry for the confusion – we did not mean to say we cannot refine without a metal. If we want to add something to the density, we cannot refine it without suggesting a metal or solvent.  The site without anything added will refine without any issues but in the absence of additional verification, we cannot be sure of the identity of the ions. We are confident of the metal binding site – but not confident of the exact metal bound.  We used Sodium as our first hypothesis.

We don’t think the scattering factors will help in the identification of the ions. Servalcat as part of CCP-EM can produce diLerence maps and we believe that for identification of ions, it will require higher resolution (<2.5 Å) but at this resolution, we can say that there is a nonprotein density but not more than that. We were using “electron” (which we believe is default with phenix.real_space_refine). The refinement was performed using standard protocols and appropriate scattering factors (Phenix version 1.19x), and we have previously used similar refinement protocols for other maps/models (Example -Vinothkumar KR, Arya CK, Ramanathan G, Subramanian R. 2021. Comparison of CryoEM and X-ray structures of dimethylformamidase. Progress in Biophysics and Molecular Biology, CryoEM microscopy developments and their biological applications 160:66–78. doi:10.1016/j.pbiomolbio.2020.06.008).

To convince the reviewer of the quality of the maps, we have added figures that show the model-to-map fit of all of the main secondary structural elements in both the unliganded and the Neu5Ac bound forms.

- I certainly understand the authors' reluctance to not model the entirety of protein densities; however, I think it would be useful to highlight these densities in the global context of the protein. A common way to show this is to show the density proximal to protein chains in one color, and the remaining densities in a contrasting color (Figure 1 somewhat demonstrates this but it is di=icult to tell). I think this would be a nice figure to show the presence and location of unmodeled densities.

We have modified supplementary figure 3 to include unmodelled densities in panels G and H for both structures.

- Small detail, "uniform" is misspelled as "unifrom" in supplementary Figure 3. 

Thank you.  Corrected.

Author response:

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

We appreciate the positive assessment and agree that the experimental data offer valuable insights into HBV capsid assembly inhibition. Based on the reviewers' suggestions, we have clarified the cryo-EM data and added structural and mechanistic details throughout the manuscript, which we believe significantly enhance its overall clarity and impact. The manuscript now better reflects a promising strategy to interfere with the HBV life cycle. We have carefully addressed all comments to improve both the clarity and quality of the manuscript.

Response to Public Reviews

We greatly appreciate the insightful comments and suggestions from the reviewers. Below, we provide responses to the points raised in the public reviews.

Reviewer #1 (Public Review):

Summary:

In this paper, the authors present an interesting strategy to interfere with the HBV life cycle: the preparation of geranyl and peptides' dimers that could impede the correct assembly of hepatitis B core protein HBc into viable capsids. These dimers are of different nature, depending on the HBc site the authors plan to target. A preliminary study with geranyl dimers (targeting a hydrophobic site of HBc) was first investigated. The second series deals with peptide-PEG linker-peptide dimers, targeting the tips of HBc dimer spikes.

Strengths:

This work is very well conducted, combining ITC experiments (for determination of dimers' KD), cellular effects (thanks to the grafting of previously developed dimers with polyarginine-based cell penetrating peptide) HBV infected HEK293 cells and Cryo-EM studies.

The findings of these research teams unambiguously demonstrated the interest of such dimeric structures in impeding the correct HBV life cycle and thus, could bring solutions in the control of its development. Ultimately, a new class of HBV Capside Assembly Modulators could arise from this study.

There is no doubt that this work could bring very interesting information for people working on VHB.

Weaknesses:

Some minor corrections must be made, especially for a more precise description of the strategy and the chemical structure of the designed new VHB capsid assembly modulators.

We are grateful for the positive feedback on the experimental design, the combination of ITC, cellular effects, and Cryo-EM studies, and the potential for developing new classes of HBV Capsid Assembly Modulators (CAMs). In the revised version we have clarified the design rationale for the choice of the PEG linker length in the Supplementary Information, linking it to the structural measurements of the capsid. Chemical structures and detailed molecular formulas were added and terms have been corrected. A scrambled dimeric peptide served as a negative control, which showed no binding, confirming the specificity of our designed peptide and ruling out non-specific interactions from other elements of the molecules such as the linkers. Finally, we have revised the nomenclature for the geranyl dimers to better reflect the chemical structure. All figures, including Figure 3, have been updated to high-resolution. All mentioned typos have been corrected. Consultation dates have been added to the website references. HPLC terminology was corrected.

Reviewer #2 (Public Review):

Summary:

Vladimir Khayenko et al. discovered two novel binding pockets on HBc with in vitro binding and electron microscopy experiments. While the geranyl dimer targeting a central hydrophobic pocket displayed a micromolar affinity, the P1-dimer binding to the spike tip of HBc has a nanomolar affinity. In the turbidity assay and at the cellular level, an HBc aggregation from peptide crosslinking was demonstrated.

Strengths:

The study identifies two previously unexplored binding pockets on HBc capsids and develops novel binders targeting these sites with promising affinities.

Weaknesses:

While the in vitro and cellular HBc aggregation effects are demonstrated, the antiviral potential against HBV infection is not directly evaluated in this study.

Thank you for recognizing the innovative approach of our work and the potential for developing novel antivirals targeting HBc. We have now included additional discussion on potential future experiments aimed at evaluating the compounds' effects on cellular physiology and viral infectivity.

Reviewer #3 (public Review):

Summary:

HBV is a continuing public health problem and new therapeutics would be of great value. Khayenko et al examine two sites in the HBc dimer as possible targets for new therapeutics. Older drugs that target HBc bind at a pocket between two HBc dimers. In this study Khayenko et al examine sites located in the four helix bundle at the dimer interface.

The first site is a pocket first identified as a triton100 binding site. The authors suggest it might bind terpenes and use geraniol as an example. They also test a decyl maltose detergent and a geraniol dimer intended for bivalent binding. The KDs were all in the 100µM range. Cryo-EM shows that geraniol binds the targeted site.

The second site is at the tip of the spike. Peptides based on a 1995 study (reference 43) were investigated. The authors test a core peptide, two longer peptides, and a dimer of the longest peptide. A deep scan of the longest monomer sequence shows the importance of a core amino acid sequence. The dimeric peptide (P1-dimer) binds almost 100 fold better than the monomer parent (P1). Cryo-EM structures confirm the binding site. The dimeric peptide caused HBc capsid aggregation When HBc expressing cells were treated with active peptide attached to a cell penetrating peptide, the peptide caused aggregation of HBc antigen mirroring experiments with purified proteins.

Strengths:

The two sites have not been well investigated. This paper marks a start. The small collection of substrates investigated led to discovery of a dimeric peptide that leads to capsid aggregation, presumably by non-covalent crosslinking. The structures determined could be very useful for future investigations.

Weaknesses:

In this draft, the rational for targets for the triton x100 site is not well laid out. The target molecules bind with KDs weaker that 50µM. The way the structural results are displayed, one cannot be sure of the important features of binding site with respect to the the substrate. The peptide site and substrates are better developed, but structural and mechanistic details need to be described in greater detail.

We appreciate the reviewer’s positive comments on identifying and targeting previously unexplored sites on HBc, and the potential utility of our dimeric peptides in future studies. We have revised the Results section to better explain the rationale behind targeting the hydrophobic binding site. Additionally, the structures have been revised for clearer presentation, and we now emphasize the key features of the binding site and the role of substrate specificity.

Recommendations For The Authors:

Reviewer #1 (Recommendations For The Authors):

For clarity, the chemical structure of SLLGRM peptide, geraniol and HAP molecules must be indicated, preferably in Fig. 1 (at least in the Supplementary Information section).

We have now included the chemical structures of the SLLGRM peptide, geraniol, and HAP molecules for clarity in Figure 1 and in the main manuscript to ensure they are easily accessible for reference and to provide further detail and context.

In the same idea, in Fig. 1 (and in the text): The molecular formula of heteroaryldihydropyrimidine HAP must be clearly indicated, as the nature of the heteroatom (S, O, N?) in this "heteroaryl" derivative is not indicated.

The full molecular formula of HAP (((2S)-1-[[(4R)-4-(2-chloranyl-4-fluoranyl-phenyl)-5-methoxycarbonyl-2-(1,3-thiazol-2-yl)-1,4-dihydropyrimidin-6-yl]methyl]-4,4-bis(fluoranyl)-pyrrolidine-2-carboxylic acid), is now included the figure legend.

with a polyethylene glycol (PEG) linker that could bridge the distance of 38 Å between the two opposing hydrophobic pockets": what is the rationale of the design of this linker? Authors must explain briefly why/how they have chosen this linker length and nature (please indicate a reference for the appropriate choice of PEG linker). Same remarks for dimers targeting the capsid spike tips, having 50 angstroms PEG linkers. So, the choice of the linker length must be clearly explained and not be only mentioned in the sentence of the discussion part "Using our structural knowledge of the capsid, particularly the distances between the spikes.

We have now better clarified the rationale for the design of the PEG linker length. The linker lengths were specifically chosen based on structural knowledge of the capsid, particularly the measured distances between the spike tips (60 Å) and the hydrophobic pockets (40 Å). In the Supplementary Information (Supplementary Figure 1), we now clearly explain how these measurements guided the choice of PEG linker length, allowing for optimal bridging and interaction between the binding sites. This supplementary figure now explicitly connects the design rationale to the specific structural features of the capsid.

I do not agree with the authors when they claim a "nanomolar affinity of 312 nM". To me, a nanomolar affinity would require several of few tens of nanoM (but not three hundreds) ... So, please correct with "sub-micromolar affinity of 312 nM" and all the other parts of the manuscript (title and caption of Figure 3..., "the peptide dimer (P1dC) with nanomolar affinity" "nanomolar levels"...).

We thank the Rev#1 for pointing this out. Since the term "nanomolar affinity" can indeed be interpreted as referring to the lower end of the nanomolar range, rather than values close to 300 nM we have revised the manuscript to refer to the "sub-micromolar affinity" where applicable. This change has been made throughout the manuscript, including the subtitles and figure captions, and the text.

The drug design strategy was to combine two peptides showing low affinity, attached by a PEG linker with an appropriate length and appears obvious to me. But a control experiment is anyway missing: the peptide-PEG linker derivative (not the dimer peptide-PEG linker-peptide...) should have been evaluated for an unambiguous proof of concept of these dimeric peptides. To my opinion, for the publication of this work, these experiments should be brought (eg, when describing the affinities of SLLGR dimers). I agree that Cryo-EM experiments bring evidences of the dimer binding but the affinity values for (peptide-PEG linker) derivatives would bring an additional proof (as the PEG flexible linkers was not resolved by Cryo-EM).

Thank you for your thoughtful comment regarding the use of a monovalent control for the peptide-PEG linker. A scrambled dimeric peptide serves as a negative control. In ITC it showed no binding at all. Thereby ruling out possibly unspecific interactions mediated by the introduced PEG linker or handle itself.

Given the complete lack of binding with the scrambled dimeric peptide, we believe this thoroughly excludes the need for an additional monovalent control, as it provides strong evidence that the observed binding is driven specifically by the designed peptide sequence and not by the linker or other structural components. We have now made this clarification more explicit in the revised manuscript to avoid any ambiguity. We hope this addresses your concern, and we appreciate your suggestion to further strengthen the rigor of the work. Despite its identical charge, molecular weight and atom composition the scrambled control did not cause HBc aggregation in living cells, thus indicating sequence specific action of the aggregating dimer.

The nomenclature of the dimers must be modified because there is no logic between the name "long dimer" and the chemical structure. Particularly, the number of ethylene glycol motifs must be indicated: authors have to find an appropriate nomenclature indicating both the linker length and nature (small molecule or peptide) of the bivalent parts (and hence, do not mention anymore "short geranyl dimer" "long geranyl dimer").

Thank you for your valuable suggestion regarding the nomenclature of the dimers. We agree that the terms "short geranyl dimer" and "long geranyl dimer" do not fully reflect the chemical structure of the molecules. In response, we have revised the nomenclature to provide a clearer indication of both the linker length and the nature of the bivalent parts. We now refer to the dimers as (Geranyl)₂-Lys for the dimer with two geranyl groups attached to lysine and (Geranyl-PEG3)₂-Lys for the dimer with a PEG3 linker (three ethylene glycol units) between the lysine amine and the geranyl groups. These revised names more accurately describe the structural differences and should avoid any ambiguity.

Lines 198-199: "Among these, the dimerized P1 exhibited a higher 198 occupation of the binding site, as illustrated in Supplementary Figure 9." But in Supp. Fig. 9, dimer P1dC (10) is described. As the text above is describing P1-dimer (9), the Supp. Fig. 9 must be provided, if available. If not, please modify this conclusion accordingly. In the text, when mentioning dimerized P1 peptide, authors must indicate with which compound it deals: (9) or (10)?

Thank you for your careful reading of the manuscript and for pointing out the discrepancy. In Supplementary Figure 9, the dimer described is P1dC, not P1d. The text has been revised to clarify this. We appreciate your attention to detail.

Please note that the graphic quality of Figure 3 is bad as it results in pixelized drawings (especially for the chemical structures).

Thank you for your feedback regarding the quality of Figure 3. We have now updated all figures, including Figure 3, to high-resolution PNG format with 300-500 dpi to ensure optimal graphic quality. This should resolve the pixelization issue, particularly for the chemical structures.

Minor typos: "clinical studies, a third are CAMs.[6]" "to the spike base hydrophobic pocket" "geraniol affinity to the central hydrophobic pocket, we designed"

We have corrected the punctuation in the mentioned sentences and appreciate your careful review of the manuscript.

Concerning the citation of a website (references 5 and 6), I guess that the consultation date should be mentioned.

We have now updated the references accordingly, including the consultation dates.

In the Materials and Methods part, Peptide synthesis paragraph, authors must write "semi-preparative HPLC.

It’s now corrected to "semi-preparative HPLC".

In the supplementary information file, 1H and 13C NMR spectrum for the small molecule "Short Geranyl Dimer (SGD)" should be provided.

The purity and identity of this Geranyl derivate were confirmed through UV detection in LC-MS and supported by the mass spectra, which provide robust and clear evidence of the compound's structure and well-accepted method for confirming the structure in this context. While we understand the value of NMR in structural analysis, we believe that additional analytical evidence is not critical for this study.

Reviewer #2 (Recommendations For The Authors):

Overall, this study presents an innovative approach to target the HBV core protein and paves the way for developing new classes of antivirals with a distinct mechanism of action. The findings expand the current knowledge of druggable sites on HBc capsids and provide promising lead compounds. Future studies exploring the antiviral effects and optimizing the binders for therapeutic applications would be valuable next steps.

We sincerely thank the reviewer for the positive assessment of our work and for highlighting its innovative approach to targeting the HBV core protein. We appreciate your recognition of the study's potential in paving the way for developing new classes of antivirals with distinct mechanisms of action. Below, we provide responses to each of the points raised.

The significance of the central hydrophobic pocket as a target may require additional experiments for validation. Currently, the substrate binding activity is relatively low and appears to have a non-significant impact on HBc.

We agree that the central hydrophobic pocket exhibits relatively weak binding affinity with the ligands tested in this study. However, we have provided additional structural evidence and affinity data to support its relevance as a druggable site. In recognition of the weak affinity of these small molecules, we expanded our focus to include peptide-based binders, which yielded higher affinities, particularly when dimerized.

It might be more effective to present Figure 1B after summarizing all the results.

We understand the reviewer’s suggestion. However, we decided to highlight and summarize the major findings early in the manuscript. We included Figure 1B at the beginning to allow readers to quickly grasp the core concepts and outcomes of our study.

The labels for P1/P2 are presented in Figure 1A, yet their definitions are not provided until the second part of the Results section.

We appreciate the reviewer’s observation. While see a benefit of showing three trackable sites on HBV early and as an overview but we also agree that the early presentation of P1/P2 could lead to some confusion. To resolve this, we have revised the figure to introduce only on the minimal peptide to avoid any ambiguity. The full dimer sequences and names are introduced later.

Further investigation of the cytotoxic potential of peptide-induced HBc aggregation is necessary.

Investigating the cytotoxicity together with infectivity is an important future direction but outside the scope of this study. We now elaborate on this point in the discussion.

Reviewer #3 (Recommendations For The Authors):

Two sites in the dimer interface are shown to bind ligands. It is not shown that filling these regions will change infection. The exhaustive studies by Bruss showed point mutations directly alter infection and would be of value to discuss.

We thank Rev#3 for this very helpful comment. We now highlight how point mutations in these regions were shown to affect HBV infectivity. Thereby providing a link between our findings and how ligand binding might influence the viral life cycle.

It is not shown whether the two sites interact. Molecular dynamics by Hadden or Gumbart may be informative. The failure to look for a connection between these sites is an oversight.

We thank Rev#3 for the insightful suggestion to explore potential interactions between the two binding sites. We acknowledge that molecular dynamics (MD) simulations, such as those performed by Gumbart et al. and Hadden et al., could indeed provide valuable insights into the structural dynamics and potential cooperativity between these sites. Indeed, molecular dynamics of the HBV capsid by Perilla and Hadden has demonstrated significant flexibility in the capsid spikes and their interactions with neighboring subunits suggesting that the dynamics of binding sites could influence ligand accessibility and potential crosstalk.

We believe that our own previous structural studies together with data in this work provide substantial experimental evidence on this topic. In Makbul et al. 2021a (doi.org/10.3390/microorganisms9050956) we observed that peptide binding (particularly P2) did not stabilize the spikes; instead, the upper part of the spikes exhibited considerable wobbling. This variability mirrored the conformational diversity reported in MD simulations. Using local classification, we noted that the variability in the spike's upper region was greater when P2 was bound than in its absence. Additionally, in Makbul et al. 2021b (doi.org/10.3390/v13112115), we showed that peptide binding had little effect on the hydrophobic pocket beneath the mobile spike region, located in the more rigid part of the capsid. While we observed F97 in the D-monomer adopting two alternate rotamer orientations upon P2 binding this was not exclusive to P2, as similar changes were noted in the L60V mutant even without bound peptide.

We have updated the manuscript to briefly discuss this crosstalk, that provides additional context to our findings. Interestingly, only TX100—but not geraniol—completely flipped F97 into an alternate orientation, forming a new π-π stacking interaction with the mobile region of the spike. This finding suggests that interactions within the hydrophobic pocket are transmitted based on ligand specific interactions to the tips of the spikes. Thus, supporting and refining the concept of a crosstalk between binding sites, primarily initiated from the hydrophobic pocket in a ligand specific fashion.

The logic for proposing a terpene ligand is strained. Comparisons are made to HBs and the HDV delta antigen. However, HBs is myristoylated not farnesylated and delta antigen binds HBs not HBc.

We have revised the text to clarify the rationale for testing terpenes as ligands, focusing instead on the specific properties of the hydrophobic pocket targeted by geraniol.

The authors suggest larger terpenes as binding agents, but there does not appear to be room for a longer molecule in the binding site. The authors do not discuss whether a longer molecule could be modeled in the site based on their density.

We appreciate this observation and agree that the potential for larger terpenes to bind this site is not obvious from the structural data presented in this work. We have now included a more detailed visualization (Fig2D) and discussion of the hydrophobic binding pocket, based on the density observed in the presented geraniol structure and the previous triton structure and discuss its implications of the binding of larger hydrophobic molecules into the site (Fig 2D).

The authors note that the structure could explain molecular details of this site, but these are not discussed. A more complete analysis of the geraniol protein is necessary, including an estimate of the resolution of that density.

We agree that a more complete analysis of the hydrophobic binding site was warranted. We have now expanded the discussion of the structural details of this binding site based on the geraniol-bound structure, the density and occupancy accounted by this ligand. These additional details (Fig 2C,D and Fig 5) should provide a clearer understanding of the binding interactions observed.

The dimeric geraniol is marginally better binding than the monomer, two-fold, but this could be due to doubling the number of geraniols per ligand or due to an undefined interaction of the extended molecule with the surface of the capsid. A geraniol linker should be tested.

The modest improvement in binding may indeed only reflect the doubled number of geraniols rather than linker-mediated avidity effects. Interaction of the linker with the capsid surface is ruled-out by the scrambled control that included the same linkers but did not show any capacity to bind.

Is the enhanced binding of dimer due to bivalent binding of dimer to one capsid? Is it a chance interaction of the linker with the surface of HBc, which is easily tested? Is it an avidity effect due to aggregation of capsids?

Thank you for this insightful question. Our data suggest that the enhanced binding is due to bivalent interactions. To address the possibility of non-specific interactions from either the handle or the linker, we included a scrambled dimeric peptide as a negative control, which showed no binding. This rules out non-specific interactions from the linker or handle. Given this, we believe an additional monovalent control is unnecessary, as the scrambled control confirms that the binding is driven by the geraniol and peptide warheads alone. We have clarified this in the revised manuscript and appreciate your suggestion to strengthen the study.

The experimental analysis of point mutation of P1 is not analyzed beyond stating that it shows the importance of the core peptide sequence. Is there rationale for the effect of R3 to E and K10 to E mutation?

We appreciate the reviewer's curiosity and request for a more detailed discussion of the P1 deep mutational scan data and its implications. The observed low mutation tolerance of the core peptide sequence SLLGRM regarding HBc binding is highly consistent with our prior structural data and binding studies in solutions (https://doi.org/10.3390/microorganisms9050956) as well as the results from the original phage library screening (M. R. Dyson, K. Murray, Proceedings of the National Academy of Sciences 1995, 92, 2194–2198), and the binding data presented here. Notably, the data set does not suggest that additional binding interfaces contribute to the aggregation seen with N-terminal elongated P1 and P2 versus the non-aggregating shorter SLLGRM. While the positional scan largely aligns with previous phage binding hierarchy and quantified ligands, we were previously prompted by surprising affinity gains for positive to negative amino exchanges in related peptides in same way as Rev#3: Specifically, “SLLGEM” has been predicted previously and here to show enhanced affinity over “SLLGRM”. Quantification in solution, however, could not confirm this enhanced HBV binding affinity (Makbul et al. 2021 Microorganisms), which could not be recapitulated by in solution quantification. In the revised version of the manuscript we now highlight the possible limited predictive power of this assay for positions where positively charged residues are exchanged by negatively charged residues (Figure legend of Fig 3D).

The fluctuations in Figure 3B could be largely magnification of noise due to changing the y-axis. The fluctuations can be characterized as standard variation, excluding the injections, to allow a quantitative judgment.

Isothermal titration calorimetry heat fluctuations without injections are now shown in the supplementary information scaled to the same y-axis (Supplementary Figure 3D). 

Molecular graphics throughout are too small and poorly labeled.

We have revised the molecular graphics throughout the manuscript to increase their size and improve labeling for clarity. All figures are now provided in 500dpi.

In Figure 2, compounds 1 and 2 are pyrophosphates. The label in the figure should be corrected.

Thank you for pointing this out. These compounds were removed for clarity.

In the introduction, the phrase "discontinuation frequently leads to relapse" should be changed to something less ambiguous.

Thank you for highlighting this point regarding the phrasing in the introduction. We have revised the statement to more accurately reflect the clinical situation by specifying that stopping treatment often results in viral rebound and disease recurrence in many patients. This adjustment clarifies the intended meaning and addresses the ambiguity you identified. We hope this revision better aligns with the clinical context of HBV management and improves the overall clarity of the manuscript.

Define "functional cure" in the introduction.

Thank you for your suggestion to clarify the term 'functional cure.' We have revised the manuscript and instead of ”functional cure” we mention the goal of sustained viral suppression without detectable HBV DNA and loss of hepatitis B surface antigen (HBsAg) without the need for continuous therapy. This should provide greater clarity for readers and improve the overall comprehensibility of the introduction.

The sentence beginning line 92 is not clear unless one has already read the paper. Figure 1 is not well described.

Thank you for your valuable feedback regarding the clarity of this sentence and the legend of Figure 1. We have revised the text and legend to provide more context and improve the flow for readers who are unfamiliar with the specifics of the study. The revised version now clearly explains the targeted binding sites and the purpose of the bivalent binders at the beginning of the results section.

In line 235 the meaning is not clear. What is in excess? Is there free CPP in solution? Is it the charge on the CPP?

We have clarified the passage as requested.

When describing peptide-induced aggregation, Figures 5 and 6, figure 1B is never referred to. Figure 1B would work better as part of Figure 6.

We understand the reviewer’s suggestion. However, we decided to highlight and summarize the major findings and the underlying hypothesis early in the manuscript. We included Figure 1B at the beginning to allow readers to quickly grasp a core concept and outcome of our study.

We now however refer to Figure 1B and together with all the other changes hope that we have improved the clarity and quality of the manuscript.

We appreciate your constructive feedback and the opportunity to further refine the work.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Rigor in the design and application of scientific experiments is an ongoing concern in preclinical (animal) research. Because findings from these studies are often used in the design of clinical (human) studies, it is critical that the results of the preclinical studies are valid and replicable. However, several recent peer-reviewed published papers have shown that some of the research results in cardiovascular research literature may not be valid because their use of key design elements is unacceptably low. The current study is designed to expand on and replicate previous preclinical studies in nine leading scientific research journals. Cardiovascular research articles that were used for examination were obtained from a PubMed Search. These articles were carefully examined for four elements that are important in the design of animal experiments: use of both biological sexes, randomization of subjects for experimental groups, blinding of the experimenters, and estimating the proper size of samples for the experimental groups. The findings of the current study indicate that the use of these four design elements in the reported research in preclinical research is unacceptably low. Therefore, the results replicate previous studies and demonstrate once again that there is an ongoing problem in the experimental design of preclinical cardiovascular research.

Strengths:

This study selected four important design elements for study. The descriptions in the text and figures of this paper clearly demonstrate that the rate of use of all four design elements in the examined research articles was unacceptably low. The current study is important because it replicates previous studies and continues to call attention once again to serious problems in the design of preclinical studies, and the problem does not seem to lessen over time.

Weaknesses:

The current study uses both descriptive and inferential statistics extensively in describing the results. The descriptive statistics are clear and strong, demonstrating the main point of the study, that the use of these design elements is quite low, which may invalidate many of the reported studies. In addition, inferential statistical tests were used to compare the use of the four design elements against each other and to compare some of the journals. The use of inferential statistical tests appears weak because the wrong tests may have been used in some cases. However, the overall descriptive findings are very strong and make the major points of the study.

We sincerely appreciate the reviewer’s comments and detailed feedback and their recognition of the importance of this work in replicating previous studies and calling attention to the problems in preclinical study design. In response to the reviewer’s suggestions, we have recalculated our inferential statistics. In place of our previous inferential statistics, we have used an alternative correction calculation for p-values (Holm-Bonferroni corrections) and used median-based linear model analyses and nonparametric Kruskal-Wallis tests that are more appropriate for analyzing this dataset. Our overall trends in results remain the same.

Reviewer #2 (Public Review):

Summary

This study replicates a 2017 study in which the authors reviewed papers for four key elements of rigor: inclusion of sex as a biological variable, randomization of subjects, blinding outcomes, and pre-specified sample size estimation. Here they screened 298 published papers for the four elements. Over a 10 year period, rigor (defined as including any of the 4 elements) failed to improve. They could not detect any differences across the journals they surveyed, nor across models. They focused primarily on cardiovascular disease, which both helps focus the research but limits the potential generalizability to a broader range of scientific investigation. There is no reason, however, to believe rigor is any better or worse in other fields, and hence this study is a good 'snapshot' of the progress of improving rigor over time.

Strengths

The authors randomly selected papers from leading journals, e.g., PNAS). Each paper was reviewed by 2 investigators. They pulled papers over a 10-year period, 2011 to 2021, and have a good sample of time over which to look for changes. The analysis followed generally accepted guidelines for a structured review.

Weaknesses

The authors did not use the exact same journals as they did in the 2017 study. This makes comparing the results complicated. Also, they pulled papers from 2011 to 2021, and hence cannot assess the impact of their own prior paper.

The authors write "the proportion of studies including animals of both biological sexes generally increased between 2011 and 2021, though not significantly (R2= 0.0762, F(1,9)= 0.742, p= 0.411 (corrected p=8.2". This statement is not rigorous because the regression result is not statistically significant. Their data supports neither a claim of an increase nor a decrease over time. A similar problem repeats several times in the remainder of their results presentation.

I think the Introduction and the Discussion are somewhat repetitive and the wording could be reduced.

Impact and Context

Lack of reproducibility remains an enormous problem in science, plaguing both basic and translational investigations. With the increased scrutiny on rigor, and requirements at NIH and other funding agencies for more rigor and transparency, one would expect to find increasing rigor, as evidenced by authors including more study design elements (SDEs) that are recommended. This review found no such change, and this is quite disheartening. The data implies that journals-editors and reviewers-will have to increase their scrutiny and standards applied to preclinical and basic studies. This work could also serve as a call to action to investigators outside of cardiovascular science to reflect on their own experiences and when planning future projects.

We sincerely appreciate the reviewer’s insights and comments and recognition of our work contributing to the growing body of evidence on the lack of rigor in preclinical cardiovascular research study design. Regarding the weaknesses the reviewer noted; the referenced 2017 publication details a study by Ramirez et al, and was not conducted by our group. Our study aimed to expand upon their findings by using a more recent timeframe and an alternative list of highly respected cardiovascular research journals. We have now better clarified this distinction in the manuscript. We have also addressed our phrasing regarding the lack of statistical significance in the increase of the proportion of studies including animals of both sexes from 2011-2021.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Many of the methods in this study were strong or adequate. Although the descriptive statistics appear solid, there are significant problems that need to be addressed in the selection and use of inferential statistics.

(1) One of the design elements that was studied was sample size estimation. This is usually done by a power analysis. The authors should consider what group size for the examined journals is adequate for their statistics to be valid. Or they could report the power of their studies to achieve a given meaningful difference.

We thank the reviewer for this excellent observation. We unfortunately failed to conduct an a priori power analysis. Previous research (Gupta, et al. 2016) suggests that post-hoc power calculations should not be carried out after the study has been conducted. We acknowledge the importance of establishing a sufficient sample size to draw sound conclusions based on an adequate effect size, and we regret that we did not carry out the appropriate estimations. We are very appreciative of the reviewer’s suggestions and aim to implement such an appropriate study design element in future studies.

Gupta KK, Attri JP, Singh A, Kaur H, Kaur G. Basic concepts for sample size calculation: Critical step for any clinical trials!. Saudi J Anaesth. 2016;10(3):328-331. doi:10.4103/1658-354X.174918

(2) A Bonferroni correction was used extensively. Because of its use, the corrected p values often appear much too high. The Bonferroni test becomes much too conservative for more than 3 or 4 tests. I suggest using a different test for multiple comparisons.

We thank the reviewer for their insightful suggestion. We have updated all p-values to reflect a Holm-Bonferroni correction instead. All p-values have been corrected and updated.

(3) The use of the chi-square test for categorical data is appropriate. However, the t-test and multiple regression tests are designed for continuous variables. Here, it appears that they were used for the nominal variables (Table 1). For these nominal data, other nonparametric tests should be used.

We thank the reviewer for this valuable insight. We have updated our statistical analysis methods and now use nonparametric Kruskal-Wallis tests to analyze differences in SDE reporting across journals, instead of chi-square test. Our reported p-values have been adjusted accordingly.

(4) It is not clear exactly when each test is used. The stats section in Methods should better delineate when each test is used. In addition, it would be helpful to include the test used in the figure legends.

We thank the reviewer for bringing up this important point. We have now updated the methods section to better delineate which tests were used, and also included the specific tests in the figure legends.

(5) You will need to rewrite some sections of the text to reflect the changes due to changing your use of statistics.

We have rewritten the sections of the text to reflect the changes in our use of statistics.

Here are a few comments on the presentation.

(1) Some of the figure legends are almost impossible to read. They are too congested.

We thank the reviewer for pointing this out. We have edited the figure legends to make them more readable. We will also attach a pdf with the graphs to allow for easier formatting.

(2) Also, is it possible to drop some of the panels in Figure 1?

The panels in figure 1 have been rearranged to make them more readable. We believe that each panel provides valuable visual summaries of our data, that will aid readers in understanding our results.

(3) It is not mandatory that values of y-axis on the graphs go up 100% (Figs 2 and 3). Using a maximum value of 100% clumps the lines visually. I suggest a max value on the y-axis of the graph of 50% or 60%. That will spread the lines better visually so differences can better be seen.

We thank the reviewer for considering the experience of our paper’s readers. The y-axes of Figures 2 and 3 have been truncated to 50%. The trend lines in each Figure now appear more separated and differences can better be seen.

Reviewer #2 (Recommendations For The Authors):

The authors did not use the exact same journals as they did in the 2017 study. This makes comparing the results complicated. Also, they pulled papers from 2011 to 2021, and hence cannot assess the impact of their own prior paper.

We appreciate the reviewer’s concern in maintaining consistency with the paper published by Ramirez, et al. in 2017. To clarify, our efforts focused on providing a replication study that expanded upon the original Ramirez publication - which we have no affiliation with. For our study, we used different academic journals than those used by Ramirez, et al, and also a different time-frame. We have updated the language in the manuscript to better-clarify the purpose and parameters of our study relative to the previous, unaffiliated, study.

The authors write "the proportion of studies including animals of both biological sexes generally increased between 2011 and 2021, though not significantly (R2= 0.0762, F(1,9)= 0.742, p= 0.411 (corrected p=8.2". This statement is not rigorous because the regression result is not statistically significant. Their data supports neither a claim of an increase nor a decrease over time. A similar problem repeats several times in the remainder of their results presentation.

Thank you for bringing this information to our attention. We agree with the concern regarding the statement, “the proportion of studies including animals of both biological sexes generally increased between 2011 and 2021, though not significantly (R2= 0.0762, F(1,9)= 0.742, p= 0.411 (corrected p=8.2.” We have rephrased the statement. Our updated Holm-Bonferroni corrected p-value is now noted in this more appropriately worded description of our results. Lastly, we have addressed the wording and redundancy seen in both the introduction and discussion and have made both more concise.

I think the Introduction and the Discussion are somewhat repetitive and the wording could be reduced.

We thank the reviewer for bringing this to our attention. We have addressed the redundancy across the Introduction and the Discussion. We have also altered the wording to reflect a more concise explanation of our study.

The 'trends' are not statistically significant. A non-significant trend does not exist and no claim of a 'trend' is justified by the data.

We thank the reviewer for this observation. We have updated the phrasing of ‘trends’ in all areas of the manuscript.

Author response:

Public Reviews: 

Reviewer #1 (Public review): 

Summary: 

Authors of this article have previously shown the involvement of the transcription factor Zinc finger homeobox-3 (ZFHX3) in the function of the circadian clock and the development/differentiation of the central circadian clock in the suprachiasmatic nucleus (SCN) of the hypothalamus. Here, they show that ZFHX3 plays a critical role in the transcriptional regulation of numerous genes in the SCN. Using inducible knockout mice, they further demonstrate that the deletion Of Zfhx3 induces a phase advance of the circadian clock, both at the molecular and behavioral levels. 

Strengths: 

- Inducible deletion of Zfhx3 in adults 

- Behavioral analysis 

- Properly designed and analyzed ChIP-Seq and RNA-Seq supporting the conclusion of the behavioral analysis 

Weaknesses: 

- Further characterization of the disruption of the activity of the SCN is required. 

(1) We thank the reviewer for their valuable inputs. Indeed, a comprehensive behavioral assessment of mice of this genotype was executed in Wilcox et al. ;2017 study. In Wilcox et al.; 2017, Figure 4, 6-h phase advance (jetlag) clearly showed faster reentrainment in ZFHX3-KO mice when compared to the controls.

- The description of the controls needs some clarification. 

(2) We agree with the reviewer and will modify the text to clearly describe the controls wherever mentioned.

Reviewer #2 (Public review): 

Summary: 

ZFHX3 is a transcription factor expressed in discrete populations of adult SCN and was shown by the authors previously to control circadian behavioral rhythms using either a dominant missense mutation in Zfhx3 or conditional null Zfhx3 mutation using the Ubc-Cre line (Wilcox et al., 2017). In the current manuscript, the authors assess the function of ZFHX3 by using a multi-omics approach including ChIPSeq in wildtype SCNs and RNAseq of SCN tissues from both wildtype and conditional null mice. RNAseq analysis showed a loss of oscillation in Bmal1 and changes in expression levels of other clock output genes. Moreover, a phase advance gene transcriptional profile using the TimeTeller algorithm suggests the presence of a regulatory network that could underlie the observed pattern of advanced activity onset in locomotor behavior in knockout mice. 

In figure1, the authors identified the ZFHX3 bound sites using ChIPseq and compared the loci with other histone marks that occur at promoters, TSS, enhancers and intergenic regions. And the analysis broadly points to a role for ZFHX3 in transcriptional regulation. The vast majority of nearly 40000 peaks overlapped H3K4me3 and K27ac marks, active promoters which also included genes falling under the GO category circadian rhythms. However, no significant differential ZFHX3 bound peaks were detected between ZT3 and ZT15. In these experiments, it is not clear if and how the different ChIP samples (ZFHX3 and histone PTM ChIPs) were normalized/downsampled for analysis. Moreover, it seems that ZFHX3 binding or recruitment has little to do with whether the promoters are active.

(3) We thank the reviewer for their valuable comment. Different ChIP samples. (ZFHX3 and histone PTM ChIPs) were treated in the same manner from preprocessing (quality control by FastQC, Trimming, Alignment to mm10 genome and Peak calling) using MACS2 as mentioned in Methods. The data was normalized using bamCoverage tools and bigwig files were generated for visual inspection using USCS Genome Browser. These additional details will be added to Methods. Finally, BEDTools was employed to study overlapping peaks between ZFHX3 and histone PTMs.

We agree that, alone, the current data does not make any claim for ZFHX3 being crucial for promoter to be active. Our data clearly suggests that a vast majority of ZFHX3 genomic binding in the SCN was observed at active promoters marked by H3K4me3 and H3K27ac and potentially regulating gene transcription. 

Based on a enrichment of ARNT domains next to K4Me3 and K27ac PTMs, the authors propose a model where the core-clock TFs and ZFHX3 interact. If the authors develop other assays beyond just predictions to test their hypothesis, it would strengthen the argument for role in circadian transcription in the SCN. It would be important in this context to perform a ChIP-seq experiment for ZFHX3 in the knockout animal (described from Figure 2 onwards) to eliminate the possibility of non-specific enrichment of signal from "open chromatin'. Alternatively, a ChIPseq analysis for BMAL1 or CLOCK could also strengthen this argument to identify the sites co-occupied by ZFHX3 and core-clock TFs. 

(4a) We agree that follow-up experiments such as BMAL1/CLOCK ChIPseq suggested by the reviewer will further confirm the proposed interaction of ZFHX3 with core-clock TFs. However, this is beyond the scope of the current study. 

(4b) Again, conducting complementary ChIPseq in ZFHX3 knockout mice will strengthen the findings, but conducting TF-ChIPseq in a specific brain tissue such as the SCN (unlike peripheral tissues such as liver) does not only warrant use of multiple animals per sample but is also technically challenging and time-consuming to ensure specificity of the sample. For these reasons, datasets such as ours on the SCN are uncommon. Furthermore, in this particular context, we are certain that, based on current dataset, the ZFHX3 peaks (narrow) we observed were well-defined and met the specified statistical criteria mitigating any risk of signal arising from non-specific enrichment from open-chromatin regions. 

Next, they compared locomotor activity rhythms in floxed mice with or without tamoxifen treatment. As reported before in Wilcox et al 2017, the loss of ZFHX3 led to a shorter free running period and reduced amplitude and earlier onset of activity. Overall, the behavioral data in Figure 2 and supplementary figure 2 has been reported before and are not novel.

(5) We recognise that a detailed circadian behavior assessment from adult mice lacking ZFHX3 has been conducted previously by Nolan lab (Wilcox et al; 2017). In the current study, however, we used a separate cohort of mice, to focus on the behavioral advance noted in 24-h LD cycle and generate a more refined assessment. Importantly, these mice were also used for transcriptomic studies as detailed in Figure 3, which we consider to be a positive feature of our experimental design: behavior and molecular analyses were performed on the same animals. 

Next, the authors performed RNAseq at 4hr intervals on wildtype and knockout animals maintained in light/dark cycles to determine the impact of loss of ZFHX3. Overall transcriptomic analysis indicated changes in gene expression in nearly 36% of expressed genes, with nearly half being upregulated while an equal fraction was downregulated. Pathways affected included mostly neureopeptide neurotransmitter pathways. Surprisingly, there was no correlation between the direction in change in expression and TF binding since nearly all the sites were bound by ZFHX3 and the active histone PTMs. The ChIP-seq experiment for ZFHX3 in the UBC-Cre+Tam mice again could help resolve the real targets of ZFHX3 and the transcriptional state in knockout animals. 

(6) We agree with the reviewer that most of the differentially expressed genes showed ZFHX3 binding at active promoter sites. That said, the current dataset is in line with recently published ZFHX3-CHIPseq data by Baca et al; 2024 [PMID: 38412861] in human neural stem cells and Hu et al; 2024 [PMID: 38871709] in human prostate cancer cells that clearly suggests ZFHX3 binds at active promoters and act as chromatin remodellers/mediators that modulate gene transcription depending on the accessory TFs assembled at target genes. Therefore, finding no correlation in the direction of change in expression is not striking.  

To determine the fraction of rhythmic transcripts, Using dryR, the authors categorise the rhythmic transcriptome into modules that include genes that lose rhythmicity in the KO, gain rhythmicity in the KO or remain unaffected or partially affected. The analysis indicates that a large fraction of the rhythmic transcriptome is affected in the KO model. However, among core-clock genes only Bmal1 expression is affected showing a complete loss of rhythm. The authors state a decrease in Clock mRNA expression (line 294) but the panel figure 4A does not show this data. Instead it depicts the loss in Avp expression - {{ misstated in line 321 ( we noted severe loss in 24-h rhythm for crucial SCN neuropeptides such as Avp (Fig. 3a).}} 

(7a) Indeed, among the core-clock genes rhythmic expression is lost after ZFHX3 knockout only for Bmal1. However, given the mice were rhythmic (as assessed by wheel-running activity) in LD conditions, the observed 24-h gene expression rhythm in the majority of core-clock genes (Pers and Crys)  is consistent with behavior data,  and suggests towards a molecular clock with plausible scenarios as explained at line 439. That said, the unique and well-defined changes (amplitude and phase) observed as demonstrated in Figure 5 highlights a model in which ZFHX3 exerts differential control, for example in case of Per2 noted advance in molecular rhythm (~2-h), but no such change in Cry, presents an opportunity to delineate further the regulation of TTFL genes. 

(7b) Line 294 states- loss of Bmal1 rhythm and reduction in Clock mRNA . Figure 4a is in support of former. We shall revise the text for clarity. 

(7c) As rightly pointed out by the reviewer, line 321 is referring to loss of Avp expression and we shall correct the typo by replacing “Figure 3a to 4a”. Thank you.  

However, core-clock genes such as Pers and Crys show minor or no change in expression patterns while Per2 and Per3 show a ~2hr phase advance. While these could only weakly account for the behavioral phase advance, the authors used TimeTeller to assess circadian phase in wildtype and ZFHX3 deficient mice. This approach clearly indicated that while the clock is not disrupted in the knockout animals, the phase advance can be correctly predicted from a network of gene expression patterns. 

Strengths: 

The authors use a multiomic strategy in order to reveal the role of the ZFHX3 transcription factor with a combination of TF and histone PTM ChIPseq, time-resolved RNAseq from wildtype and knockout mice and modeling the transcriptomic data using TimeTeller. The RNAseq experiments are nicely controlled and the analysis of the data indicates a clear impact on gene-expression levels in the knockout mice and the presence of a regulatory network that could underlie the advanced activity onset behavior. 

Weaknesses: 

It is not clear whether ZFHX3 has a direct role in any of the processes and seems to be a general factor that marks H3K4me3 and K27ac marked chromatin. Why it would specifically impact the core-clock TTFL clock gene expression or indeed daily gene expression rhythms is not clear either. Details for treatment of different ChIP samples (ZFHX3 and histone PTM ChIPs) on data normalization for analysis are needed. The loss of complete rhythmicity of Avp and other neuropeptides or indeed other TFs could instead account for the transcriptional deregulation noted in the knockout mice.

(8) We thank the reviewer for the constructive feedback.  The current data suggests ZFHX3 acts as a mediating factor, occupying targeted active promoter sites and regulating gene expression by partnering with other key TFs in the SCN. Please see point 7 for clarification. The binding sites of ZFHX3 clearly showed enrichment for E-box(CACGTG) motif bound by CLOCK/BMAL1 along with binding sites for key SCN-specific TFs such as RFX (please see Supplementary Fig1). Our data thereby shows that it affects both core-clock and clock output genes (at varied levels) thereby exercising a pervasive control over the SCN transcriptome. 

For treatment of ChIP samples please see point 4. We followed ENCODE guidelines strictly.

Author response:

We sincerely appreciate the insightful feedback and constructive suggestions provided by the reviewers. We thank reviewers for their valuable support in improving our manuscript.

In response to the public reviews raised by reviewers, we plan to make the following revisions:

(1) Most metadata have been rectified through collaborative review of original literature sources rather than automated processes. We intend to incorporate a detailed discussion on this matter in the revised manuscript.

(2) We will include a corrections table for entries to provide clarity and transparency regarding any amendments made.

(3) Additional references will be included to elucidate the rationale behind the selection of interact residues definition methods and the set threshold. The threshold is not fixed. In fact, we utilized a 5Å cutoff in current version, listing all residues with distances less than 5Å alongside the corresponding distances. The researchers could screen the residues through distance according to their custom cutoff. To offer researchers flexibility, we will also provide interact residues and corresponding distances with higher cutoffs for custom screening. These enhancements will be detailed in the revised manuscript.

(4)We acknowledge the importance of expanding the database to include a wider range of experimental information and complexes with diverse target sizes. Regrettably, immediate updates to address these limitations are not feasible at this time. Thus, we will give an illustration in the later detail response to reviewers.

Author response:

We very much appreciate the reviewers’ and editor’s overall positive responses to our manuscript "Evolution of lateralized gustation in nematodes".

Reviewer #1:

The mechanism of lsy-6-independent establishment of ASEL/R asymmetry in P. pacificus remains uncharacterized. 

We thank the reviewer for recognizing the novel contributions of our work in revealing the existence of alternative pathways for establishing neuronal lateral asymmetry despite the absence of the lsy-6 miRNA in a divergent nematode species. We are certainly encouraged now to search for genetic factors that abolish asymmetric expression of gcy-22.3.

Reviewer #2:

(1) The authors observe only weak attraction of C. elegans to NaCl. These results raise the question of whether the weak attraction observed is the result of the prior salt environment experienced by the worms. More generally, this study does not address how prior exposure to gustatory cues shapes gustatory responses in P. pacificus. Is salt sensing in P. pacificus subject to the same type of experience-dependent modulation as salt sensing in C. elegans? 

Proposed revision: For our live imaging experiments, we had not considered if starved P. pacificus animals in the presence of salt may exhibit responses different from a well-fed state. However, we will venture to address the effect of experience-dependent modulation in P. pacificus chemotaxis behavior using NH4Cl.

(2) A key finding of this paper is that the Ppa-CHE-1 transcription factor is expressed in the Ppa-AFD neurons as well as the Ppa-ASE neurons, despite the fact that Ce-CHE-1 is expressed specifically in Ce-ASE. However, additional verification of Ppa-AFD neuron identity is required. Based on the image shown in the manuscript, it is difficult to unequivocally identify the second pair of CHE-1-positive head neurons as the Ppa-AFD neurons. Ppa-AFD neuron identity could be verified by confocal imaging of the CHE-1-positive neurons, co-expression of Ppa-che-1p::GFP with a likely AFD reporter, thermotaxis assays with Ppa-che-1 mutants, and/or calcium imaging from the putative Ppa-AFD neurons. 

We are happy to provide additional evidence to confirm Ppa-AFD neuron identity since the expression of Ppa-CHE-1 in non-ASE amphid neurons is one of the major differences between the two nematode specie

Proposed revision: We will provide results showing the Ppa-ttx-1::gfp reporter expression in finger-like neuronal endings and Ppa-_TTX-1::ALFA co-localization with _Ppa-che-1::gfp in the putative AFD neurons and discuss the possible role of Ppa-CHE-1 in AFD differentiation. We attempted to obtain AFD markers using several reporter strains. However, Ppa-gcy-8.1p::gfp(csuEx101) (PPA24212) showed no expression while Ppa-gcy-8.2p::gfp(csuEx100) (PPA41407) showed only expression in pharyngeal cells.

(4) The authors show that silencing Ppa-ASE has a dramatic effect on salt chemotaxis behavior. However, these data lack control with histamine-treated wild-type animals, with the result that the phenotype of Ppa-ASE-silenced animals could result from exposure to histamine dihydrochloride. This is an especially important control in the context of salt sensing, where histamine dihydrochloride could alter behavioral responses to other salts. 

Proposed revision: Thank you for noticing this oversight. The control for histamine-treated wild-type worms in the Ppa-ASE silencing experiments was inadvertently left out in the original submission. Because the HisCl transgene is on a randomly segregating transgene array, we have scored worms with and without the transgene expressing the co-injection marker (Ppa-egl-20p::rfp expressed in the tail) to show that the presence of the transgene is necessary for the knockdown of NH4Br attraction.

We will also address most of the other more minor suggestions and clarifications sought by the reviewers.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

In this paper Kawasaki et al describe a regulatory role for the PIWI/piRNA pathway in rRNA regulation in Zebrafish. This regulatory role was uncovered through a screen for gonadogenesis defective mutants, which identified a mutation in the meioc gene, a coiled-coil germ granule protein. Loss of this gene leads to redistribution of Piwil1 from germ granules to the nucleolus, resulting in silencing of rRNA transcription.

Strengths:

Most of the experimental data provided in this paper is compelling. It is clear that in the absence of meioc, PiwiL1 translocates in to the nucleolus and results in down regulation of rRNA transcription. the genetic compensation of meioc mutant phenotypes (both organismal and molecular) through reduction in PiwiL1 levels are evidence for a direct role for PiwiL1 in mediating the phenotypes of meioc mutant.

Weaknesses:

Questions remain on the mechanistic details by which PiwiL1 mediated rRNA down regulation, and whether this is a function of Piwi in an unperturbed/wildtype setting. There is certainly some evidence provided in support of the natural function for piwi in regulating rRNA transcription (figure 5A+5B). However, the de-enrichment of H3K9me3 in the heterozygous (Figure 6F) is very modest and in my opinion not convincingly different relative to the control provided. It is certainly possible that PiwiL1 is regulating levels through cleavage of nascent transcripts. Another aspect I found confounding here is the reduction in rRNA small RNAs in the meioc mutant; I would have assumed that the interaction of PiwiL1 with the rRNA is mediated through small RNAs but the reduction in numbers do not support this model. But perhaps it is simply a redistribution of small RNAs that is occurring. Finally, the ability to reduce PiwiL1 in the nucleolus through polI inhibition with actD and BMH-21 is surprising. What drives the accumulation of PiwiL1 in the nucleolus then if in the meioc mutant there is less transcription anyway?

Despite the weaknesses outlined, overall I find this paper to be solid and valuable, providing evidence for a consistent link between PIWI systems and ribosomal biogenesis. Their results are likely to be of interest to people in the community, and provide tools for further elucidating the reasons for this link.

The amount of cytoplasmic rRNA in piwi+/- was increased by 26% on average (figure 5A+5B), the amount of ChiP-qPCR of H3K9 was decreased by about 26% (Figure 6F), and ChiP-qPCR of Piwil1 was decreased by 35% (Figure 6G), so we don't think there is a big discrepancy. On the other hand, the amount of ChiP-qPCR of H3K9 in meioc^mo/mo was increased by about 130% (Figure 6F), while ChiP-qPCR of Piwil1 was increased by 50%, so there may be a mechanism for H3K9 regulation of Meioc that is not mediated by Piwil1. As for what drives the accumulation of Piwil1 in the nucleolus, although we have found that Piwil1 has affinity for rRNA (Fig. 6A), we do not know what recruits it. Significant increases in the 18-35nt small RNA of 18S, 28S rRNA and R2 were not detected in meioc^mo/mo testes enriched for 1-8 cell spermatogonia, compared with meioc^+/mo testes. The nucleolar localization of Piwil1 has revealed in this study, which will be a new topic for future research.

Reviewer #2 (Public review):

Summary:

In this study, the authors report that Meioc is required to upregulate rRNA transcription and promote differentiation of spermatogonial stem cells in zebrafish. The authors show that upregulated protein synthesis is required to support spermatogonial stem cells' differentiation into multi-celled cysts of spermatogonia. Coiled coil protein Meioc is required for this upregulated protein synthesis and for increasing rRNA transcription, such that the Meioc knockout accumulates 1-2 cell spermatogonia and fails to produce cysts with more than 8 spermatogonia. The Meioc knockout exhibits continued transcriptional repression of rDNA. Meioc interacts with and sequesters Piwil1 to the cytoplasm. Loss of Meioc increases Piwil1 localization to the nucleolus, where Piwil1 interacts with transcriptional silencers that repress rRNA transcription.

Strengths:

This is a fundamental study that expands our understanding of how ribosome biogenesis contributes to differentiation and demonstrates that zebrafish Meioc plays a role in this process during spermatogenesis. This work also expands our evolutionary understanding of Meioc and Ythdc2's molecular roles in germline differentiation. In mouse, the Meioc knockout phenocopies the Ythdc2 knockout, and studies thus far have indicated that Meioc and Ythdc2 act together to regulate germline differentiation. Here, in zebrafish, Meioc has acquired a Ythdc2-independent function. This study also identifies a new role for Piwil1 in directing transcriptional silencing of rDNA.

Weaknesses:

There are limited details on the stem cell-enriched hyperplastic testes used as a tool for mass spec experiments, and additional information is needed to fully evaluate the mass spec results. What mutation do these testes carry? Does this protein interact with Meioc in the wildtype testes? How could this mutation affect the results from the Meioc immunoprecipitation?

Stem cell-enriched hyperplastic testes came from wild-type adult sox17::GFP transgenic zebrafish. Sperm were found in these hyperplastic testes, and when stem cells were transplanted, they self-renewed and differentiated into sperm. It is not known if the hyperplasias develop due to a genetic variant in the line. We will add the following comment.

“The stem cell-enriched hyperplastic testes, which are occasionally found in adult wildtype zebrafish, contain cells at all stages of spermatogenesis. Hyperplasia-derived SSCs self-renewed and differentiated in the same manner as SSCs of normal testes in transplants of aggregates mixed with normal testicular cells.”

Reviewer #3 (Public review):

Summary:

The paper describes the molecular pathway to regulate germ cell differentiation in zebrafish through ribosomal RNA biogenesis. Meioc sequesters Piwil1, a Piwi homolog, which suppresses the transcription of the 45S pre-rDNA by the formation of heterochromatin, to the perinuclear bodies. The key results are solid and useful to researchers in the field of germ cell/meiosis as well as RNA biosynthesis and chromatin.

Strengths:

The authors nicely provided the molecular evidence on the antagonism of Meioc to Piwil1 in the rRNA synthesis, which supported by the genetic evidence that the inability of the meioc mutant to enter meiosis is suppressed by the piwil1 heterozygosity.

Weaknesses:

(1) Although the paper provides very convincing evidence for the authors' claim, the scientific contents are poorly written and incorrectly described. As a result, it is hard to read the text. Checking by scientific experts would be highly recommended. For example, on line 38, "the global translation activity is generally [inhibited]", is incorrect and, rather, a sentence like "the activity is lowered relative to other cells" is more appropriate here. See minor points for more examples.

Thank you for pointing that out. I will correct the parts pointed out.

(2) In some figures, it is hard for readers outside of zebrafish meiosis to evaluate the results without more explanation and drawing.

We will refine Figure 1A and add schema of spermatogonia culture system in a supplemental figure. 

(3) Figure 1E, F, cycloheximide experiments: Please mention the toxicity of the concentration of the drug in cell proliferation and viability.

When testicular tissue culture was performed at 0.1, 1, 10, 100, 250, and 500mM, abnormal strong OP-puro signals including nuclei were found in cells at 10mM or more. We will add the results in the Supplemental Material. In addition, at 1mM, growth was perturbed in fast-growing 32≤-cell cysts of spermatogonia, but not in 1-4-cell spermatogonia, as described in L122-125.

Author response:

Reviewer #1 (Public review):

Summary:

By way of background, the Jiang lab has previously shown that loss of the type II BMP receptor Punt (Put) from intestinal progenitors (ISCs and EBs) caused them to differentiate into EBs, with a concomitant loss of ISCs (Tian and Jiang, eLife 2014). The mechanism by which this occurs was activation of Notch in Put-deficient progenitors. How Notch was upregulated in Put-deficient ISCs was not established in this prior work. In the current study, the authors test whether a very low level of Dl was responsible. But co-depletion of Dl and Put led to a similar phenotype as depletion of Put alone. This result suggested that Dl was not the mechanism. They next investigate genetic interactions between BMP signaling and Numb, an inhibitor of Notch signaling. Prior work from Bardin, Schweisguth and other labs has shown that Numb is not required for ISC self-renewal. However the authors wanted to know whether loss of both the BMP signal transducer Mad and Numb would cause ISC loss. This result was observed for RNAi depletion from progenitors and for mad, numb double mutant clones. Of note, ISC loss was observed in 40% of mad, numb double mutant clones, whereas 60% of these clones had an ISC. They then employed a two-color tracing system called RGT to look at the outcome of ISC divisions (asymmetric (ISC/EB) or symmetric (ISC/ISC or EB/EB)). Control clones had 69%, 15% and 16%, respectively, whereas mad, numb double mutant clones had much lower ISC/ISC (11%) and much higher EB/EB (37%). They conclude that loss of Numb in moderate BMP loss of function mutants increased symmetric differentiation which lead caused ISC loss. They also reported that Numb¹⁵ and numb⁴ clones had a moderate but significant increase in ISC-lacking clones compared to control clones, supporting the model that Numb plays a role in ISC maintenance. Finally, they investigated the relevance of these observation during regeneration. After bleomycin treatment, there was a significant increase in ISC-lacking clones and a significant decrease in clone size in numb⁴ and Numb¹⁵ clones compared to control clones. Because bleomycin treatment has been shown to cause variation in BMP ligand production, the authors interpret the numb clone under bleomycin results as demonstrating an essential role of Numb in ISC maintenance during regeneration.

Strengths:

(i) Most data is quantified with statistical analysis

(ii) Experiments have appropriate controls and large numbers of samples

(iii) Results demonstrate an important role of Numb in maintaining ISC number during regeneration and a genetic interaction between Mad and Numb during homeostasis.

Weaknesses:

(i) No quantification for Fig. 1

Thank you for your suggestion. Quantification of Fig.1 will be added.  

(ii) The premise is a bit unclear. Under homeostasis, strong loss of BMP (Put) leads to loss of ISCs, presumably regardless of Numb level (which was not tested). But moderate loss of BMP (Mad) does not show ISC loss unless Numb is also reduced. I am confused as to why numb does not play a role in Put mutants. Did the authors test whether concomitant loss of Put and Numb leads to even more ISC loss than Put-mutation alone.

Thank you for your comment. We have tested the genetic interaction between punt and numb using punt RNAi and numb RNAi driven by esg^ts. According to the results in this study and our previously published data, punt mutant clone or esg^ts> punt RNAi could induce a rapid loss of ISC (whin 8 days). We did not observe further enhancement of stem cell loss phenotype caused punt RNAi by numb RNAi.

(iii) I think that the use of the word "essential" is a bit strong here. Numb plays an important role but in either during homeostasis or regeneration, most numb clones or mad, numb double mutant clones still have ISCs. Therefore, I think that the authors should temper their language about the role of Numb in ISC maintenance.

Thank you. We will revise the language.

Reviewer #2 (Public review):

Summary:

This work assesses the genetic interaction between the Bmp signaling pathway and the factor Numb, which can inhibit Notch signalling. It follows up on the previous studies of the group (Tian, Elife, 2014; Tian, PNAS, 2014) regarding BMP signaling in controlling stem cell fate decision as well as on the work of another group (Sallé, EMBO, 2017) that investigated the function of Numb on enteroendocrine fate in the midgut. This is an important study providing evidence of a Numb-mediated back up mechanism for stem cell maintenance.

Strengths:

(1) Experiments are consistent with these previous publications while also extending our understanding of how Numb functions in the ISC.

(2) Provides an interesting model of a "back up" protection mechanism for ISC maintenance.

Weaknesses:

(1) Aspects of the experiments could be better controlled or annotated:

(a) As they "randomly chose" the regions analyzed, it would be better to have all from a defined region (R4 or R2, for example) or to at least note the region as there are important regional differences for some aspects of midgut biology.

Thank you. Since we mainly focus on region 4, we have added the clarification in the manuscript.

(b) It is not clear to me why MARCM clones were induced and then flies grown at 18{degree sign}C? It would help to explain why they used this unconventional protocol.

To avoid spontaneous clone, we kept the flies under 18°C.

(2) There are technical limitations with trying to conclude from double-knockdown experiments in the ISC lineage, such as those in Figure 1 where Dl and put are both being knocked down: depending on how fast both proteins are depleted, it may be that only one of them (put, for example) is inactivated and affects the fate decision prior to the other one (Dl) being depleted. Therefore, it is difficult to definitively conclude that the decision is independent of Dl ligand.

In our hand, Dl-RNAi is very effective and exhibited loss of N pathway activity as determined by the N pathway reporter Su(H)-lacZ (Fig. 1D). Therefore, the ectopic Su(H)-lacZ expression in Punt Dl double RNAi (fig. 1E) is unlikely due to residual Dl expression. Nevertheless, we will change the statement “BMP signaling blocks ligand-independent N activity” to” Loss of BMP signaling results in ectopic N pathway activity even when Dl is depleted”

(3) Additional quantification of many phenotypes would be desired.

(a) It would be useful to see esg-GFP cells/total cells and not just field as the density might change (2E for example).

We focused on R4 region for quantification where the cell density did not exhibit apparent change in different experimental groups. In addition, we have examined many guts for quantification. It is unlikely that the difference in the esg+ cell number is caused by change in cell density.

(b) Similarly, for 2F and 2G, it would be nice to see the % of ISC/ total cell and EB/total cell and not only per esgGFP+ cell.

Unfortunately, we didn’t have the suggested quantification. However, we believe that quantification of the percentage of ISC or EB among all progenitor cells, as we did here, provides a faithful measurement of the self-renewal status of each experimental group.

(c) Fig1: There is no quantification - specifically it would be interesting to know how many esg+ are su(H)lacZ positive in Put- Dl- condition compared to WT or Put- alone. What is the n?

Quantification will be added.

(d) Fig2: Pros + cells are not seen in the image? Are they all DllacZ+?

Anti-Pros and anti-E(spl)mβ-CD2 were stained in the same channel (magenta).  Pros+ is nuclear dot-like staining, while CD2 outlined the cell membrane of EB cell.

(e) Fig3: it would be nice to have the size clone quantification instead of the distribution between groups of 2 cell 3 cells 4 cell clones.

Thank you for your suggestion. In this study, we have quantified the clone size of each clone and calculated the average size for each genotype. However, the frequency distribution analysis was chosen because it highlights the significance of the clone size differences among genotypes.

(f) How many times were experiments performed?

All experiments are performed 3 times.

(4) The authors do not comment on the reduction of clone size in DSS treatment in Figure 6K. How do they interpret this? Does it conflict with their model of Bleo vs DSS?

numb⁴ clone containing guts treated with DSS exhibited a slight reduction of clone size, evident by a higher percentage of 2-cell clones and lower percentage of > 8 cell clones. This reduction is less significant in guts containing numb¹⁵ clones. However, the percentage of Dl⁺-containing clones is similar between DSS and mock-treated guts. It is possible that ISC proliferation is lightly reduced due to numb⁴ mutation or the genetic background.

(5) There is probably a mistake on sentence line 314 -316 "Indeed, previous studies indicate that endogenous Numb was not undetectable by Numb antibodies that could detect Numb expression in the nervous system".

We will make a correction of the sentence.

Reviewer #3 (Public review):

Summary:

The authors provide an in-depth analysis of the function of Numb in adult Drosophila midgut. Based on RNAi combinations and double mutant clonal analyses, they propose that Numb has a function in inhibiting Notch pathway to maintain intestinal stem cells, and is a backup mechanism with BMP pathway in maintaining midgut stem cell mediated homeostasis.

Strengths:

Overall, this is a carefully constructed series of experiments, and the results and statistical analyses provides believable evidence that Numb has a role, albeit weak compared to other pathways, in sustaining ISC and in promoting regeneration especially after damage by bleomycin, which may damage enterocytes and therefore disrupt BMP pathway more. The results overall support their claim.

The data are highly coherent, and support a genetic function of Numb, in collaborating with BMP signaling, to maintain the number and proliferative function of ISCs in adult midguts. The authors used appropriate and sophisticated genetic tools of double RNAi, mutant clonal analysis and dual marker stem cell tracing approaches to ensure the results are reproducible and consistent. The statistical analyses provide confidence that the phenotypic changes are reliable albeit weaker than many other mutants previously studied.

Weaknesses:

In the absence of Numb itself, the midgut has a weak reduction of ISC number (Fig. 3 and 5), as well as weak albeit not statistically significant reduction of ISC clone size/proliferation. I think the authors published similar experiments with BMP pathway mutants. The mad^1-2 allele used here as stated below may not be very representative of other BMP pathway mutants. Therefore, it could be beneficial to compare the number of ISC number and clone sizes between other BMP experiments to provide the readers with a clearer picture of how these two pathways individually contribute (stronger/weaker effects) to the ISC number and gut homeostasis.

Thank you for your comment. We have tested other components of BMP pathway in our previously study (Tian et al., 2014). More complete loss of BMP signaling (for example, Put clones, Put RNAi, Tkv/Sax double mutant clones or double RNAi) resulted in ISC loss regardless of the status of numb, suggesting a more predominant role of BMP signaling in ISC self-renewal compared with Numb. We speculate that the weak stem cell loss phenotype associated with numb mutant clones in otherwise wild type background could be due to fluctuation of BMP signaling in homeostatic guts.

The main weakness of this manuscript is the analysis of the BMP pathway components, especially the mad^1-2 allele. The mad RNAi and mad^1-2 alleles (P insertion) are supposed to be weak alleles and that might be suitable for genetic enhancement assays here together with numb RNAi. However, the mad^1-2 allele, and sometimes the mad RNAi, showed weakly increased ISC clone size. This is kind of counter-intuitive that they should have a similar ISC loss and ISC clone size reduction.

We used mad^1-2 and mad RNAi here to test the genetic interaction with numb because our previous studies showed that partial loss of BMP signaling under these conditions did not cause stem cell loss, therefore, may provide a sensitized background to determine the role of Numb in ISC self-renewal. The increased proliferation of ISC/ clone size in associated with mad^1-2 and mad RNAi is due to the fact that the reduction of BMP signaling in either EC or EB will non-autonomously induce stem cell proliferation. However, in mad numb double mutant clones, there was a reduction in clone size, which correlated with loss of ISC.

A much stronger phenotype was observed when numb mutants were subject to treatment of tissue damaging agents Bleomycin, which causes damage in different ways than DSS. Bleomycin as previously shown to be causing mainly enterocyte damage,  and therefore disrupt BMP signaling from ECs more likely. Therefore, this treatment together with loss of numb led to a highly significant reduction of ISC in clones and reduction of clone size/proliferation. One improvement is that it is not clear whether the authors discussed the nature of the two numb mutant alleles used in this study and the comparison to the strength of the RNAi allele. Because the phenotypes are weak and more variable, the use of specific reagents is important.

Numb¹⁵ is a null allele, and the nature of numb⁴ has not been elucidated. According to Domingos, P.M. et al., numb¹⁵ induced a more severe phenotype than numb⁴ did. Consistently, we also found that more numb¹⁵ mutant clones were void of stem cell than numb⁴.

Furthermore, the use of possible activating alleles of either or both pathways to test genetic enhancement or synergistic activation will provide strong support for the claims.

Activation of BMP (Tkv^CA) also induced stem cell tumor (Tian et al., 2014), which is not suitable for synergistic activation experiment.

Author response:

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

eLife Assessment

This study offers a useful treatment of how the population of excitatory and inhibitory neurons integrates principles of energy efficiency in their coding strategies. The analysis provides a comprehensive characterisation of the model, highlighting the structured connectivity between excitatory and inhibitory neurons. However, the manuscript provides an incomplete motivation for parameter choices. Furthermore, the work is insufficiently contextualized within the literature, and some of the findings appear overlapping and incremental given previous work.

We are genuinely grateful to the Editors and Reviewers for taking time to provide extremely valuable suggestions and comments, which will help us to substantially improve our paper. We decided to do our very best to implement all suggestions, as detailed in the point-by-point rebuttal letter below. We feel that our paper has improved considerably as a result. 

Public Reviews:

Reviewer #1 (Public Review): 

Summary: Koren et al. derive and analyse a spiking network model optimised to represent external signals using the minimum number of spikes. Unlike most prior work using a similar setup, the network includes separate populations of excitatory and inhibitory neurons. The authors show that the optimised connectivity has a like-to-like structure, leading to the experimentally observed phenomenon of feature competition. They also characterise the impact of various (hyper)parameters, such as adaptation timescale, ratio of excitatory to inhibitory cells, regularisation strength, and background current. These results add useful biological realism to a particular model of efficient coding. However, not all claims seem fully supported by the evidence. Specifically, several biological features, such as the ratio of excitatory to inhibitory neurons, which the authors claim to explain through efficient coding, might be contingent on arbitrary modelling choices. In addition, earlier work has already established the importance of structured connectivity for feature competition. A clearer presentation of modelling choices, limitations, and prior work could improve the manuscript.

Thanks for these insights and for this summary of our work.  

Major comments:

(1) Much is made of the 4:1 ratio between excitatory and inhibitory neurons, which the authors claim to explain through efficient coding. I see two issues with this conclusion: (i) The 4:1 ratio is specific to rodents; humans have an approximate 2:1 ratio (see Fang & Xia et al., Science 2022 and references therein); (ii) the optimal ratio in the model depends on a seemingly arbitrary choice of hyperparameters, particularly the weighting of encoding error versus metabolic cost. This second concern applies to several other results, including the strength of inhibitory versus excitatory synapses. While the model can, therefore, be made consistent with biological data, this requires auxiliary assumptions.

We now describe better the ratio of numbers of E and I neurons found in real data, as suggested. The first submission already contained an analysis of how the optimal ratio of E vs I neuron numbers depends in our model on the relative weighting of the loss of E and I neurons and on the relative weighting of the encoding error vs the metabolic cost in the loss function (see Fig. 7E). We revised the text on page 12 describing Fig. 7E. 

To allow readers to form easily a clear idea of how the weighting of the error vs the cost may influence the optimal network configuration, we now present how optimal parameters depend on the weighting in a systematic way, by always including this type of analysis when studying all other model parameters (time constants of single E and I neurons, noise intensity, metabolic constant, ratio of mean I-I to E-I connectivity). These results are shown on the Supplementary Fig. S4 A-D and H, and we comment briefly on each of them in Results sections (pages 9, 10, 11 and 12) that analyze each of these parameters.  

Following this Reviewer’s comment, we now included a joint analysis of network performance relative to the ratio of E-I neuron numbers and the ratio of mean I-I to E-I connectivity (Fig. 7J). We found a positive correlation between optima values of these two ratios. This implies that a lower ratio of E-I neuron numbers, such as a 2:1 ratio in human cortex mentioned by the reviewer, predicts lower optimal ratio of I-I to E-I connectivity and thus weaker inhibition in the network. We made sure that this finding is suitably described in revision (page 13).

(2) A growing body of evidence supports the importance of structured E-I and I-E connectivity for feature selectivity and response to perturbations. For example, this is a major conclusion from the Oldenburg paper (reference 62 in the manuscript), which includes extensive modelling work. Similar conclusions can be found in work from Znamenskiy and colleagues (experiments and spiking network model; bioRxiv 2018, Neuron 2023 (ref. 82)), Sadeh & Clopath (rate network; eLife, 2020), and Mackwood et al. (rate network with plasticity; eLife, 2021). The current manuscript adds to this evidence by showing that (a particular implementation of) efficient coding in spiking networks leads to structured connectivity. The fact that this structured connectivity then explains perturbation responses is, in the light of earlier findings, not new.

We agree that the main contribution of our manuscript in this respect is to show how efficient coding in spiking networks can lead to structured connectivity implementing lateral inhibition similar to that proposed in the recent studies mentioned by the Reviewer. We apologize if this was not clear enough in the previous version. We streamlined the presentation to make it clearer in revision.  We nevertheless think it useful to report the effects of perturbations within this network because these results give information about how lateral inhibition works in our network. Thus, we kept presenting it in the revised version, although we de-emphasized and simplified its presentation. We now give more emphasis to the novelty of the derivation of this connectivity rule from the principles of efficient coding (pages 4 and 6). We also describe better (page 8) what the specific results of our simulated perturbation experiments add to the existing literature.

(3) The model's limitations are hard to discern, being relegated to the manuscript's last and rather equivocal paragraph. For instance, the lack of recurrent excitation, crucial in neural dynamics and computation, likely influences the results: neuronal time constants must be as large as the target readout (Figure 4), presumably because the network cannot integrate the signal without recurrent excitation. However, this and other results are not presented in tandem with relevant caveats.

We improved the Limitations paragraph in Discussion, and also anticipated caveats in tandem with results when needed, as suggested. 

We now mention the assumption of equal time constants between the targets and readouts in the Abstract. 

We now added the analysis of the network performance and dynamics as a function of the time constant of the target (t_x) to the Supplementary Fig S5 (C-E). These results are briefly discussed in text on page 13. The only measure sensitive to t_x is the encoding error of E neurons, with a minimum at t_x =9 ms, while I neurons and metabolic cost show no dependency. Firing rates, variability of spiking as well as the average and instantaneous balance show no dependency on t_x. We note that t_x = t, with t=1/l the time constant of the population readout (Eq. 9), is an assumption we use when we derive the model from the efficiency objective (Eq. 18 to 23). In our new and preliminary work (Koren, Emanuel, Panzeri, Biorxiv 2024), we derived a more general class of models where this assumption is relaxed, which gives a network with E-E connectivity that adapts to the time constant of the stimulus. Thus, the reviewer is correct in the intuition that the network requires E-E connectivity to better integrate target signals with a different time constant than the time constant of the membrane. We now better emphasize this limitation in Discussion (page 16).

(4) On repeated occasions, results from the model are referred to as predictions claimed to match the data. A prediction is a statement about what will happen in the future – but most of the “predictions” from the model are actually findings that broadly match earlier experimental results, making them “postdictions”.

This distinction is important: compared to postdictions, predictions are a much stronger test because they are falsifiable. This is especially relevant given (my impression) that key parameters of the model were tweaked to match the data.

We now comment on every result from the model as either matching earlier experimental results, or being a prediction for experiments. 

In Section “Assumptions and emergent properties of the efficient E-I network derived from first principles”, we report (page 4) that neural networks have connectivity structure that relates to tuning similarity of neurons (postdiction). 

In Section “Encoding performance and neural dynamics in an optimally efficient E-I network” we report (page 5) that in a network with optimal parameters, I neurons have higher firing rate than E neurons (postdiction), that single neurons show temporally correlated synaptic currents (postdiction) and that the distribution of firing rates across neurons is log-normal (postdiction). 

In Section “Competition across neurons with similar stimulus tuning emerging in efficient spiking networks” we report (page 6)  that the activity perturbation of E neurons induces lateral inhibition on other E neurons, and that the strength of lateral inhibition depends on tuning similarity (postdiction). We show that activity perturbation of E neurons induces lateral excitation in I neurons (prediction). We moreover show that the specific effects of the perturbation of neural activity rely on structured E-I-E connectivity (prediction for experiments, but similar result in Sadeh and Clopath, 2020). We show strong voltage correlations but weak spike-timing correlations in our network (prediction for experiments, but similar result in Boerlin et al. 2013). 

In Section “The effect of structured connectivity on coding efficiency and neural dynamics”, we report (page 7) that our model predicts a number of differences between networks with structured and unstructured (random) connectivity. In particular, structured networks differ from unstructured ones by showing better encoding performance, lower metabolic cost, weaker variance over time in the membrane potential of each neuron, lower firing rates and weaker average and instantaneous balance of synaptic currents.

In Section “Weak or no spike-triggered adaptation optimizes network efficiency”, we report (page 9) that our model predicts better encoding performance in networks with adaptation compared to facilitation. Our results suggest that adaptation should be stronger in E compared to I (PV+) neurons (postdiction). In the same section, we report (page 10) that our results suggest that the instantaneous balance is a better predictor of model efficiency than average balance (prediction).

In Section “Non-specific currents regulate network coding properties”, we report (page 10) that our model predicts that more than half of the distance between the resting potential and firing threshold is taken by external currents that are unrelated to feedforward processing (postdiction). We also report (page 11) that our model predicts that moderate levels of uncorrelated (additive) noise is beneficial for efficiency (prediction for experiments, but similar results in Chalk et al., 2016, Koren et al., 2017, Timcheck et al. 2022).

In Section “Optimal ratio of E-I neuron numbers and of mean I-I to E-I synaptic efficacy coincide with biophysical measurements”, we predict the optimal ratio of E to I neuron numbers to be 4:1 (postdiction) and the optimal ratio of mean I-I to E-I connectivity to be 3:1 (postdiction). Further, we report (page 13) that our results predict that a decrease in the ratio of E-I neuron numbers is accompanied with the decrease in the ratio of mean I-I to E-I connectivity. 

Finally, in Section “Dependence of efficient coding and neural dynamics on the stimulus statistics”, we report (page 13) that our model predicts that the efficiency of the network has almost no dependence on the time scale of the stimulus (prediction). 

Reviewer #2 (Public Review):

Summary:

In this work, the authors present a biologically plausible, efficient E-I spiking network model and study various aspects of the model and its relation to experimental observations. This includes a derivation of the network into two (E-I) populations, the study of single-neuron perturbations and lateral-inhibition, the study of the effects of adaptation and metabolic cost, and considerations of optimal parameters. From this, they conclude that their work puts forth a plausible implementation of efficient coding that matches several experimental findings, including feature-specific inhibition, tight instantaneous balance, a 4 to 1 ratio of excitatory to inhibitory neurons, and a 3 to 1 ratio of I-I to E-I connectivity strength. It thus argues that some of these observations may come as a direct consequence of efficient coding.

Strengths:

While many network implementations of efficient coding have been developed, such normative models are often abstract and lacking sufficient detail to compare directly to experiments. The intention of this work to produce a more plausible and efficient spiking model and compare it with experimental data is important and necessary in order to test these models.

In rigorously deriving the model with real physical units, this work maps efficient spiking networks onto other more classical biophysical spiking neuron models. It also attempts to compare the model to recent single-neuron perturbation experiments, as well as some longstanding puzzles about neural circuits, such as the presence of separate excitatory and inhibitory neurons, the ratio of excitatory to inhibitory neurons, and E/I balance. One of the primary goals of this paper, to determine if these are merely biological constraints or come from some normative efficient coding objective, is also important.

Though several of the observations have been reported and studied before (see below), this work arguably studies them in more depth, which could be useful for comparing more directly to experiments.

Thanks for these insights and for the kind words of appreciation of the strengths of our work.  

Weaknesses:

Though the text of the paper may suggest otherwise, many of the modeling choices and observations found in the paper have been introduced in previous work on efficient spiking models, thereby making this work somewhat repetitive and incremental at times. This includes the derivation of the network into separate excitatory and inhibitory populations, discussion of physical units, comparison of voltage versus spike-timing correlations, and instantaneous E/I balance, all of which can be found in one of the first efficient spiking network papers (Boerlin et al. 2013), as well as in subsequent papers. Metabolic cost and slow adaptation currents were also presented in a previous study (Gutierrez & Deneve 2019). Though it is perfectly fine and reasonable to build upon these previous studies, the language of the text gives them insufficient credit.

We indeed built our work on these important previous studies, and we apologize if this was not clear enough. We thus improved the text to make sure that credit to previous studies is more precisely and more clearly given (see detailed reply for the list of changes made). 

To facilitate the understanding on how we built on previous work, we expanded the comparison of our results with the results of Boerlin et al. (2013) about voltage correlations and uncorrelated spiking (page 7), comparison with the derivation of physical units of Boerlin et al. (2013) (page 3), discussion of how results on the ratio of the number of E to I neurons relate  to Calaim et al (2022) and Barrett et al. (2016) (page 16), and comment on the previous work by Gutierrez and Deneve about adaptation (page 8).  

Furthermore, the paper makes several claims of optimality that are not convincing enough, as they are only verified by a limited parameter sweep of single parameters at a time, are unintuitive and may be in conflict with previous findings of efficient spiking networks. This includes the following. 

Coding error (RMSE) has a minimum at intermediate metabolic cost (Figure 5B), despite the fact that intuitively, zero metabolic cost would indicate that the network is solely minimizing coding error and that previous work has suggested that additional costs bias the output. 

Coding error also appears to have a minimum at intermediate values of the ratio of E to I neurons (effectively the number of I neurons) and the number of encoded variables (Figures 6D, 7B). These both have to do with the redundancy in the network (number of neurons for each encoded variable), and previous work suggests that networks can code for arbitrary numbers of variables provided the redundancy is high enough (e.g., Calaim et al. 2022). 

Lastly, the performance of the E-I variant of the network is shown to be better than that of a single cell type (1CT: Figure 7C, D). Given that the E-I network is performing a similar computation as to the 1CT model but with more neurons (i.e., instead of an E neuron directly providing lateral inhibition to its neighbor, it goes through an interneuron), this is unintuitive and again not supported by previous work. These may be valid emergent properties of the E-I spiking network derived here, but their presentation and description are not sufficient to determine this.

With regard to the concern that our previous analyses considered optimal parameter sets determined with a sweep of a single parameter at a time, we have addressed this issue in two ways. First, we presented (Figure 6I and 7J and text on pages 11 and 13) results of joint sweeps of variations of pairs of parameters whose joint variations are expected to influence optimality in a way that cannot be understood varying one parameter at a time. These new analyses complement the joint parameter sweep of the time constants of single E and I neurons (t_r^E and t_r^I) that has already been presented in Fig. 5A (former Fig. 4A). Second, we conducted, within a reasonable/realistic range of possible variations of each individual parameter, a Monte-Carlo random joint sampling (10000 simulations with 20 trials each) of all 6 model parameters that we explored in the paper. We presented these new results on Fig. 2 and discuss it on pages 5-6. 

The Reviewer is correct in stating that the error (RMSE) exhibits a counterintuitive minimum as a function of the metabolic constant despite the fact that, intuitively, for vanishing metabolic constant the network is solely minimizing the coding error (Fig. 6B). In our understanding, this counterintuitive finding is due to the presence of noise in the membrane potential dynamics. In the presence of noise, a non-vanishing metabolic constant is needed to suppress “inefficient” spikes purely induced by noise that do not contribute to coding and increase the error. This gives rise to a form of “stochastic resonance”, where the noise improves detection of the signal coming from the feedforward currents. We note that the metabolic constant and the noise variance both appear in the non-specific external current (Eq. 29f in Methods), and, thus, a covariation in their optimal values is expected. Indeed, we find that the optimal metabolic constant monotonically increases as a function of the noise variance, with stronger regularization (larger beta) required to compensate for larger variability (larger sigma) (Fig. 6I). Finally, we note that a moderate level of noise (which, in turn, induces a non-trivial minimum of the coding error as a function of beta) in the network is optimal. The beneficial effect of moderate levels of noise on performance in networks with efficient coding has been shown in different contexts in previous work (Chalk et al. 2016, Koren and Deneve, 2017). The intuition is that the noise prevents the excessive synchronization of the network and insufficient single neuron variability that decrease the performance. The points above are now explained in the revised text on page 11.

The Reviewer is also correct in stating that the network exhibits an optimal performance for intermediate values of the number of I neurons and the number of encoded features. In our understanding, the optimal number of encoded features of M=3 arises simply because all the other parameters were optimized for those values of M. The purpose of those analyses was not to state that a network optimally encodes only a given number of features, but how a network whose parameters are optimized for a given M perform reasonably well when M is varied. We clarify this on page 13 of Results in Discussion on page 16. In the same Discussion paragraph we refer also to the results of Calaim et al mentioned by the Reviewer. 

To address the concern about the comparison of efficiency between the E-I and the 1CT model, we took advantage of the Reviewer’s suggestions to consider this issue more deeply. In revision, we now compare the efficiency of the 1CT model with the E population of the E-I model (Fig. 8H). This new comparison changes the conclusion about which model is more efficient, as it shows the 1CT model is slightly more efficient than the E-I model. Nevertheless, the E-I model performance is more robust to small variations of optimal parameters, e.g., it exhibits biologically plausible firing rates for non-optimal values of the metabolic constant. See also the reply to point 3 of the Public Review of Reviewer 2 for more detail. We added these results and the ensuing caveats for the interpretation of this comparison on Page 14, and also revised the title of the last subsection of Results.  

Alternatively, the methodology of the model suggests that ad hoc modeling choices may be playing a role. For example, an arbitrary weighting of coding error and metabolic cost of 0.7 to 0.3, respectively, is chosen without mention of how this affects the results. Furthermore, the scaling of synaptic weights appears to be controlled separately for each connection type in the network (Table 1), despite the fact that some of these quantities are likely linked in the optimal network derivation. Finally, the optimal threshold and metabolic constants are an order of magnitude larger than the synaptic weights (Table 1). All of these considerations suggest one of the following two possibilities. One, the model has a substantial number of unconstrained parameters to tune, in which case more parameter sweeps would be necessary to definitively make claims of optimality. Or two, parameters are being decoupled from those constrained by the optimal derivation, and the optima simply corresponds to the values that should come out of the derivation.

We thank the reviewer for bringing about these important questions.

In the first submission, we presented both the encoding error and the metabolic cost separately as a function of the parameters, so that readers could get an understanding of how stable optimal parameters would be to the change of the relative weighting of encoding error and metabolic cost. We specified this in Results (page 5) and we kept presenting separately encoding and metabolic terms in the revision.

However, we agree that it is important to present the explicit quantification on how the optimal parameters may depend on g_L. In the first submission, we showed the analysis for all possible weightings in case of two parameters for which we found this analysis was the most relevant – the ratio of neuron numbers (Fig. 7E, Fig. 6E in first submission) and the optimal number of input features M (see last paragraph on page 13 and Fig. 8D). We now show this analysis also for the rest of studied model parameters in the Supplementary Fig. S4 (A-D and H). This is discussed on pages 9, 10,11 and 12.

With regard to the concern that the scaling of synaptic weights should not be controlled separately for each connection type in the network, we agree and we would like to clarify that we did not control such scaling separately. Apologies if this was not clear enough. From the optimal analytical solution, we obtained that the connectivity scales with the standard deviation of decoding weights (s_w^E and s_w^I) of the pre and postsynaptic populations (Methods, Eq. 32). We studied the network properties as a function of the ratio of average I-I to E-I connectivity (Fig. 7 F-I; Supplementary Fig. S4 D-H), which is equivalent to the ratio of standard deviations s_w^I /s_w^E (see Methods, Eq. 35). We clarified this in text on page 12.

Next, it is correct that our synaptic weights are an order of magnitude smaller than the metabolic constant. We analysed a simpler version of the network that has the coding and dynamics identical to our full model (Methods, Eq. 25) but without the external currents. We found that the optimal parameters determining the firing threshold in such a simpler network were biologically implausible (see Supplementary Text 2 and Supplementary Table S1). We considered as another simple solution the rescaling of the synaptic efficacy such as to have biologically plausible threshold. However, that gave implausible mean synaptic efficacy (see Supplementary Text 2).  Thus, to be able to define a network with biologically plausible firing threshold and mean synaptic efficacy, we introduced the non-specific external current. After introducing such current, we were able to shift the firing threshold to biologically plausible values while keeping realistic values of mean synaptic efficacy. Biologically plausible values for the firing threshold are around 15 -– 20 mV above the resting potential (Constantinople and Bruno, 2013), which is the value that we have in our model. A plausible value for the average synaptic strength is between a fraction of one millivolt to a couple of millivolts (Constantinople & Bruno, 2013, Campagnola et al. 2022), which also corresponds to values that the synaptic weights take. The above results are briefly explained in the revised text on page 4.

Finally, to study the optimality of the network when changing multiple parameters at a time, we added a new analysis with Monte-Carlo random joint sampling (10.000 parameter sets with 20 trials for each set) of all 6 model parameters that we explored in the paper. We compared (Fig 2) the so-obtained results of each simulation with those obtained from the understanding gained from varying one or two parameters at a time (optimal parameters reported in Table 1 and used throughout the paper).  We found (Fig. 2) that the optimal configuration in Table 1 was never improved by any other simulations we performed, and that the first three random simulations that came the closest to the optimal one of Table 1 had stronger noise intensity but also stronger metabolic cost than the configuration on Table 1. The second, third and fourth configurations had longer time constants of both E and I single neurons (adaptation time constants). Ratio of E-I neuron numbers and of I-I to E-I connectivity in the second, third and fourth best configuration were either jointly increased or decreased with respect to our configuration. These results are reported on Fig. 2 and in Tables 2-3 and they are discussed in Results (page 5).

Reviewer #3 (Public Review):

Summary:

In their paper the authors tackle three things at once in a theoretical model: how can spiking neural networks perform efficient coding, how can such networks limit the energy use at the same time, and how can this be done in a more biologically realistic way than previous work?

They start by working from a long-running theory on how networks operating in a precisely balanced state can perform efficient coding. First, they assume split networks of excitatory (E) and inhibitory (I) neurons. The E neurons have the task to represent some lower dimensional input signal, and the I neurons have the task to represent the signal represented by the E neurons. Additionally, the E and I populations should minimize an energy cost represented by the sum of all spikes. All this results in two loss functions for the E and I populations, and the networks are then derived by assuming E and I neurons should only spike if this improves their respective loss. This results in networks of spiking neurons that live in a balanced state, and can accurately represent the network inputs.

They then investigate in-depth different aspects of the resulting networks, such as responses to perturbations, the effect of following Dale's law, spiking statistics, the excitation (E)/inhibition (I) balance, optimal E/I cell ratios, and others. Overall, they expand on previous work by taking a more biological angle on the theory and showing the networks can operate in a biologically realistic regime.

Strengths:

(1) The authors take a much more biological angle on the efficient spiking networks theory than previous work, which is an essential contribution to the field.

(2) They make a very extensive investigation of many aspects of the network in this context, and do so thoroughly.

(3) They put sensible constraints on their networks, while still maintaining the good properties these networks should have.

Thanks for this summary and for these kind words of appreciation of the strengths of our work.  

Weaknesses:

(1) The paper has somewhat overstated the significance of their theoretical contributions, and should make much clearer what aspects of the derivations are novel. Large parts were done in very similar ways in previous papers. Specifically: the split into E and I neurons was also done in Boerlin et al (2008) and in Barrett et al (2016). Defining the networks in terms of realistic units was already done by Boerlin et al (2008). It would also be worth it to discuss Barrett et al (2016) specifically more, as there they also use split E/I networks and perform biologically relevant experiments.

We improved the text to make sure that credit to previous studies is more precisely and more clearly given (see rebuttal to the specific suggestions of Reviewer 2 for a full list).

We apologize if this was not clear enough in the previous version. 

With regard to the specific point raised here about the E-I split, we revised the text on page 2. With regard to the realistic units, we revised the text on page 3. Finally, we commented on relation between our results and results of the study by Barrett et al. (2016) on page 16.

(2) It is not clear from an optimization perspective why the split into E and I neurons and following Dale's law would be beneficial. While the constraints of Dale's law are sensible (splitting the population in E and I neurons, and removing any non-Dalian connection), they are imposed from biology and not from any coding principles. A discussion of how this could be done would be much appreciated, and in the main text, this should be made clear.

We indeed removed non-Dalian connections because Dale’s law is a major constraint for biological plausibility. Our logic was to consider efficient coding within the space of networks that satisfy this (and other) biological plausibility constraints. We did not intend to claim that removing the non-Dalian connections was the result of an analytical optimization. We clarified this in revision (page 4).

(3) Related to the previous point, the claim that the network with split E and I neurons has a lower average loss than a 1 cell-type (1-CT) network seems incorrect to me. Only the E population coding error should be compared to the 1-CT network loss, or the sum of the E and I populations (not their average). In my author recommendations, I go more in-depth on this point.

We carefully considered these possibilities and decided to compare only the E population of the E-I model with the 1-CT model. On Fig.8G (7C of the first submission), E neurons have a slightly higher error and cost compared to the 1CT network. In the revision, we compared the loss of E neurons of the E-I model with the loss of the 1-CT model. Using such comparison, we found that the 1CT network has lower loss and is more efficient compared to E neurons of the E-I model. We revised Figure 8H and text on page 14 to address this point. 

(4) While the paper is supposed to bring the balanced spiking networks they consider in a more experimentally relevant context, for experimental audiences I don't think it is easy to follow how the model works, and I recommend reworking both the main text and methods to improve on that aspect.

We tried to make the presentation of the model more accessible to a non-computational audience in the revised paper. We carefully edited the text throughout to make it as accessible as possible. 

Assessment and context:

Overall, although much of the underlying theory is not necessarily new, the work provides an important addition to the field. The authors succeeded well in their goal of making the networks more biologically realistic, and incorporating aspects of energy efficiency. For computational neuroscientists, this paper is a good example of how to build models that link well to experimental knowledge and constraints, while still being computationally and mathematically tractable. For experimental readers, the model provides a clearer link between efficient coding spiking networks to known experimental constraints and provides a few predictions.

Thanks for these kind words. We revised the paper to make sure that these points emerge more clearly and in a more accessible way from the revised paper.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Referring to the major comments:

(1) Be upfront about particular modelling choices and why you made them; avoid talk of a "striking/surprising", etc. ability to explain data when this actually requires otherwise-arbitrary choices and auxiliary assumptions. Ideally, this nuance is already clear from the abstract.

We removed all the "striking/surprising" and similar expressions from the text. 

We added to the Abstract the assumption of equal time constants of the stimulus and of the membrane of E and I neurons and the assumption of the independence of encoded stimulus features.

In revision, we performed additional analyses (joint parameter sweeps, Monte-Carlo joint sampling of all 6 model parameters) providing additional evidence that the network parameters in Table 1 capture reasonably well the optimal solution. These are reported on Figs. 2, 6I and 7J and in Results (pages 5, 11 and 13). See rebuttal to weaknesses of the public review of the Referee 2 for details.

(2) Make even more of an effort to acknowledge prior work on the importance of structured E-I and I-E connectivity.

We have revised the text (page 4) to better place our results within previous work on structured E-I and I-E connectivity.

(3) Be clear about the model's limitations and mention them throughout the text. This will allow readers to interpret your results appropriately.

We now comment more on model's limitations, in particular the simplifying assumption about the network's computation (page 16), the lack of E-E connectivity (page 3), the absence of long-term adaptation (page 10), and the simplification of only having one type of inhibitory neurons (page 16). 

(4) Present your "predictions" for what they are: aspects of the model that can be made consistent with the existing data after some fitting. Except in the few cases where you make actual predictions, which deserve to be highlighted.

We followed the suggestion of the reviewer and distinguished cases where the model is consistent with the data (postdictions) from actual predictions, where empirical measurements are not available or not conclusive. We compiled a list of predictions and postdictions in response to the point 4 of Reviewer 1. In revision, we now comment about every property of the model as either reproducing a known property of biological networks (postdiction) or being a prediction. We improved the text in Results on pages 4, 5, 6, 7, 9, 10, 11, 12 and 13 to accommodate these requests.

Minor comments and recommendations

It's a sizable list, but most can be addressed with some text edits.

(1) The image captions should give more details about the simulations and analyses, particularly regarding sample sizes and statistical tests. In Figure 5, for example, it is unclear if the lines represent averages over multiple signals and, if so, how many. It's probably not a single realization, but if it is, this might explain the otherwise puzzling optimal number of three stimuli. Box plots visualize the distribution across simulation trials, but it's not clear how many. In Figure 7d, a star suggests statistical significance, but the caption does not mention the test or its results; the y-axis should also have larger limits.

All statistical results were computed on 100 or 200 simulation trials, depending on the figure, with duration of the trial of 1 second of simulated time. To compute statistical results in Fig. 1, we used 10 trials with duration of 10 seconds for each trial. Each trial consisted of M independent realizations of Ornstein-Uhlenbeck (OU) processes as stimuli, independent noise in the membrane potential and an independent draw of tuning parameters, such that the results are general over specific realization of these random variables. Realizations of the OU processes were independent across stimulus dimensions and across trials. We added this information in the caption of each figure. 

The optimal number of M=3 stimuli is the result of measuring the performance of the network in 100 simulation trials (for each parameter value), thus following the same procedure as for all other parameters. Boxplots on Fig. 8G-H were also generated from results computed in 100 simulation trials, which we have now specified in the caption of the figure, together with the statistical test used for assessing the significance (twotailed t-test). We also enlarged the limits of Fig. 8H (7D in the previous version).

(2) The Oldenburg paper (reference 62) finds suppression of all but nearby neurons in response to two- photon stimulation of small neural ensembles (instead of single neurons, as in Chettih & Harvey). This isn't perfectly consistent with the model's results, even though the Oldenburg experiments seem more relevant given the model's small size, and strong connectivity/high connection probability between similarly tuned neurons. What might explain the potential mismatch?

We sincerely apologize for not having been precise enough on this point when comparing our model against Chettih & Harvey and Oldenburg et al. We corrected the sentence (page 6) to remove the claim that our model reproduces both. 

We speculate that the discrepancy between perturbing our model and the Oldenburg data may arise from the lack of E-E connectivity in our model. Synaptic connections between E neurons with similar selectivity could create an enhancement instead of suppression between neuronal pairs with very similar tuning. We added a sentence about this in the section with perturbation experiments “Competition across neurons with similar stimulus tuning emerging in efficient spiking networks” (page 7) where we discuss this limitation of our model. We feel that this example shows the utility to derive some perturbation results from our model, as not all networks with some degree of lateral inhibition will show the same perturbation results. Comparing our model's perturbation with real data perturbation results has thus some value to better appreciate strengths and limitations of our approach. 

(3) "Previous studies optogenetically stimulated E neurons but did not determine whether the recorded neurons were excitatory or inhibitory " (p. 11). I believe Oldenburg et al. did specifically image excitatory neurons.

The reviewer is correct about Oldenburg et al. imaging specifically excitatory neurons. We have revised this part of the Discussion (page 15). 

(4) The authors write that efficiency is particularly achieved where adaptation is stronger in E compared to I neurons (p. 7; Figure 4). Although this would be consistent with experimental data (the I neurons in the model seem akin to fast-spiking Pv+ cells), I struggle to see it in the figure. Instead, it seems like there are roughly two regimes. If either of the neuronal timescales is faster than the stimulus timescale, the optimisation fails. If both are at least as slow, optimisation succeeds.

We agree with the reviewer that the adaptation properties of our inhibitory neurons are compatible with Pv+ cells. What is essential for determining the dynamical regime of the network is less the relation to the time constant of the stimulus (t_x) but rather the relation between the time constant of the population readout (t, which is also the membrane time constant) and the time constant of the single neuron (t_r^y for y=E and y=I; see Eq. 23, 25 or 29e). The relation between t and t_r^y determines if single neurons generate spike-triggered adaptation (t_r^y > t) or spike-triggered facilitation (t_r^y < t; see Table 4). In regimes with facilitation in either E or I neurons (or both), the network performance strongly deteriorates compared to regimes with adaptation (Fig. 5A). 

Beyond adaptation leading to better performance, we also found different effects of adaptation in E and I neurons. We acknowledge that the difference of these effects was difficult to see from the Fig. 4B in the first submission. We have now replotted results from previously shown Fig. 4B to focus on the adaptation regime only, (since the Fig. 5A already establishes that this is the regime with better performance). We also added figures showing the differential effect of adaptation in E and I cell type on the firing rate and on the average loss (Fig. 5C-D). Fig. 5B and C (top plots) show that with adaptation in E neurons, the error and the loss increase more slowly than with adaptation in I neurons. Moreover, the firing rate in both cell types decreases with adaptation in E neurons, while this is not the case with adaptation in I neurons (Fig. 5D). These results are added to the figure panels specified above and discussed in text on page 9.

To clarify the relation between neuronal and stimulus timescale, we now also added the analysis of network performance as a function of the time constant of the stimulus t_x (Supplementary Fig. S5 C-E). We found that the model's performance is optimal when the time constant of the stimulus is close to the membrane time constant t. This result is expected, because the equality of these time constants was imposed in our analytical derivation of the model (t_x  = t). We see a similar decrease in performance for values of t_x  that are faster and slower with respect to the membrane time constant (Supplementary Fig. S5C, top). These results are added to the figure panels specified above and discussed in text on page 13.

(5) A key functional property of cortical interneurons is their lower stimulus selectivity. Does the model replicate this feature?

We think that whether I neurons are less selective than E neurons is still an open question. A number of recent empirical studies reported that the selectivity of I neurons is comparable to the selectivity of E neurons (see., e.g., Kuan et al. Nature 2024, Runyan et al. Neuron 2010, Najafi et al. Neuron 2020). In our model, the optimal solution prescribes a precise structure in recurrent connectivity (see Eq. 24 and Fig. 1C(ii)) and structured connectivity endows I neurons with stimulus selectivity. To show this, we added plots of example tuning curves and the distribution of the selectivity index across E and I neurons (Fig. 8E-F) and described these new results in Results (page 14). Tuning curves in our network were similar to those computed in a previous work that addressed stimulus tuning in efficient spiking networks (Barrett et al. 2016). We evaluated tuning curves using M=3 constant stimulus features and we varied one of the features while the two others were kept fixed. We provided details on how the tuning curves and the selectivity index were computed in a new Methods subsection (“Tuning curves and selectivity index”) on page 50.

(6) The final panels of Figure 4 are presented as an approach to test the efficiency of biological networks. The authors seem to measure the instantaneous (and time-averaged) E-I balance while varying the adaptation parameter and then correlate this with the loss. If that is indeed the approach (it's difficult to tell), this doesn't seem to suggest a tractable experiment. Also, the conclusion is somewhat obvious: the tighter the single neuron balance, the fewer unnecessary spikes are fired. I recommend that the authors clearly explain their analysis and how they envision its application to biological data.

We indeed measured the instantaneous (and time-averaged) E-I balance while varying the adaptation parameters and then correlating this with the loss. We did not want to imply that the latter panels of Figure 4 are a means to test the efficiency or biological networks or that we are suggesting new and possibly unfeasible experiments. We see it as a way to better conceptually understand how spike triggered adaptation helps the network’s coding efficiency, by tightening the E I balance in a way that it reduces the number of unnecessary spikes. We apologize if the previous text was confusing in this respect.   We have now removed the initial paragraph of former Results Subsection (including removing the subsection title) and added new text about different effect of adaptation in E and I neurons on Page 9. We also thoroughly revised Figure 5.

(7) The external stimuli are repeatedly said to vary (or be tracked) across "multiple time scales", which might inadvertently be interpreted as (i) a single stimulus containing multiple timescales or (ii) simultaneously presented stimuli containing different timescales. These scenarios are potential targets for efficient coding through neuronal adaptation (reference 21 in the manuscript and Pozzorini et al. Nat. Neuro. 2013), but they are not addressed in the current model. I recommend the authors clarify their statements regarding timescales (and if they're up for it, acknowledge this as a limitation).

We thank the reviewer for bringing up this interesting point. To address the second point raised by the Reviewer (simultaneously presented stimuli containing multiple timescales), we performed new analyses to test the model with simultaneously presented stimuli that have different timescales. We found that the model encodes efficiently such stimuli.  We tested the case with a 3-dimensional stimulus where each dimension is an Ornstein-Uhlenbeck process with a different time constant. More precisely, we kept the time constant in the first dimension fixed (at 10 ms), and varied the time constant in the second and third dimension such that the time constant in the third dimension is doubled with respect to the second dimension. We plotted the encoding error in every stimulus dimension for E and I neurons (Fig. 8B, left plot) as well as the encoding error and the metabolic cost averaged across stimulus dimensions (Fig. 8B, right plot). The results are briefly described with text on page 13.

Regarding the case i) (single stimulus containing multiple timescales), we considered two possibilities. One possibility is that timescales of the stimulus are separable, and in this case a single stimulus containing several time scales can be decomposed in several stimuli with a single time scale each. As we assign a new set of weights for each dimension of the decomposed stimulus, this case is similar to the case ii) that we already addressed. Another possibility is that timescales of the stimulus cannot be separated. This case is not covered in the present analysis and we listed it among the limitations of the model. We revised the text (page 13) around the question of multiple time scales and included the citation of Pozzorini et al. (2013). 

(8) It is claimed that the model uses a mixed code to represent signals, citing reference 47 (Rigotti et al., Nature 2013). But whereas the model seems to use linear mixed selectivity, the Rigotti reference highlights the virtues of nonlinear mixed selectivity. In my understanding, a linearly mixed code does not enjoy the same benefits since it’s mathematically equivalent to a non-mixed code (simply rotate the readout matrix). I recommend that the authors clarify the type of selectivity used by their model and how it relates to the paper(s) they cite.

The reviewer is correct that our selectivity is a linear mixing of input variables, and differs from the selectivity in Rigotti et al. (2013) which is non-linear. We revised the sentence on page 4 to clarify better that the mixed selectivity we consider is linear and we removed Rigotti’s citation. 

(9) Reference 46 is cited as evidence that leaky integration of sensory features is a relevant computation for sensory areas. I don’t think this is quite what the reference shows. Instead, it finds certain morphological and electrophysiological differences between single pyramidal neurons in the primary visual cortex compared to the prefrontal cortex. Reference 46’ then goes on to speculate that these are differences relevant to sensory computation. This may seem like a quibble, but given the centrality of the objectivee function in normative theories, I think it's important to clarify why a particular objective is chosen.

We agree that our reference of Amatrudo et al was not the best reference and that the previous text was confusing. We thus tried to improve on its clarity. We looked at the previous theoretical efficient coding papers introducing this leaky integration and we could not find in the previous theoretical work a justification of this assumption based on experimental papers. However, there is evidence that neurons in sensory structures, and in cortical association areas respond to time varying sensory evidence by summing stimuli over time with a weight that decreases steadily going back in time from the time of firing, which suggests that neurons integrate time-varying sensory features. In many cases, these integration kernels decay approximately exponentially going back in time, and several models explaining successfully perceptual readouts of neural activity work assuming leaky integration. This suggests that the mathematical approximation of leaky integration of sensory evidence, though possibly simplistic, is reasonable.  We revised the text in this respect (page 2).  

(10) The definition of the objective function uses beta as a tuning parameter, but later parts of the text and figures refer to a parameter g_L which might only be introduced in the convex combination of Eq. 40a.

This is correct. Parameter optimization has been performed on a weighted sum of the average encoding error and cost as given by the Eq. 39a (40a in first submission), with the weighting g_L for the error versus the cost, and not the beta that is part of the objective in Eq.10. The convex combination in Eq. 39a allowed us to find a set of optimal parameters that is within biologically realistic parameter ranges, which includes realistic values for the firing threshold. The average encoding error and metabolic cost (the two terms on the right-hand side of Eq. 39a, without weighting with g_L) in our network are of the same order (see Fig 8G for the E-I model where these values are plotted separately for the optimal network). Weighing the cost with optimal beta that is in the range of ~10 would have yielded a network that optimizes almost exclusively the metabolic cost and would bias the results towards solutions with poor encoding accuracy.

To document more fully how the choice of weighting of the error with the cost (g_L) affects the optimal parameters, we now added new analysis (Fig. 8D and Supplementary Fig. S4 A-D and H) showing optimal parameters as a function of this weighting. We commented on these results in the text on pages 9-11 and 12. For further details, please see also the reply to point 1 or Reviewer 1.

(11) Figure 1J: "In E neurons, the distribution of inhibitory and of net synaptic inputs overlap". In my understanding, they are in fact identical, and this is by construction. It might help the reader to state this.

We apologize for an unclear statement. In E neurons, net synaptic current is the sum of the feedforward current and of recurrent inhibition (Eq. 29c and Eq. 42). With our choice of tuning parameters that are symmetric around zero and with stimulus features that have vanishing mean, the mean of the feedforward current is close to zero. Because of this, the mean of the net current is negative and is close to the mean of the inhibitory current. We have clarified this in the text (page 5).

(12) A few typos:

-  p1. "Minimizes the encoding accuracy" should be "maximizes..."

-  p1: "as well the progress" should be something like "as well as the progress"

-  p.11 In recorded neurons where excitatory or inhibitory. ", "where" should be "were" - Fig3: missing parentheses (B)

-  Fig4B: the 200 ticks on the y-scale are cut off.

-  Panel Fig. 5a: "stimulus" should be "stimuli".

-  Ref 24 "Efficient andadaptive sensory codes" is missing a space.

-  p. 26: "requires" should be "required".

-  On several occasions, the article "the" is missing.

We thank the reviewer for kindly pointing out the typos that we now corrected.

Reviewer #2 (Recommendations For The Authors):

I would like to give the authors more details about the two main weaknesses discussed above, so that they may address specific points in the paper. First, there is the relation to previous work. Several published articles have presented very similar results to those discussed here, including references 5, 26, 28, 32, 33, 42, 43, 48, and an additional reference not cited by the authors (Calaim et al. 2022 eLife e73276). This includes:

(1) Derivation of an E-I efficient spiking network, which is found in refs. 28, 42, 43, and 48. This is not reflected in the text: e.g., "These previous implementations, however, had neurons that did not respect Dale's law" (Introduction, pg. 1); "Unlike previous approaches (28, 48), we hypothesize that E and I neurons have distinct normative objectives...". The authors should discuss how their derivation compares to these.

We have now fully clarified on page 3 that our model builds on the seminal previous works that introduced E-I networks with efficient coding (Supplementary text in Boerlin et al. 2013, Chalk et al. 2016, Barrett et al. 2016). 

(2) Inclusion of a slow adaptation current: I believe this also appears in a previous paper (Gutierrez & Deneve 2019, ref. 33) in almost the exact same form, and is again not reflected in the text: "The strength of the current is proportional to the difference in inverse time constants ... and is thus absent in previous studies assuming that these time constants are equal (... ref. 33). Again, the authors should compare their derivation to this previous work.

We thank the reviewer for pointing this out. We sincerely apologize if our previous version did not recognize sufficiently clearly that the previous work of Gutierrez and Deneve (eLife 2019; ref 33) introduced first the slow adaptation current that is similar to spike-triggered adaptation in our model. We have made sure that the revised text recognizes it more clearly. We also explained better what we changed or added with respect to this previous work (see revised text on page 8). 

The work by Gutierrez and Deneve (2019) emphasizes the interplay between single neuron property (an adapting current in single neurons) and network property (networklevel coding through structured recurrent connections). They use a network that does not distinguish E and I neurons. Our contribution instead focuses on the adaptation in an E-I network. To improve the presentation following the Reviewer’s comment, we now better emphasize the differential effect of adaptation in E and in I neurons in revision (Fig. 5 B-D). Moreover, Gutierrez and Deneve studied the effect of adaptation on slower time scales (1 or 2 seconds) while we study the adaptation on a finer time scale of tens of milliseconds. The revised text detailed this is reported on Page 8.

(3) Background currents and physical units: Pg. 26: "these models did not contain any synaptic current unrelated to feedforward and recurrent processing" and "Moreover previous models on efficient coding did not thoroughly consider physical units of variables" - this was briefly described in ref. 28 (Boerlin et al. 2013), in which the voltage and threshold are transformed by adding a common constant, and additional aspects of physical units are discussed.

It is correct that Boerlin et al (2013) suggested adding a common constant to introduce physical units. We now revised the text to make clearer the relation between our results and the results of Boerlin et al. (2013) (page 3). In our paper, we built on Boerlin et al. (2013) and assigned physical units to computational variables that define the model's objective (the targets, the estimates, the metabolic constant, etc.). We assigned units to computational variables in such a way that physical variables (such as membrane potential, transmembrane currents, firing thresholds and resets) have the correct physical units.  We have now clarified how we derived physical units in the section of Results where we introduce the biophysical model (page 3) and specified how this derivation relates to the results in Boerlin et al. (2013).

(4) Voltage correlations, spike correlations, and instantaneous E/I balance: this was already pointed out in Boerlin et al. 2013 (ref 28; from that paper: "Despite these strong correlations of the membrane potentials, the neurons fire rarely and asynchronously") and others including ref. 32. The authors mention this briefly in the Discussion, but it should be more prominent that this work presents a more thorough study of this well-known characteristic of the network.

We agree that it would be important to comment on how our results relate to these results in Boerlin et al. (2013). It is correct that in Boerlin et al. (2013) neurons have strong correlations in the membrane potentials, but fire asynchronously, similarly to what we observe in our model. However, asynchronous dynamics in Boerlin et al. (2013) strongly depends on the assumption of instantaneous synaptic transmission and time discretization, with a “one spike per time bin” rule in numerical implementation. This rule enforces that at most one spike is fired in each time bin, thus actively preventing any synchronization across neurons. If this rule is removed, their network synchronizes, unless the metabolic constant is strong enough to control such synchronization to bring it back to asynchronous regime (see ref. 36). Our implementation does not contain any specific rule that would prevent synchronization across neurons. We now cite the paper by Boerlin and colleagues and briefly summarize this discussion when we describe the result of Fig. 3D on page 7. 

(5) Perturbations and parameters sweep: I found one previous paper on efficient spiking networks (Calaim et al. 2022) which the authors did not cite, but appears to be highly relevant to the work presented here. Though the authors perform different perturbations from this previous study, they should ideally discuss how their findings relate to this one. Furthermore, this previous study performs extensive sweeps over various network parameters, which the authors might discuss here, when relevant. For example, on pg. 8, the authors write “We predict that, if number of neurons within the population decreases, neurons have to fire more spikes to achieve an optimal population readout” – this was already shown in Calaim et al. 2022 Figure 5, and the authors should mention if their results are consistent.

We apologize for not being aware of Calaim et al. (2022) when we submitted the first version of our paper. This important study is now cited in the revised version. We have now, as suggested, performed sweeps of multiple parameters inspired by the work of Calaim. This new analysis is described extensively in reply to Weaknesses in the Public Review of reviewer 2 and is found in Fig 2, 6I and 7J and described on pages 5,11 and 13.

The Reviewer is also correct that the compensation mechanism that applies when changing the ratio of E-I neuron numbers is similar to the one described in Barrett et al. (2016) and related to our claim “if number of neurons within the population decreases, neurons have to fire more spikes to achieve an optimal population readout”. We have now added (page 11) that this prediction is consistent with the finding of Barrett et al. (2016).

With regard to the dependence of optimal coding properties on the number of neurons, we have tried to better describe similarities and differences with our work and that of Calaim et al as well as with the work of Barrett et al. (2016) which reports highly relevant results. These additional considerations are summarized in a paragraph in Discussion (page 16).

(6) Overall, the authors should distinguish which of their results are novel, which ones are consistent with previous work on efficient spiking networks, and which ones are consistent in general with network implementations of efficient and sparse coding. In many of the above cases, this manuscript goes into much more depth and study of each of the network characteristics, which is interesting and commendable, but this should be made clear. In clarifying the points listed above, I hope that the authors can better contextualize their work in relation to previous studies, and highlight what are the unique characteristics of the model presented here.

We made a number of clarifications of the text to provide better contextualization of our model within existing literature and to credit more precisely previous publications. This includes commenting on previous studies that introduced separate objective functions of E and I neurons (page 2), spike-triggered adaptation (page 8), physical units (page 3), and changes in the number of neurons in the network (page 16). 

Next, there are the claims of optimal parameters. As explained on pg. 35 (criterion for determining optimal model parameters), it appears to me that they simply vary each parameter one at a time around the optimal value. This argument appears somewhat circular, as they would need to know the optimal parameters before starting this sweep. In general, I find these optimality considerations to be the most interesting and novel part of the paper, but the simulations are relatively limited, so I would ask the authors to either back them up with more extensive parameter sweeps that consider covariations in different parameters simultaneously (as in Calaim et al. 2022). Furthermore, the authors should make sure that they are not breaking any of the required relationships between parameters necessary for the optimization of the loss function. Again, some of the results (such as coding error not being minimized with zero metabolic cost) suggests that there might be issues here. 

We thank the reviewer for this insightful suggestion. We have now added a joint sweep of all relevant model parameters using Monte-Carlo parameter search with 10.000 iterations. We randomly drew parameter configurations from predetermined parameter ranges that are detailed in the newly added Table 2. Parameters were sampled from a uniform distribution. We varied all the six model parameters studied in the paper (metabolic constant, noise intensity, time constant of single E and I neurons, ratio of E to I neurons and ratio of the mean I-I to E-I connectivity).  We now present these results on a new Figure 2. We did not find any set of parameters with lower loss than the parameters in Table 1 when the weighting of the error with the cost was in the following range: 0.4<g_L<0.81 (Fig. 2C). While our large but finite Monte-Carlo random sampling does not fully prove that the configuration we selected as optimal (on Table 1) is a global optimum, it shows that this configuration is highly efficient. Further, and as detailed in the rebuttal to the Weaknesses of the Public Review of Referee 2, analyses of the near optimal solutions are compatible with the notion (resulting from the join parameter sweep studies that we added to Figures 6 and 7) that network optimality may be influenced by joint covariations in parameters. These new results are reported in Results (page 5, 11 and 13) and in Figure 2, 6I an 7J.

Some more specific points:

(1) In general, I find it difficult to understand the scaling of the RMSE, cost, and loss values in Figures 4-7. Why are RMSE values in the range of 1-10, whereas loss and cost values are in the range of 0-1? Perhaps the authors can explicitly write the values of the RMSE and loss for the simulation in Figure 1G as a reference point.

Encoding error (RMSE), metabolic cost (MC) and average loss for a well performing network are within the range of 1-10 (see Fig. 8G or 7C in the first submission). To ease the visualization of results, we normalized the cost and the loss on Figs. 6-8 in order to plot them on the same figure (while the computation of the optima is done following the Eq. 39 and is without normalization). We have now explicitly written the values of RMSE, MC and the average loss (non-normalized) for the simulation in Fig. 1D on page 5, as suggested by the reviewer. We have also revised Fig. 4 and now show the absolute and not the relative values of the RMSE and the MC (metabolic cost). 

(2) Optimal E-I neuron ratio of 4:1 and efficacy ratio of 3:1: besides being unintuitive in relation to previous work, are these two optimal settings related to one another? If there are 4x more excitatory neurons than inhibitory neurons, won't this affect the efficacy ratio of the weights of the two populations? What happens if these two parameters are varied together?

Thanks for this insightful point. Indeed, the optima of these two parameters are interdependent and positively correlated - if we decrease the E-I neuron ratio, the optimal efficacy ratio decreases as well. To better show this relation we added figures with 2dimensional parameter search (Fig. 7J) where we varied jointly the two ratios. The red cross on the right figure marks the optimal ratios used as optimal parameters in our study. These finding are discussed on page 13.

(3) Optimal dimensionality of M=[1,4]: Again, previous work (Calaim et al. 2022) would suggest that efficient spiking networks can code for arbitrary dimensional signals, but that performance depends on the redundancy in the network - the more neurons, the better the coding. From this, I don't understand how or why the authors find a minimum in Figure 7B. Why does coding performance get worse for small M?

We optimized all model parameters with M=3 and this is the reason why M=3 is the optimal number of inputs when we vary this parameter. Our network shows a distinct minimum of the encoding error as a function of the stimulus dimensionality for both E and I neurons (Fig. 8C, top). This minimum is reflected in the minimum of the average loss (Fig. 8C, bottom). The minimum of the loss is shifted (or biased) by the metabolic cost, with strong weighting of the cost lowering the optimal number of inputs. This is discussed on pages 13-14.

Here are a list of other, more minor points, that the authors can consider addressing to make the results and text more clear:

(1) Feedforward efficient coding models: in the introduction (pg. 1) and discussion (pg. 11) it is mentioned that early efficient coding models, such as that of Olshausen & Field 96, were purely feedforward, which I believe to be untrue (e.g., see Eq. 2 of O&F 96). Later models made this even more explicit (Rozell et al. 2008). Perhaps the authors can either clarify what they meant by this, or downplay this point.

We sincerely apologize for the oversight present in the previous version of the text. We agree with the reviewer that the model in Olshausen and Field (1996) indeed defines a network with recurrent connections, and the same type of recurrent connectivity has been used by Rozell et al. (2008, 2013). The structure of the connectivity in Olshausen and Field (as well as in Rozell et al (2008)) is closely related to the structure of connectivity that we derived in our model. We have corrected the text in the introduction (page 1) to remove these errors.

(2) Pg. 2 - The authors state: "We draw tuning parameters from a normal distribution...", but in the methods, it states that these are then normalized across neurons, so perhaps the authors could add this here, or rephrase it to say that weights are drawn uniformly on the hypersphere.

We rephrased the description of how weights were determined (page 2).

(3) Pg. 2 - "We hypothesize the time-resolved metabolic cost to be proportional to the estimate of a momentary firing rate of the neural population" - from what I can see, this is not the usual population rate, which would be an average or sum of rates across the population.

Indeed, the time-dependent metabolic cost is not the population rate (in the sense of the sum of instantaneous firing rates across neurons), but is proportional to it by a factor of 1/t. More precisely, we can define the instantaneous estimate of the firing rate of a single neuron i as z_i(t) = 1/t_r r_i(t) with r_i(t) as in Eq. 7. We have clarified this in the revised text on page 3. 

(4) Pg. 3: "The synaptic strength between two neurons is proportional to their tuning similarity if the tuning similarity is positive" - based on the figure and results, this appears to be the case for I-E, E-I, and I-I connections, but not for E-E connections. This should be clarified in the text. Furthermore, one reference given in the subsequent sentence (Ko et al. 2011, ref. 51), is specifically about E-E connections, so doesn't appear to be relevant here.

We have now specified that the Eq. 24 does not describe E-E connections. We also agree that the reference (Ko et al. 2011) did not adequately support our claim and we thus removed it and revised the text on page 3 accordingly.

(5) Pg. 3: "the relative weight of the metabolic cost over the encoding error controls the operating regime of the network" and "and an operating regime controlled by the metabolic constant" - what do you mean by operating regime here?

We used the expression “operating regime” in the sense of a dynamical regime of the network.  However, we agree that this expression may be confusing and we removed it in revision. 

(6) Pg. 3: "Previous studies interpreted changes of the metabolic constant beta as changes to the firing thresholds, which has less biological plausibility" - can the authors explain why this is less plausible, or ideally provide a reference for it?

In biological networks, global variables such as brain state can strongly modulate the way neural networks respond to a feedforward stimulus. These variables influence neural activity in at least two distinct ways. One is by changing non-specific synaptic inputs to neurons, which is a network-wide effect (Destexhe and Pare, Nature Reviews Neurosci. 2003). This is captured in our model by changing the strength of the mean and fluctuations in the external currents. Beyond modulating synaptic currents, another way of modulating neural activity is by changing cell-intrinsic factors that modulate the firing threshold in biological neurons (Pozzorini et al. 2013). Previous studies on spiking networks with efficient coding interpreted the effect of the metabolic constant as changes to the firing threshold (Koren and Deneve, 2017, Gutierrez and Deneve 2019), which corresponds to cell-intrinsic factors. Here we instead propose that the metabolic constant modulates the neural activity by changing the non-specific synaptic input, homogeneously across all neurons in the network. Interpreting the metabolic constant as setting the mean of the non-specific synaptic input was necessary in our model to find an optimal set of parameters (as in Table 1) that is also biologically plausible. We revised the text accordingly (page 4).

(7) Pg. 4: Competition across neurons: since the model lacks E-E connectivity, it seems trivial to conclude that there is competition through lateral inhibition, and it can be directly determined from the connectivity. What is gained from running these perturbation experiments?

We agree that a reader with a good understanding of sparse / efficient coding theory can tell that there is competition across neurons with similar tuning already from the equation for the recurrent connectivity (Eq. 24). However, we presume that not all readers can see this from the equations and that it is worth showing this with simulations.

Following the reviewer's comment, we have now downplayed the result about the model manifesting lateral inhibition in general on page 6. We have also removed its extensive elaboration in Discussion.

One reason to run perturbation experiments was to test to what extent the optimal model qualitatively replicates empirical findings, in particular, single neuron perturbation experiments in Chettih and Harvey, 2019, without specifically tuning any of the model parameters. We found that the model reproduces qualitatively the main empirical findings, without tuning the model to replicate the data. We revised the text on page 5 accordingly.

Further reason to run these experiments was to refine predictions about the minimal amount of connectivity structure that generates perturbation response profiles that are qualitatively compatible with empirical observations. To establish this, we did perturbation experiments while removing the connectivity structure of a particular connectivity sub-matrices (E-I, I-I or I-E; Fig. S3 F). This allowed us to determine which connectivity matrix has to be structured to observe results that qualitatively match empirical findings. We found that the structure of E-I and I-E connectivity is necessary, but not the structure of I-I connectivity. Finally, we tested partial removal of the connectivity structure where we replaced the precise (and optimal) connectivity structure and imposed a simpler connectivity rule. In the optimal connectivity, the connection strength is proportional to the tuning similarity. A simpler connectivity rule, in contrast, only specifies that neurons with similar tuning share a connection, and beyond this the connection strength is random. Running perturbation experiments in such a network obeying a simpler connectivity rule still qualitatively replicated empirical results from Chettih and Harvey (2019). This is shown on the Supplementary Fig. S2F on described on page 8.

(8) Pg. 4: "the optimal E-I network provided a precise and unbiased estimator of the multidimensional and time-dependent target signal" - from previous work (e.g., Calaim et al. 2022), I would guess that the estimator is indeed biased by the metabolic cost. Why is this not the case here? Did you tune the output weights to remove this bias?

Output weights were not tuned to remove the bias. On Fig. 1H in the first submission we plotted the bias for the network that minimizes the encoding error. We forgot to specify this in the text and figure caption, for which we apologize. We now replaced this figure with a new one (Fig. 1E) where we plot the bias of the network minimizing the average loss (with parameters as in Table 1). The bias of the network minimizing the error is close to zero, B^E = 0.02 and B^I = 0.03.  The bias of the network minimizing the loss is stronger and negative, B^E = -0.15 and B^I=-0.34. In the text of Results, we now report the bias of both networks (i.e., optimizing the encoding error and optimizing the loss). We also added a plot showing trial-averaged estimates and a time-dependent bias in each stimulus dimension (Supplementary figure S1 F). Note that the network minimizing the encoding error requires a lower metabolic constant (β = 6) than the network optimizing the loss (β=14), however, the optimal metabolic cost in both networks is nonzero. We revised the text and explained these points on page 5.

(9) Pg. 4: "The distribution of firing rates was well described by a log-normal distribution" - I find this quite interesting, but it isn't clear to me how much this is due to the simulation of a finitetime noisy input. If the neurons all have equal tuning on the hypersphere, I would expect that the variability in firing is primarily due to how much the input correlates with their tuning. If this is true, I would guess that if you extend the duration of the simulation, the distribution would become tighter. Can you confirm that this is the stationary distribution of the firing rates?

We now simulated the network with longer simulation time (10 seconds of simulated time instead of 2 seconds used previously) and also iterated the simulation across 10 trials to report a result that is general across random draws of tuning parameters (previously a single set of tuning parameters was used). The reviewer is correct that the distribution of firing rates of E neurons has become tighter with longer simulation time, but distributions remain log-normal. We also recomputed the coefficient of variation (CV) using the same procedure. We updated these plots on Fig. 1F.

(10) Pg. 4: "We observed a strong average E-I balance" - based on the plots in Figure 1J, the inputs appear to be inhibition-dominated, especially for excitatory neurons. So by what criterion are you calling this strong average balance?

The reviewer is correct about the fact that the net synaptic input to single neurons in our optimal network shows excess inhibition and the network is inhibition-dominated, so we revised this sentence (page 5) accordingly.  

(11) Pg. 4: Stronger instantaneous balance in I neurons compared to E neurons - this is curious, and I have two questions: (1) can the authors provide any intuition or explanation for why this is the case in the model? and (2) does this relate to any literature on balance that might suggest inhibitory neurons are more balanced than excitatory neurons?

In our model, I neurons receive excitatory and inhibitory synaptic currents through synaptic connections that are precisely structured. E neurons receive structured inhibition and a feedforward current. The feedforward current consists of M=3 independent OU processes projected on the tuning vectors of E neurons w_i^E. We speculate that because the synaptic inhibition and feedforward current are different processes and the 3 OU inputs are independent, it is harder for E neurons to achieve the instantaneous balance that would be as precise as in I neurons. While we think that the feedforward current in our model reflects biologically plausible sensory processing, it is not a mechanistic model of feedforward processing. In biological neurons, real feedforward signals are implemented as a series of complex feedforward synaptic inputs from downstream areas, while the feedforward current in our model is a sum of stimulus features, and is thus a simplification of a biological process that generates feedforward signals. We speculate that a mechanistic implementation of the feedforward current could increase the instantaneous balance in E neurons.  Furthermore, the presence of EE connections could potentially also increase the instantaneous balance in E neurons. We revised the Discussion about these important questions that lie on the side of model limitations and could be advanced in future work. We could not find any empirical evidence directly comparing the instantaneous balance in E versus I neurons.  We have reported these considerations in the revised Discussion (page 16).

(12) Pg. 5, comparison with random connectivity: "Randomizing E-I and I-E connectivity led to several-fold increases in the encoding error as well as to significant increases in the metabolic cost" and Discussion, pg. 11: "the structured network exhibits several fold lower encoding error compared to unstructured networks": I'm wondering if these comparisons are fair. First, regarding activity changes that affect the metabolic cost - it is known that random balanced networks can have global activity control, so it is not straightforward that randomizing the connectivity will change the metabolic cost. What about shuffling the weights but keeping an average balance for each neuron's input weights? Second, regarding coding error, it is trivial that random weights will not map onto the correct readout. A fairer comparison, in my opinion, would at least be to retrain the output weights to find the best-fitting decoder for the threedimensional signal, something more akin to a reservoir network.

Thank you for raising these interesting questions. The purpose of comparing networks with and without connectivity structure was to observe causal effects of the connectivity structure on the neural activity. We agree that the effect on the encoding error is close to trivial, because shuffling of connectivity weights decouples neural dynamics from decoding weights. We have carefully considered Reviewer's suggestions to better compare the performance of structured and unstructured networks. 

In reply to the first point, we followed the reviewer's suggestion and compared the optimal network with a shuffled network that matched the optimal network in its average balance. This was achieved by increasing the metabolic constant, decreasing the noise intensity and slightly decreasing the feedforward stimulus (we did not find a way to match the net current in both cell types by changing a single parameter). As we compared the metabolic cost between the optimal and the shuffled network with matched average balance, we still found lower metabolic cost in the optimal network, even though the difference was now smaller. We replaced Fig. 3B from the first submission with these new results in Fig. 4B and commented on them in the text (page 7).

In reply to the second point, we followed reviewer’s suggestion and compared the encoding error (RMSE) of the optimal network and the network with shuffled connectivity where decoding weights are trained such as to optimally reconstruct the target signal. As suggested, we now analyzed the encoding error of the networks using decoding weights trained on the set of spike trains generated by the network using linear least square regression to minimize the decoding error. For a fair and quantitative comparison and because we did not train decoding weights of our structured model, we performed this same analysis using spike trains generated by networks with structured and shuffled recurrent connectivity. We found that the encoding error is smaller in the E population and much smaller in the I population in the structured compared to the random network. Decoding weights found numerically in the optimal network approach uniform distribution of weights that we used in our model (Fig. 4A, right). In contrast, decoding weights obtained from the random network do not converge to a uniform distribution, but instead form a much sparser distribution, in particular in I neurons (Supplementary Fig. S3 A). These additional results reported in the above mentioned figures are discussed in text on page 14.  

(13) Pg. 5: "a shift from mean-driven to fluctuation-driven spiking" and Pg. 11 "a network structured as in our efficient coding solution operates in a dynamical regime that is more stimulus-driven, compared to an unstructured network that is more fluctuation driven" - I would expect that the balanced condition dictates that spiking is always fluctuation driven. I'm wondering if the authors can clarify this.

We agree with the reviewer that networks with and without connectivity structure are fluctuation-driven, because in a mean-driven network the mean current must be suprathreshold (Ahmadian and Miller, 2021), which is not the case of either network. We removed the claim of the change from mean to fluctuation driven regime in the revised paper. We are grateful to the Reviewer for helping us tighten the elaboration of our findings.

(14) Pg. 5: "suggesting that variability of spiking is independent of the connectivity structure" - the literature of balanced networks argues against this. Is this not simply because you have a noisy input? Can you test this claim?

We thank the reviewer for the suggestion. We tested this claim by measuring the coefficient of variation in networks receiving a constant stimulus. In particular, we set the same strength in each of the M=3 stimulus dimensions and set the stimulus amplitude such as to match the firing rate of the optimal network in response to the OU stimulus. We computed the coefficient of variation in 200 simulation trials.  The removal of connectivity structure did not cause significant change of the coefficient of variation in a network driven by a constant stimulus (Fig. 4E). These additional results are discussed in text on page 7. 

We also taken the suggestion about variability of spiking being independent of the connectivity structure. We removed this claim in the revision, because we only tested a couple of specific cases where the connectivity is structured with respect to tuning similarity (fully structured, fully unstructured and partially unstructured networks). This is not exhaustive of all possible structures that recurrent connectivity may have.

(15) Pg. 6: "we also removed the connectivity structure only partially, keeping like-to-like connectivity structure and removing all structure beyond like-to-like" - can you clarify what this means, perhaps using an equation? What connectivity structure is there besides like-to-like?

In the optimal model, the strength of the synapse between a pair of neurons is proportional to the tuning similarity of the two neurons, Y_ij proportional to J_ij for Y_ij >0 (see Eq. 24 and Fig. 1C(ii)). Besides networks with optimal connectivity, we also tested networks with a simpler connectivity rule. Such a simpler rule prescribes a connection if the pair of neurons has similar tuning (Y_ij >0), and no connection otherwise. The strength of the connection following this simpler connectivity rule is otherwise random (and not proportional to pairwise tuning similarity Y_ij as it is in the optimal network). We clarified this in the revision (page 8), also by avoiding the term “like-to-like” for the second type of networks, which could indeed be prone to confusion.

(16) Pgs. 6-7: "we indeed found that optimal coding efficiency is achieved with weak adaptation in both cell types" and "adaptation in E neurons promotes efficient coding because it enforces every spike to be error- correcting" - this was not clear to me. First, it appears as though optimal efficiency is achieved without adaptation nor facilitation, i.e., when the time constants are all equal. Indeed, this is what is stated in Table 1. So is there really a weak adaptation present in the optimal case? Second, it seems that the network already enforces each spike to be errorcorrecting without adaptation, so why and how would adaptation help with this?

We agree with the Reviewer that the network without adaptation in E and I neurons is already optimal. It is also true that most spikes in an optimal network should already be error-correcting (besides some spikes that might be caused by the noise). However, regimes with weak adaptation in E neurons remain close to optimality. Spike-triggered facilitation, meanwhile, ads spikes that are unnecessary and decrease network efficiency. We revised the Fig.5 (Fig. 4 in first submission) and replaced 2-dimensional plots in Fig.4 C-F with plots that show the differential effect of adaptation in E neurons (top) and in I neurons (bottom plots) for the measures of the encoding error (RMSE), the efficiency (average loss) and the firing rate (Fig. 5B-D). On the new Fig. 5C it is evident that the loss of E and I population grows slowly with adaptation in E neurons (top) while it grows faster with adaptation in I neurons (bottom). These considerations are explained in revised text on page 9.

(17) Pg. 7: "adaptation in E neurons resulted in an increase of the encoding error in E neurons and a decrease in I neurons" - it would be nice if the authors could provide any explanation or intuition for why this is the case. Could it perhaps be because the E population has fewer spikes, making the signal easier to track for the I population?

We agree that this could indeed be the case. We commented on it in revision (page 9).

(18) Pg. 7: "The average balance was precise...with strong adaptation in E neurons, and it got weaker when increasing the adaptation in I neurons (Figure 4E)" - I found the wording of this a bit confusing. Didn't the balance get stronger with larger I time constants?

By increasing the time constant of I neurons, the average imbalance got weaker (closer to zero) in E neurons (Fig. 5G, left), but stronger (further away from zero) in I neurons (Fig. 5G, right). We have revised the text on page 9 to make this clearer.

(19) Pg. 7: Figure 4F is not directly described in the text.

We have now added text (page 9) commenting on this figure in revision.

(20) Pg. 8: "indicating that the recurrent network dynamics generates substantial variability even in the absence of variability in the external current" -- how does this observation relate to your earlier claim (which I noted above) that "variability of spiking is independent of connectivity structure"?

We agree that the claim about variability of spiking being independent of connectivity structure was overstated and we thus removed it. The observation that we wanted to report is that both structured and unstructured networks have very similar levels of variability of spiking of single neurons. The fact that much of the variability of the optimal network is generated by recurrent connections is not incompatible. We revised the related text (page 11) for clarity.

(21) Pg. 9: "We found that in the optimally efficient network, the mean E-I and I-E synaptic efficacy are exactly balanced" - isn't this by design based on the derivation of the network?

True, the I-E connectivity matrix is the transpose of the E-I connectivity matrix, and their means are the same by the analytical solution. This however remains a finding of our study. We have clarified this in the revised text (page 12).

(22) Pg. 30, eq. 25: the authors should verify if they include all possible connectivity here, or if they exclude EE connectivity beforehand.

We now specify that the equation for recurrent connectivity (Eq. 24, Eq. 25 in first submission) does not include the E-E connectivity in the revised text (page 41).

Reviewer #3 (Recommendations For The Authors):

Essential

(1)  Currently, they measure the RMSE and cost of the E and I population separately, and the 1CT model. Then, they average the losses of the E and I populations, and compare that to the 1CT model, with the conclusion that the 1CT model has a higher average loss. However, it seems to me that only the E population should be compared to the 1CT model. The I population loss determines how well the I population can represent the E population representation (which it can do extremely well). But the overall coding accuracy of the network of the input signal itself is only represented by the E population. Even if you do combine the E and I losses, they should be summed, not averaged. I believe a more fair conclusion would be that the E/I networks have generally slightly worse performance because of needing to follow Dale's law, but are still highly efficient and precise nonetheless. Of course, I might be making a critical error somewhere above, and happy to be convinced otherwise!

We carefully considered the reviewer's comment and tested different ways of combining the losses of the E and I population. We decided to follow the reviewer's suggestion and to compare the loss of the E population of the E-I model with the loss of the one cell type model. As evident already from the Fig. 8G, such comparison indeed changes the result to make the 1CT model more efficient. Also, the sum of losses of E and I neurons results in the 1CT model being more efficient than the E-I model. Note, however, the robustness of the E-I model to changes in the metabolic constant (Fig. 6C, top). The firing rates of the E-I model stay within physiological ranges for any value of the metabolic constant, while the firing rate of the 1CT model skyrocket for the metabolic constant that is lower than optimal (Fig. 8I).

We added to Results (page 14) a summary of these findings.

(2) The methods and main text should make much clearer what aspects of the derivation are novel, and which are not novel (see review weaknesses for specifics).

We specified these aspects, as discussed in more detail in the above reply to point 4 of the public review of Reviewer 1.

Request:

If possible, I would like to see the code before publication and give recommendations on that (is it easy to parse and reproduce, etc.)

We are happy to share the computer code with the reviewer and the community. We added a link to our public repository containing the computer code that we used for simulations and analysis to the preprint and submission (section “Code availability” on page 17). 

Suggestions:

(1) I believe that for an eLife audience, the main text is too math-heavy at the beginning, and it could be much simplified, or more effort could be made to guide the reader through the math.

We tried to do our best to improve the clarity of description of mathematical expressions in the main text.

(2) Generally vector notation makes network equations for spiking neurons much clearer and easier to parse, I would recommend using that throughout the paper (and not just in the supplementary methods).

We now use vector notation throughout the paper whenever we think that this improves the intelligibility of the text. 

(3) In the discussion or at the end of the results adding a clear section summarizing what the minimal requirements or essential assumptions are for biological networks to implement this theory would be helpful for experimentalists and theorists alike.

We have added such a section in Discussion (page 15). 

(5) I think the title is a bit too cumbersome and hard to parse. Might I suggest something like 'Efficient coding and energy use in biophysically realistic excitatory-inhibitory spiking networks' or 'Biophysically constrained excitatory-inhibitory spiking networks can efficiently implement efficient coding'.

We followed reviewer’s suggestion and changed the title to “Efficient coding in biophysically realistic excitatory-inhibitory spiking networks.”

(6) How the connections were shuffled exactly was not clear to me in how it was described now. Did they just take the derived connectivity, and shuffle the connections around? I recommend a more explicit methods section on it (I might have missed it).

Indeed, the connections of the optimal network were randomly shuffled, without repetition, between all neuronal pairs of a specific connectivity matrix. This allows to preserve all properties of the distribution of connectivity weights and only removes the structure of the connectivity, which is precisely what we wanted to test. We now added a section in Methods (“Removal of connectivity structure”) on pages 51-52 where we explain how the connectivity structure is removed.

(7) Figure 1 sub-panel ordering was confusing to read (first up down, then left right). Not sure if re- arranging is possible, but perhaps it could be A, B, and C at the top, with subsublabels (i) and (ii). Might become too busy though.

We followed this suggestion and rearranged the Fig. 1 as suggested by the reviewer. 

(8) Equation 3 in the main text should specify that 'y' stands for either E or I.

This has been specified in the revision (page 3). 

(9) Figure 1D shows a rough sketch of the types of connectivities that exist, but I would find it very useful to also see the actual connection strengths and the effect of enforcing Dale's law.

We revised this figure (now Fig. 1B (ii)) and added connection strengths as well as a sketch of a connection that was removed because of Dale’s law.

(10) The main text mentions how the readout weights are defined (normal distributions), but I think this should also be mentioned in the methods.

Agreed. We indeed had Methods section “Parametrization of synaptic connectivity (page 46), where we explain how readout weights are defined. We apologize if a call on this section was not salient enough in the first submission. We made sure that the revised main text contains a clear pointer to this Methods section for details. 

(11) The text seems to mix ‘decoding weights’ and ‘readout weights’.

Thanks for this suggestion to use consistent language. We opted for ‘decoding weights’ and removed ‘readout weights’.

(12) The way the paper is written makes it quite hard to parse what are new experimental predictions, and what results reproduce known features. I wonder if some sort of 'box' is possible with novel predictions that experimentalists could easily look at and design an experiment around.

We now revised the text. We clarified for every property of the model if this property is a prediction of facts that were not yet experimentally tested or if it accounts for previously observed properties of biological neurons. Please see the reply to point 4 of Reviewer 1. 

(13) Typo's etc.:

Page 5 bottom -- ("all") should have one of the quotes change direction (common latex typo, seems to be the only place with the issue).

We thank the reviewer for pointing out this typo that has been removed in revision.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors investigated the anatomical features of the synaptic boutons in layer 1 of the human temporal neocortex. They examined the size of each synapse, the macular or perforated appearance, the size of the synaptic active zone, the number and volume of the mitochondria, and the number of synaptic and dense core vesicles, also differentiating between the readily releasable, the recycling, and the resting pool of synaptic vesicles. The coverage of the synapse by astrocytic processes was also assessed, and all the above parameters were compared to other layers of the human temporal neocortex. The authors conclude that the subcellular morphology of the layer 1 synapses are suitable for the functions of the neocortical layer, i.e. the synaptic integration within the cortical column. The low glial coverage of the synapses might allow increased glutamate spillover from the synapses, enhancing synaptic crosstalk within this cortical layer.

Strengths:

The strengths of this paper are the abundant and very precious data about the fine structure of the human neocortical layer 1. Quantitative electron microscopy data (especially that derived from the human brain) are very valuable since this is a highly time- and energy-consuming work. The techniques used to obtain the data, as well as the analyses and the statistics performed by the authors are all solid, strengthen this manuscript, and mainly support the conclusions drawn in the discussion.

We would like to thank reviewer#1 for his very positive comments on our manuscript stating that such data about the fine structure of the human neocortex are are highly relevant.

Weaknesses:

There are several weaknesses in this work. First, the authors should check and review extensively for improvements to the use of English. Second, several additional analyses performed on the existing data could substantially elevate the value of the data presented. Much more information could be gained from the existing data about the functions of the investigated layer, of the cortical column, and about the information processing of the human neocortex. Third, several methodological concerns weaken the conclusions drawn from the results.

We would like to thank the reviewer for his critical and thus helpful comments on our manuscript. We took the first comment of the reviewer concerning the English and have thus improved our manuscript by rephrasing and shortening sentences. Secondly, according to the reviewer several additional analyses should be performed on the existing data, which could substantially elevate the value of the data presented. We will implement some of the suggestions in the improved version of the manuscript where appropriate. We will address a more detailed answer to the reviewer’s queries in her/his suggestions to the authors (see below). However, the reviewer states himself: “The techniques used to obtain the data, as well as the analyses and the statistics performed by the authors are all solid, strengthen this manuscript, and mainly support the conclusions drawn in the discussion”.

Reviewer #2 (Public review):

Summary:

The study of Rollenhagen et al. examines the ultrastructural features of Layer 1 of the human temporal cortex. The tissue was derived from drug-resistant epileptic patients undergoing surgery, and was selected as far as possible from the epilepsy focus, and as such considered to be non-epileptic. The analyses included 4 patients with different ages, sex, medication, and onset of epilepsy. The manuscript is a follow-on study with 3 previous publications from the same authors on different layers of the temporal cortex:

Layer 4 - Yakoubi et al 2019 eLife

Layer 5 - Yakoubi et al 2019 Cerebral Cortex

Layer 6 - Schmuhl-Giesen et al 2022 Cerebral Cortex.

They find, that the L1 synaptic boutons mainly have a single active zone, a very large pool of synaptic vesicles, and are mostly devoid of astrocytic coverage.

Strengths:

The manuscript is well-written and easy to read. The Results section gives a detailed set of figures showing many morphological parameters of synaptic boutons and glial elements. The authors provide comparative data of all the layers examined by them so far in the Discussion. Given that anatomical data in the human brain are still very limited, the current manuscript has substantial relevance. The work appears to be generally well done, the EM and EM tomography images are of very good quality. The analysis is clear and precise.

We would like to thank the reviewer for his very positive evaluation of our paper and the comments that such data have a substantial relevance, in particular in the human neocortex. In contrast to reviewer#1, this reviewer’s opinion is that the manuscript is well written and easy to read.

Weaknesses:

One of the main findings of this paper is that "low degree of astrocytic coverage of L1 SBs suggests that glutamate spillover and as a consequence synaptic cross-talk may occur at the majority of synaptic complexes in L1". However, the authors only quantified the volume ratio of astrocytes in all 6 layers, which is not necessarily the same as the glial coverage of synapses. In order to strengthen this statement, the authors could provide 3D data (that they have from the aligned serial sections) detailing the percentage of synapses that have glial processes in close proximity to the synaptic cleft, that would prevent spillover.

We agree with the reviewer that we only quantified the volume ratio of the astrocytic coverage but not necessarily the percentage of synapses that may or not contribute to the formation of the ‘tripartite’ synapse. As suggested, we will re-analyze our material with respect to the percentage of coverage for individual synaptic boutons in each layer and will implement the results in the improved version of the manuscript. However, since this is a completely new analysis that is time-consuming we would like to ask the reviewer for additional time to perform this task.

A specific statement is missing on whether only glutamatergic boutons were analyzed in this MS, or GABAergic boutons were also included. There is a statement, that they can be distinguished from glutamatergic ones, but it would be useful to state it clearly in the Abstract, Results, and Methods section what sort of boutons were analyzed. Also, what is the percentage of those boutons from the total bouton population in L1?

We would like to thank the reviewer for this comment. Although our title clearly states, we focused on quantitative 3D-models of excitatory synaptic boutons, we will point out that more clearly in the Methods and Result chapters. Our data support recent findings by others (see for example Cano-Astorga et al. 2023, 2024; Shapson-Coe et al. 2024) that have evaluated the ratio between excitatory vs. inhibitory synaptic boutons in the temporal lobe neocortex, the same area as in our study, which was between 10-15% inhibitory terminals but with a significant layer and region specific difference. We will include the excitatory vs. inhibitory ratio and the corresponding citations in the Results section.

Synaptic vesicle diameter (that has been established to be ~40nm independent of species) can properly be measured with EM tomography only, as it provides the possibility to find the largest diameter of every given vesicle. Measuring it in 50 nm thick sections results in underestimation (just like here the values are ~25 nm) as the measured diameter will be smaller than the true diameter if the vesicle is not cut in the middle, (which is the least probable scenario). The authors have the EM tomography data set for measuring the vesicle diameter properly.

We partially disagree with the reviewer on this point. Using high-resolution transmission electron microscopy, we measured the distance from the outer-to-outer membrane only on those synaptic vesicles that were round in shape with a clear ring-like structure to avoid double counts and discarded all those that were only partially cut according to criteria developed by Abercrombie (1946) and Boissonnat (1988). We assumed that within a 55±5 nm thick ultrathin section (silver to gray interference contrast) all clear-ring-like vesicles were distributed in this section assuming a vesicle diameter between 25 to 40nm. For large DCVs, double-counts were excluded by careful examination of adjacent images and were only counted in the image where they appeared largest.

In addition, we have measured synaptic vesicles using TEM tomography and came to similar results. We will address this in Material and Methods that both methods were used.

It is a bit misleading to call vesicle populations at certain arbitrary distances from the presynaptic active zone as readily releasable pool, recycling pool, and resting pool, as these are functional categories, and cannot directly be translated to vesicles at certain distances. Indeed, it is debated whether the morphologically docked vesicles are the ones, that are readily releasable, as further molecular steps, such as proper priming are also a prerequisite for release.

We thank the reviewer for this comment. However, nobody before us tried to define a morphological correlate for the three functionally defined pools of synaptic vesicles since synaptic vesicles normally are distributed over the entire nerve terminal. As already mentioned above, after long and thorough discussions with Profs. Bill Betz, Chuck Stevens, Thomas Schikorski and other experts in this field we tried to define the readily releasable (RRP), recycling (RP) and resting pools by measuring the distance of each synaptic vesicle to the presynaptic density (PreAZ). Using distance as a criterion, we defined the RRP including all vesicles that were located within a distance (perimeter) of 10 to 20 nm from the PreAZ that is less than an average vesicle diameter (between 25 to 40 nm). The RP was defined as vesicles within a distance of 60-200 nm away, still quite close but also rapidly available on demand and the remaining ones beyond 200 nm were suggested to belong to the resting pool. This concept was developed for our first publication (Sätzler et al. 2002) and this approximation since then is very much acknowledged by scientist working in the field of synaptic neuroscience and computational neuroscientist. We were asked by several labs worldwide whether they can use our data of the perimeter analysis for modeling. We agree that our definition of the three pools can be seen as arbitrary but we never claimed that our approach is the truth but nothing as the truth. Concerning the debate whether only docked vesicles or also those very close the PreAZ should constitute the RRP we have a paper in preparation using our perimeter analysis, EM tomography and simulations trying to clarify this debate. Our preliminary results suggest that the size of the RRP should be reconsidered.

Tissue shrinkage due to aldehyde fixation is a well-documented phenomenon that needs compensation when dealing with density values. The authors cite Korogod et al 2015 - which actually draws attention to the problem comparing aldehyde fixed and non-fixed tissue, still the data is non-compensated in the manuscript. Since all the previous publications from this lab are based on aldehyde fixed non-compensated data, and for this sake, this dataset should be kept as it is for comparative purposes, it would be important to provide a scaling factor applicable to be able to compare these data to other publications.

We thank the reviewer for his suggestion. However, for several reasons we did not correct for shrinkage caused by aldehyde fixation. There are papers by Eyre et al. (2007) and the mentioned paper by Korogod et al. 2015 that have demonstrated that cryo-fixation reveals larger numbers of docked synaptic vesicles, a smaller glial volume, and a less intimate glial coverage of synapses and blood vessels compared to chemical fixation. Other structural subelements such as active zone size and shape and the total number of synaptic vesicles remained unaffected. In two further publications Zhao et al. (2012a, b) investigating hippocampal mossy fiber boutons using cryo-fixation and substitutions came to similar results with respect to bouton and active zone size and number and diameter of synaptic vesicles compared to aldehyde-fixation as described by Rollenhagen et al. 2007 for the same nerve terminal. This was one of the reasons not correcting for shrinkage. In addition, all cited papers state that chemical fixation in general provides a much better ultrastructural preservation of tissue samples when compared with cryo-fixation and substitution where optimal preservation is only regional within a block of tissue and therefore less suitable for large-scale ultrastructural analyses as we performed.

Reviewer #3 (Public review):

Summary:

Rollenhagen et al. offer a detailed description of layer 1 of the human neocortex. They use electron microscopy to assess the morphological parameters of presynaptic terminals, active zones, vesicle density/distribution, mitochondrial morphology, and astrocytic coverage. The data is collected from tissue from four patients undergoing epilepsy surgery. As the epileptic focus was localized in all patients to the hippocampus, the tissue examined in this manuscript is considered non-epileptic (access) tissue.

Strengths:

The quality of the electron microscopic images is very high, and the data is analyzed carefully. Data from human tissue is always precious and the authors here provide a detailed analysis using adequate approaches, and the data is clearly presented.

We are very thankful to the reviewer upon his very positive comments about our data analysis and presentation.

Weaknesses:

The study provides only morphological details, these can be useful in the future when combined with functional assessments or computational approaches. The authors emphasize the importance of their findings on astrocytic coverage and suggest important implications for glutamate spillover. However, the percentage of synapses that form tripartite synapses has not been quantified, the authors' functional claims are based solely on volumetric fraction measurements.

We thank the reviewer for his critical comments on our findings concerning the layer-specific astrocytic coverage as also suggested by reviewer#2. As already stated above we will analyze the astrocytic coverage and the layer-specific percentage of astrocytic contribution to the ‘tripartite’ synapse in more detail. We are, however, a bit puzzled about the comment that structural anatomists usually receive that our study only provides morphological details. Our thorough analysis of structural and synaptic parameters of synaptic boutons underlie and might even predict the function of synaptic boutons in a given microcircuit or network and will thus very much improve our understanding and knowledge about the functional properties of these structures, in particular in the human brain where such studies are still quite rare. The main goal of our studies in the human neocortex was the quantitative morphology of synaptic boutons and thus the synaptic organization of the cortical column, layer by layer which to our knowledge is the first such detailed study undertaken in the human brain. Our efforts have set a golden standard in the analysis of synaptic boutons embedded in different microcircuits und is meanwhile internationally very well accepted.

The distinction between excitatory and inhibitory synapses is not clear, they should be analyzed separately.

As already stated above in response to reviewer#1 our study focused on excitatory synaptic boutons since they represent the majority of synapses. However, in the improved version of our manuscript in the Material and Method section we included a paragraph with structural criteria to distinguish excitatory from inhibitory terminals (see also our comment to reviewer#1 concerning this point) including appropriate citations.

The text connects functional and morphological characteristics in a very direct way. For example, connecting plasticity to any measurement the authors present would be rather difficult without any additional functional experiments. References to various vesicle pools based on the location of the vesicles are also more complex than suggested in the manuscript. The text should better reflect the limitations of the conclusions that can be drawn from the authors' data.

We thank the reviewer for this comment. However, it has been shown by meanwhile numerous publications that the shape and size of the active zone together with the pool of synaptic vesicles and the astrocytic coverage critically determines synaptic transmission and synaptic strength, but can also contribute to the modulation of synaptic plasticity (see also citations within the text). It has been shown that synaptic boutons can switch upon certain stimulation conditions to different modes of release (uni- vs. multiquantal, uni- vs multivesicular release) and from asynchronous to synchronous release leading also to the modulation of synaptic short- and long-term plasticity. To the second comment: When we started with our first paper about the Calyx of Held – principal neuron synapse in the MNTB (Sätzler et al. 2002) we tried to define a morphological correlate for the three functionally defined pools. As already mentioned above in our reply to the other two reviewers, this is rather difficult since synaptic vesicles are normally distributed over the entire nerve terminal. After long and thorough discussions with Bill Betz, Chuck Stevens and other leading scientist in the field of synaptic neuroscience, we together with Bert Sakmann tried to define a morphological correlate for the functionally defined pools using a perimeter analysis. We defined the readily releasable pool as vesicles 10 to 20 nm away from the presynaptic active zone, the recycling pool as those in 60-200 nm distance and the remaining as those belonging to the resting pool. However, it has been shown by capacitance measurements (see for example Hallermann et al 2003), FM1-43 investigations (see for example Henkel et al. 1996) and high-resolution electron microscopy (see for example Schikorski and Stevens 2001; Schikorski 2014) that our estimate of the RRP nearly perfectly matches with the functionally defined pools at hippocampal and cortical synapses (Silver et al. 2003). In addition, in one of our own papers (Rollenhagen et al. 2018) we also estimated the RP functionally from trains of EPSPs using an exponential fit analysis and came to similar results upon its size using the perimeter analysis.

Of course, as stated by the reviewer the scenario could be more complex, using other criteria but we never claimed that our morphologically defined pools are the truth but nothing as the truth but we believe it offers a quite good approximation.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

Abstract:

Avoid the numerous abbreviations in the abstract. The paragraph describing the results obtained in this study is too short. Include more results, such as the size of the active zone, the proportion of perforated synapses, the ratio of synapses terminating on dendrites/spines, the percentage of volume occupied by mitochondria, etc. In the last paragraph, compare the layer-specific data to other layers of the neocortex before writing the concluding sentence.

To meet the word limits of the abstract (150 words) defined by eLife we had to use abbreviations. We followed the suggestions by the reviewer and expanded our abstract by adding the proportion of macular vs. perforated active zone and the percentage of mitochondria within an SB. However, we did not include the comparison of structural parameters in the Abstract since this is discussed thoroughly in the MS at other places (see Results and Discussion).

Results:

First of all, wonderful data! Lots of work, very valuable quantitative electron microscopy results.

Main concerns:

Adding several analyses would give much more information about the cortical synaptic organization. It would be very useful to differentiate between excitatory and inhibitory terminals (and give their ratio) and include this information in all different analyses, such as in the SV number, SV pool analysis, mitochondrion analysis, etc., that would give functional information as well. You have all the data for this, and you know how to differentiate between inhibitory and excitatory synapses, it can be done. We could see the possible morphological differences between excitatory and inhibitory synapses (maybe one is larger/has more SVs, etc. than the other). Based on these possible differences conclusions could be drawn about functional hypotheses, such as one or the other is more efficient in inducing postsynaptic potentials, excitation or inhibition is more pronounced in layer 1, etc. Furthermore, looking at the ratio of perforated synapses, we could gain information about the formation of new synapses. Maybe there is a difference between excitatory and inhibitory circuits in this point of view.

To the first point: Since our focus was on excitatory synaptic boutons as already stated in the title we have not analyzed inhibitory SBs. To do so, we have to re-analyze our complete data which is time-consuming and an additional workload. However, we can give a ratio excitatory vs. inhibitory synaptic boutons which was between 10-15% but with layer-specific differences. Our finding are in good agreement with a recent publication in Science by the Lichtman group (Shapson-Coe et al. 2024) and work by the DeFelipe group (Cano-Astorga et al. 2023, 2024) estimating the number of inhibitory boutons in different layers of the temporal lobe neocortex as we did by 10-15%. We included a small paragraph about inhibitory synapses, their percentage and included the citations in our Results section. Concerning the ratio between macular, non-perforated vs. perforated active zones we stated the majority of synaptic boutons were of the macular, non-perforated type (~75%; see improved version of the MS). If perforated, this was found predominantly on the postsynaptic site, but quite rare in L1 SBs. Since GABAergic terminals had only a small or no clearly visible PSD this would be hard to look at.

To the last point, it has been demonstrated that the number of dense core vesicles and their fusion with the presynaptic density could be a critical factor in the build-up of the active zone. In addition, the findings of the Geinismann group suggesting that perforated synapses are more efficient than non-perforated ones is nowadays very controversially discussed since other factors such as size of the active zone (see for example Matz et al. 2010; Holderith et al. 2012) and the astrocytic coverage contribute to synaptic efficacy and strength.

Related to this topic: although in the case of rat CA1 pyramidal cells all inhibitory synapses terminated on dendritic shafts (Megias et al., Neuroscience 2001), please be aware that both excitatory and inhibitory synapses can terminate on both dendritic shafts and spines in humans (inhibitory synapses are though rare on spines, usually less than 10%, but they do exist, see for example Wittner et al, Neuroscience, 2001). Please, define the excitatory/inhibitory nature of the synapses based on morphological features (not on their postsynaptic target), i.e., flattened vesicles and thin postsynaptic density for GABAergic synapses, whereas larger, round vesicles and thick postsynaptic density for glutamatergic synapses. Anyway, the ratio of excitatory and inhibitory synapses on dendrites and spines in the two sublamina would also give useful information about the synaptic organization of the human neocortical layer 1.

We are aware that not all terminals targeting on spines are excitatory, in turn it has been shown that not all terminals on shafts were inhibitory as long thought (Silver et al. 2003). However, as stated by the reviewer their abundancy on spines is rather low. At the moment it is rather unclear which functional impact inhibitory terminals on spines have, despite a local inhibition (see for example Kubota et al. eLife 2015), and thus their role is rather speculative since excitatory synapses are the predominant class on dendritic spines. As already stated above the ratio of excitatory vs. inhibitory terminals is between 10-15% and not significantly different between the two sublaminae. We are willing to add this in the results section (see in the improved version of the manuscript).

(2) About the glial coverage: Please, specify how glial elements were determined. What were the morphological features specific to astroglial processes? In Figure 5, how could we know whether the glial element marked by green is not a spine neck? The lack of morphological features specific to glial processes makes this analysis weak. The most accurate would be to make it with the aid of GFAP staining. I know this is not possible with your existing data, but at least, provide information on how glial processes were identified.

We used the criteria first described by Peters et al. (1991) and Ventura and Harris (1999) identifying astrocytic profiles by their irregular stellate shape, relatively clear cytoplasm, numerous glycogen granules and bundles of intermediate filaments. After more than 20 years of structural investigations, we hope that the reviewers will believe us that we can identify astrocytic processes at the high-resolution TEM level. In some of our publications (Rollenhagen et al. 2007; 2015; 2018; Yakoubi et al. 2019a) we have used glutamine synthetase pre-embedding immunhistochemistry to identify astrocytic processes, but a disadvantage of this method is the reduction of the ultrastructural preservation of the tissue. We have included the criteria to identify astrocytic processes of glial coverage in our manuscript together with the two citations (see improved version of the manuscript).

(3) The authors state that the total number of SVs was very variable. How was the distribution of the number of SVs? Homogenous distribution suggests that different types of synapses cannot be distinguished based on their morphological features, whereas distribution with more than one peak would suggest that different types of synapses are present in L1, and that they can be differentiated by their appearance (number of SVs, for example). This might be also related to the type of synapse (i.e., excitatory or inhibitory). The same applies to the number of RP and resting pool SVs.

To look for differences in structural and synaptic parameters that can further classify synaptic boutons we have performed a hierarchical cluster and multivariance analysis. However, it turned out that according to structural and functional parameters no further classification into subtypes could be done.

(4) The authors should check and review extensively for improvements to the use of English. The Results and Discussion sections contain many sentences which are not easy to understand. They have either a too complicated structure, or they are incomplete and hard to follow. Few examples: "The RRP/PreAZ at p20 nm criterium was on average 19.05 {plus minus} 17.23 SVs (L1a: 25.04 {plus minus} 21.09 SVs and L1b: 13.07 {plus minus} 13.87SVs) and thus nearly 2-fold larger for L1a." If you take out the parenthesis, the sentence has no meaning. "The majority of SBs in L1 of the human TLN had a single at most three AZs that could be of the non-perforated macular or perforated type comparable with results for other layers in the human TLN but by ~1.5-fold larger than in rodent and non-human primates." Rephrase these types of sentences, please.

We partially agree with the reviewer. We have improved our manuscript by rephrasing and shortening sentences.

Other suggestions:

(1) Put the synaptic density part after the description of the neuronal and synaptic composition part, it is more logical this way (i.e., first qualitative description, the distinction between sublayers, then quantitative data). Please write down in the description of the neuronal and synaptic composition part how L1a and L1b were differentiated (see also my comment on Figure 1).

We agree with the reviewer and did the change according to the suggestion. For a better understanding, we have also expanded the neuronal and synaptic description of the two sublaminae in L1.

(2) Introduce a list of abbreviations at the beginning, that would help.

It is quite unusual to provide a list of abbreviations in eLife. However, when used first the full meaning of the abbreviations is now given.

(3) What is cleft width? Usually, it refers to the distance between the pre- and the postsynaptic membrane, but here, I think it refers to the size (diameter) of the active zone. Please, clarify in the Result section (as it appears earlier than the Methods section, where it is explained). I would probably use the expression "synaptic cleft size" instead of "synaptic cleft width" to avoid misunderstanding.

We thank the reviewer for the suggestion and used synaptic cleft size for better clarity and have transferred the sentence from the Material and Methods to the Results section.

(4) The description of the different SVs (RRP, RP, etc.) is not clear in lines 236-242. What does it mean, that RRP vesicles are located {less than or equal to}10 nm and {less than or equal to}20 nm from the active zone? Explain, why the two different distance criteria were used. Furthermore, how were the vesicles located at p20-p60 defined? Why were these vesicles not considered in the determination of the different pools?

As stated in the public review to the reviewers concern we have tried to define a morphological correlate to the three functionally defined pools. After thorough discussions, with leading scientists in the field of synaptic neuroscience we have decided to use the distance of individual vesicles from the PreAZ and sort vesicles upon these criteria. One can argue that this approach is random, however, these distance criteria were described by Rizzoli and Betz (2004, 2005) and Denker and Rizzoli (2010). As also stated in the public review there is still a controversial discussion whether only docked or omega-shaped SVs constitute the RRP. We decided that also those very close within 10 and 20 nm away from the PreAZ, which is less than a SV diameter may also contribute to the RRP since it was shown that SVs are quite mobile.

(5) Please, explain how the number of docked vesicles can be 3x larger in L1b, than the number of vesicles located at p10? Docked vesicles are the closest (with the membrane touching the PreAZ)... if this comes from the fact that another pool of boutons was used for the EM tomography analysis, then the entire pool of boutons analyzed, then it means that the selection of boutons for the EM tomography is highly biased. This also implies that EM tomography data are most probably not valid for the entire L1b. The difference might also come from the different ratios of dendrite/spine synapses included in the two different analyses. In this case, it would be helpful to distinguish between synapses terminating on dendrites/spines and analyse them separately (same as for inhibitory/excitatory, which is not exactly the same as dendrite/spine!). Different n numbers of synapses are given in the text (n=25, 25, 25 25) and in Table 2 (n=91, 98, 87, and 84) for the analysis of the docked vesicles, please, correct this.

This is a correct value and thus there is a nearly 3-fold difference. The TEM tomography was carried out on the same blocks that have been used for our 3D-volume reconstructions. To carry out TEM tomography we had to use thicker sections (250 nm) to look for complete SBs as we also did in our serial sections, but of course, we could not quantify the same SBs. The completeness of SBs was one of our main criteria to reconstruct structural and synaptic parameters. The second was that the synaptic cleft was cut perpendicular. Only SBs that met these criteria were chosen for further quantitative analysis. In this respect we are of course biased in both methods.

Secondly, as already stated we did not quantify inhibitory terminals in serial sections. However, we did not find significant differences between shaft vs. spine synapses.

Finally, in Table 2 the total number of ‘docked’ SVs is given analyzed from the total number of SBs analyzed.

Discussion:

Please include the recent findings of human L1 neurons, including the "rosehip" cells in the L1 neuronal network, see Boldog et al., Nat Neurosci 2018. It would be also useful to consider in the discussion the human-specific cortical synchrony and integration phenomena derived from in vitro data (Mansvelder, Lein, Tamas, Wittner, Larkum, Huberfeld labs, etc.), and how the synaptic morphology can be related to these.

We thank the reviewer and include the reference in our chapter functional significance.

Figures and Tables:

Figure 1: In the legend, it is written that CR cells are marked by an asterisk, but on the figure it is marked by arrowheads. H: I would put the dashed line slightly lower, just above the two neuronal cell bodies. Now it looks like in the middle of the astrocytic layer. One of the asterisks marking the CR cell is not above the nucleus of that cell. I: the gabaergic neuron is outside of the framed area. I would delete the frame, anyway, the arrowheads and the asterisk are enough to show what the authors want to show.

We have changed the Figure according to the suggestions raised by the reviewer.

Figure 3: The transparent yellow is not visible. It is a bit disturbing that the contours of the boutons are not visible, I would make the transparent yellow stronger (less transparent). The SVs in green/magenta will be still visible.

We wanted to highlight the internal subelements of SBs and thus made the covering transparent but we think it is still visible.

Figure 6C: The data concerning other layers than L1 are most probably taken from other publications of the research group. One is cited (for L6), but not the others. Please correct this, or if not, then write this in the Results and Methods.

We changed the citation in the improved version of the manuscript. We overlooked that the values for L4 and L5 were already published in Schmuhl-Giesen et al. 2022.

Table 1: What does central and lateral cleft width mean in Table 1? Furthermore, please, give the name for abbreviations CV and IQR in Tables 1 and 2.

The measurements of the synaptic cleft are now described in detail in the Results section. We now have given the full names for CV and IQR in the legends of tables 1 and 2.

Supplemental Figures 1 and 2: Why Hu01 and Hu02 are twice? What is the difference? Based on the figure legend, it is L1a and L1b? If yes, please, indicate on the figure or in the legend.<br /> Supplemental Table 1: What is TLE in the case of Hu_04? If it is temporal lobe epilepsy, then why age at epilepsy onset is missing?

Yes, Hu01 and Hu02 were selected for both L1a and L1b in separate serial sections preparations each. We indicated this now in the figure legend. Concerning Hu_04, unfortunately we do not have any further information about the medical background of the patient.

Supplemental Table 1 (Patient table), that there are many abbreviations explained which do not appear in the table (lBAZ: Brivaracetam CBZ: Carbamazepine; CLB: Clobazam; ESL: Eslicarbazepin; GGL: Ganglioglioma, etc.), please check and correct.

We have removed the unnecessary abbreviations.

Other minor suggestions:

What is Pr? Please, give the name a first appearance (line 368).

We explained Pr (release probability) when used for the first time.

Give the name for t-LDT, please (lines 442-443).

We explained t-LTD (timing-dependent long-term depression) when used for the first time.

Typo in line 169: DCW instead of DCV (dense core vesicle), DCV is used in the figure legends.

We changed DCW to DCV.

Typo in line 190: Yokoubi instead of Yakoubi (reference).

We changed Yokoubi to Yakoubi.

Typo in line 237: Rizzoloi instead of Rizzoli (reference).

We changed Rizzoloi to Rizzoli.

Line 229-230: One reference is not inserted properly - Piccolo and Bassoon.

The reference of Schoch and Gundelfinger and Murkherjee to the build-up of the active zone and the role of DCV containing Piccolo and Bassoon are properly cited in the text.

Typo in line 398: exit instead of exist.

Corrected

Typo in line 700: Reynolds (1063) instead of 1963.

Corrected

Reviewer #2 (Recommendations for the authors):

Abstract:

The last sentence seems far-fetched, and unrelated to the manuscript. How mostly single active zone boutons can "mediate, integrate and synchronize contextual and cross-modal information, enabling flexible and state-dependent processing of feedforward sensory inputs from other layers of the cortical column"? Which of the anatomical findings of the manuscript led to these conclusions?

According to the review by Schuman et al. (2021) layer 1 is regarded as a layer that mediate, integrate and synchronize contextual and cross-modal information, enabling flexible and state-dependent processing of feedforward sensory inputs from other layers of the cortical column to which the structural quantitative 3D- models of SBs contribute since they are an integral element connecting neurons and building networks.

I am also puzzled by the authors' statement in more than one place of the manuscript that "L1a can be characterized as a predominantly astrocytic sublamina". If the L1 contains the lowest measured volume ratio of glial processes (Figure 6), then this description does not seem to hold. Please rephrase.

The reviewer is right and we rephrased the sentences for more clarity in the improved version of our manuscript.

Results:

The authors find large inter-patient variability in the synapse density at L1, which raises the issue of what were the criteria to include certain patients in the analyses. Apparently, these are different from the ones analysed in their previous papers, and all the provided parameters were different (sex, age, medication, onset of epilepsy), and any of them can result in altered synapse density.

First, we have not used all patients for this study. Secondly, it was not possible to use all patients for all six layers.

It would be useful to add a panel for Figure 1 with synapse density across the different layers, as they provide this data in the Discussion.

We implemented a Supplementary Table 1 with the synaptic density values over all layers compared in the Discussion.

I cannot find Source Data 1 in the manuscript although it is referred to in more than 1 place (e.g. page 5 line 100).

Source data were uploaded when our manuscript was submitted directly to eLife as Supplemental Material. However, as stated by bioRxiv ‘any Supplemental Materials associated with this manuscript have not been transferred to bioRxiv to avoid the posting of potentially sensitive information’ all source data have not been uploaded to the preprint server.

Page 5 line 100 the correct value is 7.3*107 or rather 108?

We corrected the value in the improved version of the MS.

It would be nice to put the synapse density values into context by comparing them to e.g. mouse, rat, or monkey data.

Since we are working on the human temporal lobe neocortex we avoided to compare those data with those estimated in experimental animals. In addition as discussed by DeFelipe et al. (1999) different methods were used to quantify synaptic density in experimental animals so these results are difficult to compare.

Page 5 Line 117 CR-cells stands for Cayal-Retzius cells?

CR-cells is the abbreviation for Cajal-Retzius cells.

Page 6 Line 146 repeated sentence.

We deleted the repeated sentence.

Page 7 Line 154 "file-scale TEM" ??

We replaced file-scale by fine-scale.

Page 7 Line 164 "GABAergic synapses identified by the smaller more spherical SVs". With this fixation condition, GABAergic vesicles are more ovoid than glutamatergic ones. What were the criteria to distinguish them?

To our knowledge in meanwhile numerous publications using the same fixation inhibitory terminals contain more spherical and smaller and not roundish synaptic vesicles and showed no clear prominent PSDs as described in our paper. We have addressed that more clearly in the results section of the improved version of the MS.

Page 8 line 197 "The majority (~98%) of SBs in L1a and L1b had only a single (Figures 2C-E, 3A-C, E) at most two or three AZs" is in striking contrast with the other statement from page 7 Line 163 "Numerous SBs in both sublaminae were seen to establish either two or three synaptic contacts on the same spine or dendrite". Which of these statements is valid? Please provide exact quantification for this statement and decide which one is true.

It is true that the majority of synaptic boutons had a single active zone. However, for example on a spine not only a single but also two or three SBs can be found. We have rephrased this sentence for more clarity.

Page 9 Line 206 "L1 AZs did not show a large variability in size as indicated by the low SD, CV, and variance (Table 1)" Is this inter-patient variance of mean values? As in Supplementary Figure 1, both the SBs volume and PreAZ area show large variability in a given patient sample. Only the inter-patient variability of mean values seems low. Please state it clearly throughout the MS for other datasets as well.

For clarity concerning the variability between patients and structural parameters we have generated box plots (Suppl. Figures 1 and 2).

Page 9 Line 208 data is on Figure 5A and not 8A.

We thank the reviewer and corrected the citation of the Figure

Page 12 Line 295 how can the number of docked vesicles for L1b be larger than the one measured by the perimeter p10 nm? This later should contain the docked and PreAZ membrane proximal pool as well. This difference is even larger if we assume, that at EM tomography only partial AZs were analysed in a 200 nm thick section, not the entire AZ as for the perimeter measurement. Can the authors provide density estimates by dividing the docked / p10 nm vesicle numbers with the AZ area and comparing them?

This is a result comparing both methods. To the second concern: As stated in the text only synaptic boutons were the active zone can be followed from the beginning to its end and were the synaptic cleft was cut perpendicular were included in the TEM tomography sample as we also did in our 3D-volume reconstructions.

Methods:

Page 25 Line 624 While the PSD area can be equivocally measured, due to the dense appearance of the PSD on the EM images, the PreAZ is more difficult to outline due to lack of evident anatomical markers except the synaptic cleft (the dense material is much thinner). That is why in many publications the PreAZ area is considered to be identical to the PSD area. What are the anatomical criteria used here for the PreAZ? Why do the authors correct the PSD area, which is easy to measure with the PreAZ area that is much less certain to outline?

As stated in material and Methods both the pre- and postsynaptic densities are not defined by placing a closed contour in both densities because one can’t be certain that the dense accumulation of particles defining both areas since the impregnation (staining) and contrast of both structures critically depends on the uranyl and lead staining which could led to misinterpretation due to different staining results. That’s why we have drawn a contour line from the beginning to the end of the presynaptic density and extrapolated that for the postsynaptic density (for details see Material and Methods). In our samples both the pre- and postsynaptic densities were always clearly visible in those boutons further analyze.

Page 26 Line 640 vesicle density measurement: All the synaptic vesicles that are in the 50 nm thick section in their entirety are missed, and there are methods based on EM tomography to correct these estimations. One can not assume, that the error caused by "double counts" of vesicles cancels for the lost ones. There are stereological methods to estimate both types of error please include them and correct the values.

We would like to point out that the whole body of our work to structural analysis of vesicle pools is based on image data stemming from transmission electron microscopy (TEM) generating a projection of the entire volume of the ultra-thin section and NOT from scanning electron microscopy (SEM) where only a small volume close to the surface of the section would be captured. Operating in TEM mode ensures that no vesicle is missed only because it is embedded in its entirety in the section as postulated by the reviewer. Hence, EM tomography, which is basically a TEM operating from different incident angles in relation to the specimen or section, does not provide any advantage in detecting these vesicles. It does, however, help to better position a 3D object within the section volume itself and therefore allows to detect objects that could overlap from one viewing angle by using another angle. As the average vesicle diameter is of similar size compared to the section thickness, the possibility of a complete overlap to happen, however, is almost zero. And as we only count clear ring-like structures, a stereological correction factor calculated according to Abercrombie (1946) would underestimate real counts (see also Saetzler et al. 2002). If there is, however, relevant literature on "methods based on EM tomography" and "stereological methods to estimate both types of error" (over- and underestimates) that we are missing out on, we would appreciate the reviewer providing us with the corresponding references so that we can include such calculations in our paper.

Page 27 Line 664 and 665 "sections" are still tissue blocks, as sectioning comes after if the process is correctly written. Please correct.

We have corrected this according to the reviewer’s comment.

Page 43 Figure 4 D Data for L1b is missing, only the correlation line is visible.

Corrected in a new Figure.

Page 44 Figure 5 C arrowheads are in the correct places? Some of them do not seem to point to the edge of the synapse.

We carefully checked the Figure and adjusted the arrowheads.

Figure 5 E lower arrowhead labels something, that is difficult to identify but does not seem to be a vesicle.

We agree with the reviewer on this point and changed the figure accordingly.

Figure 5 F, the upper vesicle is at least 10 nm apart from the PreAZ membrane. Did the authors consider it as docked (indicated with arrowhead, according to the legend it labels docked vesicles)?

We agree with the reviewer on this point and changed the figure accordingly.

Page 45 Figure 6 B one of the 2 synaptic boutons (sb), sb2 has a tangential active zone that precludes the identification of the pre- and post-synaptic membranes, still 2 "docked vesicles" are labeled. How were they classified as docked? Please remove these tangential synapses from the dataset, as membranes can not be identified.

The reviewer is right that the active zone is tangentially cut, however, the two vesicles are associated with the AZ. In addition, we did not use this AZ for vesicle data analysis.

Page 46 Line 1124 interneuron axon labelled in green not brown.

Corrected as suggested by the reviewer.

Line 1129 SStC is missing.

Changed according to the reviewer’s comment.

Page 48 Table 2 Number of docked vesicles Median values are rounded to integer values? If yes why?

The statistic package used rounded to the given values.

Page 51 Supplementary Table 1 Hu_04 Histopathology, what does TLE stands for?

TLE: temporal lobe epilepsy. We included the abbreviation in the legend of Supplementary Table1, that is now table 2.

Reviewer #3 (Recommendations for the authors):

(1) Reanalysis of astrocytic coverage based on the % of synapses that form tripartite synapses.

We have reanalyzed the data concerning this point (new Figure 6D).

(2) Segregation of excitatory and inhibitory synapses.

We have now included a paragraph in our results section to distinguish between excitatory and inhibitory synapses.

(3) Better explanation of the limits of the study to assess functional parameters.

We disagree with the reviewer on this point and have not included an explanation concerning the limits of this study.

Author response:

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

eLife Assessment 

This useful study uses high-field fMRI to test the hypothesized involvement of subcortical structure, particularly the striatum, in WM updating. It overcomes limitations in prior work by applying high-field imaging with a more precise definition of ROIs. Thus, the empirical observations are of use to specialists interested in working memory gating or the reference back task specifically. However, evidence to support the broader implications, including working memory gating as a construct, is incomplete and limited by the ambiguities in this task and its connection to theory. 

We would like to express our gratitude to the editor and the reviewers for their time and effort in providing insightful and valuable comments. We greatly value the critical perspective on the relationship between fMRI contrasts and the PBWM model. We hope to have addressed all the last critical points and changed the manuscript according to the reviewers’ suggestions. Furthermore, we would like to point out that the behavioral results section was edited, as a double-check of the results section revealed some erroneous descriptive statistics.

Public Reviews:

Reviewer #1:

Summary: 

Trutti and colleagues used 7T fMRI to identify brain regions involved in subprocesses of updating the content of working memory. Contrary to past theoretical and empirical claims that the striatum serves a gating function when new information is to be entered into working memory, the relevant contrast during a reference-back task did not reveal significant subcortical activation. Instead, the experiment provided support for the role of subcortical (and cortical) regions in other subprocesses. 

Strengths: 

The use of high-field imaging optimized for subcortical regions in conjunction with the theory-driven experimental design mapped well to the focus on a hypothetical striatal gating mechanism. 

Consideration of multiple subprocesses and the transparent way of identifying these, summarized in a table, will make it easy for future studies to replicate and extend the present experiment.   

Weaknesses: 

The reference-back paradigm seems to only require holding a single letter in working memory (X or O; Figure 1). It remains unclear how such low demand on working memory influences associated fMRI updating responses. It is also not clear whether reference-switch trials with 'same' response truly tax working-memory updating (and gate opening), as the working-memory content/representation does not need to be updated in this case. These potential design issues, together with the rather low number of experimental trials, raise concerns about the demonstrated absence of evidence for striatal gate opening. 

We acknowledge that a limitation of our study is that the task involved relatively low working memory demands. It remains to be clarified whether the same neural mechanisms would be engaged under a higher working memory load, and this is an important consideration for future research.

We also fully agree that it is uncertain whether reference-switch trials requiring a ‘same’ (or ‘match’ ) response truly engage working memory updating (or gate opening), as the working memory content or representation does not need to be altered in these cases. This concern is addressed in detail in the discussion section titled “No Support for Striatal Gate Opening” (see second paragraph).

Regarding our references to dopamine, we completely agree with the reviewer about the speculative nature of these discussions. In response, we thoroughly reviewed the manuscript and made revisions where necessary to ensure that we consistently emphasize the speculative nature of our commentary on dopamine and dopaminergic pathways.

Finally, we acknowledge the concerns about the design and the relatively low number of trials. However, our fMRI analyses of other reference-back task contrasts did reveal activity in the striatum and other subcortical ROIs. This suggests that our scanning protocol and task design are sufficiently sensitive to detect striatal activity, even with the limited number of trials.

The authors provide a motivation for their multi-step approach to fMRI analyses. Still, the three subsections of fMRI results (3.2.1; 3.2.2; 3.3.3) for 4 subprocesses each (gate opening, gate closing, substitution, updating mode) made the Results section complex and it was not always easy to understand why some but not other approaches revealed significant effects (as the midbrain in gate opening). 

We thank the reviewer for this important remark and the opportunity to clarify our approach. We conducted whole-brain general linear models (GLMs) to generate a comprehensive wholebrain map of brain activity for each contrast. However, the whole-brain statistical parametric mappings (SPMs) involve data smoothing, which–while improving signal detection–reduces spatial precision. This is especially problematic in smaller or closely adjacent regions, where spatial blurring can merge distinct activations or make localized signals appear more widespread.

Additionally, the statistical thresholds in whole-brain analyses may detect weak or borderline significant effects, whereas ROI-wise GLMs, which assume uniform behavior across the entire region, may miss the same effects if the signal is weak or inconsistent across the ROI.

Since our primary focus was on the subcortex, we relied more heavily on ROI-wise GLMs, which were limited to subcortical regions. We prioritized findings that were supported by either the ROI-wise GLMs or by both GLM analyses. For instance, the midbrain activations found in our whole-brain analysis but not in the ROI analysis may result from smoothing (where activation from neighboring regions spreads into midbrain voxels) or from functional heterogeneity within the ROI, which can obscure localized activations when averaged in the ROI-wise GLMs. Inferences from each GLM approach, along with their discrepancies, are discussed for each contrast throughout the discussion, with additional details on the clusterbased ROI analysis in the discussion section titled “Dopaminergic involvement in working memory substitution” (see third paragraph).

We acknowledge that the results section may seem complex, and we apologize for any inconvenience this may cause.

Reviewer #2:

Summary: 

The study reported by Trutti et al. uses high-field fMRI to test the hypothesized involvement of subcortical structure, particularly striatum, in WM updating. Specifically, participants were scanned while performing the Reference Back task (e.g., Rac-Lubashevsky and Kessler, 2016), which tests constructs like working memory gate opening and closing and substitution. While striatal activation was involved in substitution, it was not observed in gate opening. This observation is cited as a challenge to cortico-striatal models of WM gating, like PBWM (Frank and O'Reilly, 2005). 

Strengths: 

While there have been prior fMRI studies of the reference back task (Nir-Cohen et al., 2020), the present study overcomes limitations in prior work, particularly with regard to subcortical structures, by applying high-field imaging with a more precise definition of ROIs. And, the fMRI methods are careful and rigorous, overall. Thus, the empirical observations here are useful and will be of interest to specialists interested in working memory gating or the reference back task specifically. 

Weaknesses: 

I am less persuaded by the more provocative points regarding the challenge it presents to models like PBWM, made in several places by the paper. As detailed below, issues with conceptual clarity of the main constructs and their connection to models, like PBWM, along with some incomplete aspects of the results, make this stronger conclusion less compelling. 

(1) The relationship of the Nir-Cohen et al. (2020) task analysis of the reference back task, with its contrasts like gate opening and closing, and the predictions of PBWM is far from clear to me for several reasons. 

First, contrasts like gate opening and gate closing make strong finite state assumptions. As far as I know, this is not an assumption of PBWM, certainly not for gate opening. At a minimum, PBWM is default closed because of the tonic inhibition of cortico-thalamic dynamics by the globus pallidus. Indeed, this was even noted in the discussion of this paper, which seems to acknowledge this discrepancy, but then goes on to conclude that they have challenged the PBWM model anyway.  

We thank the reviewer for this remark and agree that the reference-back task contrasts do not perfectly align with the predictions of the PBWM model. In the discussion section "No support for striatal gate opening," we note that our data support the PBWM model by emphasizing the central role of the basal ganglia in working memory processes. However, we acknowledge that it may not have been sufficiently clear in the manuscript that the way the reference-back task is operationalised does not allow for a precise test of the PBWM's gating predictions. To address this, we have revised the manuscript to shift focus away from framing it as a direct challenge to the PBWM model. Below, some edits are highlighted.

‘This contrasts with the findings of Nir-Cohen et al. (2020) and raises questions about the relationship between the gate opening process in the reference back task and the indirect striatal gating mechanism described in the PBWM model (Frank et al., 2001; Hazy et al., 2007; O’Reilly & Frank, 2006) and other neurocomputational theories (Hazy et al., 2007; Jongkees, 2020). According to these models, a dopaminergic signal in the striatum is required to trigger gating. Although the orthogonal contrasts in the referenceback task are intended to isolate working memory subprocesses inspired by models of working memory, the two gating contrasts do not fully capture the gating mechanism as originally proposed in neurocomputational models (Frank et al., 2001; Hazy et al., 2007; O’Reilly & Frank, 2006).’ (line 721-730)

‘Another explanation for the lack of enhanced striatal activity in gate opening challenges the conceptualization of the gating mechanism in the reference-back task, which does not accurately map onto the PBWM predictions.’ (line 746)

‘Moreover, despite the lack of striatal involvement during gate opening, our findings do not rule out the possibility that the PBWM model's predictions about striatal gating in working memory are correct, given the misalignment between the gate opening contrast and the PBWM’s proposal regarding striatal gating. It remains unclear whether the absence of striatal activation during gate opening trials is specific to low-demand tasks, like the reference-back task, which does not require as much gating compared to high working memory-demand tasks involving preparation for updating. Or whether the gate opening contrast does not sufficiently capture the PBWM proposed gating mechanism. Further investigation is needed to determine whether (dopamine-driven) striatal gating occurs in high-demand working memory tasks, where the gating process plays a more critical role.’

Second, as far as I know, PBWM emphasizes go/no-go processes around constructs of input- and output-gating, rather than state shifts between gate opening and closing. While this relationship is less clear in reference back, substituting task-relevant items into working memory does appear to be an example of input gating, as modeled by PBWM. Thus, it is not clear to me why the substitution contrast would not be more of a test of input gating than the gate opening contrast, which requires assumptions that are not clear are required by the model, as noted above. 

We fully agree with the reviewer, which is why we proposed that neural mechanisms involving the midbrain and striatum are more likely to be observed in the substitution contrast rather than the gate opening contrast.

Third, PBWM relies on striatal mechanisms to solve the problem of selective gating, inputting, or outputting items in memory while also holding on to others. Selective gating contrasts with global gating, in which everything in memory is gated or nothing. The reference back task is a test of global gating. It is an important distinction because non-striatal mechanisms that can solve global gating, cannot solve selective gating. Indeed, this limitation of non-striatal mechanisms was the rationale for PBWM adding striatum. The connectivity of the striatum with the cortex permits this selectivity. It is not clear that the reference back task tests these selective demands in the first place. That limitation in this task was the rationale behind the recent Rac-Lubashevsky and Frank (2022) paper using the reference back 2 procedure that modifies the original reference back for selective gating. 

We thank the reviewer for highlighting this excellent reference. We believe it holds exciting potential for future high-field fMRI studies that explore the neural mechanisms underlying selective gating.

So, if the primary contribution of the paper is to test PBWM, as suggested by the first line of the abstract, then it is not clear that the reference back task in general, or the gate opening contrast in particular, is the best test of these predictions. Other contrasts (substitution), or indeed, tasks (reference back 2) would have been better suited. 

We agree with the reviewer that the gate opening contrast may not be the optimal test for the PBWM model predictions. However, previous studies have found evidence of striatal gateopening mechanisms using the reference-back task, which cannot be overlooked. We hypothesized that striatal mechanisms are likely active only when working memory content requires replacement, as seen in the substitution contrast in line with the PBWM model. Additionally, the reference-back 2 task (Rac-Lubashevsky & Frank, 2021) had not yet been published when we began data collection. Exploring this task in future studies, particularly with a 7 T fMRI protocol optimized for subcortical regions, would be an exciting avenue for further investigation.

Finally, in response to the reviewer’s remark, we have revised the abstract to remove the emphasis on challenging the PBWM model.

(2) In general, observations of univariate activity in the striatum have been notoriously variable in the context of WM. Indeed, Chatham et al. (2014) who tested working memory output gating - notably in a direct test of the predictions of PBWM - noted this variability. They too did not observe univariate activation in the striatum associated with selective output gating. Rather they found evidence of increased connectivity between the striatum and cortex during selective output gating. They argued that one account of this difference is that striatal gating dynamics emerge from the balance between the firing of both Go and NoGo cell populations that decide whether to gate or not. It is not always clear how this balance should relate to univariate activation in the striatum. Thus, the present study might also test cortico-striatal connectivity, rather than relying exclusively on univariate activation, in their test of striatal involvement in these WM constructs. 

We appreciate the reviewer’s insightful observation regarding the variability of univariate activity in the striatum, particularly in the context of working memory and the challenges noted by Chatham et al. (2014). We agree that striatal gating dynamics likely reflect a balance between Go and NoGo cell populations, which may not always manifest in univariate activation alone. In line with the reviewer’s suggestion, examining cortico-striatal connectivity could provide a more comprehensive understanding of striatal involvement in working memory processes, particularly selective gating.

While our current study focused primarily on univariate activity, we recognize the importance of connectivity-based approaches and plan to incorporate functional connectivity analyses in future studies to further explore these dynamics. Such an approach, especially when combined with ultra-high-field fMRI, may offer valuable insights into the interaction between the striatum and cortex during working memory tasks.

(3) It is concerning that there was no behavioral cost for comparison switch vs. repeat trials. This differs from with prior observations from the reference back (e.g., Nir-Cohen et al., 2020), and in general, is odd given the task switch/cue interpretation component. This failure to observe a basic behavioral effect raises a concern about how participants approached this task and how that might differ from prior reports of the reference back. If they were taking an unusual strategy, it further complicates the interpretation of these results and the implications they hold for theory. 

We understand the reviewer’s concern regarding the lack of behavioral response time costs for comparison switch versus repeat trials, which does indeed differ from previous findings in studies such as Nir-Cohen et al. (2020). It is possible that this results from our fMRI task design, such as increased inter-trial intervals compared to behavioral studies. While this is certainly a point of concern, we believe that the neural data still provide valuable insights into the mechanisms underlying working memory gating despite the absence of a clear behavioral effect.

In future studies, we aim to increase the number of trials and more closely align our task design with previous studies to mitigate this issue. We agree that further investigation is necessary to ensure the robustness of these effects and their theoretical implications.

In summary, the present observations are useful, particularly for those interested in the reference back task. For example, they might call into question verbal theories and task analyses of the reference back task that tie constructs like gate-opening to striatal mechanisms. However, given the ambiguities noted above, the broader implications for models like PBWM, or indeed, other models of working memory gating, are less clear.

Author response:

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

Weaknesses (Reviewer 1):

The role of Fgf signaling in gliogenesis and Foxg1 in neurogenesis is well known. It is not clear if Fgf18 is a direct target of Foxg1.

We agree with the reviewer- Fgf signaling is an established pro-gliogenic pathway (Duong et al 2019) and Foxg1 overexpression is known to promote neurogenesis in cultured neural stem cells (Branacaccio et al 2019). Our study links these two mechanisms, as the Reviewer has summarized: (a) we demonstrate that FOXG1 works via modulating Fgf signaling cell-autonomously within progenitors by regulating the levels of Fgfr3. (b) Loss of Foxg1 in postmitotic neurons results in the upregulation of Fgf ligand expression (possibly via indirect mechanisms) and this non-cell autonomously increases Fgf signaling in progenitors_. Our study is entirely performed _in vivo.

Revision: We have revised the manuscript to reflect that Fgf18 may be an indirect target of FOXG1 in postmitotic neurons.

Weaknesses (Reviewer 2):

It wasn't clear to me why the authors chose postnatal day 14 to examine the effects of Foxg1 deletion at E15 - this is a long time window, giving time for indirect consequences of Foxg1 deletion to influence development and thereby potentially complicating the interpretation of findings. For example, the authors show that there is no increased proliferation of astrocytes or death of neurons lacking Foxg1 shortly after cre-mediated deletion, but it remains formally possible (if perhaps unlikely) that these processes could be affected later during the time window. The rationale underlying the choice of this time point should be explained.

I don't agree with the statement in the very last sentence of the results section that "neurogenesis is not possible in the absence of [Foxg1]" as there are multiple reports in the literature demonstrating the presence of neurons in Foxg1-/- mice (eg: Xuan et al., 1995; Hanashima et al., 2002, Martynoga et al., 2005, Muzio and Mallamaci 2005). Perhaps the statement refers specifically to late-born cortical neurons. This point also arises in the discussion section.

Revisions:

(a) We have revised the manuscript to explain why we chose postnatal day 14 to examine the effects of Foxg1 deletion at E15.

●  We have examined the transcriptomic dysregulation after Foxg1 deletion at E17.5, which is a reasonable period to identify potential direct targets. Furthermore, FOXG1 occupies the Fgfr3 locus in ChIP-seq performed at E15.5. Together, these support the interpretation that Fgfr3 is a direct target of Foxg1.

● As the Reviewer notes, we have investigated the possibility of increased proliferation of astrocytes and death of neurons and found no evidence suggesting these phenomena occur in the 3 days after loss of Foxg1. Cortical neurons are postmitotic and differentiated by E18.5, the stage at which we examined CC3 staining and found no difference in cell death in control and mutants (Supplementary Figure S2C, C’). The majority of progenitors (PAX6+ve cells) that lose Foxg1 at E15.5 express the gliogenic transcription factor NFIA by E18.5 (Figure 2C, C’), but hardly any express intermediate (neurogenic) progenitor marker TBR2 (Supplementary Figure S2B, B’). It is therefore unlikely that neurons are born from Foxg1 mutant progenitors and then die at a later stage.

● The cellular consequences of loss of Foxg1 require additional time to detect e.g. it takes ~ 5 days for GFAP to be detected in astrocytes once they are born. The P14 timepoint permits the assessment of oligogenesis which begins after astrogliogenesis and therefore permits a comprehensive assessment of the lineage of E15.5 Foxg1 null progenitors.

(b) Thank you for pointing out that the last sentence of the results section implied (incorrectly) that ALL neurogenesis is not possible in the absence of Foxg1 We have modified this (and the discussion) to reflect that this applies to E14/15 progenitors and late-born cortical neurons.

Recommendations for the authors (Reviewer 2):

(c) We thank the reviewer for this suggestion. We will modify the schematic (Figure 7) to remove any ambiguity regarding Foxg1 expression.

Author response:

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

Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.

Reviewer #3 (Public review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

As in previous revisions, there remains concerning ambiguity in the methodology used for microbiota sequence analysis and it would be difficult to replicate the analysis in any meaningful way. In this revision, concerns about the rigor and reproducibility of this component of the manuscript have been increased. Readers should be cautious with interpretation of this data.

(1) In previous versions of the manuscript it would appear the correct bioproject accession was listed but, the actual link went to an unrelated project. The updated accession link appears to contain raw data; however, the authors state they used an Illumina HiSeq 2500. This would be an unusual choice for V3-V4 as it would not have read lengths long enough to overlap. Inspection of the first sample (SRR19164796) demonstrates that this is absolutely not the raw data, as there is a ~400 nt forward read, and a 0 length reverse read. All quality scores are set to 30. There is no logical way to go from HiSeq 2500 raw data and read lengths to what was uploaded to the SRA and it was certainly not described in the manuscript.

What we uploaded to the SRA was Contigs files for sample, we have modified the description on line 694.

(2) No multiple testing correction was applied to the microbiome data.

The alpha diversity indexes were tested using T-test and wilcox test, and we showed the result of T-test in FigureS1B. The p-values were corrected for multiple testing using the Benjamini-Hochberg method, we have modified the description on line 322.

---------

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

Public Reviews:

Reviewer #2 (Public Review):

Ma X. et al proposed that A. muciniphila was a key strain that promotes the proliferation and differentiation of intestinal stem cells through acting on the Wnt/β-catenin signaling pathway. They used various models, such as piglet model, mouse model and intestinal organoids to address how A. muciniphila and B. fragilis offer the protection against ETEC infection. They showed that FMT with fecal samples, A. muciniphila or B. fragilis protected piglets and/or mice from ETEC infection, and this protection is manifested as reduced intestinal inflammation/bacterial colonization, increased tight junction/Muc2 proteins, as well as proper Treg/Th17 cells. Additionally, they demonstrated that A. muciniphila protected basal-out and/or apical-out intestinal organoids against ETEC infection via Wnt signaling.

Comments on revised version:

Please add proper references to indicate the invasion of ETEC into organoids after 1 h of infection.

We have added references on line 211.

References:

Xiao K, Yang Y, Zhang Y, Lv QQ, Huang FF, Wang D, Zhao JC, Liu YL. 2022. Long-chain PUFA ameliorate enterotoxigenic Escherichia coli-induced intestinal inflammation and cell injury by modulating pyroptosis and necroptosis signaling pathways in porcine intestinal epithelial cells. Br. J. Nutr. 128(5):835-850.

Qian MQ, Zhou XC, Xu TT, Li M, Yang ZR, Han XY. 2023. Evaluation of Potential Probiotic Properties of Limosilactobacillus fermentum Derived from Piglet Feces and Influence on the Healthy and E. coli-Challenged Porcine Intestine. Microorganisms. 11(4).

Reviewer #3 (Public Review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

After an additional revision, the bioinformatics section of the methods has changed significantly from previous versions and now indicates a third sequencer was used instead: Ion S5 XL. Important parameters required to replicate analysis have still not been provided. Inspection of the SRA data indicates a mix of Illumina MiSeq and Illumina HiSeq 2500. It is now unclear which sequencing technology was used as authors have variably reported 4 different sequencers for these samples. Appropriate metadata was not provided in the SRA, although some groups may be inferred from sample names. These changing descriptions of the methodologies and ambiguity in making the data available create concerns about rigor of study and results.

Due to confusing the sequencing method of this experiment with other experiment samples, we apologize for the multiple incorrect modifications of the method description. We have modified the method for microbiome sequencing technology on line 304. The sequencing technology is Illumina HiSeq 2500. The SRA metadata can be viewed at https://www.ncbi.nlm.nih.gov/sra/PRJNA837047. The sample names ep1-6 and ef1-6 were correspond to the EP and EF groups, respectively.

Recommendations For the Authors:

As in the previous revision:

-provide important parameters required to replicate analysis

-ensure that reporting of sequencing technology is correct as data listed on SRA appears to be derived from Illumina sequencers, and was deposited indicating as such.

-update SRA metadata such that experimental groups are clear and match the nomenclature used in the manuscript (Particularly for samples which are labelled [A-Z][0-9]

- The multiple testing correction wasn’t applied.

-Due to confusing the sequencing method of this experiment with other experiment samples, we apologize for the multiple incorrect modifications of the method description. We have modified the method for microbiome sequencing technology on line 304. The sequencing technology is Illumina HiSeq 2500.

- The SRA metadata can be viewed at https://www.ncbi.nlm.nih.gov/sra/PRJNA837047. The sample names ep1-6 and ef1-6 were correspond to the EP and EF groups, respectively.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

The authors investigate the effects of aging on auditory system performance in understanding temporal fine structure (TFS), using both behavioral assessments and physiological recordings from the auditory periphery, specifically at the level of the auditory nerve. This dual approach aims to enhance understanding of the mechanisms underlying observed behavioral outcomes. The results indicate that aged animals exhibit deficits in behavioral tasks for distinguishing between harmonic and inharmonic sounds, which is a standard test for TFS coding. However, neural responses at the auditory nerve level do not show significant differences when compared to those in young, normal-hearing animals. The authors suggest that these behavioral deficits in aged animals are likely attributable to dysfunctions in the central auditory system, potentially as a consequence of aging. To further investigate this hypothesis, the study includes an animal group with selective synaptic loss between inner hair cells and auditory nerve fibers, a condition known as cochlear synaptopathy (CS). CS is a pathology associated with aging and is thought to be an early indicator of hearing impairment. Interestingly, animals with selective CS showed physiological and behavioral TFS coding similar to that of the young normal-hearing group, contrasting with the aged group's deficits. Despite histological evidence of significant synaptic loss in the CS group, the study concludes that CS does not appear to affect TFS coding, either behaviorally or physiologically.

We agree with the reviewer’s summary.

Strengths:

This study addresses a critical health concern, enhancing our understanding of mechanisms underlying age-related difficulties in speech intelligibility, even when audiometric thresholds are within normal limits. A major strength of this work is the comprehensive approach, integrating behavioral assessments, auditory nerve (AN) physiology, and histology within the same animal subjects. This approach enhances understanding of the mechanisms underlying the behavioral outcomes and provides confidence in the actual occurrence of synapse loss and its effects. The study carefully manages controlled conditions by including five distinct groups: young normal-hearing animals, aged animals, animals with CS induced through low and high doses, and a sham surgery group. This careful setup strengthens the study's reliability and allows for meaningful comparisons across conditions. Overall, the manuscript is well-structured, with clear and accessible writing that facilitates comprehension of complex concepts.

Weaknesses:

The stimulus and task employed in this study are very helpful for behavioral research, and using the same stimulus setup for physiology is advantageous for mechanistic comparisons. However, I have some concerns about the limitations in auditory nerve (AN) physiology. Due to practical constraints, it is not feasible to record from a large enough population of fibers that covers a full range of best frequencies (BFs) and spontaneous rates (SRs) within each animal. This raises questions about how representative the physiological data are for understanding the mechanism in behavioral data. I am curious about the authors' interpretation of how this stimulus setup might influence results compared to methods used by Kale and Heinz (2010), who adjusted harmonic frequencies based on the characteristic frequency (CF) of recorded units. While, the harmonic frequencies in this study are fixed across all CFs, meaning that many AN fibers may not be tuned closely to the stimulus frequencies.

We chose the stimuli for the AN recordings to be identical to the stimuli used in the behavioral evaluation of the perceptual sensitivity. Only with this approach can we directly compare the response of the population of AN fibres with perception measured in behaviour. We will address this more clearly in the revision.

If units are not responsive to the stimulus further clarification on detecting mistuning and phase locking to TFS effects within this setup would be valuable.

It is unclear to us what the reviewer alludes to. We ask to rephrase the question.

Given the limited number of units per condition-sometimes as few as three for certain conditions - I wonder if CF-dependent variability might impact the results of the AN data in this study and discussing this factor can help with better understanding the results. While the use of the same stimuli for both behavioral and physiological recordings is understandable, a discussion on how this choice affects interpretation would be beneficial. In addition a 60 dB stimulus could saturate high spontaneous rate (HSR) AN fibers, influencing neural coding and phase-locking to TFS. Potentially separating SR groups, could help address these issues and improve interpretive clarity.

In the discussion of a revised version of the manuscript, we will point out the pros and cons of using fixed-level stimuli that were not adjusted in frequency to the BF.

A deeper discussion on the role of fiber spontaneous rate could also enhance the study. How might considering SR groups affect AN results related to TFS coding? While some statistical measures are included in the supplement, a more detailed discussion in the main text could help in interpretation. We do not think that it will be necessary to conduct any statistical analysis in addition to that already reported in the supplement.

We will consider moving some supplementary information back into the main manuscript when revising.

Although Figure S2 indicates no change in median SR, the high-dose treatment group lacks LSR fibers, suggesting a different distribution based on SR for different animal groups, as seen in similar studies on other species. A histogram of these results would be informative, as LSR fiber loss with CS-whether induced by ouabain in gerbils or noise in other animals-is well documented (e.g., Furman et al., 2013).

We will add information on the distribution when revising.

Although ouabain effects on gerbils have been explored in previous studies, since these data already seems to be recorded for the animal in this study, a brief description of changes in auditory brainstem response (ABR) thresholds, wave 1 amplitudes, and tuning curves for animals with cochlear synaptopathy (CS) in this study would be beneficial. This would confirm that ouabain selectively affects synapses without impacting outer hair cells (OHCs). For aged animals, since ABR measurements were taken, comparing hearing differences between normal and aged groups could provide insights into the pathologies besides CS in aged animals. Additionally, examining subject variability in treatment effects on hearing and how this correlates with behavior and physiology would yield valuable insights. If limited space maybe a brief clarification or inclusion in supplementary could be good enough.

We do indeed have data on ABR amplitudes and the wave 1 growth functions but only in response to broadband clicks. For more frequency-specific information, mass-potential recordings are available, obtained before and after ouabain treatment. Regarding neural tuning, we did not obtain full frequency-threshold curves but do have bandwidths for response curves recorded close to threshold. We are in the process of analyzing all these data further and will consider how to best incorporate them into the manuscript, to address the reviewer’s concerns.

Another suggestion is to discuss the potential role of MOC efferent system and effect of anesthesia in reducing efferent effects in AN recordings. This is particularly relevant for aged animals, as CS might affect LSR fibers, potentially disrupting the medial olivocochlear (MOC) efferent pathway. Anesthesia could lessen MOC activity in both young and aged animals, potentially masking efferent effects that might be present in behavioral tasks. Young gerbils with functional efferent systems might perform better behaviorally, while aged gerbils with impaired MOC function due to CS might lack this advantage. A brief discussion on this aspect could potentially enhance mechanistic insights.

Our provisional response below will be integrated in similar form into the Discussion.

Olivocochlear efferent activity is a potential modulator of OHC gain (by medial olivocochlear neurons, MOC) and afferent activity (by lateral olivocochlear neurons, LOC). Beyond this general observation it is, however, difficult to speculate about its specific role in the TFS1 test, as almost nothing is known about efferent activity under naturalistic conditions in a behaving animal (reviewed by Lauer et al., 2022). We note, however, that efferent activity is believed to be reduced under general anesthesia (reviewed by Guinan, 2011, DOI 10.1007/978-1-4419-7070-1_3) and possibly abnormal in other ways, considering the potential top-down inputs to the efferent neurons from extensive brain networks (reviewed by Schofield, 2011, DOI 10.1007/978-1-4419-7070-1_9; Romero and Trussell, 2022, DOI: 10.1016/j.heares.2022.108516). Thus, it is reasonable to assume a reduced efferent influence in our auditory-nerve data, compared to the behavioral test situation. In contrast, we assume more comparable efferent influences in young-adult and old gerbils. It was recently shown that, despite age-related losses in both MOC and LOC cochlear innervation, this basically reflected the loss of efferent target structures (OHC and type-I afferents), with the surviving cochlear circuitry remaining largely normal (Steenken et al., 2024, DOI: 10.3389/fnsyn.2024.1422330). The main difference was an increased proportion of OHC without any efferent innervation, predominantly in low-frequency cochlear regions (Steenken et al., 2024). Such OHC are thus not under efferent control, and they are more numerous (about 10 – 30%) in old gerbils.

Lastly, although synapse counts did not differ between the low-dose treatment and NH I sham groups, separating these groups rather than combining them with the sham might reveal differences in behavior or AN results, particularly regarding the significance of differences between aged/treatment groups and the young normal-hearing group. For maximizing statistical power, we combined those groups in the statistical analysis. These two groups did not differ in synapse number and had quite similar ABR wave 1 growth functions.

Reviewer #2 (Public review):

Summary:

Using a gerbil model, the authors tested the hypothesis that loss of synapses between sensory hair cells and auditory nerve fibers (which may occur due to noise exposure or aging) affects behavioral discrimination of the rapid temporal fluctuations of sounds. In contrast to previous suggestions in the literature, their results do not support this hypothesis; young animals treated with a compound that reduces the number of synapses did not show impaired discrimination compared to controls. Additionally, their results from older animals showing impaired discrimination suggest that age-related changes aside from synaptopathy are responsible for the age-related decline in discrimination.

We agree with the reviewer’s summary.

Strengths:

(1) The rationale and hypothesis are well-motivated and clearly presented.

(2) The study was well conducted with strong methodology for the most part, and good experimental control. The combination of physiological and behavioral techniques is powerful and informative. Reducing synapse counts fairly directly using ouabain is a cleaner design than using noise exposure or age (as in other studies), since these latter modifiers have additional effects on auditory function.

(3) The study may have a considerable impact on the field. The findings could have important implications for our understanding of cochlear synaptopathy, one of the most highly researched and potentially impactful developments in hearing science in the past fifteen years.

Weaknesses:

(1) My main concern is that the stimuli may not have been appropriate for assessing neural temporal coding behaviorally. Human studies using the same task employed a filter center frequency that was (at least) 11 times the fundamental frequency (Marmel et al., 2015; Moore and Sek, 2009). Moore and Sek wrote: "the default (recommended) value of the centre frequency is 11F0." Here, the center frequency was only 4 or 8 times the fundamental frequency (4F0 or 8F0). Hence, relative to harmonic frequency, the harmonic spacing was considerably greater in the present study. By my calculations, the masking noise used in the present study was also considerably lower in level relative to the harmonic complex than that used in the human studies. These factors may have allowed the animals to perform the task using cues based on the pattern of activity across the neural array (excitation pattern cues), rather than cues related to temporal neural coding. The authors show that mean neural driven rate did not change with frequency shift, but I don't understand the relevance of this. It is the change in response of individual fibers with characteristic frequencies near the lowest audible harmonic that is important here.

The auditory filter bandwidth of the gerbil is about double that of human subjects. Because of this, the masking noise has a larger overall level than in the human studies in the filter. This precludes that the gerbils can use excitation patterns, especially in the condition with a center frequency of 1600 Hz and a fundamental of 200 Hz and in the condition with a center frequency of 3200 Hz and a fundamental of 400 Hz.

The case against excitation pattern cues needs to be better made in the Discussion. It could be that gerbil frequency selectivity is broad enough for this not to be an issue, but more detail needs to be provided to make this argument. The authors should consider what is the lowest audible harmonic in each case for their stimuli, given the level of each harmonic and the level of the pink noise. Even for the 8F0 center frequency, the lowest audible harmonic may be as low as the 4th (possibly even the 3rd). In human, harmonics are thought to be resolvable by the cochlea up to at least the 8th.

Because of the gerbil’s broader auditory filters, with the exception of the condition with center frequency of 1600 Hz and fundamental of 400 Hz harmonics are are not resolved. We will expand the topic of potential excitation pattern cues in the discussion of the revised version and add results on modeled excitation patterns to the supplement.

(2) The synapse reductions in the high ouabain and old groups were relatively small (mean of 19 synapses per hair cell compared to 23 in the young untreated group). In contrast, in some mouse models of the effects of noise exposure or age, a 50% reduction in synapses is observed, and in the human temporal bone study of Wu et al. (2021, https://doi.org/10.1523/JNEUROSCI.3238-20.2021) the age-related reduction in auditory nerve fibres was ~50% or greater for the highest age group across cochlear location. It could be simply that the synapse loss in the present study was too small to produce significant behavioral effects. Hence, although the authors provide evidence that in the gerbil model the age-related behavioral effects are not due to synaptopathy, this may not translate to other species (including human). This should be discussed in the manuscript.

Our provisional response below will be integrated in similar form into the Discussion.

The observed extent of age-related or noise-induced loss of type-I afferent synapses on IHC varies widely between species and studies. For example, in ageing CBA/CaJ mice, mean losses of between 20 and 50% of afferent synapses (depending on cochlear location and precise age) were reported (Sergeyenko et al., 2013, DOI: 10.1523/JNEUROSCI.1783-13.2013; Kobrina et al., 2020, DOI: 10.1016/j.neurobiolaging.2020.08.012). Humans showed more pronounced losses of peripheral axons, of 40–100%, again depending on cochlear location, precise age, and noise history (Wu et al., 2019, DOI: 10.1016/j.neuroscience.2018.07.053; 2021, DOI: 10.1523/JNEUROSCI.3238-20.2021). The age-related and induced synapse losses in our gerbils were in a more moderate range, around 20% (Steenken et al., 2021, DOI: 10.1016/j.neurobiolaging.2021.08.019; this study). Thus, it is possible that a more severe, induced synaptopathy would have resulted in behavioral deficits in young-adult gerbils. However, in the absence of additional noise or pharmacologically induced damage, our study provides strong evidence for other factors causing temporal processing problems with advancing age. Our 3-year-old gerbils are approximately comparable to a 60-year-old human (Castano-Gonzalez et al., 2024, DOI: 10.1016/j.heares.2024.108989) with beginning but not yet clinically relevant hearing loss (Hamann et al., 2002, DOI: 10.1016/S0378-5955(02)00454-9).

It would be informative to provide synapse counts separately for the animals who were tested behaviorally, to confirm that the pattern of loss across the group was the same as for the larger sample.

Yes, the pattern was the same for the subgroup of behaviorally tested animals. We will add this information to the revised version of the manuscript.

(3) The study was not pre-registered, and there was no a priori power calculation, so there is less confidence in replicability than could have been the case. Only three old animals were used in the behavioral study, which raises concerns about the reliability of comparisons involving this group.

The results for the three old subjects differed significantly from those of young subjects and young ouabain-treated subjects. This indicates a sufficient statistical power, since otherwise no significant differences would be observed.

Reviewer #3 (Public review):

This study is a part of the ongoing series of rigorous work from this group exploring neural coding deficits in the auditory nerve, and dissociating the effects of cochlear synaptopathy from other age-related deficits. They have previously shown no evidence of phase-locking deficits in the remaining auditory nerve fibers in quiet-aged gerbils. Here, they study the effects of aging on the perception and neural coding of temporal fine structure cues in the same Mongolian gerbil model.

They measure TFS coding in the auditory nerve using the TFS1 task which uses a combination of harmonic and tone-shifted inharmonic tones which differ primarily in their TFS cues (and not the envelope). They then follow this up with a behavioral paradigm using the TFS1 task in these gerbils. They test young normal hearing gerbils, aged gerbils, and young gerbils with cochlear synaptopathy induced using the neurotoxin ouabain to mimic synapse losses seen with age. In the behavioral paradigm, they find that aging is associated with decreased performance compared to the young gerbils, whereas young gerbils with similar levels of synapse loss do not show these deficits. When looking at the auditory nerve responses, they find no differences in neural coding of TFS cues across any of the groups.

However, aged gerbils show an increase in the representation of periodicity envelope cues (around f0) compared to young gerbils or those with induced synapse loss. The authors hence conclude that synapse loss by itself doesn't seem to be important for distinguishing TFS cues, and rather the behavioral deficits with age are likely having to do with the misrepresented envelope cues instead.

We agree with the reviewer’s summary.

The manuscript is well written, and the data presented are robust. Some of the points below will need to be considered while interpreting the results of the study, in its current form. These considerations are addressable if deemed necessary, with some additional analysis in future versions of the manuscript.

Spontaneous rates - Figure S2 shows no differences in median spontaneous rates across groups. But taking the median glosses over some of the nuances there. Ouabain (in the Bourien study) famously affects low spont rates first, and at a higher degree than median or high spont rates. It seems to be the case (qualitatively) in Figure S2 as well, with almost no units in the low spont region in the ouabain group, compared to the other groups. Looking at distributions within each spont rate category and comparing differences across the groups might reveal some of the underlying causes for these changes. Given that overall, the study reports that low-SR fibers had a higher ENV/TFS log-z-ratio, the distribution of these fibers across groups may reveal specific effects of TFS coding by group.

As the reviewer points out, our sample from the group treated with a high concentration of ouabain showed very few low-spontaneous-rate auditory-nerve fibers, as expected from previous work. However, this was also true, e.g., for our sample from sham-operated animals, and may thus well reflect a sampling bias. We are therefore reluctant to attach much significance to these data distributions. We will consider moving some supplementary information back into the main manuscript when revising.

Threshold shifts - It is unclear from the current version if the older gerbils have changes in hearing thresholds, and whether those changes may be affecting behavioral thresholds. The behavioral stimuli appear to have been presented at a fixed sound level for both young and aged gerbils, similar to the single unit recordings. Hence, age-related differences in behavior may have been due to changes in relative sensation level. Approaches such as using hearing thresholds as covariates in the analysis will help explore if older gerbils still show behavioral deficits.

Unfortunately, we did not obtain behavioral thresholds that could be used here. The ABR thresholds, although not directly comparable to behavioral thresholds, suggest that our old animals had at most a moderate threshold increase in quiet. Furthermore, we want to point out that the TFS 1 stimuli had an overall level of 68 dB SPL, and the pink noise masker would have increased the threshold more than expected from the moderate, age-related hearing loss in quiet. Thus, the masked thresholds for all gerbil groups are likely similar and should have no effect on the behavioral results.

Task learning in aged gerbils - It is unclear if the aged gerbils really learn the task well in two of the three TFS1 test conditions. The d' of 1 which is usually used as the criterion for learning was not reached in even the easiest condition for aged gerbils in all but one condition for the aged gerbils (Fig. 5H) and in that condition, there doesn't seem to be any age-related deficits in behavioral performance (Fig. 6B). Hence dissociating the inability to learn the task from the inability to perceive TFS 1 cues in those animals becomes challenging.

Even in the group of gerbils with the lowest sensitivity, for the condition 400/1600 the animals achieved a d’ of on average above 1. Furthermore, stimuli were well above threshold and audible, even when no discrimination could be observed. Finally, as explained in the methods, different stimulus conditions were interleaved in each session, providing stimuli that were easy to discriminate together with those being difficult to discriminate. This approach ensures that the gerbils were under stimulus control, meaning properly trained to perform the task. Thus, an inability to discriminate does not indicate a lack of proper training.

Increased representation of periodicity envelope in the AN - the mechanisms for increased representation of periodicity envelope cues is unclear. The authors point to some potential central mechanisms but given that these are recordings from the auditory nerve what central mechanisms these may be is unclear. If the authors are suggesting some form of efferent modulation only at the f0 frequency, no evidence for this is presented. It appears more likely that the enhancement may be due to outer hair cell dysfunction (widened tuning, distorted tonotopy). Given this increased envelope coding, the potential change in sensation level for the behavior (from the comment above), and no change in neural coding of TFS cues across any of the groups, a simpler interpretation may be -TFS coding is not affected in remaining auditory nerve fibers after age-related or ouabain induced synapse loss, but behavioral performance is affected by altered outer hair cell dysfunction with age.

A similar point is made by Reviewer #1. As indicated above, we do have limited data on neural bandwidths and will explore if these are sufficient to address the reviewers’ questions about potential, age-related changes in neural tuning in our sample. Previous work found no substantial OHC losses (Tarnowski et al., 1991, DOI: 10.1016/0378-5955(91)90142-V; Adams and Schulte, 1997, DOI: 10.1016/S0378-5955(96)00184-0; Steenken et al., 2024, DOI: 10.3389/fnsyn.2024.1422330) nor any deterioration in neural frequency tuning (Heeringa et al., 2020, DOI: 10.1523/JNEUROSCI.2784-18.2019), in quiet-aged gerbils of similar age as the ones used here.

Emerging evidence seems to suggest that cochlear synaptopathy and/or TFS encoding abilities might be reflected in listening effort rather than behavioral performance. Measuring some proxy of listening effort in these gerbils (like reaction time) to see if that has changed with synapse loss, especially in the young animals with induced synaptopathy, would make an interesting addition to explore perceptual deficits of TFS coding with synapse loss.

This is an interesting suggestion that we will explore in the revision of the manuscript. Reaction times were recorded for responses that can be used as a proxy for listening effort.

Author response:

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

Public Reviews:

Reviewer #2 (Public Review):

Summary:

This computational modeling study addresses the observation that variable observations are interpreted differently depending on how much uncertainty an agent expects from its environment. That is, the same mismatch between a stimulus and an expected stimulus would be less significant, and specifically would represent a smaller prediction error, in an environment with a high degree of variability than in one where observations have historically been similar to each other. The authors show that if two different classes of inhibitory interneurons, the PV and SST cells, (1) encode different aspects of a stimulus distribution and (2) act in different (divisive vs. subtractive) ways, and if (3) synaptic weights evolve in a way that causes the impact of certain inputs to balance the firing rates of the targets of those inputs, then pyramidal neurons in layer 2/3 of canonical cortical circuits can indeed encode uncertainty-modulated prediction errors. To achieve this result, SST neurons learn to represent the mean of a stimulus distribution and PV neurons its variance.

The impact of uncertainty on prediction errors in an understudied topic, and this study provides an intriguing and elegant new framework for how this impact could be achieved and what effects it could produce. The ideas here differ from past proposals about how neuronal firing represents uncertainty. The developed theory is accompanied by several predictions for future experimental testing, including the existence of different forms of coding by different subclasses of PV interneurons, which target different sets of SST interneurons (as well as pyramidal cells). The authors are able to point to some experimental observations that are at least consistent with their computational results. The simulations shown demonstrate that if we accept its assumptions, then the authors’ theory works very well: SSTs learn to represent the mean of a stimulus distribution, PVs learn to estimate its variance, firing rates of other model neurons scale as they should, and the level of uncertainty automatically tunes the learning rate, so that variable observations are less impactful in a high uncertainty setting.

Strengths:

The ideas in this work are novel and elegant, and they are instantiated in a progression of simulations that demonstrate the behavior of the circuit. The framework used by the authors is biologically plausible and matches some known biological data. The results attained, as well as the assumptions that go into the theory, provide several predictions for future experimental testing. The authors have taken into account earlier review comments to revise their paper in ways that enhance its clarity.

Weaknesses:

One weakness could be that the proposed theory does rely on a fairly large number of assumptions. However, there is at least some biological support for these. Importantly, the authors do lay out and discuss their key assumptions in the Discussion section, so readers can assess their validity and implications for themselves.

Thank you very much, we are very satisfied with this public review.

Reviewer #4 (Public Review):

Summary:

Wilmes and colleagues develop a model for the computation of uncertainty modulated prediction errors based on an experimentally inspired cortical circuit model for predictive processing. Predictive processing is a promising theory of cortical function. An essential aspect of the model is the idea of precision weighting of prediction errors. There is ample experimental evidence for prediction error responses in cortex. However, a central prediction of the theory is that these prediction error responses are regulated by the uncertainty of the input. Testing this idea experimentally has been difficult due to a lack of concrete models. This work provides one such model and makes experimentally testable predictions.

Strengths:

The model proposed is novel and well-implemented. It has sufficient biological accuracy to make useful and testable predictions.

Weaknesses:

One key idea the model hinges on is that stimulus uncertainty is encoded in the firing rate of parvalbumin positive interneurons. This assumption, however, is rather speculative and there is no direct evidence for this.

Thank you very much for this nice description. With regard to the weakness: it is true that the key idea hinges on uncertainty being encoded in the firing of inhibitory neurons. If it turns out that these inhibitory neurons are not PV neurons, however, the theory does not break down. The suggestion of PV neurons is fueled by the observation that PV neurons implement shunting and hence divisive inhibition and by the connectivity of PVs in the circuit. We discuss this in the discussion section: "To provide experimental predictions that are immediately testable, we suggested specific roles for SSTs and PVs, as they can subtractively and divisively modulate pyramidal cell activity, respectively. In principle, our theory more generally posits that any subtractive or divisive inhibition could implement the suggested computations. With the emerging data on inhibitory cell types, subtypes of SSTs and PVs or other cell types may turn out to play the proposed role."

Recommendations for the authors:

Reviewer #4 (Recommendations For The Authors):

(1) Line numbers would simplify reviewing.

We will add line numbers to our next submission.

(2) The existence of positive and negative PE was already suggested by Rao & Ballard.

We added the citation to the sentence "Because baseline firing rates are low in layer 2/3 pyramidal cells () positive and negative prediction errors were suggested to be represented by distinct neuronal populations [44,66],[...]" in the section "Computation of UPEs in cortical microcircuits".

(3) wekk should probably read well.

Indeed, thank you. We fixed it.

(4) Figure 4. legends A-C are mixed up. What are the two values of ¦s-u¦ in F and I - the same as in D and F.

Thank you, we fixed this.

(5) "representation neurons, the activity of which reflects the internal model". For consistency with the original definitions this should read "the activity of which reflects the internal representation". The internal "model" is the synaptic weights (or transformation between areas) - the activity of representation neurons (as the name implies) is the internal "representation".

Thank you, we changed it.

(6) "Mice trained in a predictable environment [...] [4]." This should read "reared" in an unpredictable environment, etc. Relatedly, the problem with this argument is that, the referenced paper argues that the mice never learned to predict and the reduced PE responses are a consequence of a reduction in prediction strength (these mice never - in life - had experience of visuomotor coupling). Better evidence might be the acute changes observed in normal mice (see e.g. Figure 3B in https://pubmed.ncbi.nlm.nih.gov/22681686/ However, another finding from the paper referenced is that in mice reared without visuomotor coupling, MM responses of SST interneurons are unchanged, while those in PV interneurons are completely absent. Would the authors model come to similar results if trained in an environment with (very) high uncertainty and then tested in a low uncertainty environment?

Thank you for pointing us to Figure 3B of Keller et al. 2012. We are now citing this result as it is indeed better evidence.

Thank you very much for your illuminating question and for pointing out that a mouse that never experienced a predictable visual flow may not have formed a model of the visual flow, and hence may not have any prediction about its visual experience. We haven’t considered this scenario in our paper before. So far, we only considered scenarios, in which it is possible to learn a prediction, i.e. to infer the mean from the sensory input. We now consider this other scenario in which the mouse that was reared in an unpredictable environment did not form a prediction and compare SST (1) and PV (2) activity in this mouse to one that learned to form a prediction, and added it to the section "Predictions for different cell types":

"Second, prediction error activity seems to decrease in less predictable, and hence more uncertain, contexts: in mice reared in a predictable environment [where locomotion and visual flow match, 42], error neuron responses to mismatches in locomotion and visual flow decreased with each day of experiencing these unpredictable mismatches. Third, the responses of SSTs and PVs to mismatches between locomotion and visual flow [4] are in line with our model (note that in this experiment the mismatches are negative prediction errors as visual flow was halted despite ongoing locomotion): In this study, SST responses decreased during mismatch, i.e. when the visual flow was halted, and there was no difference between mice reared in a predictable or unpredictable environment. In line with these observations, the authors concluded that SST responses reflected the actual visual input. In our model negative PE circuit, SSTs also reflect the actual stimulus input, which in our case was a whisker stimulus (SST rates in Fig. 6C and I reflect the stimuli (black and grey bar) in A and G, respectively) and SST rates are the same for high and low uncertainty (corresponding to mice reared in a predictable or unpredictable environment). In the same study, PV responses were absent towards mismatches in animals reared in an unpredictable environment [4]. The authors argued that mice reared in an unpredictable environment did not learn to form a prediction. In our model, the missing prediction corresponds to missing predictive input from the auditory domain (e.g. due to undeveloped synapses from the predictive auditory input). If we removed the predictive input in our model, PVs in the negative PE circuit would also be silent as they would not receive any of the excitatory predictive inputs."

(7) "Our model further posits the existence of two distinct subtypes of SSTs in positive and negative error circuits." There is some evidence for this: Figure 5a in https://pubmed.ncbi.nlm.nih.gov/36747710/

Thank you, we added this citation to the corresponding section.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The focus of this manuscript was to investigate the role of Cldn9 in the development of the mammalian cochlea. The main rationale of the study is the fact that cochlear hair cells do not regenerate, so when damaged they are lost forever, causing irreparable hearing loss. The authors have attempted to address this problem by inducing the ectopic production of additional hair cells and testing whether they acquire the morphological and functional characteristics of native hair cells. They show that downregulation of Cldn9 using a well-established genetic manipulation of transgenic mice led to the production of extra numerary inner hair cells, which were able to survive for several months. By performing a large battery of experiments, the authors were able to determine that the native and ectopic inner hair cells have comparable morphological and physiological characteristics. There are several conclusions highlighted by the authors in different parts of the manuscript, including the key role of Cldn9 in coordinating embryonic and postnatal development, the differentiation of supporting cells into inner hair cells, and the possible use of Cldn9 to induce inner hair cell differentiation following deafness induced by hair cell loss.

Strengths:

Several of the conclusions in this study are well supported by the experimental work.

Weaknesses:

Some aspects of the data and its interpretation needs better explanation and requires further investigation.

(1) The Results section is the most difficult part to read and understand. It contains a very limited, and in some places confusing and repetitive, description of the data. Statistical analysis is missing for some of the key data (e.g., ABRs), and in some places the text contradicts the data presented in the figures (e.g., Figure 8). I am sure carefully revising the text would clarify some of these issues.

We thank the reviewer for the suggestion. We revised parts of the results section and added the statistical analysis to the ABRs and DPOAE (lines 151-159; Page 29, lines 846-880). 

(2) One puzzling finding that is not addressed in the manuscript is the lack of functional benefit from these additional inner hair cells. In fact, it appears to be detrimental based on the increased ABR thresholds. Maybe it would be useful to analyze the wave 1 characteristics.

We thank the reviewer for the suggestion. We added the wave 1 characteristics as S8.

(3) It is not clear what direct evidence there is, apart from some immunostaining, indicating that the ectopic inner hair cells derive from the supporting cells. This part would benefit from a more careful consideration and maybe an attempt at a more direct experimental approach.

We thank the reviewer for the suggestion. We intend to investigate the origin of the ectopic inner hair cells using (for example, a qRT-PCR, sm FISH, etc.) in our future study.

(4) One point that should be made clear throughout the manuscript is that the ectopic inner hair cells are generated in a cochlea that is undergoing normal maturation. Thus, there is no guarantee that modulating the expression levels of Cldn9 in a deaf mouse lacking hair cells would produce the same result as that shown in this study. My guess is that it probably won't, but I am sure this could be tested (maybe in the future) using the excellent experimental approach applied in this study.

That is a great point. We will explore it in our future experiments.

Reviewer #2 (Public Review):

Summary:

The generation of functional extranumerary inner hair cells (IHCs) in postnatal mice, particularly with virus-mediated knockdown of Cldn9 mRNA expression in the neonatal cochlear duct, is an important observation. It is significant because not many studies exist that report molecular manipulations of the neonatal organ of Corti that result in the generation of new hair cells that remain functional and appear to be intact for an extended time, here more than one year. Overall, this is a carefully conducted study; the observations are clear, and the methods are solid. Two independent methods for reducing the expression of Cldn9 mRNA were used: a conditional transgenic model and AAV-mediated knockdown with shRNA. The lack of a functional explanation of how the reduced expression of Cldn9 specifically leads to the formation of extranumerary IHCs leaves open questions. For example, it is not clear whether there is indeed a fate change happening and whether Cldn9 reduction affects developmental processes. The discussion of how Cldn9 reduction potentially affects Notch signaling, without hard evidence, is handwaving.

Strengths:

It is a very interesting observation and somewhat unexpected in its specificity for inner hair cells. Using two different approaches to manipulate Cldn9 expression provides a strong experimental foundation. The study is conducted quantitatively and with care.

Weaknesses:

The lack of mechanistic insight results in an open-ended story where at least the potential interaction of Cldn9 reduction with known and well-characterized signaling pathway components should have been investigated. This missed opportunity limits the scope of the study and should be addressed: How does Cldn9 downregulation affect the expression levels of other known genes linked to hair cell production and cell fate decisions? Quantitative RT-PCR works well for the authors, and comparing the expression of Notch or other known pathway components could provide mechanistic insight.

We thank the reviewer for the suggestion. We did quantitative RT-PCR to compare the expression of Notch or other known pathway components in our future work. Besides, we used smFISH with ccnd1 probe and cdkn1b probe to detect cyclin D1 and cyclin-dependent kinase inhibitor 1B (p27) separately in the mouse cochlea. GAPDH was selected as a reference gene. The quantification results showed no significant difference between Cldn9^+/T mice and Cldn9^+/+ mice at P2, P7, and P14.

It is unclear how P21 inner hair cells were identified for the patch-clamp experiments shown in Fig 4E-H. This is a challenging endeavor without the possibility of using specific markers.

We did not have a specific marker for IHCs. However, one with experience in hair bundle morphology and knowledge of their location in the epithelia can identify IHCs from the upright microscope.

Please also address the numerous minor points outlined below; it will improve the paper's readability.

Thanks. Please find the point-to-point answers below.

Please include page numbers and line numbers in a revised manuscript.

We include page numbers and line numbers in a revised manuscript.

Reviewer #3 (Public Review):

This important study by Chen et al help in advancing our knowledge about the regulation of inner hair cell (IHC) development and revealed the role of Cldn9 in IHC embryonic and postnatal induction by transdifferentiation from the supporting cells. The authors developed an inducible doxycycline (dox)-tet-OFF-Cldn9 transgenic mice to regulate expression levels of Cldn9 and show that downregulation of Cldn9 resulted in additional, although incomplete row of IHCs immediately adjacent to the original IHC row. These induced extra IHCs had similar well developed hair bundles, able to mechanotransduce and were innervated by auditory neurons resembling wild-type IHCs. In addition, the authors knock down Cldn9 postnatally using shRNA injections in P1-7 mice with similar induction of extranumerary IHC next to the original row of IHCs. The conclusions of this paper are mostly well supported by the data, but some data analysis needed to be clarified and some crucial controls should be provided to improve the confidence in the presented results. There is a great potential for practical use of these valuable findings and new knowledge on IHC developmental regulation to design Cldn9 gene therapy in the future.

The described by Chen et al mechanisms of extra hair cell generation by suppression of the tight junction protein Cldn9 expression level are very interesting and previously unknown. In particular, the generation of extra IHCs postnatally using downregulation of Cldn9 by shRNA could potentially be very useful as a replacement of HCs lost after noise-induced trauma, ototoxic agents, or other environmental trauma. On the other hand, the replacement of lost hair cells due to various genetic mutations by inducing a supernumerary IHCs with the same abnormalities would not be reasonable.

The authors show that postnatally generated ectopic IHCs are viable and mechanotransducive, but it would be nice to show the maturation steps of ectopic IHC during this postnatal period. For example, stereocilia bundles of the ectopic hair cells should mature later than the original IHCs. A few days after viral delivery of shRNA, you should be able to observe immature IHC bundles that unequivocally will define newly generated IHCs. Unfortunately, the authors show only examples of already mature ectopic IHCs at P21 and in 5-6 weeks old mice and at relatively low resolution. Also, during maturation, IHCs usually have transient axo-somatic synapses that are not present in mature IHCs. It would be great to see if, in 5-6 weeks old mouse, the ectopic IHCs still have axo-somatic synapses or not, and if the majority of the ectopic IHCs have innervation. Some of the data in this study would benefit from showing corresponding controls and some - from higher resolution imaging.

We appreciate the reviewer's suggestion. The objective of the paper is to report the phenomenon and present the coarse features of the Cldn9-mediated induced ectopic hair cells. The systematic details are for future studies, which are ongoing and out of the current scope.

In the mammalian cochlea, each HC is separated from the next by intervening supporting cells, forming an invariant and alternating mosaic along the cochlea's length. Cochlear supporting cells in some conditions can divide and trans-differentiate into HCs, serving as a potential resource for HC differentiation, using transcription and other developmental signaling factors.

However, when ectopic hair cells are generated from supporting cell trans-differentiation, the intricate mosaic of the organ of Corti is altered, which could by itself lead to hearing issues. In case of downregulation of Cldn9, the extra row of IHCs seems to be positioned immediately adjacent to the original IHC row. It is not clear if the newly formed unusual junctions between the ectopic and original IHCs are sufficiently tight to prevent leakage of the endolymph to the basolateral surface of IHCs. Also, it is not clear if the other organ of Corti tight junctions could lose their tightness due to the downregulation of Cldn9, which could over time affect the endocochlear potential as shown by this study and hearing abilities.

There was a slightly increased ABR threshold (5 dB -15 dB) (Fig. 4A) and a decrease in the magnitude of the EP and the rise in the K⁺ concentration in the endolymph and perilymph of Cldn9+/T mice compared to from age-matched littermates (S10) indicated there might be a compromised epithelium tight junction. The downregulation of Cldn9 affected the endocochlear potential and hearing abilities ((Fig. 4A, S10) after 2m, suggesting an age-dependent effect. The effective downregulation of Cldn9 would require proper titration of Cldn9 levels to induce extra hair cells with intact epithelial integrity; work may require additional studies.

Importantly, CLDN9 immunofluorescence staining data that show cytoplasmic staining of supporting cells should be revisited and the organ of Corti schematics showing CLDN9 expression should be corrected, considering that CLDN9 localizes to the tight junctions of the reticular lamina as was shown by immunoEM in this study and described in previous publications (Kitajiri et al., 2004; Nakano et al., 2009, Ramzan et al., 2021). While the current version of the manuscript will interest scientists working in the inner ear development and regeneration field, it could be more valuable to hearing researchers outside this immediate field and perhaps developmental biologists and cell biologists after proper revision.

We appreciate the reviewer's comments. We were concerned about the observation, but the results were consistent. Indeed, that was the motivation for performing the immunoEM (S3). A follow-up report may address it further.

Recommendations for the authors:

Reviewer #2 (Recommendations For The Authors):

Please address the points I made about the presentation (word choice, inconsistencies in labeling, etc). It ultimately helps a reader to understand and to follow your logic. This is an important observation.

We corrected the inconsistencies in labeling and addressed the points you suggested.

Making the extra effort to investigate a possible interaction between Cldn9 and Notch signaling would substantially increase the significance of the work.

Thanks for the suggestions. We will explore it in our future work.

Minor points:

Some sentences would benefit from revision:

- The abstract argues that hearing loss is incurable because mammalian hair cells are terminally differentiated (3rd sentence). This is not accurate.

Mammalian HCs are terminally differentiated by birth, making HC loss challenging to replace.

- The second sentence of the second paragraph of the introduction, "Cochlear SCs can divide and trans-differentiate into HCs, serving as a potential resource for HC differentiation, using transcription and developmental signaling factors (White et al., 2006)," should be referenced in the context of the animal's age. This feature of supporting cells is transient and only observed in neonatal mice. The following sentences in the same paragraph would also benefit from being placed into the same context when appropriate.

We thank the reviewer for the suggestion. These sentences have been corrected.

- Introduction: "But functional features of the newly developed HC are circumspect." The authors probably meant "circumspect," but is this the appropriate word? Also, please use the plural of HC = HCs.

The sentence has been corrected to “but the functional features of the newly developed HCs are circumspect”.

- Introduction: Isn't an essential function of tight junctions in the organ of Corti the separation of fluid-filled spaces? Perhaps additional functions of tight junction proteins are unclear, but at least this one function appears clear.

We thank the reviewer for the suggestion. We added the “additional” before the “function” in this sentence.

- Introduction: "using shRNA injection in postnatal (P) days (P1-7) mice." This is a rather vague statement that could be better defined. Perhaps mention that the injections targeted the round window and that an AAV-based method was used. Also, it is not clear from the methods whether the injection needle pierced the round window. Please clarify. Likewise, the methods state that these experiments were conducted in P1-P15 mice, but the main text says P1-P7. Later, in the results section and in the figure legend for Fig 7, the mice are between P1-P7 and P14; the figure itself is labeled with P1 and P14. However, data is presented (Fig 6) for injections at P2, P4, P7, and P14. In the text referring to Fig 6B in the results section, it is stated, "By contrast, the P14-21 inner ear transfected with Cldn9-shRNA produced no detectable increase..." Only data for P2, P4, P7, and P14 injections are presented. These are minor issues, but please check the inconsistencies because they make it difficult to follow.

We corrected this sentence to “Analogous additional putative IHCs differentiation was observed when Cldn9-shRNA was injected through the round window to postnatal (P) days (P2-7, and P14) mice…”.  The label in Fig 7A has been changed to P2-7, and the text referring to Fig 6B in the result section has been changed to “the P14 inner ear transfected with Cldn9-shRNA produced no detectable increase...".

- Last statement of the Introduction: "making Cldn9 a viable target for generating transformed IHCs." It is not clear what transformed IHCs are.

We replaced the transformed with supernumerary.

- To understand the Southern Blot analysis in Fig 1E, the location of BstAPI and BamHI restriction sites and the probe need to be illustrated in Fig 1D.

The restriction sites BstAPI, (Bst), and BamHI (Bam) are indicated (Fig. 1D).

- Please define the purple arrows and arrowheads in Fig 1D. What do the different colors for the backbone mean? I see red and green, but also orange and yellow in the floxed allele. In Fig 1F, is "Knock-in" synonymous with homozygote? Would it be clearer to use the nomenclature Cldn9(T/T), Cldn9(T/+), and Cldn9(+/+), which is used later in the text?

We have made the changes as requested.

- Results, first paragraph: "Results of RT-PCR..." This refers to quantitative RT-PCR; please add the word "quantitative."

Thanks. We added “quantitative” to the sentence.

- Results and Fig S1. Is the strong upregulation of Cldn9 mRNA (S1A) also reflected in stronger Cldn9 immunoreactivity?

Yes, the strong upregulation of Cldn9 mRNA showed higher cldn9 immunoreactivity.

- Results, Fig 1. Please add a schematic drawing showing all elements of the inducible gene expression cassette in the final transgenic allele, and please illustrate how the system works. This helps the reader to understand the strong Cldn9 mRNA upregulation in Cldn9(T/T) mice, where expression is likely driven by the CMV promoter and reciprocally, in the presence of doxycycline, the suppression of transcription by binding of the tTA-dox protein to the TRE elements of the modified CMV promoter. Is this a correct assumption?

Yes, this is a correct assumption

- Results, about Fig S3. Why is it important to investigate Cldn6 and ILDR1 levels in the context of Cldn9 downregulation? Also, that is meant with "no comparative differences in others?". If a potential compensatory effect is suspected, why are the authors not systematically characterizing the expression of other tight junction proteins with quantitative RT-PCR? The results shown in S3 are anecdotal, without proper quantification, and lack context.

The goal is to examine the potential compensatory changes in other TJ proteins. It was not to examine all possible TJ proteins localized in the inner ear.

Results, section headed with "Downregulation of..." First sentence. Fig. 2A-C à Fig. 2A-E.

Thanks. We corrected the sentence “5-week-old mice Cldn9^+/T cochleae displayed a notable row of ectopic HCs (Fig. 2A-C).” to “5-week-old mice Cldn9^+/T cochleae displayed a notable row of ectopic HCs (Fig. 2A-E).”

The same section: "were negatively labeled with anti-prestin antibody." Consider "were not labeled with antibody to prestin." Likewise, a few sentences below, please consider rephrasing "the ectopic HCs ... reacted positively to otoferlin antibodies". Also, "...expressed multiple CtBP2 labeling..." - this reads like an incomplete sentence.

Thanks for the suggestions. We have corrected the three sentences mentioned.

The phrase "putative ectopic" lacks clarity because "putative" could refer to "ectopic" (like an adverb). Consider swapping the two words and writing "ectopic putative IHCs" or simply "ectopic IHCs."

Thanks for the suggestions. We replaced the “putative ectopic IHCs” with “ectopic IHCs” in all contexts.

Please use more precise figure labels when referring to a specific figure panel. For example, "Additionally, the ectopic HCs show IHC bundle features (Fig. 2)," - Bundles are shown in Fig 2D and Fig 2E. Please check all instances where a full figure is mentioned, but the specific reference is to a panel of the figure. Another example, "... using quantitative RT-PCR (S7)..." would be more specific if Fig S7A is referred to.

Thanks for the suggestions. We checked all instances and corrected the labels. Thanks!

"IHC counts at different ages (P2-P21) and the cochlear frequency segments (4-32 kHz) demonstrate..."- the figure shows data for 8 kHz and 32 kHz; please revise: "segments (8 kHz and 32 kHz) demonstrate."

This sentence has been revised based on your suggestion. Thanks!

Please add a legend to Fig. 3C (like the one shown in Fig. 2F).

Thanks for the reminder. The legend for Fig. 3C was modified.

Fig 4A and Fig 4B. It is impossible to distinguish the open/closed circles and the many lines. Please consider a different format or an extended supplemental figure. Also, drawing a line connection between the 32 kHz and click data points in 4A is inappropriate.

Instead of the open/closed circles, the dashed line means Cldn9^+/+ mice, and solid lines represent Cldn9^+/T mice. We added the line labels. The line connecting between 32 kHz and click data points was removed.

Fig 4, legend. Please define BHB and BHC levels.

BHB and BHC are defined.

The paragraph "Synaptic features of PE IHCs match original IHCs" is confusing because it states the following: "The synapses between the IHCs and auditory neurons at the apical, middle, and basal cochlear locations from 5-week-old Cldn9+/+ and Cldn9+/T mice show substantial differences." The meaning of the heading, therefore, does not match what is ultimately shown and discussed.

We have changed the title to “Synaptic features of ectopic IHCs and original IHCs”.

Moreover, no actual features of synapses are investigated; CtBP2/Homer pairs were used to identify afferent synapses, which this reviewer would argue provides a reasonable estimate of the number of synapses where pre- and post-synaptic markers are detected in close vicinity. It would be helpful to describe the method for counting juxtaposed CtBP2 and Homer-labeled puncta with more detail.

The method section now includes more information about the synapse count, which this reviewer would argue provides a reasonable estimate of the number of synapses where pre- and post-synaptic markers are detected in close proximity.

The final concluding sentence of the section also suggests that synaptic transmission from PE IHCs might be compromised because significant differences in synapse numbers were identified. It would be important to mention this.

Thanks for the reminder. We added this information to the final concluding sentence.

Fig. 5C, 5D; legend. Is "co-expressed" the right word choice? Consider "colocalized" or "juxtaposed".

The "co-expressed" has been replaced with "colocalized".

Voltage-clamp recordings of P21 inner hair cell mechanoelectrical transduction currents. This reviewer cannot identify a previous publication describing the details of this method on P21 cochlear inner hair cells; this seems like an excellent methodological advance.

Yes, we can record data from older mice. Thanks for pointing it out.

"Transfection in vivo of Cldn9 shRNA," the P14-21 inner ear transfected with Cldn9-shRNA." Plus, additional use of the word "transfection." Transfection generally means the introduction of plain nucleic acid into cells. The word refers to methods that do not use viruses. In contrast, "transduction" is the term used for virus-mediated gene transfer. The authors used AAVs. Please correct for appropriate scientific terminology.

Thanks for the clarification. This information has been corrected accordingly.

"A slight decline in the amplitude of the EP and a substantial rise in perilymph K+ was detected in 8-month-old Cldn9+/T (S7)." Probably Fig. S8A,B is meant.

Yes, it referred to Fig. S8 A, B. We corrected it in the result section. Thanks!

Heading "Discussions" -> "Discussion"

The focus of the second part of the discussion on potential interactions between Cldn9 suppression and known signaling pathways is essential. The logic that is presented with respect to Notch signaling, however, is not clear and misleading. For example, it is not obvious what is meant by "Cldn9 subserves the signaling catalyst to activate NICD cascades" and whether this statement is supported by any published data.

The statement was a suggestion and has been qualified with a “may” clause (line 299).

The authors might consider discussing whether the observed effect caused by Cldn9 elimination is a specific role of the Cldn9 protein itself or is an epiphenomenon resulting from cytomechanical changes in the developing and maturing organ of Corti. This would add a potential Notch-independent component for a possible interpretation of the observations.

We state lines 302-304 “Alternatively, Cldn9 levels disruption may alter the mechanical properties of the developing and maturing organ of Corti that may trigger ectopic IHC differentiation, an epiphenomenon independent of the Notch signaling“.

Methods:

"Deletion of the selection marker in the tTA cassette by crossing the F1 mouse with the embryonic Cre line (B6.129S4-Meox2tm1(cre)Sor/J)." This sentence seems to be incomplete.

Thanks for pointing it out. This sentence has been rewritten.

"Images were captured under a confocal microscope." Consider writing "with a confocal microscope".

This sentence has been corrected. Thanks!

RNA extraction and... How many mice were used per experiment? 10-15 or just 10?

The mice number for the RNA extraction is between 10 and 15. Thanks

Reviewer #3 (Recommendations For The Authors):

Below are my suggestions, questions, and criticisms.

(1) The red outline on Fig1A schematic does not correspond to the previously published expression pattern of CLDN9 in the organ of Corti reticular lamina tight junctions (Kitajiri et al, 2004, Nakano et al., 2009, Ramzan et al., 2021). Also, there are no tight junctions all around the pillar cells. The tight junctions are restricted to the sites of tight attachments between two cells. The immunofluorescence staining using CLDN9 antibody looks rather cytoplasmic (Fig 1 and Fig S1) than associated with the tight junctions as it was shown by immunoEM data here and reported previously (Kitajiri et al, 2004; Nakano et al, 2009; Ramzan et al, 2021). Please correct the schematic and explain your data.

We have redrawn the diagram (Fig. 7).

(2) The CLDN9 staining in Figure 1, B and C, highlights the cytoplasm of the supporting cells, and hair cells devoid of the staining. From the images in Fig. S1C, it also looks like CLDN9 is present only in supporting cells and not in hair cells? How would the authors reconcile their data with Cldn9 expression data from the gEAR database and Ramzan et al.'s 2021 RNAscope data? Please provide the validation of the antibody used in this study.

We recognize the reviewer’s concern but RNA and protein levels are not always in parallel.

(3) Figure 1D. The dash lines from the targeting vector to the wt allele seem to indicate a recombination event. Please do not show the recombination event, instead just show what part of the targeting vector was incorporated to replace wt Cldn9. There is no description in the figure 1 legend what purple arrows and arrowheads mean and what yellow and orange line segments in the floxed allele schematic indicate. Please also show where the BstAPI and BamHI restriction enzyme sites are.

We have provided supplement Fig 1., and have noted the BstAPI and BamHI restriction enzyme sites in Fig. 1D.

(4) What does the organ of Corti that has 40-to-55-fold increase in Cldn9 mRNA expression looks like before dox treatment? Any abnormalities at all? How is CLDN9 protein localization looks in the Cldn9+/T untreated mice? Do they have normal number of IHCs? Cldn9+/T untreated mice should be used as another control at least in Figure S1. What does the organ of Corti that has a 40-to-55-fold increase in Cldn9 mRNA expression look like before dox treatment? Are there any abnormalities at all?

The untreated Cldn9^+/T mice can grow normally but are not fertile. So, we used a very low concentration of dox water (0.1 mg/ml) instead of normal water to keep the breeding pairs. The protein level increased in the Cldn9^+/T mice compared with Cldn9^+/+mice. With 0.1 mg/ml dox water, they also showed ectopic IHCs.

(5) It is interesting that decline of 0.4-0.6-fold in mRNA level leads to about 8-fold decrease in protein level based on your immunoEM data on tight junctions of IHC with supporting cells. Do you observe the same effect in OHC-SC tight junctions, or the decrease was observed selectively around IHCs?

The reviewer is alluding to matching RNA and protein levels. It appears that for Clnd9 one cannot expect a closely matched relationship.

(6) The quality of the immunoEM data is great, but a control of secondary antibody alone staining in wt and Cldn9+/T dox treated should be shown and compared to the Cldn9+/T treated sample.

We thank the reviewer for raising the issue. Secondary antibodies are used as a control in all immunoEMs in the laboratory. We opted not to show negative results.

(7) The authors observed a decrease in Cldn6 expression albeit not quantitative in response to Cldn9 downregulation. How were the immunofluorescence signals compared and evaluated? Please provide a detailed description of the method used. Did the authors used the same image acquisition parameters? Was the Cldn9 and Cldn6 immunostaining done using same protocol with the same aliquot and dilution of the secondary antibodies, etc.? The staining for CLDN6 seems to be concentrated in the cytoplasm of supporting cells, and not in the tight junctions, similar to CLDN9 immunoreactivity shown in Fig. S1C and to the ILDR1 pattern of staining in Fig. S3. How can the authors explain this? How were the antibodies validated?

The Cldn9 and Cldn6 immunostaining were done using the same protocol with the same aliquot and dilution of the secondary antibodies.

(8) CLDN14 is also expressed in the organ of Corti tight junctions. What happened to this TJ protein during CLDN9 downregulation?

We detected Cldn14 with immunostaining in the Cldn9+/T mice and Cldn9+/+ mice fed with 0.25 mg/ml dox water, and the results showed increased expression of Cldn14 in Cldn9+/T mice. Detail alterations of other TJ proteins have been reserved for future studies. 

(9) When supernumerary IHCs were observed in Cldn9+/T mice, have the authors noticed a corresponding decrease in supporting cells surrounding IHCs? Quantification of the IHCs supporting cells would be useful. Do the ectopic IHCs have apical tight junctions with original IHCs or they are surrounded by supporting cells?

We quantified the SCs around the IHCs but did not detect significant differences among the groups.

(10) The authors indicated that viable PE IHCs were observed in 15 months old Cldn9+/T dox treated mice. How stereocilia bundles look in these ectopic hair cells? Are they preserved similar to the original IHCs or degenerated? It is hard to see this in Fig 3, phalloidin panel. High-resolution SEM would show this better.

For the remaining ectopic IHCs in 15 months, we did not detect apparent differences in hair bundles compared with the original IHCs.

(11) Interestingly, the authors indicate that the highest number of the ectopic IHCs were developed in the apical turn and the higher elevation of ABR threshold was also observed at low frequencies end. This may indicate that extra IHCs do not help hearing function.

The extra IHCs showed along the whole cochlea, even though it is more obvious in the apical turn. The declined hearing may have resulted from the leakage of the endolymph K+ to the perilymph and EP decline.

(12) No age-matched wt control is shown for decreased expression of Cldn9 after shRNA injection at P2 (Fig. 6A).

As indicated earlier, we opted to state but did not show negative results.

(13) Figure 6C. The better- quality SEM images showing a longer stretch of IHCs are needed to convince readers that there are ectopic IHCs that are well preserved in 5-6 weeks old mice in all cochlear turns after GFP-Cldn9 shRNA treatment at P2-P7.

In S4, we showed that there are ectopic IHCs along the cochlear axis.

(14) Do scrambled shRNA control samples had some ectopic IHCs? This control is missing in Fig.6D.

No scrambled shRNA controls did not show ectopic IHCs. We have stated it.

(15) Figure 7B, lower schematic. There are no known continuous tight junctions and CLDN9 expression around the OHCs and IHCs. CLDN9 is known to be concentrated at the reticular lamina tight junctions which separate the endolymph from perilymph. Please, correct all schematics accordingly.

We have made the changes as requested.

Minor comments:

(1) Page 1, Abstract. I would not say "making HC loss incurable" since recent gene therapy results show some advances in this direction. Please rephrase more accurately.

We have made the changes as requested.

(2) Page 4, Results, line 5; please rephrase "PCR of tail tissue samples performed genotyping."

It has been corrected to “The genotyping was performed by the PCR with the tail tissue.”

(3) Fig. 1 legend, panel B, replace "showing IHC stained myosin7a" with "showing IHC stained by myosin7a". Also, in the same sentence, "phalloidin, actin (green) antibodies," Phalloidin is not an antibody; please change this.

Thanks. We have corrected this information.

(4) Fig 2C, IHC label obscures the view of IHCs, please move this label out and use an arrow to point to IHCs.

We have made the changes as requested.

(5) Figure 4, title. Replace "currents elicited original" with "current elicited from original".

This sentence has been corrected. Thanks.

(6) Figure 4, panel A. It is hard to see the open symbols on the graph. Are they associated with the dash lines? Please make them more visible or indicate what dash lines are. "ABR threshold for (n=12)" should be "ABR threshold for Cldn9+/+(n=12)"?

Yes, they are associated with the dash lines. We added the labels for the solid lines and dash lines. "ABR threshold for (n=12)" was corrected to "ABR threshold for Cldn9+/+(n=12)."

(7) Figure 4, legend. "Within each wt and heterozygote mice, there was no significant shift...". Do you mean within each group of mice? Also "Mean DPOAE threshold for 2-8 mos (n=9) was tested,..." Do you mean (n=9) for each group or what group?

Yes, "Within each wt and heterozygote mice, there was no significant shift..." has been revised. The number of mice in each group for the DPOAE test was clarified in the Fig. 4B legend. Thanks.

(8) Please label the X axis in Figure 4D.

The X-axis has been labeled (Time (s))

(9) Figure 4 B, do the colors of the lines indicate the same age groups as in Fig 4A? Do the dash lines associate with open symbols? Please state this clearly in the figure's legend.

Yes. We added this information in Fig. 4B legend.

(10) Figure 4D. Please label the X axis of the fluorescence intensity graph.

The X-axis has been labeled (Time (s))

(11) Figure 4G, legend. Replace "(mean +std)" with "(mean +SD)" for consistency here and in Figure 5 legend.

Thanks. We replaced "(mean +std)" with "(mean +SD) in the legend of Fig. 4G and Fig.5 and Fig.6.

(12) Figure 5B, legend. Replace "makers" with "markers".

Thanks. This information was corrected.

(13) Figure 6A, legend. There is no downregulation of Cldn9 by shRNA shown in "S5". Do the authors mean Figure S7? Please, correct "S5" to "Fig. S7".

This information was corrected. Thanks.

(14) Figure 6A, legend. There is no reduced CLDN9 protein expression shown in Fig. 1C. Do the authors mean Fig. 6A, third panel? Please correct the phrase "reduced protein expression (Fig. 1C) is shown in the 3rd Panel (Cldn9, red)" accordingly, and do not capitalize "p" in the "3rd Panel".

This information was corrected. Thanks (line 917-918).

(15) Also there, replace "The right Panel shows two rows of IHCs (marked HC marker, Myo7a (cyan), and the merged photomicrograph" with "The right panel shows the merged image with two rows of IHCs stained with HC marker Myo7a (cyan) and the expression of Ad-GFP-mCldn9 shRNA (green) in the adjacent row of supporting cells". Please indicate in what cells Ad-GFP-mCldn9 shRNA (green) is expressed. It looks like only one row of supporting cells has this green signal.

This information was corrected.

(16) Figure 6B, legend. Replace "Examples of photomicrographs of sections of the whole-mount cochlea of P2, P4, P7, and P14 Cldn9 shRNA injected mice" with "Examples of phalloidin stained whole-mount organ of Corti samples from cochleae of the wild-type mice injected at P2, P4, P7 and P14 with Cldn9 shRNA"

This sentence has been modified based on your suggestions. Thanks!

(17) Replace "action labeling" with "actin labeled."

Thanks!  The "action labeling" has been replaced with "actin labeled." Line 924

(18) Figure 6C. Insert "C" before SEM images description in the legend. The authors stated that SEM images of "5-6-wks-old mice" are shown. Please indicate the exact age of mice shown on each image and at what age these mice received the virus injection.

Thanks!  The “C” has been added. We have noted that the SEM images are from 5-week-old mice" in the legend, and the virus was injected at P2.

(19) Figure 6D, legend. Last sentence: move "are significantly different" and insert this between "IHCs" and "at P2 apex".

This information was corrected.

(20) Figure S7, legend. Replace "(sram)" with "(scram)" as in the figure itself. Also, Indicate the age of samples at the harvesting time for imaging and the age at injection of Cldn9 shRNA.

"(sram)" has been replaced with "(scram)". The age of samples at the harvesting time for imaging and the age at injection of Cldn9 shRNA are indicated.

(21) Figure S8. Replace "4 mos-old" and "8 mos-old" with "4 months-old" and "8 months-old" everywhere in the legend and in the figure labels.

We have made the changes as suggested.

(22) Page 8, 5th lane from the bottom. Change "EP and K+ concentration endolymph" to "EP and K+ concentration of the endolymph".

It has been corrected. Thanks.

(23) Page 8, next to the last sentence before the Discussion. Wrong figure number, please replace "(S7)" with "Fig. S8".

It has been corrected. Thanks.

Author response:

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

Joint Public Review:

Summary:

The authors aimed to identify the neural sources of behavioral variation in fruit flies deciding between odor and air, or between two odors.

Strengths:

- The question is of fundamental importance.

- The behavioral studies are automated, and high-throughput.

- The data analyses are sophisticated and appropriate.

- The paper is clear and well-written aside from some initially strong wording.

- The figures beautifully illustrate their results.

- The modeling efforts mechanistically ground observed data correlations.

Weaknesses:

- The correlations between behavioral variations and neural activity/synapse morphology are relatively weak, and sometimes overstated in the wording that describes them.

We sincerely thank the reviewers for these evaluations.

Recommendations for the authors:

Line 56: "We hypothesize that as sensory cues are encoded and transformed to produce motor outputs, their representation in the nervous system becomes increasingly idiosyncratic and predictive of individual behavioral responses". This seems obvious a priori. The sensory stimuli are the same, but the motor responses are different. Along the way there has to be a progression from same to different. Is there an alternative hypothesis? If so, perhaps state the alternative.

We added text to the first paragraph of the introduction (lines 58-60) laying out an alternative hypothesis that individuality emerges through biomechanical differences and environmental interactions, and we have altered our motivating question to assess whether circuit elements in which activity is predictive of individual behavior exist, and if so, where (lines 60-62).

Line 157: typo "remaining"

We changed “remaining” to “remain” (line 160).

Line 163: why report r sometimes and R^2 other times? Better to use R^2 throughout.

We changed all instances of r to R², notably when reporting combined train/test statistics for calcium - behavior models (line 162). We also reframed the outputs (medians + 90% confidence intervals) of the supplemental analysis inferring the strength of the latent calcium-behavior relationship to be in terms of R² (lines 166, 173-175, 241, 252; modified text in Inference of correlation between latent calcium and behavior states in Materials and Methods; adjusted figure and caption for Figure 1 – figure supplement 9).

Line 182: "odorant". Should be "odorant receptors"?

We respectfully disagree – our ORN and PN calcium data are responses to odorants in 5 glomerulus/odorant receptor types. When we group PCA loadings by glomerulus for both ORN and PN calcium, the consistency within groups is much stronger than when we group the loadings by odorant (Figure 1 – figure supplement 8). Additionally, “odorant receptor organization” would mean the same thing as “glomerular organization,” since all ORNs expressing the same odorant receptor project to a single glomerulus.

Line 331: "harbor". Maybe more modestly "contribute to"?

We changed “harbor” to “contribute to” (line 334) and added additional moderating language that the difference in DC2 and DM2 activations in PNs explains a large portion of the individuality signal (lines 337-339).

Line 403: typo "is"

We retained “is” as the corresponding verb for “the net effect,” but we adjusted the position of the reference to Gomez-Marin and Ghazanfar, 2019 for more clarity (lines 406-408).

Author response:

Reviewer #1(Public review):

Summary:

This manuscript details the results of a small pilot study of neoadjuvant radiotherapy followed by combination treatment with hormone therapy and dalpiciclib for early-stage HR+/HER2-negative breast cancer.

Strengths:

The strengths of the manuscript include the scientific rationale behind the approach and the inclusion of some simple translational studies.

Weaknesses:

The main weakness of the manuscript is that overly strong conclusions are made by the authors based on a very small study of twelve patients. A study this small is not powered to fully characterize the efficacy or safety of a treatment approach, and can, at best, demonstrate feasibility. These data need validation in a larger cohort before they can have any implications for clinical practice, and the treatment approach outlined should not yet be considered a true alternative to standard evidence-based approaches.

I would urge the authors and readers to exercise caution when comparing results of this 12-patient pilot study to historical studies, many of which were much larger, and had different treatment protocols and baseline patient characteristics. Cross-trial comparisons like this are prone to mislead, even when comparing well powered studies. With such a small sample size, the risk of statistical error is very high, and comparisons like this have little meaning.

We greatly appreciate your evaluation of our study and fully agree with the limitations you have pointed out. We have clearly stated the limitations of the small sample size and emphasized the need for a larger population to validate our preliminary findings in the discussion section (Lines 311-316).

We acknowledge that this small sample size is not powered to characterize this regimen as a promising alternative regimen in the treatment of patients with HR-positive, HER2-negative breast cancer. Therefore, we have revised the description of this regimen to serve as a feasible option for neoadjuvant therapy in HR-positive, HER2-negative breast cancers both in the discussion (Lines 317-320) and the abstract (Lines 71-72).

We agree with you that cross-trial comparisons should be approached with caution due to differences in study designs and patient populations. In our discussion section, we acknowledge that small sample size limited the comparison of our data with historical data in the literature due to the potential bias (Lines 312-313). We clearly state that such comparisons hold limited significance (Lines 313-314) and suggest a larger population to validate our preliminary findings.

• Why was dalpiciclib chosen, as opposed to another CDK4/6 inhibitor?

Thank you for your comments. The rationale for selecting dalpiciclib over other CDK4/6 inhibitors in our study is primarily based on the following considerations:

(1) Clinical Efficacy: In several clinical trials, including DAWNA-1 and DAWNA-2, the combination of dalpiciclib with endocrine therapies such as fulvestrant, letrozole, or anastrozole has been shown to significantly extend the progression-free survival (PFS) in patients with hormone receptor-positive, HER2-negative advanced breast cancer (1-2).

(2) Tolerability and Management of Adverse Reactions: The primary adverse reactions associated with dalpiciclib are neutropenia, leukopenia, and anemia. Despite these potential side effects, the majority of patients are able to tolerate them, and with proper monitoring and management, these reactions can be effectively mitigated (1-2).

(3) Comparable pharmacodynamic with other CDK4/6 inhibitors: The combination of CDK4/6 inhibitors, including palbociclib, ribociclib, and abemaciclib, with aromatase inhibitors has demonstrated an enhanced ability to suppress tumor proliferation and increase the rate of clinical response in neoadjuvant therapy for HR-positive, HER2-negative breast cancer (3-5). Furthermore, preclinical studies have shown that dalpiciclib has comparable in vivo and in vitro pharmacodynamic activity to palbociclib, suggesting its potential effectiveness in similar treatment regimens (6).

(4) Accessibility and Regulatory Approval: Dalpiciclib has gained marketing approval in China on December 31, 2021, which facilitates the accessibility of this medication, making it a more convenient option when considering treatment plans.

References:

(1) Zhang P, Zhang Q, Tong Z, et al. Dalpiciclib plus letrozole or anastrozole versus placebo plus letrozole or anastrozole as first-line treatment in patients with hormone receptor-positive, HER2-negative advanced breast cancer (DAWNA-2): a multicentre, randomised, double-blind, placebo-controlled, phase 3 trial(J). The Lancet Oncology, 2023, 24(6): 646-657.

(2) Xu B, Zhang Q, Zhang P, et al. Dalpiciclib or placebo plus fulvestrant in hormone receptor-positive and HER2-negative advanced breast cancer: a randomized, phase 3 trial(J). Nature medicine, 2021, 27(11): 1904-1909.

(3) Hurvitz S A, Martin M, Press M F, et al. Potent cell-cycle inhibition and upregulation of immune response with abemaciclib and anastrozole in neoMONARCH, phase II neoadjuvant study in HR+/HER2− breast cancer(J). Clinical Cancer Research, 2020, 26(3): 566-580.

(4) Prat A, Saura C, Pascual T, et al. Ribociclib plus letrozole versus chemotherapy for postmenopausal women with hormone receptor-positive, HER2-negative, luminal B breast cancer (CORALLEEN): an open-label, multicentre, randomised, phase 2 trial(J). The lancet oncology, 2020, 21(1): 33-43.

(5) Ma C X, Gao F, Luo J, et al. NeoPalAna: neoadjuvant palbociclib, a cyclin-dependent kinase 4/6 inhibitor, and anastrozole for clinical stage 2 or 3 estrogen receptor–positive breast cancer(J). Clinical Cancer Research, 2017, 23(15): 4055-4065.

(6) Long F, He Y, Fu H, et al. Preclinical characterization of SHR6390, a novel CDK 4/6 inhibitor, in vitro and in human tumor xenograft models(J). Cancer science, 2019, 110(4): 1420-1430.

• The eligibility criteria are not consistent throughout the manuscript, sometimes saying early breast cancer, other times saying stage II/III by MRI criteria.

criteria in our manuscript. We deeply apologize for any confusion caused by these inconsistencies. We have revised the term from “early-stage HR-positive, HER2-negative breast cancer” to “early or locally advanced HR-positive, HER2-negative breast cancer” (Lines 128 and 150). The term “early or locally advanced” encompasses two different stages of breast cancer, whereas “Stage II/III by MRI criteria” refers to specific stages within the TNM staging system.

• The authors should emphasize the 25% rate of conversion from mastectomy to breast conservation and also report the type and nature of axillary lymph node surgery performed. As the authors note in the discussion section, rates of pathologic complete response/RCB scores are less prognostic for hormone-receptor-positive breast cancer than other subtypes, so one of the main rationales for neoadjuvant medical therapy is for surgical downstaging. This is a clinically relevant outcome.

We appreciate your constructive comments. Based on your suggestions, we have made the following revisions and additions to the article.

The breast conservation rate serves as a secondary endpoint in our study (Line 62 and 179). We have highlighted the significant 25% conversion rate from mastectomy to breast conservation in both the results (Lines 229-230) and discussion sections (Lines 290-292).

In our study, all patients underwent lymph node surgery, including sentinel lymph node biopsy or axillary lymph node dissection. Among them, 58.3% of patients (7/12) underwent sentinel lymph node biopsies.

We agree with your point that the prognostic value of pathologic complete response/RCB score is lower for hormone receptor-positive breast cancer compared to other subtypes, we have revised the discussion section to clarify that one of the principal objectives for neoadjuvant therapy in this patient population is to facilitate downstaging and enhance the rate of breast conservation (Lines 289-290). And also emphasized that this neoadjuvant therapeutic regiment appeared to improve the likelihood of pathological downstaging and achieve a margin-free resection, particularly for those with locally advanced and high-risk breast cancer (Lines 293-295).

Reviewer #2 (Public review):

Firstly, as this is a single-arm preliminary study, we are curious about the order of radiotherapy and the endocrine therapy. Besides, considering the radiotherapy, we also concern about the recovery of the wound after the surgery and whether related data were collected.

Thanks for the comments. The treatment sequence in this study is to first administer radiotherapy, followed by endocrine therapy. A meta-analysis has indicated that concurrent radiotherapy with endocrine therapy does not significantly impact the incidence of radiation-induced toxicity or survival rates compared to a sequential approach (1). In light of preclinical research suggesting enhanced therapeutic efficacy when radiotherapy is delivered prior to CDK4/6 inhibitors, we have opted to administer radiotherapy before the combination therapy of CDK4/6 inhibitors and hormone therapy (2).

In our study, we collected data on surgical wound recovery. All 12 patients had Class I incisions, which healed by primary intention. The wounds exhibited no signs of redness, swelling, exudate, or fat necrosis.

References:

(1) Li Y F, Chang L, Li W H, et al. Radiotherapy concurrent versus sequential with endocrine therapy in breast cancer: A meta-analysis(J). The Breast, 2016, 27: 93-98.

(2) Petroni G, Buqué A, Yamazaki T, et al. Radiotherapy delivered before CDK4/6 inhibitors mediates superior therapeutic effects in ER+ breast cancer(J). Clinical Cancer Research, 2021, 27(7): 1855-1863.

Secondly, in the methodology, please describe the sample size estimation of this study and follow up details.

Thanks for pointing out this crucial omission. Sample size estimation for this study and follow-up details have been added in the methodology section. The section on sample size estimation has been revised to state in Statistical analysis: “This exploratory study involves 12 patients, with the sample size determined based on clinical considerations, not statistical factors (Lines 210-211).” The section on follow up has been revised to state in Procedures section “A 5-year follow-up is conducted every 3 months during the first 2 years, and every 6 months for the subsequent 3 years. Additionally, safety data are collected within 90 days after surgery for subjects who discontinue study treatment (Lines 169-172).”

Thirdly, in Table 1, the item HER2 expression, it's better to categorise HER2 into 0, 1+, 2+ and FISH-.

Thank you very much for pointing out this issue. The item HER2 expression in Table 1 has been revised from “negative, 1+, 2+ and FISH-” to “0, 1+, 2+ and FISH-”.

Author response:

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

Reviewer #1 (Public review):

Summary:

Lodhiya et al. demonstrate that antibiotics with distinct mechanisms of action, norfloxacin and streptomycin, cause similar metabolic dysfunction in the model organism Mycobacterium smegmatis. This includes enhanced flux through the TCA cycle and respiration as well as a build-up of reactive oxygen species (ROS) and ATP. Genetic and/or pharmacologic depression of ROS or ATP levels protect M. smegmatis from norfloxacin and streptomycin killing. Because ATP depression is protective, but in some cases does not depress ROS, the authors surmise that excessive ATP is the primary mechanism by which norfloxacin and streptomycin kill M. smegmatis. In general, the experiments are carefully executed; alternative hypotheses are discussed and considered; the data are contextualized within the existing literature.

We thank the reviewer for the very comprehensive summary of the study.

Strengths:

The authors tackle a problem that is both biologically interesting and medically impactful, namely, the mechanism of antibiotic-induced cell death.

Experiments are carefully executed, for example, numerous dose- and time-dependency studies; multiple, orthogonal readouts for ROS; and several methods for pharmacological and genetic depletion of ATP.

There has been a lot of excitement and controversy in the field, and the authors do a nice job of situating their work in this larger context.

Inherent limitations to some of their approaches are acknowledged and discussed e.g., normalizing ATP levels to viable counts of bacteria.

We thank the reviewer for the encouraging comments.

Weaknesses:

All of the experiments performed here were in the model organism M. smegmatis. As the authors point out, the extent to which these findings apply to other organisms (most notably, slow-growing pathogens like M. tuberculosis) is to be determined. To avoid the perception of overreach, I would recommend substituting "M. smegmatis" for Mycobacteria (especially in the title and abstract).

At first glance, a few of the results in the manuscript seem to conflict with what has been previously reported in the (referenced) literature. In their response to reviewers, the authors addressed my concerns. It would also be ideal to include a few lines in the manuscript briefly addressing these points. (Other readers may have similar concerns).

In the first round of review, I suggested that the authors consider removing Figs. 9 and 10A-B as I believe they distract from the main point of the paper and appear to be the beginning of a new story rather than the end of the current one. I still hold this opinion. However, one of the strengths of the eLife model is that we can agree to disagree.

We acknowledge the reviewer’s concern and have changed title of the manuscript by including Mycobacterium smegmatis instead of Mycobacteria. The abstract already mentioned the same.

In the discussion section of the revised manuscript, we have already addressed and analysed our results extensively within the context of the available literature, regardless of whether our findings aligned with or differed from previous studies. We still believe that the mentioned discussion will help suffice to explain our results to the readers.

In this manuscript we also sought to assess the bacteria's ability to counteract drug induced stresses, contributing to our understanding of how antibiotic tolerance develop in Mycobacterium smegmatis. Results presented in Figure 9 clearly demonstrate that M.smegmatis attempt to reduce respiration by decreasing flux through the complete TCA cycle, thereby mitigating ROS and ATP production in response to antibiotics.  Additionally, the bacterial response also included increased expression of the protein Eis, which is exemplar for intrinsic drug resistance, with a concomitant increase in mutation frequency, thereby hinting at the development of antibiotic tolerance followed by resistance. We still believe that these data should be included to support our observations and they make the study more comprehensive.

Reviewer #2 (Public review):

Summary:

The authors are trying to test the hypothesis that ATP bursts are the predominant driver of antibiotic lethality of Mycobacteria

Strengths:

No significant strengths in the current state as it is written.

Weaknesses:

A major weakness is that M. smegmatis has a doubling time of three hours and the authors are trying to conclude that their data would reflect the physiology of M. tuberculossi that has a doubling time of 24 hours. Moreover, the authors try to compare OD measurements with CFU counts and thus observe great variabilities.

Comments on revisions:

I am surprised that the authors simply did not repeat the study in figure one with CFU counts and repeated in triplicate. Since this is M. smegmatis, it would take no longer than two weeks to repeat this experiment and replace the figure. I understand that obtaining CFU counts is much more laborious than OD measurements but it is necessary. Your graph still says that there is 0 bacteria at time 0, yet in your legend it says you started with 600,000 CFU/ml. I don't understand why this experiment was not repeated with CFU counts measured throughout. This is not a big ask since this is M. smegmatis but it appears that the authors do not want to repeat this experiment. Minimally, fix the graph to represent the CFU.

We acknowledge the reviewer’s concern and have changed title of the manuscript by specifying Mycobacterium smegmatis instead of Mycobacteria.

It is still not clear to the authors what the reviewer mean by OD measurements. All the data presented in the entire manuscript , including in Figure 1 are solely based on CFU measurements. So, as suggested by the reviewer, all experiments are already presented in terms of CFU.

Author response:

We thank the editors and reviewers for the constructive assessment. We plan to address the comments as follows:

Reviewer #1 (Public review):

We are generating a new cohort of Lv-TGFB2 overexpressing mice in which IOP will be compared under the anesthesia conditions that are identical for diurnal and nocturnal states. Parenthetically, we used the awake (diurnal) and isoflurane (nocturnal) anesthesia to mirror the conditions in the Patel et al (2021) PNAS study.

Reviewer #2 (Public review):

We are not sure what the Reviewer means by the “difference between the message and transcript data” and are not sure whether providing evidence about the TRPV4-dependence of the expression of fibrotic genes and canonical TGFb2 pathway genes fits within the scope of our study (which focuses on the TGFB2-dependence of TRPV4 expression and IOP regulation). We propose to address this by including new data about the TGFb2- and TRPV4 dependence of TRPV4 and Piezo1 expression. We could include information about the effect of TGFB2 on fibrosis-related genes from a (submitted study) in which we used RNASeq to investigate TGFB2 and TGFB2 + HC067047-dependence of gene expression in TM cells on a confidential basis but not include it in the revised manuscript.

- Re:  b-tubulin comment  [b-tubulin associates with the plasma membrane by binding to integral membrane proteins in the plasma and organellar membranes, through palmitoylation and attachment to linker proteins and as an integral component of exocytotic vesicles (Wolff, BBA 2009; Hogerheide et al., PNAS 2017). Together with b-actin and Gapdh it is often used as a loading control to assess cellular TRPV4 protein expression (e.g., https://www.cellsignal.com/products/primary-antibodies/trpv4-antibody/65893; Grove et al., Science Signaling 2019 and Moore et al., PNAS 2013).  Our qPCR and RNASeq studies show that TGFB2 does not affect b-tubulin expression]

- We will provide a higher resolution image for Fig. 4A

- Will address the Fig 5A and 6A comment [We thank the Reviewer for noticing the ambiguity and revised Figure Legends to clarify that “pre-injection” in Figures 5B and 6B refers to IOP measurements before the intracameral injection of HC-06  not pre-injection of lentiviral constructs].

-  We will address the issue of constitutive TRPV4 activity and Piezo1 involvement in the revised Discussion.

We hope this is sufficient information at this point but would be more than happy to provide more information if needed.

Thank you, we are very impressed by the eLife review protocols.

Author response:

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

eLife Assessment:

This study provides valuable insights, addressing the growing threat of multi-drug-resistant (MDR) pathogens by focusing on the enhanced efficacy of colistin when combined with artesunate and EDTA against colistin-resistant Salmonella strains. The evidence is solid, supported by comprehensive microbiological assays, molecular analyses, and in vivo experiments demonstrating the effectiveness of this synergic combination. However, the discussion on the clinical application challenges of this triple combination is incomplete, and it would benefit from addressing the high risk associated with using three potential nephrotoxic agents in vivo.

The development of novel pharmaceutical dosage forms, pharmacokinetic, pharmacodynamic and safety analysis of the triple combination will be further conducted in our next study to provide a theoretical basis for the next clinical drug use. The discussion of potential toxicity of AS, colistin, EDTA and the triple combination have been added in line 318 to 337.

Public Reviews:

Reviewer #1 (Public Review):

(1) The study focuses on a limited number of Salmonella strains, and broader testing on various MDR pathogens would strengthen the findings.

The number of COL-resistant clinical strains that actually used was larger than that mentioned in our original article, when evaluating the antimicrobial activities of AS, EDTA, COL alone or drug combinations. But, considering that there were superfluous results of mcr-1 positive Salmonella strains, we omitted these results (Table supplement 7 and 8 in revised supplement materials) to avoid redundant data presentation in the original article. Additionally, much more gram-negative and -positive MDR bacteria, such as Klebsiella pneumoniae, Pseudomonas aeruginosa and Staphylococcus aureus will be selected for the next study including the development of novel pharmaceutical dosage forms, pharmacokinetic, pharmacodynamic and safety analysis et al.

(2) While the study elucidates several mechanisms, further molecular details could provide deeper insights into the interactions between these drugs and bacterial targets.

In our next study, further molecular details will be focused on the regulatory targets of CheA and SpvD-related pathways, as well as the precise inhibition targets of MCR protein by the triple combination, through the generation of deletion or point mutations, and analysis of intermolecular interactions.

(3) The time-kill experiment was conducted over 12 hours instead of the recommended 24 hours. To demonstrate a synergistic effect among the drugs, a reduction of at least 2 log10 in colony count should be shown in a 24-hour experiment. Additionally, clarifying the criteria for selecting drug concentrations is important to improve the interpretation of the results.

The time-kill experiment of 24 hours have been re-executed and could be used to replace the Figure 1 in the original paper. The New Figure 1 has been uploaded and the change do not affect our interpretation of the result.

Although in vitro studies have determined that with increasing dose of AS and EDTA, the antibacterial synergistic activity was gradually enhanced, and meanwhie, may also resulting in more toxic side effects. Thus, in our study, the 1/8 MICs of AS and EDTA were selected to ensure excellent antibacterial activity whereas minimize the potential toxicity. The instructions on the selection of drug concentration have been added in line 323 to 326.

(4) While the combination of EDTA, artesunate, and colistin shows promising in vitro results against Salmonella strains, the clinical application of this combination warrants careful consideration due to potential toxicity issues associated with these compounds.

The development of novel pharmaceutical dosage forms, pharmacokinetic, pharmacodynamic and safety analysis of the triple combination will be further conducted in our next study to provide a theoretical basis for the next clinical drug use.

Reviewer #2 (Public Review):

(1) The study by Zhai et al describes repurposing of artesunate, to be used in combination with EDTA to resensitize Salmonella spp. to colistin. The observed effect applied both to strains with and without mobile colistin resistance determinants (MCR). It was already known that EDTA in combination with colistin has an inhibitory effect on MCR-enzymes, but at the same time, both colistin and EDTA can contribute to nephrotoxicity, something which is also true for artesunate. Thus, the triple combination of three nephrotoxic agents has significant challenges in vivo, which is not particularly discussed in this paper.

The discussion of potential toxicity of triple combination has been added in line 318 to 337.

(2) The selection of strains is not very clear. Nothing is known about the sequence types of the strains or how representative they are for strains circulating in general. Thus, it is difficult to generalize from this limited number of isolates, although the studies done in these isolates are comprehensive.

The tested strains in this study were all COL-resistant clinical isolates, and the genome sequencing and comparative analysis of these strains have not been analyzed. The antibacterial activities of different antimicrobial drugs against the S16 and S30 strains have been measured and listed in the Table supplement 9 within revised supplement materials. Considering that the number of COL-resistant clinical strains that actually used was larger than that mentioned in our original article (see the NO.1 response to the Public Reviewer #1), we think that the results obtained in this study could be representative to some extent.

(3) Nothing is known about the susceptibility of the strains to other novel antimicrobial agents. Colistin has a limited role in the treatment of gram-negative infections, and although it can be used sometimes in combination, it is not clear why it would be combined with two other nephrotoxic agents and how this could have relevance in a clinical setting.

The antibacterial activities of different antimicrobial drugs against the S16 and S30 strains have been measured and listed in the Table supplement 9 within revised supplement materials. Additionally, the discussion of potential toxicity of triple combination has been added in line 318 to 337.

(4) It is not clear whether their transcriptomics analysis should at least be carried out in duplicate for reasons of being able to assess reproducibility. It is also not clear why the samples were incubated for 6 hours - no discussion is presented on the selection of a time point for this.

As it can be seen from the time kill curves that the survival number of bacteria started to decrease after 4 h incubation of drug combinations. If the incubation time is too short (for example less than 4 h), the differentially expressed genes can not be fully revealed, while too long incubation time (such as 8 h and 12 h) may lead to a significant CFU reduction of bacteria, and result in inaccurate sequencing results. Therefore, we selected the incubation time 6 h, at which point drugs exhibited  significant antibacterial effects and there were also enough survival bacteria in the sample for transcriptome analysis. Each sample had three replications to preserve the accuracy of results.

(5) Discussion is lacking on the reproducibility and selection of details for the methodology.

The results obtained in this paper have been repeated several times, which indicated that the detailed operation steps described in the materials and methods section were reproducibility. To avoid redundancy, we did not include too much details in the discussion section.

Reviewer #3 (Public Review):

(1) Number of strains tested.

The number of COL-resistant clinical strains that actually used was larger than that mentioned in our original article (see the NO.1 response to the Public Reviewer #1)

(2) Response to comment: Lack of data on cytotoxicity.

The pharmacokinetic, pharmacodynamic and safety analysis of the triple combination will be further conducted in our next study to provide a theoretical basis for the next clinical drug use.

Recommendations For The Authors:

Reviewer #1 (Recommendations For The Authors):

(1) Introduction:

The introduction should provide more context about the pathogen Salmonella, its significance in both human and veterinary medicine, and the impact of colistin resistance in these pathogens. Salmonella is a leading cause of foodborne illnesses worldwide, resulting in substantial morbidity and mortality. It can cause a range of diseases, from gastroenteritis to more severe systemic infections like typhoid fever and invasive non-typhoidal salmonellosis. In veterinary medicine, Salmonella infections can lead to significant economic losses in livestock industries due to illness and death among animals, as well as through the contamination of animal products.

The description has been added in the introduction section in line 47 to 53.

(2) Results and Discussion:

(1) While the combination of EDTA, artesunate, and colistin shows promising in vitro results against Salmonella, the clinical application of this combination warrants careful consideration due to potential toxicity issues associated with these compounds. Colistin is known for nephrotoxicity and neurotoxicity, limiting its use to severe cases where the benefits outweigh the risks. EDTA, as a chelating agent, can disrupt essential metal ions in the body, posing risks of metabolic imbalances. Although it has clinical applications, primarily in cases of heavy metal poisoning, its use as an adjuvant in antibiotics may present risks. Although generally well-tolerated for malaria, interactions of artesunate with other drugs and long-term safety in combined therapies require thorough investigation.

The discussion of potential toxicity of triple combination has been added in line 318 to 337.

(2) Table 1: The manuscript mentions that some strains used in the study are mcr-positive and mcr-negative. It is important to indicate in Table 1, in addition to the identification of Salmonella species, which strains are mcr-positive or mcr-negative.

The relevant information has been added in Table 1.

(3) Figure 2: What is the authors' hypothesis regarding the growth curves labeled "a" and "e" where strains JS and S16 resume growth 12 hours after treatment with AS? In the legend of Figure 2, describe what was used as the "positive control group."

The growth curves labeled “a” and “e” were in Figure 1. After incubated with AC for 8 h, the survival CFUs of JS and S16 strains showed a slightly reduction, but there were still living cells. Since the bactericidal activity of AC is not strong enough to exert sustained bactericidal activity, these remaining living cells will resume growth after treatment with AC for 12 h. The “positive control group” in the legend of Figure 2 has been indicated in line 724.

(4) What is the authors' hypothesis for the differences observed in the transcriptome and metabolome?

The changes in gene transcription level may cause corresponding changes in protein level, but these proteins are not all involved in the bacterial metabolic process. For example, MCR protein  is encoded by the COL resistance related gene mcr, which mediates the modification of lipid A, but are not involved in the cellular metabolic process. Therefore, the transcriptome change of mcr gene may affect the protein production of MCR, nor the bacterial metabolic processes, so there are differences observed in the transcriptome and metabolome.

(5) In some parts of the text, the authors state that artesunate and EDTA potentiate the action of colistin, which is a bacteriostatic drug. However, in other parts, the authors describe the effect of the AEC combination as bacteriostatic (Abstract: line 32; Results: line 179). How do the authors explain this inconsistency?

The artesunate and EDTA could be regarded as “adjuvants” for the bacteriostatic drug colistin. Adjuvants itself act no or weak antibacterial effect on bacteria. For antimicrobial drugs, the “adjuvants” are compounds that generally used in combination with antibacterial drugs to re-sensitizing bacteria that have developed drug resistance. Thus, in this paper the AEC combination could be regared as bacteriostatic.

(6) According to Brennan & Kirby (2019; doi: 10.1016/j.cll.2019.04.002), to evaluate the synergism between different drug combinations, bacterial growth curves need to be assessed over 24 hours. If the colony count is {greater than or equal to} 2 log10 lower than that of the most active antimicrobial alone, the combination is considered synergistic. Based on the growth curve results shown in Figure 1, the experiment was conducted for 12 hours, and in some cases, only a small reduction in growth was observed, even at the maximum concentration of colistin. Moreover, in some cases, the curve resumes rising between 8 and 12 hours. What is the authors' hypothesis in this case? It is important to conduct the assay over 24 hours to confirm the synergism between these drugs.

The time-kill experiment of 24 hours have been re-executed and could be used to replace the Figure 1 in the original paper. Additionally, the phenomenon that “the curve resumes rising between 8 and 12 hours” has been explained in the response to comment of “Reviewer #1 (Recommendations For The Authors), Results and Discussion, (3) Figure 2”.

(7) To prove that CheA and SpvD play a critical role in the effect of the AEC combination, deletion of these genes should be performed, and the mutant strains should be tested.

The deletion of cheA and spvD will be carried out in our next study.

(8) To demonstrate that the flagellum is no longer assembled, a transmission electron microscopy image using antibodies against flagellin should be performed, along with motility tests.

The motility assays have been performed and displayed as Figure supplement 5 in the revised supplement materials.

(9) Figure 7: In the X-axis legend, specify what "model" refers to.

The “model” refers to the PBS control group that mice were treated with PBS after the intraperitoneal injection of 100 µL bacterial solution (1.31 × 10⁵ CFU).

(10) Figure 8 Legend: In the legend of Figure 8 (line 717), are the authors referring to E. coli or Salmonella?

It referred to Salmonella, which has already been illustrated in the headline of Figure 8 in the revised manuscript.

(3) Materials and Methods:

(1) Bacterial Strains and Agents: It would be beneficial to include in the table the species of the strains used in the study, as well as the concentrations of colistin, artesunate, and EDTA utilized (lines 321 - 332).

We have ever tried to add the above information to Table 1, but the addition of this information would make the table too large and beyond the margins, which is not conducive to the layout design of the table, so we chose to display these information in the materials and methods section instead of the table.

(2) Antibacterial Activity In Vitro: Ensure clarity and well-defined ranges for the concentrations of colistin, EDTA, and artesunate used separately and in combinations (lines 335 - 344).

The drug concentrations have been listed in line 369 to 371.

(3) Time-Kill Assays: Clarify the criteria for selecting concentrations, whether based on MICs or peak and trough concentrations relevant to human and animal treatments with colistin (lines 345 - 351).

Although in vitro studies have determined that with increasing dose of AS and EDTA, the antibacterial synergistic activity was gradually enhanced, and meanwhie, may also resulting in more toxic side effects. Thus, in our study, the 1/8 MICs of AS and EDTA were selected to ensure excellent antibacterial activity whereas minimize the potential toxicity. The instructions on the selection of drug concentration have been added in line 323 to 326.

(4) General Corrections: Throughout the manuscript, correct typographical errors and consistently include the concentration values in mg/L alongside the MIC fractions. Specify the strains used for all experiments to ensure clarity. In the manuscript, the term "medication regimens" is used to describe the experimental setups involving different combinations of drugs tested in vitro. To improve accuracy and clarity, it is recommended to use the term "drug combination" instead. This term is more appropriate for in vitro experiments and will help avoid confusion with clinical treatment protocols.

The typographical errors have been checked and corrected throughout the manuscript, and the “medication regimens” have been replaced by “drug combinations”.

Reviewer #2 (Recommendations For The Authors):

Please see above for recommendations on what can be done to improve the manuscript.

While other omics analyses have been conducted herein, the authors do not comment on the genomic analysis of their own strains. It would have been a natural step to sequence all the strains used in the experiments.

Due to limited program funding, the genome sequencing and comparative analysis of these strains have not been analyzed. The antibacterial activities of different antimicrobial drugs against the S16 and S30 strains have been measured and listed in the Table supplement 9 within revised supplement materials.

Some minor comments:

(1) There are some spelling errors - e.g. "bacteria strains" instead of "bacterial strains".

The grammar and spelling errors have been corrected throughout the manuscript.

(2) I would avoid words like "unfortunately".

The word “unfortunately” has been changed.

(3) Some MIC-values in Table 1 seem incorrect - e.g. 24 mg/L. This is not a 2-log value - the value should be 32 mg/L if the dilution series has been carried out correctly.

We are so sorry for the mistake. The data has been corrected, and we also checked other data.

Reviewer #3 (Recommendations For The Authors):

Below are some suggestions.

(1) Sentences L47 & L48 "Infections with antibiotic-resistant pathogens, especially carbapenemase-producing Enterobacteriaceae, represent an impending catastrophe of a return to the pre-antibiotic era" - this is slightly exaggerated! I also wonder if we need to use Enterobacterales instead of Enterobacteriaceae.

The sentences in L47 & L48 have been changed. We googled the “carbapenemase-producing Enterobacteriaceae” and found it is a high-frequency word in numerous reports.

(2) L48. The drying up of the antibiotic discovery pipeline is NOT necessarily the reason to use colistin as a drug of last resort!

The sentence has been revised.

(3) The manuscript requires extensive English editing but has merit based on the strong compilation of data.

We have optimized and revised the writing of the whole article.

(4) I suggest the authors have some data on the cytotoxicity of AS alone, colistin alone, and both of them against eucaryotic cells (Caco-) and if possible determine IS (index selectivity). This additional experiment will strengthen the quality of the manuscript. The authors must also explain how to put such tri-therapy into practice.

The development of novel pharmaceutical dosage forms, pharmacokinetic, pharmacodynamic and safety analysis of the triple combination will be further conducted in our next study to provide a theoretical basis for the next clinical drug use. The discussion of potential toxicity of AS, colistin, EDTA and the triple combination have been added in line 318 to 337.

Author response:

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

Reviewer #1 (Public review):

Summary:

In this study, Bu et al examined the dynamics of TRPV4 channel in cell overcrowding in carcinoma conditions. They investigated how cell crowding (or high cell confluence) triggers a mechano-transduction pathway involving TRPV4 channels in high-grade ductal carcinoma in situ (DCIS) cells that leads to large cell volume reduction (or cell volume plasticity) and proinvasive phenotype.

In vitro, this pathway is highly selective for highly malignant invasive cell lines derived from a normal breast epithelial cell line (MCF10CA) compared to the parent cell line, but not present in another triple-negative invasive breast epithelial cell line (MDA-MB-231). The authors convincingly showed that enhanced TRPV4 plasma membrane localization correlates with highgrade DCIS cells in patient tissue samples.

Specifically in invasive MCF10DCIS.com cells, they showed that overcrowding or overconfluence leads to a decrease in cell volume and intracellular calcium levels. This condition also triggers the trafficking of TRPV4 channels from intracellular stores (nucleus and potentially endosomes), to the plasma membrane (PM). When these over-confluent cells are incubated with a TRPV4 activator, there is an acute and substantial influx of calcium, attesting to the fact that there are a high number of TRPV4 channels present on the PM. Long-term incubation of these over-confluent cells with the TRPV4 activator results in the internalization of the PMlocalized TRPV4 channels.

In contrast, cells plated at lower confluence primarily have TRPV4 channels localized in the nucleus and cytosol. Long-term incubation of these cells at lower confluence with a TRPV4 inhibitor leads to the relocation of TRPV4 channels to the plasma membrane from intracellular stores and a subsequent reduction in cell volume. Similarly, incubation of these cells at low confluence with PEG 3000 (a hyperosmotic agent) promotes the trafficking of TRPV4 channels from intracellular stores to the plasma membrane.

Strengths:

The study is elegantly designed and the findings are novel. Their findings on this mechanotransduction pathway involving TRPV4 channels, calcium homeostasis, cell volume plasticity, motility, and invasiveness will have a great impact in the cancer field and are potentially applicable to other fields as well. Experiments are well-planned and executed, and the data is convincing. The authors investigated TRVP4 dynamics using multiple different strategies- overcrowding, hyperosmotic stress, and pharmacological means, and showed a good correlation between different phenomena.

Weaknesses:

A major emphasis in the study is on pharmacological means to relate TRPV4 channel function to the phenotype. I believe the use of genetic means would greatly enhance the impact and provide compelling proof for the involvement of TRPV4 channels in the associated phenotype.

In this regard, I wonder if siRNA-mediated knockdown of TRPV4 in over-confluent cells (or knockout) would lead to an increase in cell volume and normalize the intracellular calcium levels back to normal, thus ultimately leading to a decrease in cell invasiveness.

We greatly appreciate the positive feedback regarding the design of our study and the novelty of our findings. We also acknowledge the valuable suggestion to complement our pharmacological approaches with genetic manipulation of TRPV4.

In response to the comment regarding siRNA-mediated knockdown or knockout of TRPV4, we fully agree that this would further substantiate our findings. In the revised manuscript, we implemented shRNA targeting TRPV4 to investigate its functional effects on intracellular calcium level changes, cell volume plasticity, and invasiveness phenotypes, assessed through singlecell motility assays under cell crowding or hyperosmotic stress. These results have been incorporated into the revised manuscript, and detailed descriptions of these findings are included below.

Using the shRNA approach that resulted in ~50% reduction of TRPV4 expression

(Supplementary Figure 6A and 6B show TRPV4 expression levels via IF and immunoblots, respectively), we examined the effect of reduced TRPV4 on intracellular calcium levels in MCF10DCIS.com cells under normal density (ND) and stress conditions (confluent; Con and hyperosmotic; PEG) using Fluo-4 AM imaging (Fig. 4S-X). We found that shRNA TRPV4 slightly decreased calcium levels in ND cells, likely due to fewer active calcium channels at the plasma membrane resulting from lower TRPV4 expression (as shown in the summary plot in Fig. 4W). With fewer active calcium channels, cells treated with shRNA TRPV4 exhibited less reduction in intracellular calcium levels under cell crowding conditions compared to control cells. Additionally, hyperosmotic stress using PEG 300 induced smaller calcium spikes in shRNA cells compared to the significant spike observed in control cells. This reduced calcium response to Con and hyperosmotic stress in shRNA cells was reflected in the decreased cell volume reduction by PEG 300 shown in Fig. 4Y. Consequently, shRNA-mediated TRPV4 reduction impaired cell volume plasticity in MCF10DCIS.com cells and abolished the pro-invasive mechanotransduction capability involving cell volume reduction, as evidenced by no increase in cell motility (both cell diffusivity and directionality) under hyperosmotic conditions (Fig. 5H-J). These findings demonstrate the critical role of TRPV4 in conferring pro-invasive

mechanotransduction capability to MCF10DCIS.com cells through cell volume reduction.

Reviewer #2 (Public review):

Summary:

The metastasis poses a significant challenge in cancer treatment. During the transition from non-invasive cells to invasive metastasis cells, cancer cells usually experience mechanical stress due to a crowded cellular environment. The molecular mechanisms underlying mechanical signaling during this transition remain largely elusive. In this work, the authors utilize an in vitro cell culture system and advanced imaging techniques to investigate how non-invasive and invasive cells respond to cell crowding, respectively.

Strengths:

The results clearly show that pre-malignant cells exhibit a more pronounced reduction in cell volume and are more prone to spreading compared to non-invasive cells. Furthermore, the study identifies that TRPV4, a calcium channel, relocates to the plasma membrane both in vitro and in vivo (patient samples). Activation and inhibition of the TRPV4 channel can modulate the cell volume and cell mobility. These results unveil a novel mechanism of mechanical sensing in cancer cells, potentially offering new avenues for therapeutic intervention targeting cancer metastasis by modulating TRPV4 activity. This is a very comprehensive study, and the data presented in the paper are clear and convincing. The study represents a very important advance in our understanding of the mechanical biology of cancer.

Weaknesses:

However, I do think that there are several additional experiments that could strengthen the conclusions of this work. A critical limitation is the absence of genetic ablation of the TRPV4 gene to confirm its essential role in the response to cell crowding.

We are deeply grateful for the positive assessment of our study and its contribution to advancing our understanding of mechanical signaling in cancer progression. We also greatly appreciate the suggestion to incorporate genetic ablation experiments to further validate the role of TRPV4 in cell crowding responses.

As noted in our response to Reviewer #1, we employed an shRNA approach to investigate the functional effects of TRPV4 knockdown on intracellular calcium level changes, cell volume plasticity, and invasiveness phenotypes. We assessed these effects using Fluo-4 AM calcium assay, single-cell volume measurements, and single-cell motility assays under cell crowding or hyperosmotic stress. These results have been incorporated into the revised manuscript and are described in detail in our response to Reviewer #1's "weaknesses" comment.

Reducing TRPV4 expression levels by shRNA diminished mechanosensing intracellular calcium changes under cell crowding and hyperosmotic conditions using PEG 300 treatment. Furthermore, a significantly reduced cell volume plasticity was observed under hyperosmotic conditions in shRNA treated cells compared to control cells (Fig. 4S-X). This diminished mechanosensing capability abolished the pro-invasive mechanotransduction effect, as assessed by single cell motility under hyperosmotic conditions (Fig. 5H-J). These findings demonstrate the critical role of TRPV4 in conferring pro-invasive mechanotransduction capability to MCF10DCIS.com cells through cell volume reduction.

Reviewer #1 (Recommendations for the authors):

The way the results or discussion section is written. It was a little confusing for me to relate to some phenomena. For example, it is not clear how TRPV4 inhibition (due to overcrowding) leads to a decrease in intercellular calcium levels, especially when TRPV4 channels were intercellular (not on the PM) to begin with (in normal density (ND) conditions). Along the same lines, how GSK219 causes a dip in calcium levels in ND cells when TRPV4 channels are primarily intercellular (Figure 4E). If most of the TRPV4 channels that are translocated to the PM in response to cell crowding are in an inactive state, how do they confer enhanced cell volume plasticity relative to non-invasive cell lines?

Thank you very much for raising this important point. We fully agree with your concern and have significantly revised the manuscript to clarify this aspect. Specifically, we have emphasized that a modest level of TRPV4 channels are constitutively active at the plasma membrane in normal density (ND) cells. This is now discussed in detail in the context of Fig. 4:

Page 14: “Considering these factors, we hypothesized that cell crowding might inhibit calcium-permeant ion channels that are constitutively active at the plasma membrane, including TRPV4, which would then lower intracellular calcium levels and subsequently reduce cell volume via osmotic water movement.”

Page 16-17: “… However, the temporal profile of Fluo-4 intensity in Fig. 4E, which corresponds to the time points marked in Fig. 4D (t₁: baseline and t₂: dip), clearly shows the dip at t₂, indicated by ΔCa (the vertical dashed line between the dip and baseline). This modest Fluo-4 dip at t₂ represents the inhibition of activity by GSK219 on a small population of constitutively active TRPV4 channels at the plasma membrane under ND conditions.

In Con cells, 1 nM GSK219 caused a smaller dip in Fluo-4 intensity compared to the one observed in ND cells, with no subsequent changes. This is likely due to fewer constitutively active TRPV4 at the plasma membrane in Con cells than in ND cells. …These findings suggest that a substantial portion of TRPV4 channels relocated to the plasma membrane under cell crowding was inactive, and some constitutively active TRPV4 channels already present in the membrane became inactive as a result of cell crowding.”

'Internalization' might be a better word than 'uptake' in the following line in the results section

"...activating TRPV4 under cell crowding conditions triggered channel uptake, indicating that TRPV4 trafficking depended on the channel's activation status."

Thank you very much for this suggestion. As recommended, we replaced ‘uptake’ with internalization’ on page 18: 

“However, in Con cells, where a large number of inactive TRPV4 channels are likely located at the plasma membrane, GSK101 treatment notably reduced plasma membrane-associated TRPV4 in a dose-dependent manner through internalization (Fig. 4O, 4Q), consistent with previous findings65. These data suggest that plasma membrane TRPV4 levels were largely

regulated by the channel activity status. Specifically, channel activation led to the internalization of TRPV4, while channel inhibition promoted the relocation of TRPV4 to the plasma membrane.”

1. Out of curiosity:

2. Is there any information on what the intercellular TRPV4 channels are doing in the cytosol and in the nucleus? Is there any role of intercellular calcium stores in the proposed pathway?

We greatly appreciate this insightful question. Although we were unable to find studies specifically exploring the roles of cytosolic TRPV4, a recent study (Reference 74) identified a role for nuclear TRPV4 in regulating calcium within the nucleus. We speculate that when TRPV4 activity is severely impaired, such as with additional TRPV4 inhibition under cell crowding conditions, some TRPV4 channels may be redirected to the nucleus. This redistribution could help maintain nuclear calcium homeostasis.

This discussion is included on page 18 of the manuscript:

“These findings suggest that further TRPV4 inhibition under crowding conditions triggers a distinct trafficking alteration. Recent studies have implicated nuclear TRPV4 in regulating nuclear Ca2+ homeostasis and Ca2+-regulated transcription74. In light of this study and our findings, TRPV4 may relocate to the nucleus as a compensatory mechanism to maintain nuclear calcium regulation. This relocation could reflect an adaptive response to preserve calcium-dependent transcriptional programs or other nuclear processes essential for cell survival under mechanical stress.”

One recommendation is to add some explanation or some minor details for the convenience of the reader. For example:

At normal or lower confluence, cells show an acute large dip in intercellular calcium when an inhibitor is applied implying that there are a few TRPV4 channels on the PM and they are constitutively active.

Thank you very much for highlighting this important point and for the helpful suggestion to improve clarity. We have significantly revised the text associated with Fig. 4 to ensure this point is clear. Specifically, we have added the following explanation on page 16:

"This modest Fluo-4 dip at t2 represents the inhibition of activity by GSK219 on a small population of constitutively active TRPV4 channels at the plasma membrane under ND conditions."

Reviewer #2 (Recommendations for the authors):

(1) Figure 1. The authors frequently change the medium to prevent acidification in overconfluent cultures. A cell viability assay should be performed to ensure that the over-confluent cells remain healthy and viable during the experiments. There are commercial kits that can be easily used to quantify the number of viable cells and the extent of cell toxicity. The number of viable cells would provide a more reliable basis for comparison between normal density and overconfluent conditions.

Thank you very much for raising this important point. We have consistently observed that cell crowding does not induce significant cell death in MCF10DCIS.com cells. To address your recommendation, we performed a viability assay using propidium iodide (PI) to selectively stain dead cells and WGA-488 to stain all live cells. Cell death was quantified under normal density (ND) conditions and at 1, 3, 5, 7, and 10 days post-confluence.

Our results indicate that cells remain similarly viable post-confluence, with minimal cell death

(~1.5%) compared to ND cells (~0.75%). These findings are summarized in Supplementary Figure 2, demonstrating that over-confluent cultures remain healthy and viable during the experiments.

(2) Figure 2. To determine whether the reduction in cell volume is reversible, over-confluent cells can be further diluted back to normal density. Additionally, the reversibility of TRPV4 channel trafficking to the plasma membrane should be assessed under these conditions in IF experiments and cell surface biotinylation.

Thank you for this suggestion. We reseeded the previously overcrowded (OC) cells at normal density and observed that their TRPV4 distribution predominantly returned to being intracellular, with only modest plasma membrane localization, as shown by line analysis (Supplementary Figure 10A-C, page 13). Furthermore, their invasiveness decreased to levels comparable to the original normal density (ND) cells (Supplementary Figure 3C and 3E, page 6). These results demonstrate the reversibility of TRPV4 trafficking changes and the increase in invasiveness under mechanical stress.

Page 6. "The enhanced invasiveness of MCF10DCIS.com cells under cell crowding was largely reversible. When OC cells were reseeded at normal density for invasion assays, their invasive cell fraction decreased to approximately 15%, slightly lower (p = 0.012) than the initial value of around 24% (Suppl. Fig. 3C, 3E)."

Page 13. “We investigated whether TRPV4 relocation to the plasma membrane induced by cell crowding is reversible, as suggested by its impact on invasiveness (Suppl. Fig. 3E). To test this, previously OC MCF10DCIS.com cells were reseeded under ND conditions. We then assessed TRPV4 localization via immunofluorescence (IF) imaging to determine if most channels returned to the cytoplasm and could be relocated to the plasma membrane under mechanical stress, such as hyperosmotic conditions. Consistent with their initial ND state, reseeded ND MCF10DCIS.com cells displayed intracellular TRPV4 distribution (Suppl. Fig. 10A). Upon exposure to hyperosmotic stress (74.4 mOsm/Kg PEG300), TRPV4 was again relocated to the plasma membrane (Suppl. Fig. 10B). These findings, quantified through line analysis (Suppl. Fig. 10C), demonstrate that the mechanosensing response of MCF10DCIS.com cells is reversible.”

(3) Figure 3B. A control using intracellular proteins such as GAPDH or Tubulin is missing. Including this control would help exclude the possibility of cell rupture or compromised cell membranes in crowded environments, which is very common in a cell crowding environment.

Thank you very much for pointing this out. The control lanes (GAPDH) were already included in the full gel results shown in Supplementary Figure 5. For the immunoprecipitation and immunoblotting of surface-biotinylated cell lysates, we did not expect to detect GAPDH; however, some GAPDH signals were still observed. As shown for MCF10DCIS.com cells, less GAPDH was detected under OC conditions, but the immunoprecipitated samples displayed significantly higher levels of TRPV4 on the cell surface compared to ND cells (Supplementary Figure 5A). For the whole cell lysates, TRPV4 protein levels were comparable across different cell lines based on the immunoblot results, with consistent GAPDH signals serving as a loading control (Supplementary Figure 5B).

(4) Figure 4. To convincingly demonstrate TRPV4 relocation to the plasma membrane, IF should be performed under non-permeable conditions (i.e., without detergents like saponin). This approach ensures that only plasma membrane proteins are accessible to antibodies, reducing intracellular background. The same approach should be applied to Piezo1 and TfR.

Thank you for this suggestion. We observed that under non-permeable conditions, primary antibodies could still access intracellular proteins. To address this issue, we employed extracellular-binding TRPV4 antibodies to selectively detect TRPV4 relocation to the plasma membrane under hyperosmotic conditions (74.4 mOsm/kg PEG 300) in live MCF10DCIS.com cells, as shown in Supplementary Figure 9. These results clearly demonstrate the plasma membrane relocation of TRPV4 under hyperosmotic conditions, distinguishing it from control conditions. Unfortunately, we were unable to identify high-affinity extracellular-binding antibodies for Piezo1 and TfR. Nevertheless, our findings strongly support the mechanosensing plasma membrane relocation of TRPV4.

Essential Weakness:

Throughout the study, only TRPV4 inhibitors and activators were used to show that TRPV4 relocation is associated with intracellular calcium concentration and cell size changes. It is crucial to use TRPV4 KO or KD cells to confirm that the observed effects are specific to TRPV4 and not due to off-target effects on other proteins. Additionally, fusing a plasma membrane targeting sequence to TRPV4 to make a constitutive plasma membrane-localized construct could demonstrate the opposite effect.

Thank you very much for this important comment. As noted in our response to Reviewer #1, we employed an shRNA approach to investigate the functional effects of TRPV4 knockdown on intracellular calcium level changes, cell volume plasticity, and invasiveness phenotypes. We assessed these effects using Fluo-4 AM calcium assay, single-cell volume measurements, and single-cell motility assays under cell crowding or hyperosmotic stress. These results have been incorporated into the revised manuscript and are described in detail in our response to Reviewer #1's "weaknesses" comment.

Reducing TRPV4 expression levels by shRNA diminished mechanosensing intracellular calcium changes under cell crowding and hyperosmotic conditions using PEG 300 treatment. Furthermore, a significantly reduced cell volume plasticity was observed under hyperosmotic conditions in shRNA treated cells compared to control cells (Fig. 4S-X). This diminished mechanosensing capability abolished the pro-invasive mechanotransduction effect, as assessed by single cell motility under hyperosmotic conditions (Fig. 5H-J). These findings demonstrate the critical role of TRPV4 in conferring pro-invasive mechanotransduction capability to MCF10DCIS.com cells through cell volume reduction.

Minor Points:

The introduction section is poorly written; many results currently included in the introduction would be more appropriately placed in the discussion section. The long redundant introduction makes the article hard to read through.

Thank you very much for pointing this out. In the revised introduction, we have significantly reduced references to the results, streamlining the section to make it more concise and focused. This adjustment ensures the introduction is clearer and avoids redundancy, improving the readability of the manuscript.

Author response:

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

Reviewer #1 (Public review):

Summary:

"Neural noise", here operationalized as an imbalance between excitatory and inhibitory neural activity, has been posited as a core cause of developmental dyslexia, a prevalent learning disability that impacts reading accuracy and fluency. This is study is the first to systematically evaluate the neural noise hypothesis of dyslexia. Neural noise was measured using neurophysiological (electroencephalography [EEG]) and neurochemical (magnetic resonance spectroscopy [MRS]) in adolescents and young adults with and without dyslexia. The authors did not find evidence of elevated neural noise in the dyslexia group from EEG or MRS measures, and Bayes factors generally informed against including the grouping factor in the models. Although the comparisons between groups with and without dyslexia did not support the neural noise hypothesis, a mediation model that quantified phonological processing and reading abilities continuously revealed that EEG beta power in the left superior temporal sulcus was positively associated with reading ability via phonological awareness. This finding lends support for analysis of associations between neural excitatory/inhibitory factors and reading ability along a continuum, rather than as with a case/control approach, and indicates the relevance of phonological awareness as an intermediate trait that may provide a more proximal link between neurobiology and reading ability. Further research is needed across developmental stages and over a broader set of brain regions to more comprehensively assess the neural noise hypothesis of dyslexia, and alternative neurobiological mechanisms of this disorder should be explored.

Strengths:

The inclusion of multiple methods of assessing neural noise (neurophysiological and neurochemical) is a major advantage of this paper. MRS at 7T confers an advantage of more accurately distinguishing and quantifying glutamate, which is a primary target of this study. In addition, the subject-specific functional localization of the MRS acquisition is an innovative approach. MRS acquisition and processing details are noted in the supplementary materials using according to the experts' consensus recommended checklist (https://doi.org/10.1002/nbm.4484). Commenting on rigor the EEG methods is beyond my expertise as a reviewer.

Participants recruited for this study included those with a clinical diagnosis of dyslexia, which strengthens confidence in the accuracy of the diagnosis. The assessment of reading and language abilities during the study further confirms the persistently poorer performance of the dyslexia group compared to the control group.

The correlational analysis and mediation analysis provide complementary information to the main case-control analyses, and the examination of associations between EEG and MRS measures of neural noise is novel and interesting.

The authors follow good practice for open science, including data and code sharing. They also apply statistical rigor, using Bayes Factors to support conclusions of null evidence rather than relying only on non-significant findings. In the discussion, they acknowledge the limitations and generalizability of the evidence and provide directions for future research on this topic.

Weaknesses:

Though the methods employed in the paper are generally strong, the MRS acquisition was not optimized to quantify GABA, so the findings (or lack thereof) should be interpreted with caution. Specifically, while 7T MRS affords the benefit of quantifying metabolites, such as GABA, without spectral editing, this quantification is best achieved with echo times (TE) of 68 or 80 ms in order to minimize the spectral overlap between glutamate and GABA and reduce contamination from the macromolecular signal (Finkelman et al., 2022, https://doi.org/10.1016/j.neuroimage.2021.118810). The data in the present study were acquired at TE=28 ms, and are therefore likely affected by overlapping Glu and GABA peaks at 2.3 ppm that are much more difficult to resolve at this short TE, which could directly affect the measures that are meant to characterize the Glu/GABA+ ratio/imbalance. In future research, MRS acquisition schemes should be optimized for the acquisition of Glutamate, GABA, and their relative balance.

As the authors note in the discussion, additional factors such as MRS voxel location, participant age, and participant sex could influence associations between neural noise and reading abilities and should be considered in future studies.

We have modified Figure 2 and revised the paragraph discussing the MRS methodological limitations in accordance with Reviewer #1's recommendations. Additionally, we have included the CRLB and linewidth thresholds in the Results section. Furthermore, a new figure showing the correlations between EEG and MRS biomarkers has been added (Figure 3).

Appraisal:

The authors present a thorough evaluation of the neural noise hypothesis of developmental dyslexia in a sample of adolescents and young adults using multiple methods of measuring excitatory/inhibitory imbalances as an indicator of neural noise. The authors concluded that there was not support for the neural noise hypothesis of dyslexia in their study based on null significance and Bayes factors. This conclusion is justified, and further research is called for to more broadly evaluate the neural noise hypothesis in developmental dyslexia.

Impact:

This study provides an exemplar foundation for the evaluation of the neural noise hypothesis of dyslexia. Other researcher may adopt the model applied in this paper to examine neural noise in various populations with/without dyslexia, or across a continuum of reading abilities, to more thoroughly examine evidence (or lack thereof) for this hypothesis. Notably, the lack of evidence here does not rule out the possibility for a role of neural noise in dyslexia, and the authors point out that presentation with co-occurring conditions, such as ADHD, may contribute to neural noise in dyslexia. Dyslexia remains a multi-faceted and heterogenous neurodevelopmental condition, and many genetic, neurobiological and environmental factors play a role. This study demonstrates one step toward evaluating neurobiological mechanisms that may contribute to reading difficulties.

Reviewer #2 (Public review):

Summary:

This study utilized two complimentary techniques (EEG and 7T MRI/MRS) to directly test a theory of dyslexia: the neural noise hypothesis. The authors report finding no evidence to support an excitatory/inhibitory balance, as quantified by beta in EEG and Glutamate/GABA ratio in MRS. This is important work and speaks to one potential mechanism by which increased neural noise may occur in dyslexia.

Strengths:

This is a well conceived study with in depth analyses and publicly available data for independent review. The authors provide transparency with their statistics and display the raw data points along with the averages in figures for review and interpretation. The data suggest that an E/I balance issue may not underlie deficits in dyslexia and is a meaningful and needed test of a possible mechanism for increased neural noise.

Weaknesses:

The researchers did not include a visual print task in the EEG task, which limits analysis of reading specific regions such as the visual word form area, which is a commonly hypoactivated region in dyslexia. This region is a common one of interest in dyslexia, yet the researchers measured the I/E balance in only one region of interest, specific to the language network.

Reviewer #3 (Public review):

Summary:

This study by Glica and colleagues utilized EEG (i.e., Beta power, Gamma power, and aperiodic activity) and 7T MRS (i.e., MRS IE ratio, IE balance) to reevaluating the neural noise hypothesis in Dyslexia. Supported by Bayesian statistics, their results show convincing evidence of no differences in EI balance between groups, challenging the neural noise hypothesis.

Strengths:

Combining EEG and 7T MRS, this study utilized both the indirect (i.e., Beta power, Gamma power, and aperiodic activity) and direct (i.e., MRS IE ratio, IE balance) measures to reevaluating the neural noise hypothesis in Dyslexia.

Author response:

Public Reviews:

Reviewer #1 (Public review):

When you search for something, you need to maintain some representation (a "template") of that target in your mind/brain. Otherwise, how would you know what you were looking for? If your phone is in a shocking pink case, you can guide your attention to pink things based on a target template that includes the attribute 'pink'. That guidance should get you to the phone pretty effectively if it is in view. Most real-world searches are more complicated. If you are looking for the toaster, you will make use of your knowledge of where toasters can be. Thus, if you are asked to find a toaster, you might first activate a template of a kitchen or a kitchen counter. You might worry about pulling up the toaster template only after you are reasonably sure you have restricted your attention to a sensible part of the scene.

Zhou and Geng are looking for evidence of this early stage of guidance by information about the surrounding scene in a search task. They train Os to associate four faces with four places. Then, with Os in the scanner, they show one face - the target for a subsequent search. After an 8 sec delay, they show a search display where the face is placed on the associated scene 75% of the time. Thus, attending to the associated scene is a good idea. The questions of interest are "When can the experimenters decode which face Os saw from fMRI recording?" "When can the experimenters decode the associated scene?" and "Where in the brain can the experimenters see evidence of this decoding? The answer is that the face but not the scene can be read out during the face's initial presentation. The key finding is that the scene can be read out (imperfectly but above chance) during the subsequent delay when Os are looking at just a fixation point. Apparently, seeing the face conjures up the scene in the mind's eye.

This is a solid and believable result. The only issue, for me, is whether it is telling us anything specifically about search. Suppose you trained Os on the face-scene pairing but never did anything connected to the search. If you presented the face, would you not see evidence of recall of the associated scene? Maybe you would see the activation of the scene in different areas and you could identify some areas as search specific. I don't think anything like that was discussed here.

You might also expect this result to be asymmetric. The idea is that the big scene gives the search information about the little face. The face should activate the larger useful scene more than the scene should activate the more incidental face, if the task was reversed. That might be true if the finding is related to a search where the scene context is presumed to be the useful attention guiding stimulus. You might not expect an asymmetry if Os were just learning an association.

It is clear in this study that the face and the scene have been associated and that this can be seen in the fMRI data. It is also clear that a valid scene background speeds the behavioral response in the search task. The linkage between these two results is not entirely clear but perhaps future research will shed more light.

It is also possible that I missed the clear evidence of the search-specific nature of the activation by the scene during the delay period. If so, I apologize and suggest that the point be underlined for readers like me.

We will respond to this question by acknowledging that the reviewer is right in that the delay period activation of the scene is not necessarily search-specific. We will then discuss how this possibility affects the interpretation of our results and what kind of studies would need to be conducted in order to fully establish a causal link between delay period activity and visual search performance. We will also discuss the literature on cued attention and situate our work within the context of these other studies that have used similar task paradigms to infer attentional processes. Finally, we will discuss the interpretation of delay period activity in PPA and IFJ.

Reviewer #2 (Public review):

Summary:

This work is one of the best instances of a well-controlled experiment and theoretically impactful findings within the literature on templates guiding attentional selection. I am a fan of the work that comes out of this lab and this particular manuscript is an excellent example as to why that is the case. Here, the authors use fMRI (employing MVPA) to test whether during the preparatory search period, a search template is invoked within the corresponding sensory regions, in the absence of physical stimulation. By associating faces with scenes, a strong association was created between two types of stimuli that recruit very specific neural processing regions - FFA for faces and PPA for scenes. The critical results showed that scene information that was associated with a particular cue could be decoded from PPA during the delay period. This result strongly supports the invoking of a very specific attentional template.

Strengths:

There is so much to be impressed with in this report. The writing of the manuscript is incredibly clear. The experimental design is clever and innovative. The analysis is sophisticated and also innovative. The results are solid and convincing.

Weaknesses:

I only have a few weaknesses to point out.

This point is not so much of a weakness, but a further test of the hypothesis put forward by the authors. The delay period was long - 8 seconds. It would be interesting to split the delay period into the first 4seconds and the last 4seconds and run the same decoding analyses. The hypothesis here is that semantic associations take time to evolve, and it would be great to show that decoding gets stronger in the second delay period as opposed to the period right after the cue. I don't think this is necessary for publication, but I think it would be a stronger test of the template hypothesis.

We will conduct the suggested analysis. Depending on the outcome, we will include it in supplemental materials or the main text.

Type in the abstract "curing" vs "during."

We will fix this.

It is hard to know what to do with significant results in ROIs that are not motivated by specific hypotheses. However, for Figure 3, what are the explanations for ROIs that show significant differences above and beyond the direct hypotheses set out by the authors?

We will address how each of the ROIs wdas selected based on the use of a priori networks as masks with ROIs as sub-parcels. We will explain why specific ROIs were associated with the strongest hypotheses but how the entire networks are relevant and related to existing literatures on attentional control and working memory. This content will be included in the introduction and discussion sections.

Reviewer #3 (Public review):

The manuscript contains a carefully designed fMRI study, using MVPA pattern analysis to investigate which high-level associate cortices contain target-related information to guide visual search. A special focus is hereby on so-called 'target-associated' information, that has previously been shown to help in guiding attention during visual search. For this purpose the author trained their participants and made them learn specific target-associations, in order to then test which brain regions may contain neural representations of those learnt associations. They found that at least some of the associations tested were encoded in prefrontal cortex during the cue and delay period.

The manuscript is very carefully prepared. As far as I can see, the statistical analyses are all sound and the results integrate well with previous findings.

I have no strong objections against the presented results and their interpretation.

Thank you.

Author response:

eLife Assessment

This study addresses a novel and interesting question about how the rise of the Qinghai-Tibet Plateau influenced patterns of bird migration, employing a multi-faceted approach that combines species distribution data with environmental modeling. The findings are valuable for understanding avian migration within a subfield, but the strength of evidence is incomplete due to critical methodological assumptions about historical species-environment correlations, limited tracking data, and insufficient clarity in species selection criteria. Addressing these weaknesses would significantly enhance the reliability and interpretability of the results.

We would like to thank you and two anonymous reviewers for your careful, thoughtful, and constructive feedback on our manuscript. These reviews made us revisit a lot of our assumptions and we believe the paper will be much improved as a result. In addition to minor points, we will make three main changes to our manuscript in response to the reviews. First, we will address the concerns on the assumptions of historical species-environment correlations from perspectives of both theoretical and empirical evidence. Second, we will discuss the benefits and limitations of using tracking data in our study and demonstrate how the findings of our study are consolidated with results of previous studies. Third, we will clarify our criteria for selecting species in terms of both eBird and tracking data.

Below, we respond to each comment in turn. Once again, we thank you all for your feedback.

Reviewer #1 (Public review):

Strengths:

This is an interesting topic and a novel theme. The visualisations and presentation are to a very high standard. The Introduction is very well-written and introduces the main concepts well, with a clear logical structure and good use of the literature. The methods are detailed and well described and written in such a fashion that they are transparent and repeatable.

We appreciate the reviewer’s careful reading of our manuscript, encouraging comments and constructive suggestions.

Weaknesses:

I only have one major issue, which is possibly a product of the structure requirements of the paper/journal. This relates to the Results and Discussion, line 91 onwards. I understand the structure of the paper necessitates delving immediately into the results, but it is quite hard to follow due to a lack of background information. In comparison to the Methods, which are incredibly detailed, the Results in the main section reads as quite superficial. They provide broad overviews of broad findings but I found it very hard to actually get a picture of the main results in its current form. For example, how the different species factor in, etc.

Yes, it is the journal request to format in this way (Methods follows the Results and Discussion) for the article type of short reports. As suggested, in the revision we will elaborate on details of our findings, especially the species-specific responses, in terms of (i) shifts of distribution of avian breeding and wintering areas under the influence of the uplift of the Qinghai-Tibetan Plateau, and (ii) major factors that shape current migration patterns of birds in the Plateau. We will also better reference the approaches we used in the study.

Reviewer #2 (Public review):

Summary:

The study tries to assess how the rise of the Qinghai-Tibet Plateau affected patterns of bird migration between their breeding and wintering sites. They do so by correlating the present distribution of the species with a set of environmental variables. The data on species distributions come from eBird. The main issue lies in the problematic assumption that species correlations between their current distribution and environment were about the same before the rise of the Plateau. There is no ground truthing and the study relies on Movebank data of only 7 species which are not even listed in the study. Similarly, the study does not outline the boundaries of breeding sites NE of the Plateau. Thus it is absolutely unclear potentially which breeding populations it covers.

We are very grateful for the careful review and helpful suggestions. We will revise the manuscript carefully in response to the reviewer’s comments and believe that it will be much improved as a result. Below are our point-by-point replies to the comments.

Strengths:

I like the approach for how you combined various environmental datasets for the modelling part.

We appreciate the reviewer’s encouragement.

Weaknesses:

The major weakness of the study lies in the assumption that species correlations between their current distribution and environments found today are back-projected to the far past before the rise of the Q-T Plateau. This would mean that species responses to the environmental cues do not evolve which is clearly not true. Thus, your study is a very nice intellectual exercise of too many ifs.

This is a valid concern. We will address this from both the perspectives of the theoretical design of our study and empirical evidence.

First, we agree with the reviewer that species responses to environmental cues might vary over time. Nonetheless, the simulated environments before the uplift of the plateau serve as a counterfactual state in our study. Counterfactual is an important concept to support causation claims by comparing what happened to what would have happened in a hypothetical situation: “If event X had not occurred, event Y would not have occurred” (Lewis 1973). Recent years have seen an increasing application of the counterfactual approach to detect biodiversity change, i.e., comparing diversity between the counterfactual state and real estimates to attribute the factors causing such changes (e.g., Gonzalez et al. 2023). Whilst we do not aim to provide causal inferences for avian distributional change, using the counterfactual approach, we are able to estimate the influence of the plateau uplift by detecting the changes of avian distributions, i.e., by comparing where the birds would have distributed without the plateau to where they currently distributed. We regard the counterfactual environments as a powerful tool for eliminating, to the extent possible, vagueness, as opposed to simply description of current distributions of birds. Therefore, we assume species’ responses to environments are conservative and their evolution should not discount our findings. We will clarify this in both the Introduction and Methods.

Second, we used species distribution modelling to contrast the distributions of birds before and after the uplift of the plateau under the assumption that species tend to keep their ancestral ecological traits over time (i.e., niche conservatism). This indicates a high probability for species to distribute in similar environments wherever suitable. Particularly, considering birds are more likely to be influenced by food resources (Martins et al. 2024), and the distribution of available food before the uplift (Jia et al. 2020), we believe the findings can provide valuable insights into the influence of the plateau on avian migratory patterns. Having said that, we acknowledge other factors, e.g., carbon dioxide concentrations (Zhang et al. 2022), can influence the simulations of environments and our prediction of avian distribution. We will clarify the assumptions and evidence we have for the modelling in Methods. We will further point out the direction for future studies in the Discussion.

The second major drawback lies in the way you estimate the migratory routes of particular birds. No matter how good the data eBird provides is, you do not know population-specific connections between wintering and breeding sites. Some might overwinter in India, some populations in Africa and you will never know the teleconnections between breeding and wintering sites of particular species. The few available tracking studies (seven!) are too coarse and with limited aspects of migratory connectivity to give answer on the target questions of your study.

We agree with the reviewer that establishing interconnections for birds is important for estimating the migration patterns of birds. We employed a dynamic model to assess their weekly distributions. Thus, we can track the movement of species every week, and capture the breeding and wintering areas for specific populations. That being said, we acknowledge that our approach can be subjected to the patchy sampling of eBird data. We will better demonstrate this in the main text.  

Tracking data can provide valuable insights into the movement patterns of species but are limited to small numbers of species due to the considerable costs and time needed. We aimed to adopt the tracking data to examine the influence of focal factors on avian migration patterns, but only seven species, to the best of our ability, were acquired. Moreover, similar results were found in studies that used tracking data to estimate the distribution of breeding and wintering areas of birds in the plateau (e.g., Prosser et al. 2011, Zhang et al. 2011, Zhang et al. 2014, Liu et al. 2018, Kumar et al. 2020, Wang et al. 2020, Pu and Guo 2023, Yu et al. 2024, Zhao et al. 2024). We believe the conclusions based on seven species are rigour, but their implications could be restricted by the number of tracking species we obtained. We will demonstrate how our findings on breeding and wintering areas of birds are reinforced by other studies reporting the locations of those areas. We will also add a separate caveat section to discuss the limitations stated above.

Your set of species is unclear, selection criteria for the 50 species are unknown and variability in their migratory strategies is likely to affect the direction of the effects.

We will clarify the selection criteria for the 50 species). We first obtained a full list of birds in the plateau from Prins and Namgail (2017). We then extracted species identified as full migrants in Birdlife International (https://datazone.birdlife.org/species/spcdistPOS) from the full list.

In addition, the position of the breeding sites relative to the Q-T plate will affect the azimuths and resulting migratory flyways. So in fact, we have no idea what your estimates mean in Figure 2.

We calculated the azimuths not only by the angles between breeding sites and wintering sites but also based on the angles between the stopovers of birds. Therefore, the azimuths are influenced by the relative positions of breeding, wintering and stopover sites. We will better explain this both in the Methods and legend of Figure 2.

There is no way one can assess the performance of your statistical exercises, e.g. performances of the models.

As suggested, we will add the AUC values to assess the performances of the models.

References

Gonzalez, A., J. M. Chase, and M. I. O'Connor. 2023. A framework for the detection and attribution of biodiversity change. Philosophical Transactions of the Royal Society B: Biological Sciences 378: 20220182.

Jia, Y., H. Wu, S. Zhu, Q. Li, C. Zhang, Y. Yu, and A. Sun. 2020. Cenozoic aridification in Northwest China evidenced by paleovegetation evolution. Palaeogeography, Palaeoclimatology, Palaeoecology 557:109907.

Kumar, N., U. Gupta, Y. V. Jhala, Q. Qureshi, A. G. Gosler, and F. Sergio. 2020. GPS-telemetry unveils the regular high-elevation crossing of the Himalayas by a migratory raptor: implications for definition of a “Central Asian Flyway”. Scientific Reports 10:15988.

Lewis, D. 1973. Counterfactuals. Oxford: Blackwell.

Liu, D., G. Zhang, H. Jiang, and J. Lu. 2018. Detours in long-distance migration across the Qinghai-Tibetan Plateau: individual consistency and habitat associations. PeerJ 6:e4304.

Martins, L. P., D. B. Stouffer, P. G. Blendinger, K. Böhning-Gaese, J. M. Costa, D. M. Dehling, C. I. Donatti, C. Emer, M. Galetti, R. Heleno, Í. Menezes, J. C. Morante-Filho, M. C. Muñoz, E. L. Neuschulz, M. A. Pizo, M. Quitián, R. A. Ruggera, F. Saavedra, V. Santillán, M. Schleuning, L. P. da Silva, F. Ribeiro da Silva, J. A. Tobias, A. Traveset, M. G. R. Vollstädt, and J. M. Tylianakis. 2024. Birds optimize fruit size consumed near their geographic range limits. Science 385:331-336.

Prins, H. H. T., and T. Namgail. 2017. Bird migration across the Himalayas : wetland functioning amidst mountains and glaciers. Cambridge University Press, Cambridge.

Prosser, D. J., P. Cui, J. Y. Takekawa, M. Tang, Y. Hou, B. M. Collins, B. Yan, N. J. Hill, T. Li, Y. Li, F. Lei, S. Guo, Z. Xing, Y. He, Y. Zhou, D. C. Douglas, W. M. Perry, and S. H. Newman. 2011. Wild bird migration across the Qinghai-Tibetan Plateau: a transmission route for highly pathogenic H5N1. PloS One 6:e17622.

Pu, Z., and Y. Guo. 2023. Autumn migration of black-necked crane (Grus nigricollis) on the Qinghai-Tibetan and Yunnan-Guizhou plateaus. Ecology and Evolution 13:e10492.

Wang, Y., C. Mi, and Y. Guo. 2020. Satellite tracking reveals a new migration route of black-necked cranes (Grus nigricollis) in Qinghai-Tibet Plateau. PeerJ 8:e9715.

Yu, X., G. Song, H. Wang, Q. Wei, C. Jia, and F. Lei. 2024. Migratory flyways and connectivity of brown headed gulls (Chroicocephalus brunnicephalus) revealed by GPS tracking. Global Ecology and Conservation 56:e03340.

Zhang, G.G., D.P. Liu, Y.Q. Hou, H.X. Jiang, M. Dai, F.W. Qian, J. Lu, T. Ma, L.X. Chen, and Z. Xing. 2014. Migration routes and stopover sites of Pallas’s gulls Larus ichthyaetus breeding at Qinghai Lake, China, determined by satellite tracking. Forktail 30:104-108.

Zhang, G.G., D.P. Liu, Y.Q. Hou, H.X. Jiang, M. Dai, F.W. Qian, J. Lu, Z. Xing, and F.S. Li. 2011. Migration routes and stop-over sites determined with satellite tracking of bar-headed geese (Anser indicus) breeding at Qinghai Lake, China. Waterbirds 34:112-116, 115.

Zhang, R., D. Jiang, C. Zhang, and Z. Zhang. 2022. Distinct effects of Tibetan Plateau growth and global cooling on the eastern and central Asian climates during the Cenozoic. Global and Planetary Change 218:103969.

Zhao, T., W. Heim, R. Nussbaumer, M. van Toor, G. Zhang, A. Andersson, J. Bäckman, Z. Liu, G. Song, M. Hellström, J. Roved, Y. Liu, S. Bensch, B. Wertheim, F. Lei, and B. Helm. 2024. Seasonal migration patterns of Siberian Rubythroat (Calliope calliope) facing the Qinghai–Tibet Plateau. Movement Ecology 12:54.

Author response:

Reviewer #2 (Public review):

(1) Given their results the authors conclude that upregulation of Frizzled on the plasma membrane is not sufficient to explain the stabilization of beta-catenin seen in the ZNRF3/RNF43 mutant cells. This interpretation is sound, and they suggest in the discussion that ZNRF3/RNF43-mediated ubiquitination could serve as a sorting signal to sort endocytosed FZD to lysosomes for degradation and that absence or inhibition of this process would promote FZD recycling. This should be relatively easy to test using surface biotinylation experiments and would considerably strengthen the manuscript.

Thank you for your valuable suggestions and comments. We will perform cell surface biotinylation experiments.

(2) The authors show that the FZD5 CRD domain is required for endocytosis since a mutant FZD5 protein in which the CRD is removed does not undergo endocytosis. This is perhaps not surprising since this is the site of Wnt binding, but the authors show that a chimeric FZD5CRD-FZD4 receptor can confer Wnt-dependent endocytosis to an otherwise endocytosis incompetent FZD4 protein. Since the linker region between the CRD and the first TM differs between FZD5 and FZD4 it would be interesting to understand whether the CRD specifically or the overall arrangement (such as the spacing) is the most important determinant.

Our results in Fig. 1F-G clearly show that the CRD of FZD5 specifically is both necessary and sufficient for Wnt3a/5a-induced FZD5 endocytosis, as replacing the CRD alone in FZD5 with the CRD from either FZD4 or FZD7 completely abolished Wnt-induced endocytosis, whereas replacing the CRD alone in FZD4 or FZD7 with the FZD5 CRD alone could confer Wnt-induced endocytosis.

(3) I find it surprising that only FZD5 and FZD8 appear to undergo endocytosis or be stabilized at the cell surface upon ZNRF3/RNF43 knockout. Is this consistent with previous literature? Is that a cell-specific feature? These findings should be tested in a different cell line, with possibly different relative levels of ZNRF3 and RNF43 expression.

Thank you for your comments and suggestions. Our finding that ZNRF3/RNF43 specifically regulates FZD5/8 degradation is consistent with recent published studies in which FZD5 is required for the survival of RNF43-mutant PDAC or colorectal cancer cells (Nature Medicine, 2017, PMID: 27869803) and FZD5 is required for the maintenance of intestinal stem cells (Developmental Cell, 2024, PMID: 39579768 and 39579769), and in both cases, FZDs other than FZD5/8 are also expressed but not sufficient to compensate for the function of FZD5. The mechanism by which Wnt3a/5a specifically induces FZD5/8 endocytosis and degradation is currently unknown and needs to be explored in the future. We speculate that Wnt binding to FZD5/8 may recruit another protein on the cell surface to specifically facilitate FZD5/8 endocytosis. On the other hand, we cannot exclude the possibility that Wnts other than Wnt3a/5a may induce the endocytosis and degradation of FZDs other than FZD5/8 since there are 19 Wnts and 10 FZDs in humans. We will perform flow cytometry experiments using FZD5/8-specific antibodies to examine whether Wnt3a/5a induces FZD5/8 endocytosis in more cell lines.

(4) If FZD7 is not a substrate of ZNRF3/RNF43 and therefore is not ubiquitinated and degraded, how do the authors reconcile that its overexpression does not lead to elevated cytosolic beta-catenin levels in Figure 5B?

We are currently not sure of the mechanism underlying this result. Considering that most FZDs are expressed in 293A cells, we do not know how much of the mature form of overexpressed FZD7 was presented to the plasma membrane.

(5) For Figure 5B, it would be interesting if the authors could evaluate whether overexpression of FZD5 in the ZNRF3/RNF43 double knockout lines would synergize and lead to further increase in cytosolic beta-catenin levels. As control if the substrate selectivity is clear FZD7 overexpression in that line should not do anything.

We will perform these experiments as you suggested.

(6) In Figure 6G, the authors need to show cytosolic levels of beta-catenin in the absence of Wnt in all cases.

We did not add Wnt CM in this experiment. RSPO1 activity, which relies on endogenous Wnt, has been well documented in previous studies.

(7) Since the authors show that DVL is not involved in the Wnt and ZRNF3-dependent endocytosis they should repeat the proximity biotinylation experiment in figure 7 in the DVL triple KO cells. This is an important experiment since previous studies showed that DVL was required for the ZRNF3/RNF43-mediated ubiqtuonation of FZD.

Thank you for your valuable suggestions. We will perform the proximity biotinylation experiment in DVL TKO cells.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary

This manuscript aimed to study the role of Rudhira (also known as Breast Carcinoma Amplified Sequence 3), an endothelium-restricted microtubules-associated protein, in regulating of TGFβ signaling. The authors demonstrate that Rudhira is a critical signaling modulator for TGFβ signaling by releasing Smad2/3 from cytoskeletal microtubules and how Rudhira is a Smad2/3 target gene. Taken together, the authors provide a model of how Rudhira contributes to TGFβ signaling activity to stabilize the microtubules, which is essential for vascular development.

Strengths

The study used different methods and techniques to achieve aims and support conclusions, such as Gene Ontology analysis, functional analysis in culture, immunostaining analysis, and proximity ligation assay. This study provides an unappreciated additional layer of TGFβ signaling activity regulation after ligand receptor interaction.

We thank the reviewer for acknowledging the importance of our study and providing a clear summary of our findings.

Weaknesses

(1) It is unclear how current findings provide a beVer understanding of Rudhira KO mice, which the authors published some years ago.

Our previous study demonstrated that Rudhira KO mice have a predominantly developmental cardiovascular phenotype that phenocopies TGFβ loss of function (Shetty, Joshi et al., 2018). Additionally, we found that at the molecular level, Rudhira regulates cytoskeletal organization (Jain et al., 2012; Joshi and Inamdar, 2019). Our current study builds upon these previous findings, showing an essential role of Rudhira in maintaining TGFβ signaling and controlling the microtubule cytoskeleton during vascular development. On one hand Rudhira regulates TGFβ signaling by promoting the release of Smads from microtubules, while on the other, Rudhira is a TGFβ target essential for stabilizing microtubules. Thus, our current study provides a molecular basis for Rudhira function in cardiovascular development.

(2) Why do they use HEK cells instead of SVEC cells in Figure 2 and 4 experiments?

Our earlier studies have characterized the role of Rudhira in detail using both loss and gain of function methods in multiple cell types (Jain et al., 2012; SheVy, Joshi et al., 2018; Joshi and Inamdar, 2019). As endothelial cells are particularly difficult to transfect, and because the function of Rudhira in promoting cell migration is conserved in HEK cells, it was practical and relevant to perform these experiments in HEK cells (Figures 2 and 4E).

(3) A model shown in Figure 5E needs improvement to grasp their findings easily.

We have modified Figure 5E for clarity.

Reviewer #2 (Public Review):

Summary

It was first reported in 2000 that Smad2/3/4 are sequestered to microtubules in resting cells and TGF-β stimulation releases Smad2/3/4 from microtubules, allowing activation of the Smad signaling pathway. Although the finding was subsequently confirmed in a few papers, the underlying mechanism has not been explored. In the present study, the authors found that Rudhira/breast carcinoma amplified sequence 3 is involved in the release of Smad2/3 from microtubules in response to TGF-β stimulation. Rudhira is also induced by TGF-β and is probably involved in the stabilization of microtubules in the delayed phase after TGF-β stimulation. Therefore, Rudhira has two important functions downstream of TGF-β in the early as well as delayed phase.

Strengths:

This work aimed to address an unsolved question on one of the earliest events after TGF-β stimulation. Based on loss-of-function experiments, the authors identified a novel and potentially important player, Rudhira, in the signal transmission of TGF-β.

We thank the reviewer for the critical evaluation and appreciation of our findings.

Weaknesses:

The authors have identified a key player that triggers Smad2/3 released from microtubules after TGF-β stimulation probably via its association with microtubules. This is an important first step for understanding the regulation of Smad signaling, but underlying mechanisms as well as upstream and downstream events largely remain to be elucidated.

We acknowledge that the mechanisms regulating cytoskeletal control of Smad signaling are far from clear, but these are out of scope of this manuscript. This manuscript rather focuses on Rudhira/Bcas3 as a pivot to understand vascular TGFβ signaling and microtubule connections.

(1) The process of how Rudhira causes the release of Smad proteins from microtubules remains unclear. The statement that "Rudhira-MT association is essential for the activation and release of Smad2/3 from MTs" (lines 33-34) is not directly supported by experimental data.

We agree with the reviewer’s comment. Although we provide evidence that the loss of Rudhira (and thereby deduced loss of Rudhira-MT association) prevents release of Smad2/3 from MTs (Fig 3C), it does not confirm the requirement of Rudhira-MT association for this. In light of this, we have modified the statement to ‘Rudhira associates with MTs and is essential for the activation and release of Smad2/3 from MTs”.

(2) The process of how Rudhira is mobilized to microtubules in response to TGF-β remains unclear.

Our previous study showed that Rudhira associates with microtubules, and preferentially binds to stable microtubules (Jain et al., 2012; Joshi and Inamdar, 2019). Since TGFβ stimulation is known to stabilize microtubules, we hypothesize that TGFβ stimulation increases Rudhira binding to stable microtubules. We have mentioned this in our revised manuscript.

(3) After Rudhira releases Smad proteins from microtubules, Rudhira stabilizes microtubules. The process of how cells return to a resting state and recover their responsiveness to TGF-β remains unclear.

We show that dissociation of Smads from microtubules is an early response and stabilization of microtubules is a late TGFβ response. However, we agree that the sequence of these molecular events has not been characterized in-depth in this or any other study, making it difficult to assign causal roles (eg. whether release of Smads from MTs is a pre-requisite for MT stabilization by Rudhira) or reversibility. However, the TGFβ pathway is auto regulatory, leading to increased turnover of receptors and Smads and increased expression of inhibitory Smads, which may recover responsiveness to TGFβ. Additionally, the still short turnover time of stable microtubules (several minutes to hours) may also promote quick return to resting state. We have discussed this in our revised manuscript.

Recommendations for the authors:

Reviewer #2 (Recommendations for The Authors):

(1) Overall: Duration of TGF-β stimulation in cell-based assays should be described in the legends for readers' convenience. Avoid simple bar graphs because sample numbers are only 3. A scaVer plot should be super-imposed.

Details added, as suggested. Duration of treatment is mentioned in Materials and methods section for figures 1C-D; 2A-B; 3; 4A-C; 5A-C; S2D; S3A-C; S4C, D. Bar graphs have been replaced with a bar + scatter plot. Note that, as the Excel file for data related to fig 4A was corrupted, we repeated the experiments to generate fresh data. Hence the graph had to be replaced. However, the result holds true as before.

(2) Figure 1A: This panel is too small. Gene names are almost invisible.

Modified for clarity.

(3) Figure 1B: Show TGFβRI expression by immunoblomng (re-probing) to verify that it is expressed in the rightmost lane.

TGFβRI overexpression was confirmed by qPCR in a replicate in the same experiment (Fig S2C).

(4) Figure 1C: Show expression of Rudhira. In addition, confirm the positions of molecular weight markers. Smad2 migrated slower than pSmad2.

Rudhira expression is shown in Fig S1B. Molecular weight markers have been corrected.

(5) Figure 3A: This panel shows a negative result that Smad2/3 fails to interact with Rudhira. A positive control, for example, Smad4, would make the data convincing.

Although it would be nice to have a positive control for interaction, we do not agree that a positive control of Smad4 is essential for our conclusion from this experiment, which is that ‘we were unable to detect an interaction between Rudhira and Smad2/3’.

(6) Fig. 3B: Show Rudhira blot. If possible, show that the Rudhira-MT association precedes Smad phosphorylation by a time course experiment. This is an important point but not experimentally demonstrated.

The interaction between Rudhira and microtubules with or without TGFβ is demonstrated by PLA (Fig 3E). Although important, the suggested time course experiments to assess the sequence of events are beyond the scope of this manuscript. 

(7) Figure 3E: Does the process require the type I receptor kinase activity or non-Smad signaling pathways?

Since TGFβ pathway is complex and is regulated at multiple steps, this possibility has not been tested and is beyond the scope of current study.

(8) Figure 4A: The authors did not examine if these elements are functional. Therefore, this panel can be presented as a supplementary figure.

As suggested, the panel has been moved to supplementary information.

(9) Figure 4E: The figure legend does not say that cells were TGF-β-stimulated. It remains unclear if Smad2 and Smad3 are involved in Rudhira expression as phosphorylated or non-phosphorylated forms. Therefore, the authors should show a pSmad2 blot. In the absence of TGF-β stimulation, Smad2 and Smad3 are expected to be sequestrated to microtubules and therefore not phosphorylated. In the case that cells were stimulated with TGF-β, show if Rudhira is induced by TGF-β in HEK293T cells. This is not shown in this manuscript.

This experiment was not performed under regulated conditions with or without TGFβ, hence the sensitivity to TGFβ could not be assessed. Cells were not stimulated with exogenous TGFβ, but cultured in regular medium with serum, which can have up to ~40 ng/ml of TGFβ (latent and active). Additionally, owing to severe depletion of Smad2 or Smad3 by shRNAs we expect sufficient loss of phospho-Smads2/3. 

(10) Figure S1A: Rudhira migrated at the position corresponding to 91 kD only in this panel.

Corrected the position of molecular weight marker.

(11) Line 205-206, "Since in vivo studies indicate that rudhira depletion severely affects the TGFβ pathway [11]": Refer to Reference 11. The paper does not say anything about TGFβ.

Reference corrected to Ref #14.

(12) Smad4 was previously reported to be sequestered to microtubules [Ref. 7]. Does Rudhira release Smad4 also?

This is an interesting point which could be followed up on our future studies.

(13) It would be nice if the authors examined how Rudhira causes the release of Smad2/3 from microtubules. Currently, it remains unclear whether the association of Rudhira to microtubules is required for the release of Smad2/3. Does a Rudhira mutant lacking microtubule binding fail to induce the release of Smad2/3 after TGF-β stimulation? If so, do Rudhira and Smad2/3 share the same binding site on microtubules? In that case, the mechanism can be regarded as "competitive".

This is a thoughtful experiment much beyond the scope of current manuscript. In our previous study we were able to localize the Tubulin binding sites of Rudhira primarily to its Bcas3 domain (Joshi and Inamdar, 2019), however the equivalent sites in Tubulin were not assessed. While MH2 domains of Smad2/3 bind β-tubulin, amino acids 114-243 of β-tubulin bind to Smad2/3 (Dai et al., 2007). A systematic study of these tripartite interactions including Rudhira would be an interesting follow up for our future study.

Author response:

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

Reviewer #1 (Public review):  

Summary:  

The authors show that SVZ derived astrocytes respond to a middle carotid artery occlusion

(MCAO) hypoxia lesion by secreting and modulating hyaluronan at the edge of the lesion (penumbra) and that hyaluronan is a chemoattractant to SVZ astrocytes. They use lineage tracing of SVZ cells to determine their origin. They also find that SVZ derived astrocytes express Thbs-4 but astrocytes at the MCAO-induced scar do not. Also, they demonstrate that decreased HA in the SVZ is correlated with gliogenesis. While much of the paper is descriptive/correlative they do overexpress Hyaluronan synthase 2 via viral vectors and show this is sufficient to recruit astrocytes to the injury. Interestingly, astrocytes preferred to migrate to the MCAO than to the region of overexpressed HAS2.  

Strengths:  

The field has largely ignored the gliogenic response of the SVZ, especially with regards to astrocytic function. These cells and especially newborn cells may provide support for regeneration. Emigrated cells from the SVZ have been shown to be neuroprotective via creating pro-survival environments, but their expression and deposition of beneficial extracellular matrix molecules is poorly understood. Therefore, this study is timely and important. The paper is very well written and the flow of results logical.  

Comments on revised version:  

The authors have addressed my points and the paper is much improved. Here are the salient remaining issues that I suggest be addressed.  

We appreciate the feedback by the reviewer, and we are glad that the paper is considered to be much improved. We have done our best to address the remaining issues in this 2nd revision.

The authors have still not shown, using loss of function studies, that Hyaluronan is necessary for SVZ astrogenesis and or migration to MCAO lesions.

This is true. Unfortunately, complete removal of hyaluronan (via Hyase) triggers epilepsy, already described in 1963 by James Young (Exp Neurol paper). Degradation by Hyase also provokes neuroinflammation (Soria et al., 2020 Nat Commun). Two alternatives could be 1) partial depletion with Has inhibitor 4MU (but it is also associated with increased inflammation) or 2) a Has-KO mouse, such as Has3-/- (Arranz et al., 2014), although, to our knowledge, this mouse line is not openly available. We have added a sentence in line 332 addressing this shortcoming: “Loss-of-function studies, using HA-depletion models or HA synthase (Has)deficient mice are still needed to corroborate this finding, although the inflammation associated with HA deficiency might confound interpretation.”

(1) The co-expression of EGFr with Thbs4 and the literature examination is useful.  

We thank the reviewer for the kind comment.

(2) Too bad they cannot explain the lack of effect of the MCAO on type C cells. The comparison with kainate-induced epilepsy in the hippocampus may or may not be relevant.

As stated in the previous response, we also found this interesting, and it does warrant further exploration by looking into possible direct NSC-astrocyte differentiation. But we believe that both this possible direct differentiation and the reactive status for these astrocytes are out of the scope of the study. We will not speculate about this in the discussion, either.

(3) Thanks for including the orthogonal confocal views in Fig S6D.  

(4) The statement that "BrdU+/Thbs4+ cells mostly in the dorsal area" and therefore they mostly focused on that region is strange. Figure 8 clearly shows Thbs4 staining all along the striatal SVZ. Do they mean the dorsal segment of the striatal SVZ or the subcallosal SVZ? Fig. 4b and Fig 4f clearly show the "subcallosal" area as the one analysed but other figures show the dorsal striatal region (Fig. 2a). This is important because of the well-known embryological and neurogenic differences between the regions.  

While it is true that Thbs4 is also expressed in the other subregions of the SVZ (lateral, ventral and medial), as observed in Fig 8. we chose the dorsal area because it is the subregion where we observed the larger increase in slow proliferative NSCs (Thbs4/GFAP/BrdU-positive cells) after MCAO (Fig S3). As observed in the quantifications in Fig S3, we found Thbs4/GFAP/BrdUpositive cells increase in lateral, medial and ventral SVZ, but it is not significant. Therefore, from Fig 4 onwards, we focused on the dorsal SVZ, which the reviewer mentions as “subcallosal” area. We chose the term “dorsal” as stated in single-cell studies (Cebrian-Silla et al, 2021, eLife; Marcy et al., 2023, Sci Adv) and reviews (Sequerra 2014 Front Cell Neurosci) that investigate or mention this subregion, respectively. In the abstract, we are perfectly clear stating that newborn astrocytes migrate frm both dorsal and medial areas.  

In Fig 2a, the immunofluorescence image shows medial and lateral SVZ, but at this point in the paper, we have not yet made specific subregional quantifications, and the Nestin, DCX and Thbs4 quantifications refer to the SVZ as a whole, both in the IF and in the WB (Fig 2e-g). We apologize for the confusion. We have clarified this in the text (line 119).  

(5) It is good to know that the harsh MCAO's had already been excluded.  

(6) Sorry for the lack of clarity - in addition to Thbs4, I was referring to mouse versus rat Hyaluronan degradation genes (Hyal1, Hyal2 and Hyal3) and hyaluronan synthase genes (HAS1 and HAS2) in order to address the overall species differences in hyaluronan biology thus justifying the "shift" from mouse to rat. You examine these in the (weirdly positioned) Fig. 8h,i. Please add a few sentences on mouse vs rat Thbs4 and Hyaluronan relevant genes.  

We thank the reviewer for these remarks. We have now added a sentence pointing to the similar internalization and degradation in rat and mouse (reviewed by Sherman et al., 2015). This correction is in line 233. Hyaluronan is, in evolutionary terms, a very “old” molecule, part of the “ancient” glycan-based matrix, before the evolution of proteoglycans and fibrous proteins such as collagen, laminin etc. Hence, its machinery is highly conserved across species.

We have also reorganized the panels in Fig 8, where 8h and 8i were indeed weirdly positioned. We hope that the new version of this figure is more easily readable.

(7) Thank you for the better justification of using the naked mole rat HA synthase.  

Reviewer #3 (Public review):  

Summary:  

The authors aimed to study the activation of gliogenesis and the role of newborn astrocytes in a post-ischemic scenario. Combining immunofluorescence, BrdU-tracing and genetic cellular labelling, they tracked the migration of newborn astrocytes (expressing Thbs4) and found that Thbs4-positive astrocytes modulate the extracellular matrix at the lesion border by synthesis but also degradation of hyaluronan. Their results point to a relevant function of SVZ newborn astrocytes in the modulation of the glial scar after brain ischemia. This work's major strength is the fact that it is tackling the function of SVZ newborn astrocytes, whose role is undisclosed so far.  

Strengths:  

The article is innovative, of good quality, and clearly written, with properly described Materials and Methods, data analysis and presentation. In general, the methods are designed properly to answer the main question of the authors, being a major strength. Interpretation of the data is also in general well done, with results supporting the main conclusions of this article.  

In this revised version, the points raised/weaknesses were clarified and discussed in the article.  

Recommendations for the authors:  

Reviewer #1 (Recommendations for the authors):  

Minor points:  

(1) Thanks for the clarification.  

(2) Thanks for the clarification.  

(3) The magnification is not apparent in Fig. 5.  

We had removed two brain slices (from 4 to 2) in order to increase the size of the image 2-fold. We have now further increased the TTC panel, 25% from the revised version, 125% from the original.

(4) Thanks for the clarification.  

(5) Thanks for the clarification.  

(6) Thanks for the clarification.  

(7) Thanks for the clarification.  

(8) Thanks for the clarification.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

(1) As VRMate (a component of behaviorMate) is written using Unity, what is the main advantage of using behaviorMate/VRMate compared to using Unity alone paired with Arduinos (e.g. Campbell et al. 2018), or compared to using an existing toolbox to interface with Unity (e.g. Alsbury-Nealy et al. 2022, DOI: 10.3758/s13428-021-01664-9)? For instance, one disadvantage of using Unity alone is that it requires programming in C# to code the task logic. It was not entirely clear whether VRMate circumvents this disadvantage somehow -- does it allow customization of task logic and scenery in the GUI? Does VRMate add other features and/or usability compared to Unity alone? It would be helpful if the authors could expand on this topic briefly.

We have updated the manuscript (lines 412-422) to clarify the benefits of separating the VR system as an isolated program and a UI that can be run independently. We argue that “…the recommended behaviorMate architecture has several important advantages. Firstly, by rendering each viewing angle of a scene on a dedicated device, performance is improved by splitting the computational costs across several inexpensive devices rather than requiring specialized or expensive graphics cards in order to run…, the overall system becomes more modular and easier to debug [and] implementing task logic in Unity would require understanding Object-Oriented Programming and C# … which is not always accessible to researchers that are typically more familiar with scripting in Python and Matlab.”

VRMate receives detailed configuration info from behaviorMate at runtime as to which VR objects to display and receives position updates during experiments. Any other necessary information about triggering rewards or presenting non-VR cues is still handled by the UI so no editing of Unity is necessary. Scene configuration information is in the same JSON format as the settings files for behaviorMate, additionally there are Unity Editor scripts which are provided in the VRmate repository which permit customizing scenes through a “drag and drop” interface and then writing the scene configuration files programmatically. Users interested in these features should see our github page to find example scene.vr files and download the VRMate repository (including the editor scripts).  We provided 4 vr contexts, as well as a settings file that uses one of them which can be found on the behaviorMate github page (https://github.com/losonczylab/behaviorMate) in the “vr_contexts” and “example_settigs_files” directories. These examples are provided to assist VRMate users in getting set up and could provide a more detailed example of how VRMate and behaviorMate interact.

(2) The section on "context lists", lines 163-186, seemed to describe an important component of the system, but this section was challenging to follow and readers may find the terminology confusing. Perhaps this section could benefit from an accompanying figure or flow chart, if these terms are important to understand.

We maintain the use of the term context and context list in order to maintain a degree of parity with the java code. However, we have updated lines 173-175 to define the term context for the behaviorMate system: “... a context is grouping of one or more stimuli that get activated concurrently. For many experiments it is desirable to have multiple contexts that are triggered at various locations and times in order to construct distinct or novel environments.”

a. Relatedly, "context" is used to refer to both when the animal enters a particular state in the task like a reward zone ("reward context", line 447) and also to describe a set of characteristics of an environment (Figure 3G), akin to how "context" is often used in the navigation literature. To avoid confusion, one possibility would be to use "environment" instead of "context" in Figure 3G, and/or consider using a word like "state" instead of "context" when referring to the activation of different stimuli.

Thank you for the suggestion. We have updated Figure 3G to say “Environment” in order to avoid confusion.

(3) Given the authors' goal of providing a system that is easily synchronizable with neural data acquisition, especially with 2-photon imaging, I wonder if they could expand on the following features:

a. The authors mention that behaviorMate can send a TTL to trigger scanning on the 2P scope (line 202), which is a very useful feature. Can it also easily generate a TTL for each frame of the VR display and/or each sample of the animal's movement? Such TTLs can be critical for synchronizing the imaging with behavior and accounting for variability in the VR frame rate or sampling rate.

Different experimental demands require varying levels of precision in this kind of synchronization signals. For this reason, we have opted against a “one-size fits all” for synchronization with physiology data in behaviorMate. Importantly this keeps the individual rig costs low which can be useful when constructing setups specifically for use when training animals. behaviorMate will log TTL pulses sent to GPIO pins setup as sensors, and can be configured to generate TTL pulses at regular intervals. Additionally all UDP packets received by the UI are time stamped and logged. We also include the output of the arduino millis() function in all UDP packets which can be used for further investigation of clock drift between system components. Importantly, since the system is event driven there cannot be accumulating drift across running experiments between the behaviorMate UI and networked components such as the VR system.

For these reasons, we have not needed to implement a VR frame synchronization TTL for any of our experiments, however, one could extend VRMate to send "sync" packets back to behaviorMate to log when each frame was displayed precisely or TTL pulses (if using the same ODROID hardware we recommend in the standard setup for rendering scenes). This would be useful if it is important to account for slight changes in the frame rate at which the scenes are displayed. However, splitting rendering of large scenes between several devices results in fast update times and our testing and benchmarks indicate that display updates are smooth and continuous enough to appear coupled to movement updates from the behavioral apparatus and sufficient for engaging navigational circuits in the brain.

b. Is there a limit to the number of I/O ports on the system? This might be worth explicitly mentioning.

We have updated lines 219-220 in the manuscript to provide this information: Sensors and actuators can be connected to the controller using one of the 13 digital or 5 analog input/output connectors.

c. In the VR version, if each display is run by a separate Android computer, is there any risk of clock drift between displays? Or is this circumvented by centralized control of the rendering onset via the "real-time computer"?

This risk is mitigated by the real-time computer/UI sending position updates to the VR displays. The maximum amount scenes can be out of sync is limited because they will all recalibrate on every position update – which occurs multiple times per second as the animal is moving. Moreover, because position updates are constantly being sent by behaviorMate to VRMate and VRMate is immediately updating the scene according to this position, the most the scene can become out of sync with the mouse's position is proportional to the maximum latency multiplied by the running speed of the mouse. For experiments focusing on eliciting an experience of navigation, such a degree of asynchrony is almost always negligible. For other experimental demands it could be possible to incorporate more precise frame timing information but this was not necessary for our use case and likely for most other use cases. Additionally, refer to the response to comment 3a.

Reviewer #2 (Public review):

(1) The central controlling logic is coupled with GUI and an event loop, without a documented plugin system. It's not clear whether arbitrary code can be executed together with the GUI, hence it's not clear how much the functionality of the GUI can be easily extended without substantial change to the source code of the GUI. For example, if the user wants to perform custom real-time analysis on the behavior data (potentially for closed-loop stimulation), it's not clear how to easily incorporate the analysis into the main GUI/control program.

Without any edits to the existing source code behaviorMate is highly customizable through the settings files, which allow users to combine the existing contexts and decorators in arbitrary combinations. Therefore, users have been able to perform a wide variety of 1D navigation tasks, well beyond our anticipated use cases by generating novel settings files. The typical method for providing closed-loop stimulation would be to set up a context which is triggered by animal behavior using decorators (e.g. based on position, lap number and time) and then trigger the stimulation with a TTL pulse. Rarely, if users require a behavioral condition not currently implemented or composable out of existing decorators, it would require generating custom code in Java to extend the UI. Performing such edits requires only knowledge of basic object-oriented programming in Java and generating a single subclass of either the BasicContextList or ContextListDecorator classes. In addition, the JavaFX (under development) version of behaviorMate incorporates a plugin which doesn't require recompiling the code in order to make these changes. However, since the JavaFX software is currently under development, documentation does not yet exist. All software is open-sourced and available on github.com for users interested in generating plugins or altering the source code.

We have added the additional caveat to the manuscript in order to clarify this point (Line 197-202): “However, if the available set of decorators is not enough to implement the required task logic, some modifications to the source code may be necessary. These modifications, in most cases, would be very simple and only a basic understanding of object-oriented programming is required. A case where this might be needed would be performing novel customized real-time analysis on behavior data and activating a stimulus based on the result”

(2) The JSON messaging protocol lacks API documentation. It's not clear what the exact syntax is, supported key/value pairs, and expected response/behavior of the JSON messages. Hence, it's not clear how to develop new hardware that can communicate with the behaviorMate system.

The most common approach for adding novel hardware is to use TTL pulses (or accept an emitted TTL pulse to read sensor states). This type of hardware addition  is possible through the existing GPIO without the need to interact with the software or JSON API. Users looking to take advantage of the ability to set up and configure novel behavioral paradigms without the need to write any software would be limited to adding hardware which could be triggered with and report to the UI with a TTL pulse (however fairly complex actions could be triggered this way).

For users looking to develop more customized hardware solutions that interact closely with the UI or GPIO board, additional documentation on the JSON messaging protocol has been added to the behaviormate-utils repository (https://github.com/losonczylab/behaviormate_utils). Additionally, we have added a link to this repository in the Supplemental Materials section (line 971) and referenced this in the manuscript (line 217) to make it easier for readers to find this information.

Furthermore, developers looking to add completely novel components to the UI  can implement the interface described by Context.java in order to exchange custom messages with hardware. (described  in the JavaDoc: https://www.losonczylab.org/behaviorMate-1.0.0/)  These messages would be defined within the custom context and interact with the custom hardware (meaning the interested developer would make a novel addition to the messaging API). Additionally, it should be noted that without editing any software, any UDP packets sent to behaviorMate from an IP address specified in the settings will get time stamped and logged in the stored behavioral data file meaning that are a large variety of hardware implementation solutions using both standard UDP messaging and through TTL pulses that can work with behaviorMate with minimal effort. Finally, see response to R2.1 for a discussion of the JavaFX version of the behaviorMatee UI including plugin support.

(3) It seems the existing control hardware and the JSON messaging only support GPIO/TTL types of input/output, which limits the applicability of the system to more complicated sensor/controller hardware. The authors mentioned that hardware like Arduino natively supports serial protocols like I2C or SPI, but it's not clear how they are handled and translated to JSON messages.

We provide an implementation for an I2C-based capacitance lick detector which interested developers may wish to copy if support for novel I2C or SPI. Users with less development experience wishing to expand the hardware capabilities of  behaviorMatecould also develop adapters which can be triggered  on a TTL input/output. Additionally, more information about the JSON API and how messages are transmitted to the PC by the arduino is described in point (2) and the expanded online documentation.

a. Additionally, because it's unclear how easy to incorporate arbitrary hardware with behaviorMate, the "Intranet of things" approach seems to lose attraction. Since currently, the manuscript focuses mainly on a specific set of hardware designed for a specific type of experiment, it's not clear what are the advantages of implementing communication over a local network as opposed to the typical connections using USB.

As opposed to serial communication protocols as typical with USB, networking protocols seamlessly function based on asynchronous message passing. Messages may be routed internally (e.g. to a PCs localhost address, i.e. 0.0.0..0) or to a variety of external hardware (e.g. using IP addresses such as those in the range 192.168.1.2 - 192.168.1.254). Furthermore, network-based communication allows modules, such as VR, to be added easily. behavoirMate systems can be easily expanded using low-cost Ethernet switches and consume only a single network adapter on the PC (e.g. not limited by the number of physical USB ports). Furthermore, UDP message passing is implemented in almost all modern programming languages in a platform independent manner (meaning that the same software can run on OSX, Windows, and Linux). Lastly, as we have pointed out (Line 117) a variety of tools exist for inspecting network packets and debugging; meaning that it is possible to run behaviorMate with simulated hardware for testing and debugging.

The IOT nature of behaviorMate means there is no requirement for novel hardware to be implemented  using an arduino,  since any system capable of  UDP communication can  be configured. For example, VRMate is usually run on Odroid C4s, however one could easily create a system using Raspberry Pis or even additional PCs. behaviorMate is agnostic to the format of the UDP messages, but packaging any data in the JSON format for consistency would be encouraged. If a new hardware is a sensor that has input requiring it to be time stamped and logged then all that is needed is to add the IP address and port information to the ‘controllers’ list in a behaviorMate settings file. If more complex interactions are needed with novel hardware than a custom implementation of ContextList.java may be required (see response to R2.2). However, the provided UdpComms.java class could be used to easily send/receive messages from custom Context.java subclasses.

Solutions for highly customized hardware do require basic familiarity with object-oriented programming using the Java programming language. However, in our experience most behavioral experiments do not require these kinds of modifications. The majority of 1D navigation tasks, which behaviorMate is currently best suited to control, require touch/motion sensors, LEDs, speakers, or solenoid valves,  which are easily controlled by the existing GPIO implementation. It is unlikely that custom subclasses would even be needed.

Reviewer #3 (Public review):

(1) While using UDP for data transmission can enhance speed, it is thought that it lacks reliability. Are there error-checking mechanisms in place to ensure reliable communication, given its criticality alongside speed?

The provided GPIO/behavior controller implementation sends acknowledgement packets in response to all incoming messages as well as start and stop messages for contexts and “valves”. In this way the UI can update to reflect both requested state changes as well as when they actually happen (although there is rarely a perceptible gap between these two states unless something is unplugged or not functioning). See Line 85 in the revised manuscript “acknowledgement packets are used to ensure reliable message delivery to and from connected hardware”.

(2) Considering this year's price policy changes in Unity, could this impact the system's operations?

VRMate is not affected by the recent changes in pricing structure of the Unity project.

The existing compiled VRMate software does not need to be regenerated to update VR scenes, or implement new task logic (since this is handled by the behaviorMate GUI). Therefore, the VRMate program is robust to any future pricing changes or other restructuring of the Unity program and does not rely on continued support of Unity. Additionally, while the solution presented in VRMate has many benefits, a developer could easily adapt any open-source VR Maze project to receive the UDP-based position updates from behaviorMate or develop their own novel VR solutions.

(3) Also, does the Arduino offer sufficient precision for ephys recording, particularly with a 10ms check?

Electrophysiology recording hardware typically has additional I/O channels which can provide assistance with tracking behavior/synchronization at a high resolution. While behaviorMate could still be used to trigger reward valves, either the ephys hardware or some additional high-speed DAQ would be recommended to maintain accurately with high-speed physiology data. behaviorMate could still be set up as normal to provide closed and open-loop task control at behaviorally relevant timescales alongside a DAQ circuit recording events at a consistent temporal resolution. While this would increase the relative cost of the individual recording setup, identical rigs for training animals could still be configured without the DAQ circuit avoiding unnecessary cost and complexity.

(4) Could you clarify the purpose of the Sync Pulse? In line 291, it suggests additional cues (potentially represented by the Sync Pulse) are needed to align the treadmill screens, which appear to be directed towards the Real-Time computer. Given that event alignment occurs in the GPIO, the connection of the Sync Pulse to the Real-Time Controller in Figure 1 seems confusing.

A number of methods exist for synchronizing recording devices like microscopes or electrophysiology recordings with behaviorMate’s time-stamped logs of actuators and sensors. For example, the GPIO circuit can be configured to send sync triggers, or receive timing signals as input. Alternatively a dedicated circuit could record frame start signals and relay them to the PC to be logged independently of the GPIO (enabling a high-resolution post-hoc alignment of the time stamps). The optimal method to use varies based on the needs of the experiment. Our setups have a dedicated BNC output and specification in the settings file that sends a TTL pulse at the start of an experiment in order to trigger 2p imaging setups (see line 224, specifically that this is a detail of “our” 2p imaging setup). We provide this information as it might be useful suggesting how to have both behavior and physiology data start recording at the same time. We do not intend this to be the only solution for alignment. Figure 1 indicates an “optional” circuit for capturing a high speed sync pulse and providing time stamps back to the real time PC. This is another option that might be useful for certain setups (or especially for establishing benchmarks between behavior and physiology recordings). In our setup event alignment does not exclusively occur on the GPIO.

a. Additionally, why is there a separate circuit for the treadmill that connects to the UI computer instead of the GPIO? It might be beneficial to elaborate on the rationale behind this decision in line 260.

Event alignment does not occur on the GPIO, separating concerns between position tracking and more general input/output features which improves performance and simplifies debugging.  In this sense we maintain a single event loop on the Arduino, avoiding the need to either run multithreaded operations or rely extensively on interrupts which can cause unpredictable code execution (e.g. when multiple interrupts occur at the same time). Our position tracking circuit is therefore coupled to a separate,low-cost arduino mini which has the singular responsibility of position-tracking.

b. Moreover, should scenarios involving pupil and body camera recordings connect to the Analog input in the PCB or the real-time computer for optimal data handling and processing?

Pupil and body camera recordings would be independent data streams which can be recorded separately from behaviorMate. Aligning these forms of full motion video could require frame triggers which could be configured on the GPIO board using single TTL like outputs or by configuring a valve to be “pulsed” which is a provided type customization.

We also note that a more advanced developer could easily leverage camera signals to provide closed loop control by writing an independent module that sends UDP packets to behavoirMate. For example a separate computer vision based position tracking module could be written in any preferred language and use UDP messaging to send body tracking updates to the UI without editing any of the behaviorMate source code (and even used for updating 1D location).

(5) Given that all references, as far as I can see, come from the same lab, are there other labs capable of implementing this system at a similar optimal level?

To date two additional labs have published using behaviorMate, the Soltez and Henn labs (see revised lines 341-342). Since behaviorMate has only recently been published and made available open source, only external collaborators of the Losonczy lab have had access to the software and design files needed to do this. These collaborators did, however, set up their own behavioral setups in separate locations with minimal direct support from the authors–similar to what would be available to anyone seeking to set a behaviorMate system would find online on our github page or by posting to the message board.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(4) To provide additional context for the significance of this work, additional citations would be helpful to demonstrate a ubiquitous need for a system like behaviorMate. This was most needed in the paragraph from lines 46-65, specifically for each sentence after line 55, where the authors discuss existing variants on head-fixed behavioral paradigms. For instance, for the clause "but olfactory and auditory stimuli have also been utilized at regular virtual distance intervals to enrich the experience with more salient cues", suggested citations include Radvansky & Dombeck 2018 (DOI: 10.1038/s41467-018-03262-4), Fischler-Ruiz et al. 2021 (DOI: 10.1016/j.neuron.2021.09.055).

We thank the reviewer for the suggested missing citations and have updated the manuscript accordingly (see line 58).

(5) In addition, it would also be helpful to clarify behaviorMate's implementation in other laboratories. On line 304 the authors mention "other labs" but the following list of citations is almost exclusively from the Losonczy lab. Perhaps the citations just need to be split across the sentence for clarity? E.g. "has been validated by our experimental paradigms" (citation set 1) "and successfully implemented in other labs as well" (citation set 2).

We have split the citation set as suggested (see lines 338-342).

Minor Comments:

(6) In the paragraph starting line 153 and in Fig. 2, please clarify what is meant by "trial" vs. "experiment". In many navigational tasks, "trial" refers to an individual lap in the environment, but here "trial" seems to refer to the whole behavioral session (i.e. synonymous with "experiment"?).

In our software implementation we had originally used “trial” to refer to an imaging session rather than experiment (and have made updates to start moving to the more conventional lexicon). To avoid confusion we have remove this use of “trial” throughout the manuscript and replaced with “experiment” whenever possible

(7) This is very minor, but in Figure 3 and 4, I don't believe the gavage needle is actually shown in the image. This is likely to avoid clutter but might be confusing to some readers, so it may be helpful to have a small inset diagram showing how the needle would be mounted.

We assessed the image both with and without the gavage needle and found the version in the original (without) to be easier to read and less cluttered and therefore maintained that version in the manuscript.

(8) In Figure 5 legend, please list n for mice and cells.

We have updated the Figure 5 legend to indicate that for panels C-G, n=6 mice (all mice were recorded in both VR and TM systems), 3253 cells in VR classified as significantly tuned place cells VR, and 6101 tuned cells in TM,

(9) Line 414: It is not necessary to tilt the entire animal and running wheel as long as the head-bar clamp and objective can rotate to align the imaging window with the objective's plane of focus. Perhaps the authors can just clarify the availability of this option if users have a microscope with a rotatable objective/scan head.

We have added the suggested caveat to the manuscript in order to clarify when the goniometers might be useful (see lines 281-288).

(10) Figure S1 and S2 could be referenced explicitly in the main text with their related main figures.

We have added explicit references to figures S1 and S2 in the relevant sections (see lines 443, 460  and 570)

(11) On line 532-533, is there a citation for "proximal visual cues and tactile cues (which are speculated to be more salient than visual cues)"?

We have added citations to both Knierim & Rao 2003 and Renaudineau et al. 2007 which discuss the differential impact of proximal vs distal cues during navigation as well as Sofroniew et al. 2014 which describe how mice navigate more naturally in a tactile VR setup as opposed to purely visual ones.

(12) There is a typo at the end of the Figure 2 legend, where it should say "Arduino Mini."

This typo has been fixed.

Reviewer #2 (Recommendations For The Authors):

(4) As mentioned in the public review: what is the major advantage of taking the IoT approaches as opposed to USB connections to the host computer, especially when behaviorMate relies on a central master computer regardless? The authors mentioned the readability of the JSON messages, making the system easier to debug. However, the flip side of that is the efficiency of data transmission. Although the bandwidth/latency is usually more than enough for transmitting data and commands for behavior devices, the efficiency may become a problem when neural recording devices (imaging or electrophysiology) need to be included in the system.

behaviorMate is not intended to do everything, and is limited to mainly controlling behavior and providing some synchronizing TTL style triggers. In this way the system can easily and inexpensively be replicated across multiple recording setups; particularly this is useful for constructing additional animal training setups. The system is very much sufficient for capturing behavioral inputs at relevant timescales (see the benchmarks in Figures 3 and 4 as well as the position correlated neural activity in Figures 5 and 6 for demonstration of this). Additional hardware might be needed to align the behaviorMate output with neural data for example a high-speed DAQ or input channels on electrophysiology recording setups could be utilized (if provided). As all recording setups are different the ideal solution would depend on details which are hard to anticipate. We do not mean to convey that the full neural data would be transmitted to the behaviorMate system (especially using the JSON/UDP communications that behaviorMate relies on).

(5) The author mentioned labView. A popular open-source alternative is bonsai (https://github.com/bonsai-rx/bonsai). Both include a graphical-based programming interface that allows the users to easily reconfigure the hardware system, which behaviorMate seems to lack. Additionally, autopilot (https://github.com/auto-pi-lot/autopilot) is a very relevant project that utilizes a local network for multiple behavior devices but focuses more on P2P communication and rigorously defines the API/schema/communication protocols for devices to be compatible. I think it's important to include a discussion on how behaviorMate compares to previous works like these, especially what new features behaviorMate introduces.

We believe that behaviorMate provides a more opinionated and complete solution than the projects mentioned. A wide variety of 1D navigational paradigms can be constructed in behaviorMate without the need to write any novel software. For example, bonsai is a “visual programming language” and would require experimenters to construct a custom implementation of each of their experiments. We have opted to use Java for the UI with distributed computations across modules in various languages. Given the IOT methodology it would be possible to use any number of programming languages or APIs; a large number of design decisions were made  when building the project and we have opted to not include this level of detail in the manuscript in order to maintain readability. We strongly believe in using non-proprietary and open source projects, when possible, which is why the comparison with LabView based solutions was included in the introduction. Also, we have added a reference to the autopilot reference to the section of the introduction where this is discussed.

(6) One of the reasons labView/bonsai are popular is they are inherently parallel and can simultaneously respond to events from different hardware sources. While the JSON events in behaviorMate are asynchronous in nature, the handling of those events seems to happen only in a main event loop coupled with GUI, which is sequential by nature. Is there any multi-threading/multi-processing capability of behaviorMate? If so it's an important feature to highlight. If not I think it's important to discuss the potential limitation of the current implementation.

IOT solutions are inherently concurrent since the computation is distributed. Additional parallelism could be added by further distributing concerns between additional independent modules running on independent hardware. The UI has an eventloop which aggregates inputs and then updates contexts based on the current state of those inputs sequentially. This sort of a “snapshot” of the current state is necessary to reason about when the start certain contexts based on their settings and applied decorators. While the behaviorMate UI uses multithreading libraries in Java to be more performant in certain cases, the degree to which this represents true vs “virtual” concurrency would depend on the individual PC architecture it is run on and how the operating system allocates resources. For this reason, we have argued in the manuscript that behaviorMate is sufficient for controlling experiments at behaviorally relevant timescales, and have presented both benchmarks and discussed different synchronization approaches and permit users to determine if this is sufficient for their needs.

(7) The context list is an interesting and innovative approach to abstract behavior contingencies into a data structure, but it's not currently discussed in depth. I think it's worth highlighting how the context list can be used to cover a wide range of common behavior experimental contingencies with detailed examples (line 185 might be a good example to give). It's also important to discuss the limitation, as currently the context lists seem to only support contingencies based purely on space and time, without support for more complicated behavior metrics (e.g. deliver reward only after X% correct).

To access more complex behavior metrics during runtime, custom context list decorators would need to be implemented. While this is less common in the sort of 1D navigational behaviors the project was originally designed to control, adding novel decorators is a simple process that only requires basic object oriented programming knowledge. As discussed we are also implementing a plugin-architecture in the JavaFX update to streamline these types of additions.

Minor Comments:

(8) In line 202, the author suggests that a single TTL pulse is sent to mark the start of a recording session, and this is used to synchronize behavior data with imaging data later. In other words, there are no synchronization signals for every single sample/frame. This approach either assumes the behavior recording and imaging are running on the same clock or assumes evenly distributed recording samples over the whole recording period. Is this the case? If so, please include a discussion on limitations and alternative approaches supported by behaviorMate. If not, please clarify how exactly synchronization is done with one TTL pulse.

While the TTL pulse triggers the start of neural data in our setups, various options exist for controlling for the described clock drift across experiments and the appropriate one depends on the type of recordings made, frame rate duration of recording etc. Therefore behaviorMate leaves open many options for synchronization at different time scales (e.g. the adding a frame-sync circuit as shown in Figure 1 or sending TTL pulses to the same DAQ recording electrophysiology data).  Expanded consideration of different synchronization methods has been included in the manuscript (see lines 224-238).

(9) Is the computer vision-based calibration included as part of the GUI functionality? Please clarify. If it is part of the GUI, it's worth highlighting as a very useful feature.

The computer vision-based benchmarking is not included in the GUI. It is in the form of a script made specifically for this paper. However for treadmill-based experiments behaviorMate has other calibration tools built into it (see line 301-303).

(10) I went through the source code of the Arduino firmware, and it seems most "open X for Y duration" functions are implemented using the delay function. If this is indeed the case, it's generally a bad idea since delay completely pauses the execution and any events happening during the delay period may be missed. As an alternative, please consider approaches comparing timestamps or using interrupts.

We have avoided the use of interrupts on the GPIO due to the potential for unpredictable code execution. There is a delay which is only just executed if the duration is 10 ms or less as we cannot guarantee precision of the arduino eventloop cycling faster than this. Durations longer than 10 ms would be time stamped and non-blocking. We have adjusted this MAX_WAIT to be specified as a macro so it can be more easily adjusted (or set to 0).

(11) Figure 3 B, C, D, and Figure 4 D, E suffer from noticeable low resolution.

We have converted Figure 3B, C, D and 4C, D, E to vector graphics in order to improve the resolution.

(12) Figure 4C is missing, which is an important figure.

This figure appeared when we rendered and submitted the manuscript. We apologize if the figure was generated such that it did not load properly in all pdf viewers. The panel appears correctly in the online eLife version of the manuscript. Additionally, we have checked the revision in Preview on Mac OS as well as Adobe Acrobat and the built-in viewer in Chrome and all figure panels appear in each so we hope this issue has been resolved.

(13) There are thin white grid lines on all heatmaps. I don't think they are necessary.

The grid lines have been removed from the heatmaps  as suggested.

(14) Line 562 "sometimes devices directly communicate with each other for performance reasons", I didn't find any elaboration on the P2P communication in the main text. This is potentially worth highlighting as it's one of the advantages of taking the IoT approaches.

In our implementation it was not necessary to rely on P2P communication beyond what is indicated in Figure 1. The direct communication referred to in line 562 is meant only to refer to the examples expanded on in the rest of the paragraph i.e. the behavior controller may signal the microscope directly using a TTL signal without looping back to the UI. As necessary users could implement UDP message passing between devices, but this is outside the scope of what we present in the manuscript.

(15) Line 147 "Notably, due to the systems modular architecture, different UIs could be implemented in any programming language and swapped in without impacting the rest of the system.", this claim feels unsupported without a detailed discussion of how new code can be incorporated in the GUI (plugin system).

This comment refers to the idea of implementing “different UIs”. This would entail users desiring to take advantage of the JSON messaging API and the proposed electronics while fully implementing their own interface. In order to facilitate this option we have improved documentation of the messaging API posted in the README file accompanying the arduino source code. We have added reference to the supplemental materials where readers can find a link to the JSON API implementation to clarify this point.

Additionally, while a plugin system is available in the JavaFX version of behaviorMate, this project is currently under development and will update the online documentation as this project matures, but is unrelated to the intended claim about completely swapping out the UI.

Reviewer #3 (Recommendations For The Authors):

(6) Figure 1 - the terminology for each item is slightly different in the text and the figure. I think making the exact match can make it easier for the reader.

- Real-time computer (figure) vs real-time controller (ln88).

The manuscript was adjusted to match figure terminology.

- The position controller (ln565) - position tracking (Figure).

We have updated Figure 1 to highlight that the position controller does the position tracking.

- Maybe add a Behavior Controller next to the GPIO box in Figure 1.

We updated Figure 1 to highlight that the Behavior Controller performs the GPIO responsibility such that "Behavior Controller" and "GPIO circuit" may be used interchangeably.

- Position tracking (fig) and position controller (subtitle - ln209).

We updated Figure 1 to highlight that the position controller does the position tracking.

- Sync Pulse is not explained in the text.

The caption for Figure 1 has been updated to better explain the Sync pulse and additional systems boxes

(7) For Figure 3B/C: What is the number of data points? It would be nice to see the real population, possibly using a swarm plot instead of box plots. How likely are these outliers to occur?

In order to better characterize the distributions presented in our benchmarking data we have added mean and standard deviation information the plots 3 and 4. For Figure 3B: 0.0025 +/- 0.1128, Figure 3C: 12.9749 +/- 7.6581, Figure 4C: 66.0500 +/- 15.6994, Figure 4E: 4.1258 +/- 3.2558.

Author response:

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

Reviewer #1 (Public Review):

Time periods in which experience regulates early plasticity in sensory circuits are well established, but the mechanisms that control these critical periods are poorly understood. In this manuscript, Leier and Foden and colleagues examine early-life critical periods that regulate the Drosophila antennal lobe, a model sensory circuit for understanding synaptic organization. Using early-life (0-2 days old) exposure to distinct odorants, they show that constant odor exposure markedly reduces the volume, synapse number, and function of the VM7 glomerulus. The authors offer evidence that these changes are mediated by invasion of ensheathing glia into the glomerulus where they phagocytose connections via a mechanism involving the engulfment receptor Draper.

This manuscript is a striking example of a study where the questions are interesting, the authors spent a considerable amount of time to clearly think out the best experiments to ask their questions in the most straightforward way, and expressed the results in a careful, cogent, and well-written fashion. It was a genuine delight to read this paper. I have two experimental suggestions that would really round out existing work to better support the existing conclusions and some instances where additional data or tempered language in describing results would better support their conclusions. Overall, though, this is an incredibly important finding, a careful analysis, and an excellent mechanistic advance in understanding sensory critical period biology.

We thank the reviewer for their thoughtful and constructive comments on our manuscript. In response to their critiques, we conducted several new experiments as well as additional analysis and making changes to the text. As requested, we carried out an electrophysiological analysis of VM7 PN firing in draper knockdown animals with and without odor exposure. To our surprise, loss of glial Draper fully suppresses the dramatic reduction in spontaneous PN activity observed following critical period ethyl butyrate exposure, arguing that the functional response is restored alongside OSN morphology. It also suggests that the OR42a OSN terminals are intact and functional until they are phagocytosed by ensheathing glia. In other words, glia are not merely clearing axon terminals that have already degenerated. This evidence provides additional support to the claim that the VM7 glomerulus will be an outstanding model for defining mechanism of experience-dependent glial pruning. Detailed responses to the reviewers’ comments follow below. 

Regarding the apparent disconnect between the near complete silencing of PNs versus the 50% reduction in OR42a OSN infiltration volume, we agree with the reviewer that this tracks with previous data in the field. While our Imaris pipeline is relatively sensitive, it may not pick up modest changes to terminal arbor architecture. Indeed, as described in Jindal et al. (2023) and in the Methods in this manuscript, we chose conservative software settings that, if anything, would undercount the percent change in infiltration volume. We also note that increased inhibitory LN inputs onto PNs could contribute to dramatic PN silencing we observe. While fascinating, we view LN plasticity beyond the scope of the current manuscript. We removed any mention of ‘silent synapses’ and now speculate about increased inhibition. 

Reviewer #1 (Recommendations For The Authors):

Major Elements:

(1) The authors demonstrate that loss of draper in glia can suppress many of the pruning related phenotypes associated with EB exposure. However, they do not assess electrophysiological output in these experiments, only morphology. It would be great to see recordings from those animals to see if the functional response is also restored.

We performed the experiment the reviewer requested (see Figure 4F-J). We are pleased to report that our recordings from VM7 PNs match our morphology measurements: in repo-GAL4>UAS-draper RNAi flies, there was no difference in the innervation of VM7 PNs between animals exposed to mineral oil or 15% EB from 0-2 DPE. This result is in sharp contrast to the near-total loss of OSN-PN innervation in flies with intact glial Draper signaling, and strongly validates the role we propose for Draper in the Or42a OSN critical period.

(2) There is a disconnect between physiology and morphology with a near complete loss of activity from VM7 PNs but a less severe loss of ORN synapses. While not completely incongruent (previous work in the AL showed a complete loss of attractive behavior though synapse number was only reduced 40% - Mosca et al. 2017, eLife), it is curious. Can the authors comment further? Ideally, some of these synapses could be visualized by EM to determine if the remaining synapses are indeed of correct morphology. If not, this could support their assertion of silent inputs from page 7. Further, what happens to the remaining synapses? VM7 PNs should be receiving some activity from other local interneurons as well as neighboring PNs.

We agree that on the surface, our electrophysiology results are more striking than one might expect solely from our measurements of VM7 morphology and presynaptic content. As the reviewer points out, previous studies of fly olfaction have consistently found that relatively modest shifts in glomerular volume in response to prolonged earlylife odorant exposure can be accompanied by drastic changes in physiology and behavior (in addition, we would add Devaud et al., 2003; Devaud et al., 2001; Acebes et al., 2012; and Chodankar et al., 2020, as foundational examples of this phenomenon). 

A major driver of these changes appears to be remodeling of antennal lobe inhibitory LNs (see Das et al., 2011; Wilson and Laurent, 2005; Chodankar et al., 2020), especially GABAergic inhibitory interneurons. Perhaps increased LN inhibition of chronically activated PNs, on top of the reduced excitatory inputs resulting from ensheathing glial pruning of the Or42a OSN terminal arbor, would explain the near-total loss of VM7 PN activity we observe after critical period EB exposure. However, given that the scope of our study is limited to critical-period glial biology and does not address the complex topics of LN rewiring or synapse morphology, we have removed the sentence in which we raise the possibility of “silent synapses” in order to avoid confusion. The reviewer is also correct that VM7 PNs have inputs from non-ORN presynaptic partners, including LNs and PNs. So again, perhaps increased inhibitory inputs contributes to the near-complete silencing of the PNs. Given the heterogeneity of LN populations, we view this area as fertile ground for future research. 

Language / Data Considerations:

(1) Or42a OSNs have other inputs, namely, from LNs. What are they doing here? Are they also affected?

As discussed above, the question of how LN innervation of Or42a OSNs is altered by critical-period EB exposure is an intriguing one that fully deserves its own follow-up study, and we have tried to avoid speculation about the role of LNs when discussing our pruning phenotype. We note at multiple points throughout the text the importance of LNs and refer to previous studies of LN plasticity in response to chronic odorant exposure. 

(2) In all of the measurements, what happens to synaptic density? Is it maintained? Does it scale precisely? This would be helpful to know.

We have performed the analysis as requested, which is now included in a supplement to Figure 5. We found that synaptic density shows no trend in variation across conditions and glial driver genotypes.

(3) In Figure 5, the controls for the alrm-GAL4 experiments show a much more drastic phenotype than controls in previous figures? Does this background influence how we can interpret the results? Could the response have instead hit a floor effect and it's just not possible to recover?

The reviewer is correct that following EB exposure, astrocyte vs. ensheathing glial driver backgrounds displayed modest differences in the extent of pruning by volume (0.27 for astros, 0.36 for EG). We note that the two drpr RNAi lines that we used had non-significant (but opposite) effects on the estimated size of OSN42a OSN volume in combination with the astrocyte driver, arguing against a floor effect. In addition, a recent publication by Nelson et al. (2024) replicated our findings with a different astrocyte GAL4 driver and draper RNAi line. Thus, we are confident that this result is biologically meaningful and not an artifact of genetic background. 

(4) The estimation of infiltration measurement in Figure 6 is tricky to interpret. It implies that the projections occupy the same space, which cannot be possible. I'd advocate a tempering of some of this language and consider an intensity measurement in addition to their current volume measurements (or perhaps an "occupied space" measurement) to more accurately assess the level of resolution that can be obtained via these methods.

We completely agree that our language in describing EG infiltration could have been more precise, and we modified our language as suggested. The combination of the Or42a-mCD8::GFP label we and others use, our use of confocal microscopy, and our Surface pipeline in Imaris combine to create a glomerular mask that traces the outline of the OSN terminal arbor, but is nonetheless not 100% “filled” by neuronal membrane and/or glial processes. 

(5) Do the authors have the kind of resolution needed to tell whether there is indeed Or42a-positive axon fragmentation (as asserted on p16 and from their data in figures 4, 5, 7). If the authors want to say this, I would advocate for a measurement of fragmentation / total volume to prove it - if not, I would advocate tempering of the current language.

The reviewer brings up a fair criticism: while our assertion about axon fragmentation was based on our visual observations of hundreds of EB-exposed brains, the resolution limits of confocal microscopy do not allow us to rigorously rule out fragmentation within a bundle of OSN axons. Instead, our most compelling evidence for the lack of EB-induced Or42a OSN fragmentation in the absence of glial Draper comes from our new electrophysiology data (Figure 4F-J) in repo-GAL4>UAS-draper RNAi animals. We found no difference in spontaneous release from Or42a terminals in flies exposed to mineral oil or 15% EB from 0-2 DPE, which would not be the case if there was Draper-independent fragmentation along the axons or terminal arbors upon EB exposure. We have updated our discussion of fragmentation so that our statements are based on this new evidence, and not confocal microscopy. 

(6) There is an interesting Discussion opportunity missed here. Some experiments would, ostensibly, require pupae to detect odorants within the casing via structures consistently in place for olfaction during pupation. It would be useful for the authors to discuss a little more deeply when this critical period may arise and why the experiment where pupae are exposed to EB two days before eclosion and there is no response, occurs as it does. I agree that it's clearly a time when they are not sensitive to the odorant, but that could just be because there's no ability to detect odorants at that time. Is it a question of non-sensitivity to EB or just non-sensitivity to everything?

We share the reviewer’s interest in the plasticity of the olfactory circuit during pupariation, although, as they correctly point out, it is difficult to conceive of an odorant-exposure experiment that could disentangle the barrier effects of puparium from the sensitivity of the circuit itself, and our pre-eclosion data in Figure 3A, D, G does not distinguish between the two. While an investigation into mechanism by which the critical period for ethyl butyrate exposure opens and closes is outside the scope of the present study, we would consider the physical barrier of the puparium to be a satisfactory explanation for why eclosion marks the functional opening of experiencedependent plasticity. As the reviewer suggests, we have added this important nuance to our discussion of the opening of the critical period in the corresponding paragraph of the Results, as well as to the Discussion section “Glomeruli exhibit dichotomous responses to critical period odor exposure.” 

Minor Elements:

(1) Page 6 bottom: "Or4a-mCD8::GFP" should be "Or42a-mCD8::GFP"

(2) Page 15, end of last full paragraph. Remove the "e"

Thank you for pointing out these typos. They have been corrected. 

Reviewer #2 (Public Review):

Sensory experiences during developmental critical periods have long-lasting impacts on neural circuit function and behavior. However, the underlying molecular and cellular mechanisms that drive these enduring changes are not fully understood. In Drosophila, the antennal lobe is composed of synapses between olfactory sensory neurons (OSNs) and projection neurons (PNs), arranged into distinct glomeruli. Many of these glomeruli show structural plasticity in response to early-life odor exposure, reflecting the sensitivity of the olfactory circuitry to early sensory experiences.

In their study, the authors explored the role of glia in the development of the antennal lobe in young adult flies, proposing that glial cells might also play a role in experiencedependent plasticity. They identified a critical period during which both structural and functional plasticity of OSN-PN synapses occur within the ethyl butyrate (EB)responsive VM7 glomerulus. When flies were exposed to EB within the first two days post-eclosion, significant reductions in glomerular volume, presynaptic terminal numbers, and postsynaptic activity were observed. The study further highlights the importance of the highly conserved engulfment receptor Draper in facilitating this critical period plasticity. The authors demonstrated that, in response to EB exposure during this developmental window, ensheathing glia increase Draper expression, infiltrate the VM7 glomerulus, and actively phagocytose OSN presynaptic terminals. This synapse pruning has lasting effects on circuit function, leading to persistent decreases in both OSN-PN synapse numbers and spontaneous PN activity as analyzed by perforated patch-clamp electrophysiology to record spontaneous activity from PNs postsynaptic to Or42a OSNs.

In my view, this is an intriguing and potentially valuable set of data. However, since I am not an expert in critical periods or habituation, I do not feel entirely qualified to assess the full significance or the novelty of their findings, particularly in relation to existing research.

We thank the reviewer for their insightful critique of our work. In response to their comments, we added additional physiological analysis and tempered our language around possible explanations for the apparent disconnect between the physiological and morphological critical period odor exposure. These changes are explained in more detail in the response to the public review by Reviewer 1 and also in our responses outlined below. 

Reviewer #2 (Recommendations For The Authors):

I though do have specific comments and questions concerning the presynaptic phenotype they deduce from confocal BRP stainings and electrophysiology.

Concerning the number of active zones: this can hardly be deduced from standardresolution confocal images and, maybe more importantly, lacking postsynaptic markers. This particularly also in the light of them speculating about "silent synapses". There are now tools existing concerning labeled, cell type specific expression of acetylcholine-receptor expression and cholinergic postsynaptic density markers (importantly Drep2). Such markers should be entailed in their analysis. They should refer to previous concerning "brp-short" concerning its original invention and prior usage.

We thank the reviewer for their thoughtful approach to our methodology and claims. While the use of confocal microscopy of Bruchpilot puncta to estimate numbers of presynapses is standard practice (see Furusawa et al., 2023; Aimino et al., 2022; Urwyler et al., 2019; Ackerman et al., 2021), the reviewer is correct that a punctum does not an active zone make. Bruchpilot staining and quantification is a well-validated tool for approximating the number of presynaptic active zones, not a substitute for super-resolution microscopy. We made changes to our language about active zones to make this distinction clearer. We have also removed the sentence where we discuss the possibility of “silent synapses,” which both reviewers felt was too speculative for our existing data. Finally, we are highly interested in characterizing the response of PNs and higher-order processing centers to critical-period odorant exposure as a future direction for our research. However, given the complexity of the subject, we chose to limit the scope of this study to the interactions between OSNs and glia. 

Regarding their electrophysiological analysis and the plausibility of their findings: I am uncertain whether the moderate reduction in BRP puncta at the relevant OSN::PN synapse can fully account for the significantly reduced spontaneous PN activity they report. This seems particularly doubtful in the absence of any direct evidence for postsynaptically silent synapses. Perhaps this is my own naivety, but I wonder why they did not use antennal nerve stimulation in their experiments?

We refer to previous studies of the AL indicating that moderate changes in glomerular volume and presynaptic content can translate to far more striking alterations in electrophysiology and behavior (Devaud et al., 2003; Devaud et al., 2001; Acebes et al., 2012; and Chodankar et al., 2020, Mosca et al., 2017). This literature has demonstrated that chronic odorant exposure can result in remodeling of inhibitory local interneurons to suppress over-active inputs from OSNs. While we do not address the complex subject of interneuron remodeling in the present study, we find it highly likely that there would be significant changes in interneuron innervation of PNs, independent of glial phagocytosis of OSN excitatory inputs, resulting in additional inhibition. Moving forward, we are very interested in expanding these studies to include odor-evoked changes in PN activity.  

Additional minor point: The phrase "Soon after its molecular biology was described (et al., 1999), the Drosophila melanogaster" seems somewhat misleading. Isn't the field still actively describing the molecular biology of the fly olfactory system?

We completely agree and have removed this sentence entirely.  

Reviewing Editor's Note: to enhance the evidence from mostly compelling in most facets to solid would be to add physiology to the Draper analysis.

These experiments have been completed and are presented in Figure 4F-J. 

References

Acebes A, Devaud J-M, Arnés M, Ferrús A. 2012. Central Adaptation to Odorants Depends on PI3K Levels in Local Interneurons of the Antennal Lobe. J Neurosci 32:417–422. doi:10.1523/jneurosci.2921-11.2012

Ackerman SD, Perez-Catalan NA, Freeman MR, Doe CQ. 2021. Astrocytes close a motor circuit critical period. Nature592:414–420. doi:10.1038/s41586-021-03441-2

Aimino MA, DePew AT, Restrepo L, Mosca TJ. 2022. Synaptic Development in Diverse Olfactory Neuron Classes Uses Distinct Temporal and Activity-Related Programs. J Neurosci 43:28–55. doi:10.1523/jneurosci.0884-22.2022

Chodankar A, Sadanandappa MK, VijayRaghavan K, Ramaswami M. 2020. Glomerulus-Selective Regulation of a Critical Period for Interneuron Plasticity in the Drosophila Antennal Lobe. J Neurosci 40:5549–5560. doi:10.1523/jneurosci.2192-19.2020

Das S, Sadanandappa MK, Dervan A, Larkin A, Lee JA, Sudhakaran IP, Priya R, Heidari R, Holohan EE, Pimentel A, Gandhi A, Ito K, Sanyal S, Wang JW, Rodrigues V, Ramaswami M. 2011. Plasticity of local GABAergic interneurons drives olfactory habituation. Proc Natl Acad Sci 108:E646–E654. doi:10.1073/pnas.1106411108 Devaud J, Acebes A, Ramaswami M, Ferrús A. 2003. Structural and functional changes in the olfactory pathway of adult Drosophila take place at a critical age. J Neurobiol 56:13–23. doi:10.1002/neu.10215

Devaud J-M, Acebes A, Ferrus A. 2001. Odor Exposure Causes Central Adaptation and ́Morphological Changes in Selected Olfactory Glomeruli in Drosophila. J Neurosci 21:6274–6282. doi:10.1523/jneurosci.21-16-06274.2001

Furusawa K, Ishii K, Tsuji M, Tokumitsu N, Hasegawa E, Emoto K. 2023. Presynaptic Ube3a E3 ligase promotes synapse elimination through down-regulation of BMP signaling. Science 381:1197–1205. doi:10.1126/science.ade8978

Mosca TJ, Luginbuhl DJ, Wang IE, Luo L. 2017. Presynaptic LRP4 promotes synapse number and function of excitatory CNS neurons. eLife 6:e27347. doi:10.7554/elife.27347

Nelson N, Vita DJ, Broadie K. 2024. Experience-dependent glial pruning of synaptic glomeruli during the critical period. Sci Rep 14:9110. doi:10.1038/s41598-024-59942-3

Urwyler O, Izadifar A, Vandenbogaerde S, Sachse S, Misbaer A, Schmucker D. 2019. Branch-restricted localization of phosphatase Prl-1 specifies axonal synaptogenesis domains. Science 364. doi:10.1126/science.aau9952

Wilson RI, Laurent G. 2005. Role of GABAergic Inhibition in Shaping Odor-Evoked Spatiotemporal Patterns in the Drosophila Antennal Lobe. J Neurosci 25:9069–9079.

doi:10.1523/jneurosci.2070-05.2005

Author response:

We thank the reviewers and the editor for the detailed and constructive feedback provided. We look forward to submitting a revised version of the manuscript that addresses their comments. We acknowledge that further clarification is needed about the novelty brought by our experimental setup and model in comparison to previous studies using different methodologies. We also acknowledge that more details can be included about the calibration steps and sensitivity of the model parameters. Below we detail the planned changes for the revised version regarding the points raised by the reviewers.

Reviewer #1 (Public review):

- The authors then claim that the fragmentation of aggregates due to fluid flows occurs through erosion of small pieces. Because their experimental setup does not allow them to explicitly observe this process (for example, by watching one aggregate break into pieces), they implement an idealized model to show that the nature of the changes to the size histogram agrees with an erosion process. However, in Figure 2C there is a noticeable gap between their experiment and the prediction of their model. Additionally, in a similar experiment shown in Figure S6, the experiment cannot distinguish between an idealized erosion model and an alternative, an idealized binary fission model where aggregates split into equal halves. For these reasons, this claim is weakened.

The two idealized models of fragment distribution, namely erosion and binary fission, lead to distinguishable final size distributions. We believe that our experiments support the hypothesis of the erosion mechanism. Please note that Figure 2 is concerned with the fragmentation of large colonies, whereas Figure 3 and associated Figure S6 are concerned with very small colonies of a few cells formed by aggregation of single-cell suspension. Indeed, for very small colonies of a few cells, our experimental results cannot distinguish between a binary fission model and an erosion model (Figure S6).

The situation is very different for large colonies. To address the reviewer’s concern, we will add a new figure in the Supplementary Information (SI), similar to our Figure 2C, where we will compare the erosion model with a binary fission model for large colonies fragmented under ε = 5.8 m²/s³. We already did this exercise. The results in this new supplementary figure will show that the idealized binary fission model (i.e., where every fracture event produces exactly two fragments) does not capture the experimental fragmentation behaviour of large colonies. In contrast, the idealized erosion model provides a much better prediction of the experimental results, within the experimental uncertainty and variability in colony strength, and has the notable advantage of a straightforward computational implementation.

- The fourth major result of the manuscript is displayed in Equation 8 and Figure 5, where the authors derive an expression for the ratio between the rate of increase of a colony due to aggregation vs. the rate due to cell division. They then plot this line on a phase map, altering two physical parameters (concentration and fluid turbulence) to show under what conditions aggregation vs. cell division are more important for group formation. Because these results are derived from relatively simple biophysical considerations, they have the potential to be quite powerful and useful and represent a significant conceptual advance. However, there is a region of this phase map that the authors have left untested experimentally. The lowest energy dissipation rate that the authors tested in their experiment seemed to be \dot{epsilon}~1e-2 [m^2/s^3], and the highest particle concentration they tested was 5e-4, which means that the authors never tested Zone II of their phase map. Since this seems to be an important zone for toxic blooms (i.e. the "scum formation" zone), it seems the authors have missed an important opportunity to investigate this regime of high particle concentrations and relatively weak turbulent mixing.

We agree with the reviewer that Zone (II) of Figure 5 is of great importance to dense bloom formation under wind mixing and that this parameter range was not covered by our experiments using a cone-and-plate shear flow. The measuring range of our device was motivated by engineering applications such artificial mixing of eutrophic lakes using bubble plumes, as well as preliminary experiments which demonstrated that high levels of dissipation rate were required to achieve fragmentation. The dissipation rates of our cone-and-plate experiments capture Zones (III) and (IV) and the higher end of Zone (I). However, the cone-and-plate experiments are less suitable for the lower dissipation rates of Zone (II), as indicated by the red bars in Figure 5, due to the accumulation of colonies in stagnation points.

Instead, in our revision we will more extensively discuss recent results published in the literature for evidence of aggregation-dominance at Zone (II). The experimental studies of Wu et al. (2019) and Wu et al. (2024) (full citation below) investigated the formation of Microcystis surface scum layers at high colony concentrations (high biovolume fraction) in wind-mixed mesocosms. These studies identified aggregation of colonies at rates faster than cell division, while the stable colony size decreased with mixing rate.  The parameter range of these studies fall within Zone II, and their experimental results agree with our model predictions. We will include in the reviewed version these references and a detailed discussion elucidating the parameter range covered in our experiments and the findings of other studies.

Wu, X., Noss, C., Liu, L., & Lorke, A. (2019). Effects of small-scale turbulence at the air-water interface on Microcystis surface scum formation. Water Research, 167, 115091.

Wu, H., Wu, X., Rovelli, L., & Lorke, A. (2024). Dynamics of Microcystis surface scum formation under different wind conditions: the role of hydrodynamic processes at the air-water interface. Frontiers in Plant Science, 15, 1370874.

Other items that could use more clarity:

- The authors rely heavily on size distributions to make the claims of their paper. Yet, how they generated those size distributions is not clearly shown in the text. Of primary concern, the authors used a correction function (Equation S1) to estimate the counts of different size classes in their image analysis pipeline. Yet, it is unclear how well this correction function actually performs, what kinds of errors it might produce, and how well it mapped to the calibration dataset the authors used to find the fit parameters.

We agree with the reviewer that more details of the calibration processes should be included. We will include in the revised version of the SI more details of the calibration steps and direct comparison of raw and corrected histograms of the size distribution and its associated uncertainty.

- Second, in their models they use a fractal dimension to estimate the number of cells in the group from the group radius, but the agreement between this fractal dimension fit and the data is not shown, so it is not clear how good an approximation this fractal dimension provides. This is especially important for their later derivation of the "aggregation-to-cell division" ratio (Equation 8)

We agree with the reviewer that more details on the estimation of fractal dimension are needed. The revised version of the SI will include the estimation procedure, the number of colonies analysed, and the associated uncertainty.

Reviewer #2 (Public review)

- Especially the introduction seems to imply that shear force is a very important parameter controlling colony formation. However, if one looks at the results this effect is overall rather modest, especially considering the shear forces that these bacterial colonies may experience in lakes. The main conclusion seems that not shear but bacterial adhesion is the most important factor in determining colony size. As the importance of adhesion had been described elsewhere, it is not clear what this study reveals about cyanobacterial colonies that was not known before.

As we explain in the Introduction, it is a major open question whether cyanobacterial colonies are formed mainly by cell division (after which the dividing cells remain attached to each other by the EPS layer) or mainly by the aggregation of independent cells & colonies. See for example the highly cited review of Xiao & Reynolds 2018 (our ref 17), and references therein. This question has not been resolved and is investigated in our study. We would like to emphasize several key findings that our study reveals about the mechanical behaviour of cyanobacterial colonies under flow:

(i) Quantification of mechanical strength in cyanobacterial colonies: Our results demonstrate the high mechanical strength of cyanobacterial colonies (much higher than previously thought in references 32 and 39 of the manuscript), as evidenced by the requirement of very high shear rates to achieve fragmentation. To this end, our study highlights their resilience against naturally occurring flows and bridges the gap between theoretical assumptions about colony strength and experimentally measured mechanical properties.

(ii) Validation of a hypothesis regarding colony formation: Using a fluid-mechanical approach, we confirm the findings of recent genetic studies (references 25 and 64 of the manuscript) which indicated that colony formation of cyanobacteria under natural conditions occurs predominantly via cell division rather than via the aggregation of individual cells. Only in very dense blooms and surface scums, colony formation by the aggregation of smaller colonies likely plays a role.

(iii) Practical guidelines for cyanobacterial bloom control: Our findings provide valuable insights into the design of artificial mixing systems that are used to suppress surface blooms of buoyant cyanobacteria in lakes. In these lake applications, in which we have been involved, the aim of the mixing is to disperse the colonies over the water column so that they cannot form a surface layer (i.e., the mixing intensity should overcome the flotation velocity of the colonies), which takes away the competitive advantage of buoyant cyanobacteria over nonbuoyant phytoplankton species. However, it has always been an open question whether the high shear of artificial mixing would cause colony fragmentation. An understanding of changes in colony size is relevant for the design of artificial mixing, because smaller colonies have a lower flotation velocity. Our results show that the dissipation rates that are generated by artificial mixing are sufficient to prevent aggregation of large colonies, but not high enough to induce fragmentation of division-formed colonies.

In the revised version of the manuscript, we will improve the writing to better clarify these three novel insights obtained from our study.

- The agreement between model and experiments is impressive, but the role of the fit parameters in achieving this agreement needs to be further clarified.

The influence of the fit parameters (namely the stickiness α1 and the pairs of colony strength parameters S1,q1,S2,q2) is discussed in the sections “DYNAMICAL CHANGES IN COLONY SIZE MODELED BY A TWO-CATEGORY DISTRIBUTION” and “MATERIALS AND METHODS.” We kept the discussion concise to maintain readability. However, we agree with the reviewer that additional details about the importance of the fit parameters and the sensitivity of the results to these parameters could be beneficial. In the revised version of the SI, we will include a more detailed discussion of the fit parameters.

- The article may not be very accessible for readers with a biology background. Overall, the presentation of the material can be improved by better describing their new method.

We apologize for the limited readability of the description of the experimental setup and model used. In the revised version of the manuscript, we aim to expand the description of the new methods presented here for a broader audience of biology.

Author response:

We thank all the reviewers for their insightful comments on this work.

Response to Reviewer #1:

We greatly appreciate your comments on the general reliability and significance of our work. We fully agree that it would have been ideal to have additional evidence related to the role of PEBP1 in HRI activation. Unfortunately, we have not been able to find phospho-HRI antibodies that work reliably. The literature seems to agree with this as a band shift using total-HRI antibodies is usually used to study HRI activation. However, with the cell lines showing the most robust effect with PEBP1 knockout or knockdown, we are yet to convince ourselves with the band shifts we see. This could be addressed by optimizing phos-tag gels although these gels can be a bit tricky with complex samples such as cell lysates which contain many phosphoproteins.

To address the interaction between PEBP1 and eIF2alpha more rigorously we were inspired by the insights you and reviewer #2 provided. While we are unable to do further experiments, we now think it would indeed be possible to do this with either using the purified proteins and/or CETSA WB. These experiments could also provide further evidence for the role of PEBP1 phosphorylation. Although phosphorylation of PEBP1 at S153 has been implicated as being important for other functions of PEBP1, we are not sure about its role here. It may indeed have little relevance for ISR signalling.

For the in vitro thermal shift assay, we have performed two independent experiments. While it appears that there is a slight destabilization of PEBP1 by oligomycin, the ultimate conclusion of this experiment remains incomplete as there could be alternative explanations despite the apparent simplicity of the assay due the fluorescence background by oligomycin only. We now provide a lysate based CETSA analysis which does not display the same PEBP1 stabilization as the intact cell experiment. As for the signal saturation in ATF4-luciferase reporter assay, this is a valid point.

Response to Reviewer #2:

We strongly agree that CETSA has a lot of potential to inform us about cellular state changes and this was indeed the starting point for this project. We apologize for being (too) brief with the explanations of the TPP/MS-CETSA approach and we have now added a bit more detail. With regard to the cut-offs used for the mass spectrometry analysis, you are absolutely right that we did not establish a stringent cut-off that would show the specificity of each drug treatment. Our take on the data was that using the p values (and ignoring the fold-changes) of individual protein changes as in Fig 1D, we can see that mitochondrial perturbations display a coordinated response. We now realize that the downside of this representation is that it obscures the largest and specific drug effects. As mentioned in the response to Reviewer #1, we now also think that it would be possible to obtain more evidence for the potential interaction between PEBP1 and eIF2alpha using CETSA-based assays.

Response to Reviewer #3:

Thank you for your assessment, we agree that this manuscript would have been made much stronger by having clearer mechanistic insights. As mentioned in the responses to other reviewers above, we aim to address this limitation in part by looking at the putative interaction between PEBP1 and eIF2alpha with orthogonal approaches. However, we do realize that analysis of protein-protein interactions can be notoriously challenging due to false negative and false positive findings. As with any scientific endeavor, we will keep in mind alternative explanations to the observations, which could eventually provide that cohesive model explaining how precisely PEBP1, directly or indirectly, influences ISR signalling.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors): 

The data overall are very solid, and I would only recommend the following minor changes: 

(1) Line 187 and line 268: there is perhaps a trend towards slightly increased ATF4-luc reporter with PEBP1-S153D, but it is not statistically significant, so I would tone down the wording here. 

We now modified this part to "This data is consistent with the modest increase…" .

(2) The recently discovered SIFI complex (Haakonsen 2024, https://doi.org/10.1038/s41586023-06985-7) regulates both HRI and DELE1 through bifunctional localization/degron motifs. It seems like PEBP1 also contains such a motif, which suggests a potential mechanism for enrichment near mitochondria, perhaps even in response to stress. Maybe the authors could further speculate on this in the discussion. 

While working on the manuscript, we considered the possibility that PEBP1 function could be related to SIFI complex and concluded that here is a critical difference: while  SIFI specifically acts to turn off stress response signalling, loss of PEBP1 prevents eIF2alpha phosphorylation. We did not however consider that PEBP1 could have a localization/degron motif. Motif analysis by deepmito (busca.biocomp.unibo.it) and similar tools did not identify any conventional mitochondrial targeting signal although we acknowledge that PEBP1 has a terminal alpha-helix which was identified for SIFI complex recognition. We are not sure why you think PEBP1 contains such a motif and therefore are hesitant to speculate on this further in the manuscript.

(3) Line 358: references 50 and 45 are identical. 

Thank you for spotting this. Corrected now. 

(4) Figure S1D: it looks like Oligomycin has a significant background fluorescence, which makes interpretation of these graphs difficult - do you have measurements of the compound alone that can be used to subtract this background from the data? Based on the Tm I would say it does stabilize recombinant PEBP1, and there is no quantification of the variance across the 3 replicates to say there is no difference. 

You are right, this assay is problematic due to the background fluorescence. The measurements with oligomycin only and subtracting this background results in slightly negative values and nonsensical thermal shift curves. We now additionally show quantification from two different experiments (unfortunately we ran out of reagents for further experiments), and this quantification shows that if anything, oligomycin causes mild destabilization of recombinant PEBP1. We also used lysate CETSA assay which does not show thermal stabilization of PEBP1 by oligomycin, ruling out a direct effect. We attempted to use ferrostatin1 as a positive control as it may bind PEBP1-ALOX protein complex, and it appeared to show marginal stabilization of PEBP1. 

Reviewer #2 (Recommendations for the authors): 

I have a few comments for the authors to address: 

(1) The MS-CETSA experiment is quite briefly described and this could be expanded somewhat. Not clear if multiple biological replicates are used. Is there any cutoff in data analysis based on fold change size (which correlated to the significance of cellular effects), etc? As expected from only one early timepoint (see eg PMID: 38328090), there appear to be a limited number of significant shifts over the background (as judged from Figure S1A). In the Excel result file, however (if I read it right) there are large numbers of proteins that are assigned as stabilized or destabilized. This might be to mark the direction of potential shifts, but considering that most of these are likely not hits, this labeling could give a false impression. Could be good to revisit this and have a column for what could be considered significant hits, where a fold change cutoff could help in selecting the most biologically relevant hits. This would allow Figure 1D to be made crisper when it likely dramatically overestimates the overlap between significant CETSA shifts for these drugs.  

Fair point, while we focused more on PEBP1, it is important to have sufficient description of the methods. We used duplicate samples for the MS, which is probably the most important point which was absent from the original submission as is now added to the methods. We also added slightly more description on the data analysis. While the AID method does not explicitly use log2 fold changes, it does consider the relative abundance of proteins under different temperature fractions. Since the Tm (melting temperature) for each protein can be at any temperature, we felt that if would be complicated to compare fractions where the protein stability is changed the most and even more so if we consider both significance and log2FC. Therefore, we used this multivariate approach which indicates the proteins with most likely changes across the range of temperatures. To acknowledge that most of the statistically significant changes are not the much over the background as you correctly pointed out, we now add to the main text that “However, most of these changes are relatively small. To focus our analysis on the most significant and biologically relevant changes…” We also agree that it may be confusing that the AID output reports de/stabilization direction for all proteins. In general, we are not big fans of cutoffs as these are always arbitrary, but with multivariate p value of 0.1 it becomes clear that there are only a relatively small number of hits with larger changes. We have now added to the guide in the data sheet that "Primarily, use the adjusted p value of the log10 Multivariate normal pvalue for selecting the overall statistically significant hits (p<0.05 equals  -1.30 or smaller; p<0.01 equals  -2 or smaller)". We have also added to the guide part of the table that “Note that this prediction does not consider whether the change is significant or not, it only shows the direction of change”

(2) On page 4 the authors state "We reasoned that thermal stability of proteins might be particularly interesting in the context of mitochondrial metabolism as temperature-sensitive fluorescent probes suggest that mitochondrial temperature in metabolically active cells is close to 50{degree sign}C". I don't see the relevance of this statement as an argument for using TPP/CETSA. When this is also not further addressed in the work, it could be deleted.

Deleted. We agree, while this is an interesting point, it is not that relevant in this paper. 

(3) To exclude direct drug binding to PEBP1, a thermofluor experiment is performed (Fig S1D). However, the experiment gives a high background at the lower temperatures and it could be argued that this is due to the flouroprobe binding to a hydrophobic pocket of the protein, and that oligomycin at higher concentrations competes with this binding, attenuating fluorescence. These are complex experiments and there could be other explanations, but the authors should address this. An alternative means to provide support for non-binding would be a lysate CETSA experiment, with very short (1-3 minutes) drug exposure before heating. This would typically give a shift when the protein is indicated to be CETSA responsive as in this case. 

Agree. However, we don't have good means to perform the thermofluor experiments to rule out alternative explanations. What we can say is (as discussed above for reviewer #1, point 4) that quantification from two different experiments shows that oligomycin is does not thermally stabilizing recombinant PEBP1. To complement this conclusion, we used lysate CETSA assay which does not show thermal stabilization of PEBP1 by oligomycin. In this assay we attempted to use ferrostatin1 as a positive control as it may bind PEBP1-ALOX protein complex, and it appeared to show marginal stabilization of PEBP1. But since we lack a robust positive control for these assays, some doubt will inevitably remain.

(4) The authors appear to have missed that there is already a MS-CETSA study in the literature on oligomycin, from Sun et al (PMID: 30925293). Although this data is from a different cell line and at a slightly longer drug treatment and is primarily used to access intracellular effects of decreased ATP levels induced by oligomycin, the authors should refer to this data and maybe address similarities if any.  

Apologies for the oversight, the oligomycin data from this paper eluded us at it was mainly presented in the supplementary data. We compared the two datasets and find found some overlap despite the differences in the experimental details. Both datasets share translational components (e.g. EIF6 and ribosomal proteins), but most notably our other top hit BANF1 which we mentioned in the main text was also identified by Sun et al. We have updated the manuscript text as "Other proteins affected by oligomycin included BANF1, which binds DNA in an ATP dependent manner [16], and has also identified as an oligomycin stabilized protein in a previous MS-CETA experiment [23]", citing the Sun et al paper.   

(5) The confirmation of protein-protein interaction is notoriously prone to false positives. The authors need to use overexpression and a sensitive reporter to get positive data but collect additional data using mutants which provide further support. Typically, this would be enough to confirm an interaction in the literature, although some doubt easily lingers. When the authors already have a stringent in-cell interaction assay for PEBP1 in the CETSA thermal shift, it would be very elegant to also apply the CETSA WB assay to the overexpressed constructs and demonstrate differences in the response of oligomycin, including the mutants. I am not sure this is feasible but it should be straightforward to test. 

This is a very good suggestion. Unfortunately, due to the time constraints of the graduate students (who must write up their thesis very soon), we are not able to perform and repeat such experiments to the level of confidence that we would like.

(6) At places the story could be hard to follow, partly due to the frequent introduction of new compounds, with not always well-stated rationale. It could be useful to have a table also in the main manuscript with all the compounds used, with the rationale for their use stated. Although some of the cellular pathways addressed are shown in miniatures in figures, it could be useful to have an introduction figure for the known ISR pathways, at least in the supplement. There are also a number of typos to correct. 

We agree that there are many compounds used. We have attempted to clarify their use by adding this information into the table of used compounds in the methods and adding an overall schematic to Fig S1G and a note on line 132 "(see Figure 1-figure supplement 1G for summary of drugs used to target PEBP1 and ISR in this manuscript). We have also attempted to remove typos as far as possible.

(7) EIF2a phosphorylation in S1E does not appear to be more significant for Sodium Arsenite argued to be a positive control, than CCCP, which is argued to be negative. Maybe enough with one positive control in this figure? 

This experiment was used as a justification for our 30 min time point for the proteomics. By showing the 30 min and 4 h time points as Fig 1G and Figure 1-figure supplement 1F, our point was to demonstrate that the kinetics of phosphorylation and dephosphorylation are relevant. As you correctly pointed out, the stress response induced by sodium arsenite, but also tunicamycin is already attenuated at the 4h time point. We prefer to keep all samples to facilitate comparisons.

(8) Page 7 reference to Figure S2H, which doesn't exist. Should be S3H.  

Apologies for the mistake, now corrected to Figure 2-figure supplement 1B.

(9) Finally, although the TPP labeling of the method is used widely in the literature this is CETSA with MS detection and MS-CETSA is a better term. This is about thermal shifts of individual proteins which is a very well-established biophysical concept. In contrast, the term Thermal Proteome Profiling does not relate to any biophysical concept, or real cell biology concept, as far as I can see, and is a partly misguided term. 

We changed the term TPP into MS-CETSA, but also include the term TPP in the introduction to facilitate finding this paper by people using the TPP term.

Reviewer #3 (Recommendations for the authors): 

Major Issues 

(1) The one major issue of this work is the lack of a mechanism showing precisely how PEBP1 amplifies the mitochondrial integrated stress response. The work, as it is described, presents data suggesting PEBP1's role in the ISR but fails to present a more conclusive mechanism. The idea of mitochondrial stress causing PEBP1 to bind to eIF2a, amplifying ISR is somewhat vague. Thus, the lack of a more defined model considerably weakens the argument, as the data is largely corollary, showing KO and modulation of PEBP1 definitely has a unique effect on the ISR, however, it is not conclusive proof of what the authors claim. While KO of PEBP1 diminishes the phosphorylation of eIF2a, taken together with the binding to eIF2a, different pathways could be simultaneously activated, and it seems premature to surmise that PEBP1 is specific to mitochondrial stress. Could PEBP1 be reacting to decreased ATP? Release of a protein from the mitochondria in response to stress? Is PEBP1's primary role as a modulator of the ISR, or does it have a role in non-stress-related translation? A cohesive model would tie together these separate indirect findings and constitute a considerable discovery for the ISR field, and the mitochondrial stress field.  

Thank you for your assessment, we agree that this manuscript would have been much stronger by having clearer mechanistic insights. As with any scientific endeavor, we will keep in mind alternative explanations to the observations, which could eventually provide that cohesive model explaining how precisely PEBP1, directly or indirectly, influences ISR signalling.

(2) The data relies on the initial identification of PEBP1 thermal stabilization concomitant with mitochondrial ISR induction post-treatment of several small molecules. However, the experiment was performed using a single timepoint of 30 minutes. There was no specific rationale for the choice of this time point for the thermal proteome profiling. 

The reasoning for this was explicitly stated:  "We reasoned that treating intact cells with the drugs for only 30 min would allow us to observe rapid and direct effects related to metabolic flux and/or signaling related to mitochondrial dysfunction in the absence of major changes in protein expression levels.”

Minor Issues 

(1) In Lines 163-166 the authors state "The cells from Pebp1 KO animals displayed reduced expression of common ISR genes (Figure 2F), despite upregulation of unfolded protein response genes Ern1 (Ire1α) and Atf6 genes. This gene expression data therefore suggests that Pebp1 knockout in vivo suppresses induction of the ISR". This statement should be reassessed. While an arm of the UPR does stimulate ISR, this arm is controlled by PERK, and canonically IRE1 and ATF6 do not typically activate the ISR, thus their upregulation is likely unrelated to ISR activation and does not contribute the evidence necessary for this statement. 

Apologies for the confusion, we aimed to highlight that as there is an increase in the two UPR arms, it is more likely that ISR instead of UPR is reduced. We have now changed the statement to the following:

"The cells from Pebp1 PEBP1 KO animals displayed reduced expression of common ISR genes (Figure 2F), while there was mild upregulation of the unfolded protein response genes Ern1 (Ire1α) and Atf6 genes. This gene expression data therefore suggests that the reduced expression of common ISR genes is less likely to be mediated by changes in PERK, the third UPR arm, and more likely due to suppression of ISR by Pebp1 knockout in vivo."

(2) In Lines 169 and 170 the authors state "Western blotting indicated reduced phosphorylation of eIF2α in RPE1 cells lacking PEBP1, suggesting that PEBP1 is involved in regulating ISR signaling between mitochondria and eIF2α". This conclusion is not supported by evidence. A number of pathways could be activated in these knockout cells, and simply observing an increase in p-eIF2α after knocking out PEBP1 does not constitute an interaction, as correlation doesn't mean causation. This KO could indirectly affect the ISR, with PEBP1 having no role in the ISR. While taken together there is enough circumstantial evidence in the manuscript to suggest a role for PEBP1 in the ISR, statements such as these have to be revised so as not to overreach the conclusions that can be achieved from the data, especially with no discernible mechanism.  

We have now revised this statement by removing the conclusion and stating only the observation:  "Western blotting indicated reduced phosphorylation of eIF2α in RPE1 cells lacking PEBP1 (Fig. 3A)."

Author response:

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

comprehensiveness and rigor of the study are notable. Rarely have I reviewed a manuscript reporting the results of so many orthogonal experiments, all of which support the authors' hypotheses, and of so many excellent controls.” Reviewer 2 commented: “They have elegantly demonstrated how some mutants alter each step of processing. Together with FLIM experiments, this study provides additional evidence to support their 'stalled complex hypotheses'….This is a beautiful biochemical work. The approach is comprehensive.”

Below we respond to the relatively minor concerns of Reviewer 2, which may be included with the first version of the Reviewed Preprint.

Reviewer 2:

(1) It appears that the purified γ-secretase complex generates the same amount of Aβ40 and Aβ42, which is quite different in cellular and biochemical studies. Is there any explanation for this?  

Roughly equal production of Aβ40 and Aβ42 is a phenomenon seen with purified enzyme assays, and the reason for this has not been identified. However, we suggest that what is meaningful in our studies is the relative difference between the effects of FAD-mutant vs. WT PSEN1 on each proteolytic processing step. All FAD mutations are deficient in multiple cleavage steps in γsecretase processing of APP substrate, and these deficiencies correlate with stabilization of E-S complexes.

(2) It has been reported the Aβ production lines from Aβ49 and Aβ48 can be crossed with various combinations (PMID: 23291095 and PMID: 38843321). How does the production line crossing impact the interpretation of this work?  

In the cited reports, such crossover was observed when using synthetic Aβ intermediates as substrate. In PMID 2391095 (Okochi M et al, Cell Rep, 2013), Aβ43 is primarily converted to Aβ40, but also to some extent to Aβ38. In PMID: 38843321 (Guo X et al, Science, 2024), Aβ48 is ultimately converted to Aβ42, but also to a minor degree to Aβ40. We have likewise reported such product line “crossover” with synthetic Aβ intermediates (PMID: 25239621; Fernandez MA et al, JBC, 2014). However, when using APP C99-based substrate, we did not detect any noncanonical tri- and tetrapeptide co-products of Aβ trimming events in the LC-MS/MS analyses (PMID: 33450230; Devkota S et al, JBC, 2021). In the original report on identification of the small peptide coproducts for C99 processing by γ-secretase using LC-MS/MS (PMID: 19828817; Takami M et al, J Neurosci, 2009), only very low levels of noncanonical peptides were observed. In the present study, we did not search for such noncanonical trimming coproducts, so we cannot rule out some degree of product line crossover.

(3) In Figure 5, did the authors look at the protein levels of PS1 mutations and C99-720, as well as secreted Aβ species? Do the different amounts of PS1 full-length and PS1-NTF/CTF influence FILM results?  

FLIM results depend on the degree that C99 and long Aβ intermediates are bound to γ-secretase compared to unbound C99 and Aβ. The 6E10-Alexa 488 lifetime is significantly decreased by FAD mutations compared to WT PSEN1 (Fig. 5). However, the observed decrease in lifetime with the PSEN1 FAD mutants might also be due to lower levels of C99-720 expression or higher levels of PSEN1 CTF (i.e., mature γ-secretase complexes). We checked the C99-720 fluorescence intensities in the FLIM experiments and found that C99-720 intensities are not significantly different between cells transfected with WT and those with FAD PSEN1. Furthermore, Western blot analysis shows that levels of C99-720 are not significantly low and those of PSEN1 CTF are not high in FAD PSEN1 compared to WT PSEN1 expressing cells. Although PSEN1 CTF levels trend low for PSEN1 F386S, this mutant resulted in decreased FLIM only in Aβ-rich regions. Thus, the reduced FLIM apparently reflects effects of FAD mutation on E-S complex stability. Levels of full-length PSEN1 were also determined and found not to correlated with FLIM effects, although full-length PSEN1 represents protein not incorporated into full active γ-secretase complexes and therefore does not interact with C99-720.

(4) It is interesting that both Aβ40 and Aβ42 Elisa kits detect Aβ43. Have the authors tested other kits in the market? It might change the interpretation of some published work.  

We have not tested other ELISA kits. Considering our findings, it would be a good idea for other investigators to test whatever ELISAs they use for specificity vis-à-vis Aβ43.

Author response:

Reviewer #2 (Public Review): 

Comment 1: In terms of the biological significance of this interaction, it would be good to examine (via co-immunoprecipitation) whether the CEP89/NCS-1/C3ORF14 interaction takes place upon serum starvation. Does the complex change? 

NCS1 centriolar localization requires CEP89 as no NCS1 localization was observed in CEP89 knockout cells (Figure 2L; Figure 2-figure supplement 2B). Both CEP89 and NCS1 centriolar localization were observed (Figure 2C; Figure 1D of the PMID: 36711481) in cells grown in serum containing media, although their localization was further enhanced in serum starved cells. From these results, we predict that CEP89 and NCS1 can interact and colocalize in both serum-fed and serum-depleted condition. We think it may not be easy to assess the change in interaction with the co-immunoprecipitation assay, as interactions occur in a test tube, which may not reflect the binding condition inside the cells.

Comment 2: Also, for the subdistal appendage localization of NCS-1 and C3ORF14, would this also change upon serum starvation? 

We agree that it would be interesting to see whether the subdistal appendage localization changes upon serum starvation, as NCS1 may capture the ciliary vesicle at the subdistal appendages as we discussed. However, the loss of the subdistal appendage protein, CEP128, blocks subdistal appendage localization of CEP89 [PMID: 32242819] without affecting cilium formation [PMID: 27818179]. This suggests that the subdistal appendage localization of NCS1 or C3ORF14 is likely dispensable for cilium formation.

Comment 3: For the ciliation results and the recruitment of IFT88 in CEP89 knockout cell lines, this contradicts previous work from Tanos et al (PMID: 23348840), as well as Hou et al (PMID: 36669498). A parallel comparison using siRNA, a transient knockout system, or a degron system would help understand this. A similar point goes for Figure 4, where the effect on ciliogenesis is minimal in knockout cells, but acute siRNA has been shown to have a stronger phenotype. 

Hou et al. [PMID: 36669498] investigated the role of distal appendage proteins, CEP164, CEP89, and FBF1 in the ciliated chordotonal organ of Drosophila melanogaster by generating knockout Drosophila strains. The results were markedly different from what was observed in mammalian cells. Notably, CEP164 is not required for cilium formation, and CEP89 is required for FBF1 localization in the animal. CEP89 was required for cilium formation in the cells in the ciliated chordotonal organ, of which cilium formation is dependent on IFT machinery. They did not show if IFT centriolar recruitment is affected in the CEP89 mutant cells. These differences likely reflect the divergence of the organization of distal appendage during evolution.

The ciliation phenotype of our CEP89 knockout cells are milder than what was shown in Tanos et al [PMID: 23348840], but largely consistent with the results from Bornens group, which used siRNA to deplete CEP89 [PMID: 23789104]. Besides, NCS1 knockout cells showed very similar phenotype to the CEP89 knockout cells, and relatively acute deletion of NCS1 (14 days after infection of the lenti-virus containing sgNCS1 without single-cell cloning) displayed an almost identical ciliation defect (Figure 4B-C). Thus, we believe CEP89 is only partially required for cilium formation in RPE-hTERT cells and that the differences are more technical than definitive.

Comment 4: An elegant phenotype rescue is shown in Figure 5. An interesting question would be, how does this mutant and/or the myristoylation affect the recruitment of C3ORF14? 

NCS1 is not required for the localization of C3ORF14 (Figure 2M; Figure 2- figure supplement 2C), so we can assume that the myristoylation defective mutant does not affect C3ORF14 recruitment.

Comment 5: For the EF-hand mutants, it would be good to use control mutants, from known Ca2+ binding proteins as a control for the experiment shown. 

In the Figure 5-figure supplement 1A-C, we generated a series of EF-hand mutant of NCS1 to see if the calcium binding affects the CEP89 interaction, NCS1 localization, and cilium formation. NCS1 is only protein among the calcium binding NCS family proteins that was found as a positive hit in the mass spec data of CEP89 tandem affinity purification. Therefore, we cannot use other NCS1 family proteins as a control for CEP89 binding, NCS1 localization, and cilium formation.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Using a knock-out mutant strain, the authors tried to decipher the role of the last gene in the mycofactocin operon, mftG. They found that MftG was essential for growth in the presence of ethanol as the sole carbon source, but not for the metabolism of ethanol, evidenced by the equal production of acetaldehyde in the mutant and wild type strains when grown with ethanol (Fig 3). The phenotypic characterization of ΔmftG cells revealed a growth-arrest phenotype in ethanol, reminiscent of starvation conditions (Fig 4). Investigation of cofactor metabolism revealed that MftG was not required to maintain redox balance via NADH/NAD+, but was important for energy production (ATP) in ethanol. Since mycobacteria cannot grow via substrate-level phosphorylation alone, this pointed to a role of MftG in respiration during ethanol metabolism. The accumulation of reduced mycofactocin points to impaired cofactor cycling in the absence of MftG, which would impact the availability of reducing equivalents to feed into the electron transport chain for respiration (Fig 5). This was confirmed when looking at oxygen consumption in membrane preparations from the mutant and would type strains with reduced mycofactocin electron donors (Fig 7). The transcriptional analysis supported the starvation phenotype, as well as perturbations in energy metabolism, and may be beneficial if described prior to respiratory activity data.

The data and conclusions support the role of MftG in ethanol metabolism.

We thank the reviewer for the positive evaluation of our manuscript.

Reviewer #3 (Public review):

Summary:

The work by Graca et al. describes a GMC flavoprotein dehydrogenase (MftG) in the ethanol metabolism of mycobacteria and provides evidence that it shuttles electrons from the mycofactocin redox cofactor to the electron transport chain.

Strengths:

Overall, this study is compelling, exceptionally well designed and thoroughly conducted. An impressively diverse set of different experimental approaches is combined to pin down the role of this enzyme and scrutinize the effects of its presence or absence in mycobacteria cells growing on ethanol and other substrates. Other strengths of this work are the clear writing style and stellar data presentation in the figures, which makes it easy also for non-experts to follow the logic of the paper. Overall, this work therefore closes an important gap in our understanding of ethanol oxidation in mycobacteria, with possible implications for the future treatment of bacterial infections.

Weaknesses:

I see no major weaknesses of this work, which in my opinion leaves no doubt about the role of MftG.

We thank the reviewer for the positive evaluation of our manuscript.

Reviewer #4 (Public review):

Summary:

The manuscript by Graça et al. explores the role of MftG in the ethanol metabolism of mycobacteria. The authors hypothesise that MftG functions as a mycofactocin dehydrogenase, regenerating mycofactocin by shuttling electrons to the respiratory chain of mycobacteria. Although the study primarily uses M. smegmatis as a model microorganism, the findings have more general implications for understanding mycobacterial metabolism. Identifying the specific partner to which MftG transfers its electrons within the respiratory chain of mycobacteria would be an important next step, as pointed out by the authors.

Strengths:

The authors have used a wide range of tools to support their hypothesis, including co-occurrence analyses, gene knockout and complementation experiments, as well as biochemical assays and transcriptomics studies.

An interesting observation that the mftG deletion mutant grown on ethanol as the sole carbon source exhibited a growth defect resembling a starvation phenotype.

MftG was shown to catalyse the electron transfer from mycofactocinol to components of the respiratory chain, highlighting the flexibility and complexity of mycobacterial redox metabolism.

Weaknesses:

Could the authors elaborate more on the differences between the WT strains in Fig. 3C and 3E? in Fig. 3C, the ethanol concentration for the WT strain is similar to that of WT-mftG and ∆mftG-mftG, whereas the acetate concentration in thw WT strain differs significantly from the other two strains. How this observation relates to ethanol oxidation, as indicated on page 12.

This is a good question, and we agree with the reviewer that the sum of processes leading to the experimental observations shown in Figure 3 are not completely understood. For instance, when looking at ethanol concentrations, evaporation is a dominating effect and the situation is furthermore confounded by the fact that the rate of ethanol evaporation appears to be inversely correlated to the optical density of the samples (see Figure 3E and compare media control as well as the samples of DmftG and DmftG at OD₆₀₀ = 1). Additionally, the growth rate and thus the OD₆₀₀ of all strains monitored are different at each time point, thus further complicating the analysis. This is why we assume that the rate of ethanol oxidation is mirrored more clearly by acetate formation, at least in the early phase before 48 h (Figure 3E),i.e., before acetate consumption becomes dominant in DmftG-mftG and WT-mftG. Here, we see that the rate of acetate formation is zero for media controls, low for DmftG, but high for WT as well as DmftG-mftG and WT-mftG. The latter two strains also showed an earlier starting point of growth as well as acetate formation and the following phase of acetate depletion.

All of these observations are in line with our general statement, i.d., “Parallel to the accelerated and enhanced growth described above (Figure 3A), the overexpression strains displayed higher rates of ethanol consumption as well as an earlier onset of acetate overflow metabolism and acetate consumption (Figure 3D).” We are still convinced that this summary describes the findings well and avoids unnecessary speculation.

The authors conclude from their functional assays that MftG catalyses single-turnover reactions, likely using FAD present in the active site as an electron acceptor. While this is plausible, the current experimental set up doesn't fully support this conclusions, and the language around this claim should be softened.

This is a fair point. We revised our claim accordingly. In particular, we changed:

Page 28: we added “possibly”

Page 28 we changed “single-turnover reactions” to “reactions reminiscent of a single-turnover process”.

The authors suggest in the manuscript that the quinone pool (page 24) may act as the electron acceptor from mycofactocinol, but later in the discussion section (page 30) they propose cytochromes as the potential recipients. If the authors consider both possibilities valid, I suggest discussing both options in the manuscript.

This is true. However, no change to the manuscript is necessary, since both options were discussed on page 30.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

The authors addressing some of the original recommendations is appreciated e.g. title change. Other recommendations that were not adequately addressed would mostly improve the clarity and help comprehension for the reader, but they are at the author's discretion.

Reviewer #3 (Recommendations for the authors):

Abstract: "Here, we show that MftG enzymes strictly require mft biosynthetic genes and are found in 75% of organisms harboring these genes". I read this sentence several times and I am still somewhat confused and not sure what exactly is meant here. I suggest to rephrase, e.g., to "Here, we show that in 75% of all organisms that harbour the mft biosynthetic genes, MftG enzymes are also encoded and functionally associated with these genes" (if that was meant; also the abbreviation mft should be introduced in the abstract or otherwise the full name be used).

We thank the reviewer for the good hint. We changed the sentence to “Here, we show that MftG enzymes are almost exclusively found in genomes containing mycofactocin biosynthetic genes and are present in 75% of organisms harboring these genes”.

p.3, 2nd paragraph: "Although the role of MFT in alcohol metabolism is well established, further biological roles of mycofactocin appear to exist." Mycofactocin is once written as MFN and once in full length, which is slightly confusing. Consider rephrasing, e.g., to "...further biological roles of this cofactor appear to exist".

Thank you, we adopted the suggested change.

Fig. 1: Consider adding MftG in brackets after "mycofactocin dehydrogenase" in panel B.

Good suggestion. We added (MftG) to the figure.

Fig. 3: Legend should be corrected. The color of the signs should be teal diamond for "M. smegmatis double presence of the mftG gene" and orange upward facing triangle for "Medium with 10 g L-1 of ethanol without bacterial inoculation". Aside from the coloration, the order should ideally also be identical to the one shown in the upper right part.

Thank you for the valuable hint! We corrected the legend and unified the legends in the figure caption and figure.

p.20 : It is not exactly clear to me why "semipurified cell-free extracts from M. smegmatis ∆mftG-mftGHis6 " were used here rather than the purified enzyme. Was the purification by HisTrap columns not feasible or was the protein unstable when fully purified? In any case, it would help the reader to quickly state the reason in this section.

Indeed, the problem with M. smegmatis as an expression host was a combination of low protein yield and poor binding to Ni-NTA columns. In E. coli, poor expression, low solubility or poor binding was the issue. Unfortunately, the usage of other affinity tags resulted in either poor expression or inactive protein. We have shortly mentioned the major issues on page 21 and prefer not to focus on failed attempts too much.

p. 21: "We, therefore, concluded that MftG can indeed interact with mycofactocins as electron donors but might require complex electron acceptors, for instance, proteins present in the respiratory chain." I agree. For the future it might be worthwhile to determine the redox potential of MftG, which could provide hints on the natural electron acceptor.

Thank you for the suggestion. We will consider this question in our future work.

p. 23: "In M. smegmatis, cyanide is a known inhibitor of the cytochrome bc/aa3 but not of cytochrome bd (34), therefore, the decrease of oxygen consumption when MFTs were added to the membrane fractions in combination with KCN (Figure 7), revealed that MFT-induced oxygen consumption is indeed linked to mycobacterial respiration." It might be a good idea to quickly recapitulate the functions of these cytochromes here. Also, I think it should read "bc1aa3" (also correct in legend of Fig. 8 that says "bcc-aa3").

Thank you for the good observation. We changed all instances to the correct designation (bc1-aa3).

Reviewer #4 (Recommendations for the authors):

Abstract: revise the wording "MftG enzymes strictly require mft biosynthetic genes". It should be either mftG gene with the mft biosynthetic genes or MftG enzyme with the Mft biosynthetic proteins. I also suggest replacing "require" with a more appropriate term.

This was taken care of. See above.

Page 3, end of the first paragraph; does the alcohol dehydrogenase refer to Mno/Mdo?

Partially, yes, but also to other alcohol dehydrogenases.

Page 4, radical SAM; define upon first use

Good, point, we changed “radical SAM” to radical S-adenosyl methionine (rSAM)

Page 6; Rossman fold refers to the fold and not only the FAD binding pocket.

Good point. We deleted “(Rossman fold)”

Page 11; not exactly sure what this means "the growth curve of the complemented strain, which could be dysregulated in mftG expression"

By “dysregulated” expression, we mean that the expression of mftG could be higher or lower than in the WT and could follow different regulatory signals than in the wild type. Since this phenomenon is not well understood, we would like to avoid speculative discussions.

Page 11; Figures 2E and 2C should be 3E and 3C. Likewise on page 12 Figure 2D.

Thank you very much for the valuable hint. We corrected the figure numbers as suggested.

Page 12; the last Figure 3D in the page should be 3E?

Yes, good catch, we corrected the Figure number.

Page 17, KO; define upon first use.

Good suggestion, we changed both instances of “KO” to “knockout”

Page 24; revise: "for instance. For example"

We deleted “for instance”.

Page 26; change 6.506 to 6,506

Corrected.

Page 23; "In M. smegmatis, cyanide is a known inhibitor ..." is too long and not easy to understand/follow.

Good suggestion. We simplified the sentence to “Therefore, the decrease of oxygen consumption in the presence of KCN (Figure 7) revealed…”

Page 29; "single-turnover reactions could be observed". There are no experiments to support this statement, except the results shown in Figure 7F. I suggest softening the language, as it has been done on page 21. To claim single-turnover, a proper kinetic analysis would be necessary, which is not included in the current manuscript.

This is true and has been taken care of. See above.

Figure 1; Indicate mycofactocin dehydrogenase as MftG

Done.

Figure 5A; what is the significance of comparing ∆mftG glucose with WT ethanol?

We agree, that, although the difference of the two columns is significant, this does not have any relevant meaning. Therefore, we removed the bracket with p-value in Panel A.

Make HdB-Tyl/HdB-tyloxapol usage consistent throughout the document. Likewise, re the usage of mycobacteria/Mycobacteria/Mycobacteria

Thank you for the valuable hint, we unified the usage throughout the document

Author response:

Reviewer #1:

Summary:

Beyond what is stated in the title of this paper, not much needs to be summarized. eIF2A in HeLa cells promotes translation initiation of neither the main ORFs nor short uORFs under any of the conditions tested.

Strengths:

Very comprehensive, in fact, given the huge amount of purely negative data, an admirably comprehensive and well-executed analysis of the factor of interest.

Weaknesses:

The study is limited to the HeLa cell line, focusing primarily on KO of eIF2A and neglecting the opposite scenario, higher eIF2A expression which could potentially result in an increase in non-canonical initiation events.

We thank the reviewer for the positive evaluation. As suggested by the reviewer in the detailed recommendations, we will clarify in the title, abstract and text that our conclusions are limited to HeLa cells. Furthermore, as suggested we will test the effect of eIF2A overexpression on the luciferase reporter constructs, and will upload a revised manuscript.

Reviewer #2:

Summary

Roiuk et al describe a work in which they have investigated the role of eIF2A in translation initiation in mammals without much success. Thus, the manuscript focuses on negative results. Further, the results, while original, are generally not novel, but confirmatory, since related claims have been made before independently in different systems with Haikwad et al study recently published in eLife being the most relevant.

Despite this, we find this work highly important. This is because of a massive wealth of unreliable information and speculations regarding eIF2A role in translation arising from series of artifacts that began at the moment of eIF2A discovery. This, in combination with its misfortunate naming (eIF2A is often mixed up with alpha subunit of eIF2, eIF2S1) has generated a widespread confusion among researchers who are not experts in eukaryotic translation initiation. Given this, it is not only justifiable but critical to make independent efforts to clear up this confusion and I very much appreciate the authors' efforts in this regard.

Strengths

The experimental investigation described in this manuscript is thorough, appropriate and convincing.

Weaknesses

However, we are not entirely satisfied with the presentation of this work which we think should be improved.

We thank the reviewer for the positive evaluation. We will revise the manuscript according to the reviewer's suggestions made in the detailed recommendations.

Reviewer #3:

Summary:

This is a valuable study providing solid evidence that the putative non-canonical initiation factor eIF2A has little or no role in the translation of any expressed mRNAs in cultured human (primarily HeLa) cells. Previous studies have implicated eIF2A in GTP-independent recruitment of initiator tRNA to the small (40S) ribosomal subunit, a function analogous to canonical initiation factor eIF2, and in supporting initiation on mRNAs that do not require scanning to select the AUG codon or that contain near-cognate start codons, especially upstream ORFs with non-AUG start codons, and may use the cognate elongator tRNA for initiation. Moreover, the detected functions for eIF2A were limited to, or enhanced by, stress conditions where canonical eIF2 is phosphorylated and inactivated, suggesting that eIF2A provides a back-up function for eIF2 in such stress conditions. CRISPR gene editing was used to construct two different knock-out cell lines that were compared to the parental cell line in a large battery of assays for bulk or gene-specific translation in both unstressed conditions and when cells were treated with inhibitors that induce eIF2 phosphorylation. None of these assays identified any effects of eIF2A KO on translation in unstressed or stressed cells, indicating little or no role for eIF2A as a back-up to eIF2 and in translation initiation at near-cognate start codons, in these cultured cells.

The study is very thorough and generally well executed, examining bulk translation by puromycin labeling and polysome analysis and translational efficiencies of all expressed mRNAs by ribosome profiling, with extensive utilization of reporters equipped with the 5'UTRs of many different native transcripts to follow up on the limited number of genes whose transcripts showed significant differences in translational efficiencies (TEs) in the profiling experiments. They also looked for differences in translation of uORFs in the profiling data and examined reporters of uORF-containing mRNAs known to be translationally regulated by their uORFs in response to stress, going so far as to monitor peptide production from a uORF itself. The high precision and reproducibility of the replicate measurements instil strong confidence that the myriad of negative results they obtained reflects the lack of eIF2A function in these cells rather than data that would be too noisy to detect small effects on the eIF2A mutations. They also tested and found no evidence for a recent claim that eIF2A localizes to the cytoplasm in stress and exerts a global inhibition of translation. Given the numerous papers that have been published reporting functions of eIF2A in specific and general translational control, this study is important in providing abundant, high-quality data to the contrary, at least in these cultured cells.

Strengths:

The paper employed two CRISPR knock-out cell lines and subjected them to a combination of high-quality ribosome profiling experiments, interrogating both main coding sequences and uORFs throughout the translatome, which was complemented by extensive reporter analysis, and cell imaging in cells both unstressed and subjected to conditions of eIF2 phosphorylation, all in an effort to test previous conclusions about eIF2A functioning as an alternative to eIF2.

Weaknesses:

There is some question about whether their induction of eIF2 phosphorylation using tunicamycin was extensive enough to state forcefully that eIF2A has little or no role in the translatome when eIF2 function is strongly impaired. Also, similar conclusions regarding the minimal role of eIF2A were reached previously for a different human cell line from a study that also enlisted ribosome profiling under conditions of extensive eIF2 phosphorylation; although that study lacked the extensive use of reporters to confirm or refute the identification by ribosome profiling of a small group of mRNAs regulated by eIF2A during stress.

We thank the reviewer for the positive evaluation. We will revise the manuscript according to the recommendations made in the detailed recommendations. Regarding the two points mentioned here:

(1) the reason eIF2alpha phosphorylation does not increase appreciably is because unfortunately the antibody is very poor. The fact that the Integrated Stress Response (ISR) is induced by our treatment can be seen, for instance, by the fact that ATF4 protein levels increase strongly (in the very same samples where eIF2alpha phosphorylation does not increase much, in Suppl. Fig. 5E). We will strengthen the conclusion that the ISR is indeed activated with additional experiments/data as suggested by the reviewer.

(2) We agree that our results are in line with results from the previous study mentioned by the reviewer, so we will revise the manuscript to mention this other study more extensively in the discussion.

Author response:

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

Public Review:

The overall goal of this manuscript is to understand how Notch signaling is activated in specific regions of the endocardium, including the OFT and AVC, that undergo EMT to form the endocardial cushions. Using dofetilide to transiently block circulation in E9.5 mice, the authors show that Notch receptor cleavage still occurs in the valve-forming regions due to mechanical sheer stress as Notch ligand expression and oxygen levels are unaffected. The authors go on to show that changes in lipid membrane structure activate mTOR signaling, which causes phosphorylation of PKC and Notch receptor cleavage.

The strengths of the manuscript include the dual pharmacological and genetic approaches to block blood flow in the mouse, the inclusion of many controls including those for hypoxia, the quality of the imaging, and the clarity of the text. However, several weaknesses were noted surrounding the main claims where the supporting data are incomplete.

PKC - Notch1 activation:

(1) Does deletion of Prkce and Prkch affect blood flow, and if so, might that be suppressing Notch1 activation indirectly?

To address this concern, we performed echocardiography of Prkce^+/-;Prkch^+/-, Prkce^-/-;Prkch^+/-, and Prkce^+/-;Prkch^-/- mouse hearts (Figure 3-supplement figure 2D), showing no significant effect in heartbeat and blood flow. (Line 308)

(2) It would be helpful to visualize the expression of prkce and prkch by in situ hybridization in E9.5 embryos.

We now added immunofluorescence staining results for both PKCE and PKCH as shown in Figure 3-supplement figure 2B. In E9.5 embryonic heart, PKCH is mainly expressed in the endocardium overlying AV canal and the base of trabeculae, overlapping with the expression pattern of NICD and pPKC^Ser660. PKCE is expressed in both endocardium and myocardium. In the endocardium, PKCE is mainly expressed in the endocardium overlying AV canal (Line312-314)

(2) PMA experiments: Line 223-224: A major concern is related to the conclusion that "blood flow activates Notch in the cushion endocardium via the mTORC2-PKC signaling pathway". To make that claim, the authors show that a pharmacological activation with a potent PKC activator, PMA, rescues NICD levels in the AVC in dofetilide-treated embryos. This claim would also need proof that a lack of blood flow alters the activity of mTORC2 to phosphorylate the targets of PKC phosphorylation. Also, this observation does not explain the link between PKC activity and Notch activation.

Both AKT Ser473 and PKC Ser660 are well characterized phosphorylation sites regulated by mTORC2 (Baffi TR et. al, mTORC2 controls the activity of PKC and Akt by phosphorylating a conserved TOR interaction motif. Sci Signal. 2021;14.). pAKT^Ser473 is widely used as an indicator of mTORC2 activity. Therefore, the reduced staining intensity of pAKT^Ser473 and pPKC^Ser660 observed in the dofetilide treated embryos should reflect the reduced activity of their common upstream activator mTORC2. This information is provided in Line 317-321.

As PMA is a well-characterized specific activator of PKC, we believe the rescue of NICD by PMA could explain the link between PKC activity and Notch activation.

(3) In addition, the authors hypothesise that shear stress lies upstream of PKC and Notch activation, and that because shear stress is highest at the valve-forming regions, PKC and Notch activity is localised to the valve-forming regions. Since PMA treatment affects the entire endocardium which expresses Notch1, NICD should be seen in areas outside of the AVC in the PMA+dofetilide condition. Please clarify.

As shown in Figure 3C and Figure 3-supplement figure 2B, pPKC, PKCH and PKCE expression are all confined in the AVC region. This explains PMA activates NICD specifically in the valve-forming region. This information is added in Line 312-314.

Lipid Membrane:

(1) It is not clear how the authors think that the addition of cholesterol changes the lipid membrane structure or alters Cav-1 distribution. Can this be addressed? Does adding cholesterol make the membrane more stiff? Does increased stiffness result from higher shear stress?

We do not know how exactly addition of cholesterol alters membrane structure and influence mTORC2-PKC-Notch signaling. As cholesterol is an important component of lipid raft and caveolae, it is possible that enrichment of cholesterol might alter the membrane structure to make the lipid raft structure less dependent on sheer stress. This hypothesis need to be tested in further in vitro studies. This information is added to Line 433-436.

(2) The loss of blood flow apparently affects Cav1 membrane localization and causes a redistribution from the luminal compartment to lateral cell adhesion sites. Cholesterol treatment of dofetilide-treated hearts (lacking blood flow) rescued Cav1 localization to luminal membrane microdomains and rescued NICD expression. It remains unclear how the general addition of cholesterol would result in a rescue of regionalized membrane distribution within the AVC and in high-shear stress areas.

We do not know the exact mechanism. As replied in the previous question, future cell-based work is needed to address these important questions. (Line 433-436)

(3) The authors do not show the entire heart in that rescue treatment condition (cholesterol in dofetilide-treated hearts). Also, there is no quantification of that rescue in Figure 4B. Currently, only overview images of the heart are shown but high-resolution images on a subcellular scale (such as electron microscopy) are needed to resolve and show membrane microdomains of caveolae with Cav1 distribution. This is important because Cav-1could have functions independent of caveolae.

In Figure 4C, most panels display the large part of the heart including AVC, atrium and ventricle. The images in the third column appear to be more restricted to AVC. We have now replaced these images to reveal AVC and part of the atrium and ventricle. 

The quantification has also been provided in Figure 4C. We also added a new panel of scanning EM of AVC endocardium, showing numerous membrane invaginations on the luminal surface of the endocardial cells. The size of the invaginations ranges from 50 to 100 nm, consistent with the reported size of caveolae. Dofetilide significantly reduced the number of membrane invaginations, which recovered after restore of blood flow at 5 hours post dofetilide treatment. The reduction of membrane invaginations could also be rescued by ex vivo cholesterol treatment. This information is added to Line 342-349.

Figure Legends, missing data, and clarity:

(1) The number of embryos used in each experiment is not clear in the text or figure legends. In general, figure legends are incomplete (for instance in Figure 1).

Thanks for reminding. we have now added numbers of embryos in the figure legends.

(2) Line 204: The authors refer to unpublished endocardial RNAseq data from E9.5 embryos. These data must be provided with this manuscript if it is referred to in any way in the text.

The RNAseq data of PKC isoforms is now provided in Figure3-Figure supplement 2A, Line 301-302.

(3) Figure 1 shows Dll4 transcript levels, which do not necessarily correlate with protein levels. It would be important to show quantifications of these patterns as Notch/Dll4 levels are cycling and may vary with time and between different hearts.

The Dll4 immuno-staining in Figure 1B,C is indeed Dll4 protein, not transcript. The quantification is added in Figure 1—Figure supplement 1C. Line 215.

(4) Line 212-214: The authors describe cardiac cushion defects due to the loss of blood flow and refer to some quantifications that are not completely shown in Figure 3. For instance, quantifications for cushion cellularity and cardiac defects at three hours (after the start of treatment?) are missing.

The formation of the defects is a developmental process and time dependent. To address this concern, we quantified the cushion cellularity at 5 hours post dofetilide treatment and showed that cell density significantly decreased in the dofetilide treated embryos, albeit less pronounced than the difference at E10.5. (Line 256-257)

(5) Related to Figure 5. The work would be strengthened by quantification of the effects of dofetilide and verapamil on heartbeat at the doses applied. Is the verapamil dosage used here similar to the dose used in the clinic?

We are grateful to this suggestion. The effect of dofetilide on heartbeat has already been shown in Figure 2A. We have now additionally measured the heartbeat rate of verapamil treated embryos, and provided the results in Figure 5E. For verapamil injection in mice, a single i.p. dose of 15 mg/kg was used, which is equivalent to 53 mg/m² body surface. Verapamil is used in the clinic at dosage ranging from 200 to 480 mg/day, equivalent to 3.33 - 8 mg/kg or 117 - 282 mg/m² body surface. Therefore, the dosage used in the mouse is not excessively high compared to the clinic uses. (Line 361-365) 

Overstated Claims:

(1) The authors claim that the lipid microstructure/mTORC2/PKC/Notch pathway is responsive to shear stress, rather than other mechanical forces or myocardial function. Their conclusions seem to be extrapolated from various in vitro studies using non-endocardial cells. To solidify this claim, the authors would need additional biomechanical data, which could be obtained via theoretical modelling or using mouse heart valve explants. This issue could also be addressed by the authors simply softening their conclusions.

We aggrege with the reviewer’s comment. We have now revised the statement as “Our data support a model that membrane lipid microdomain acts as a shear stress sensor and transduces the mechanical cue to activate intracellular mTORC2-PKC-Notch signaling pathway in the developing endocardium. (line 416-418) It is noteworthy that the methodology used to alter blood flow in this study inevitably affects myocardial contraction. Additional work to uncouple sheer stress with other changes of mechanical properties of the myocardium with the aid of theoretical modelling or using mouse heart valve explants is needed to fully characterize the effect of sheer stress on mouse endocardial development.” (Line 436-440)

(2) Line 263-264: In the discussion, the authors conclude that "Strong fluid shear stress in the AVC and OFT promotes the formation of caveolae on the luminal surface of the endocardial cells, which enhances PKCε phosphorylation by mTORC2." This link was shown rather indirectly, rather than by direct evidence, and therefore the conclusion should be softened. For example, the authors could state that their data are consistent with this model.

We have revised the statement as “Strong fluid shear stress in the AVC and OFT enhances PKC phosphorylation by mTORC2 possibly by maintaining a particular membrane microstructure.” (Line 372-374)

(3) In the Discussion, it says: "Mammalian embryonic endocardium undergoes extensive EMT to form valve primordia while zebrafish valves are primarily the product of endocardial infolding (Duchemin et al., 2019)." In the paper cited, Duchemin and colleagues described the formation of the zebrafish outflow tract valve. The zebrafish atrioventricular valve primordia is formed via partial EMT through Dll-Notch signaling (Paolini et al. Cell Reports 2021) and the collective cell migration of endocardial cells into the cardiac jelly. Then, a small subset of cells that have migrated into the cardiac jelly give rise to the valve interstitial cells, while the remainder undergo mesenchymal-to-endothelial transition and become endothelial cells that line the sinus of the atrioventricular valve (Chow et al., doi: 10.1371/journal.pbio.3001505). The authors should modify this part of the Discussion and cite the relevant zebrafish literature.

Thanks for valuable comments. We have now revised the statement as “Mammalian embryonic endocardium undergoes extensive EMT to form valve primordia while zebrafish atrioventricular valve primordia is formed via partial EMT and the collective cell migration of endocardial cells into the cardiac jelly followed by tissue sheet delamination.” with relevant references added. (Line 411-414)

Recommendations to the Authors:

(1) One issue that the authors could address is the organization of figures. There are several cases where positive data that are central to the conclusions are placed in the supplement and should be moved to the main figures. Places where this occurred are listed below:

- The Tie2 conditional deletion of Dll4 showing retention of NICD in the OFT and AVC regions is highly supportive of the model. The authors should consider moving these data to main Figure 1.

Thanks for the suggestion. We have reorganized the figure as requested.

- The ligand expression data in Figure 2- Supplement Figure 1 A is VERY important to the conclusions drawn from the dofetilide treatment. The authors should move these data to main Figure 2.

The ligand expression data in Figure 2- Supplement Figure 1A are now moved to Figure 2B.

- In Figure 3A - the area in the field of view should be stated in the Figure (is it the AVC?) Figure 3 - Supplement 1 proximal OFT data should be moved to main Figure 3 as it is central to the conclusions. Negative DA data can be left in the supplement. Again, for Figure 3 - Supplement 1 Stauroporine treatment data should be moved to the main figure as it is positive data that are central to the conclusions.

Thanks for the suggestion. We have reorganized the figure as requested.

(2) Antibody used for Twist1 detection is not listed in the resource table.

Twist1 is purchased from abcam, the detailed information is now available in the resource table.

(3) Missing arrowhead in Figure 4A, last row.

Sorry for the negligence. Arrowhead is now added.

(4) Line 286. "OFT" pasted on the word "endothelium".

“OFT” is now removed.

(5) Related to Figure 2C. The fast response of NICD to flow cessation was used as an argument to support post-translational modification. It is not clear why Sox9 and Twist1 expression also responds so quickly.

Sox9 and Twist1 expression does seem to respond very quickly. Whether there exists additional regulatory pathways such as Wnt, Vegf signaling that also respond to sheer stress needs to be investigated in the future.

(6) Line 200: The sentence should end with a period.

Sorry for the oversight. It is now corrected.

(7) Lines 34 to 35: the authors phrase that Notch is "allowed" to be specifically activated in the AVC and outflow tract by shear stress.

We have rephrased the statement with “enabling Notch to be specifically activated in AVC and OFT by regional increased shear stress.” Line 27

(8) Lines 96-100: At the end of the introduction, the text is copied from the abstract. New text should be written or summarized in a different way.

The last sentence of introduction is now changed to “The results uncovered a new mechanism whereby mechanical force serves as a primary cue for endocardial patterning in mammalian embryonic heart.” (Line 93-95)

(9) Line 125: The term "agreed with the Dll4 transcript.."should be replaced with a better term like "overlapped" or "was identical with".

The word “agreed” is now “overlapped”. (Line 219)

(10) Line 291: "Thus, through these sophisticated mechanisms, the developing mouse hearts may achieve three purposes:"- The English should be adjusted here since it sounds like hearts are aiming to achieve a purpose, which is unlikely what was meant by the authors.

This sentence is rephrased to “Thus, in the developing mouse hearts: (1) VEGF signaling is reduced to permit endocardial EMT; (2) Dll4 expression is reduced to prevent widespread endocardial Notch activation and make endocardium sensitive to flow; (3) a proper cushion size and shape is maintained by limiting the flanking endocardium to undergo EMT despite physically close to the field of BMP2 derived from of AVC myocardium (Figure 6).” (Line 402-406)

Author response:

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

The mice crossing scheme is unusual as you have three mice to cross to produce genotypes, while we could understand that it is possible to produce pups of desired genotypes with different mating schemes, such a vague crossing scheme is not desirable and of poor genetics practice.

We thank the reviewer for this suggestion. Indeed, our scheme is not a representation of the actual breeding scheme but just a brief explanation of lineages used for the acquisition of the triple transgenic mice. We will include the full crossing scheme into the revision.

We added to the text the explanation that all used genotypes were maintained as homozygotes and put a full breeding scheme in the supplementary figure S1A

It is worth mentioning that single knockouts seem to show a corresponding upregulation of the level of the paralogue kinase, indicating that any lack of phenotypes might be due to feedback compensation, which would be an interesting finding if confirmed; this has not been mentioned.

We thank the reviewer for raising an important point about the paralog upregulation. Indeed, our data on primary cells (supplementary 1B) suggests the upregulation of CDK19 in CDK8KO and vice versa. We will point this out in discussion. We plan to examine the data for the testis as soon as more tissues are available.

We addressed this question by performing additional western blot (added to the paper fig. 2D) and found no paralogue upregulation in testes. To do that we also manufactured novel rabbit anti-mouse CDK19 antibodies described in Materials and Methods.

The authors should clarify or present the data on where CDK8 and CDK19  as well as CcnC are expressed so as to help the readers understand which tissues both CDK might be functioning in and cause the loss of CcnC.

Due to a limited sensitivity of single cell sequencing (only ~5,000 transcripts are sequenced from total of average 500,000 transcripts per cell, so the low expressed transcripts are not sequenced in all cells) it is challenging to firmly establish CDK8/19 positive and -negative tissues from single cell data because both transcripts are minor. This image will be included in the next version.

In this version we have added staining by CDK8 and CDK19 antibodies on paraffin sections, showing expression in variety of cells. Additionally, we have analyzed Cdk8/CcnC presence in different testicular cell types by flow cytometry. Both methods show that not only spermatogonial stem cells express Cdk8 as was shown in McCleland et al. 2005, but also some 1n cells, 4n cells and a significant part of cKit^-2n cells. We added a corresponding paragraph and figures (2E-K) to the paper. We consider this a more definitive answer to the question than RNA data.

Furthermore, data for the genitourinary system in single knockouts are very sparse; data are described for fertility in Figure 1H, ploidy, and cell number in Figures 2B and C, plasma testosterone and luteinizing hormone levels in Figures 5C and 5D, and morphology of testis and prostate tissue for single Cdk8 knockout in Supplementary Figure 1C (although in this case the images do not appear very comparable between control and CDK8 KO, thus perhaps wider fields should be shown), but, for example, there is no analysis of different meiotic stages or of gene expression in single knockouts. It is worth mentioning that single knockouts seem to show a corresponding upregulation of the level of the paralogue kinase, indicating that any lack of phenotypes might be due to feedback compensation, which would be an interesting finding if confirmed; this has not been mentioned.

We agree that a description of the single KO could be beneficial, but we expect no big differences with the WT or Cre-Ert. We found neither histological differences nor changes in cell counts or ratios of cell types. Our ethical committee also has concerns about sacrificing mice without major phenotypic changes, without a well formulated hypothesis about the observed effects. We plan to add histological pictures to the next version of the article.

We have updated histological figures with new figures for iDKO and Cre+Tam mice with additional fields of view and better quality staining (2A-B).

The second major weakness is that the correlation between double knockout and reduced expression of genes involved in steroid hormone biosynthesis is portrayed as a causal mechanism for the phenotypes observed. While this is a possibility, there are no experiments performed to provide evidence that this is the case. Furthermore, there is no evidence showing that CDK8 and/or CDK19 are directly responsible for the transcription of the genes concerned.

We agree with the reviewer that the effects on CDK8/CDK19/CCNC could lead to the observed transcriptional changes in multiple indirect steps. There are, however, major technical challenges in examining the binding of transcription factors in the tissue, especially in Leydig cells which are a relatively minor population.  We will clarify it in the revision and strengthen this point in the discussion.

We have added corresponding explanation in the Discussion: “We hypothesize that all these changes are caused by disruption of testosterone synthesis in Leydig cells, although, at this point, we cannot definitively prove that the affected genes are regulated by CDK8/19 directly.”

The claim of reproductive defects in the induced double knockout of CDK8/19 resulted from the loss of CCNC via a kinase-independent mechanism is interesting but was not supported by the data presented. While the construction and analysis of the systemic induced knockout model of Cdk8 in Cdk19KO mice is not trivial, the analysis and data are weakened by the systemic effect of Cdk8 loss, making it difficult to separate the systemic effect from the local testis effect.

We agree with the reviewer that the effects on the testes could be due to the systemic loss of CDK8 rather than specifically in the testis, and we will clarify it in the revision. We will also clarify that although our results are suggestive that the effects of CDK8/19 knockout are kinase-independent, and that the loss of Cyclin C is a likely explanation for the kinase independence, but we do not claim that it is *the* mechanism.

In this version we added several caveats indicating that the proposed mechanism is likely, but not the only one possible.

Also using TAM-treated wild type as control is ok, but a better control will be TAM-treated ERT2-cre; CDK8f/f or TAM-treated ERT2 Cre CDK19/19 KO, so as to minimize the impact from the well-recognized effect of TAM.  

We used TAM-treated ERT2-cre for most of the experiments, and did not observe any major histological or physiological differences with the WT+TAM. We will make sure to present them in the revision.

The authors found that Sertoli cells re-entered the cell cycle in the inducible double knockout but stopped short of careful characterization other than increased expression of cell cycle genes.

Unfortunately, we were not able to perform satisfactory Ki67 staining to address this point.

Dko should be appropriately named iDKO (induced dKO). We will make the corresponding change.

We performed necropsy ? not the right wording here.

Colchicine-like apoptotic bodies ? what does this mean? Not clear.

We made appropriate changes - all DKO were renamed iDKO, necropsy changed to autopsy and cells designated as “apoptotic”.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Given the proprietary claims of the authors ("We have for the first time generated mice with the systemic inducible Cdk8 knockout on the background of Cdk19 constitutive knockout"), it does not appear acceptable and indeed might be misleading, to not describe the overall phenotypes of the mice. Are mice normal size/weight? Does an autopsy reveal anything other than atrophied genital tissue in males? Do the authors find a phenotype in the intestinal epithelium, as previously reported? (N.B. this could potentially clarify a discrepancy in the literature since the loss of the secretory lineages in double knockouts reported by the Firestein lab was not reproduced by intestinal organoid double knockout in the paper by the Fisher lab).

We have removed the statement “for the first time”, although to the best of our knowledge this is the fact. We did not attempt to describe all the phenotypic effects of the Cdk8/19 knockout in this paper, since some of the phenotypic observations related to mouse weight and behavior varied between different laboratories involved and require additional analysis. The effect on the urogenital system was by far the most striking histological feature observed and it was carefully addressed in this paper. Other findings require additional experiments and are out of the scope of this paper and we plan to focus on them later. As per suggestion of the reviewer we performed histological analysis of DKO intestines and found the same decrease in the Paneth and goblet cells numbers as described by Dannappel et al. We added corresponding figures (Supplemental fig. 1C) to the paper.

If the authors wish to reinforce their claims about causality of steroidogenic gene expression and phenotype, they could try rescuing the phenotype by treating mice with testosterone.

As stated in Discussion, we hypothesized that injection of testosterone would not rescue the phenotype, as the androgen receptor signaling is also affected. However we would like to perform such an experiment, but we were not able to procure testosterone pellets at this time.

If they wish to claim a direct effect of CDK8/19 on the expression of steroidogenic genes, they could also assess CDK8/19 binding to promoters of the genes analysed by ChIP.

There are big technical challenges in examining the binding of transcription factors in the primary tissue, especially in Leydig cells, a minor population, so we cannot perform such an experiment.

In order to conclude that their CDK8/19 inhibitor treatment worked, they could show target engagement by cell thermal shift assay, loss of CDK8/19 kinase-dependent gene expression, or loss of CDK8/19 substrate phosphorylation (eg interferon-induced STAT1 S727 phosphorylation) under the conditions used. Alternatively, they could show rescue with a kinase-dead allele.

As noted in public comments - we thank the reviewer for raising this concern. The target selectivity and target engagement by the inhibitors used in this study (Senexin B and SNX631-6) have been described in other models and published. CDK8/19 engagement and target selectivity of Senexin B, used in our vitro studies, have been extensively characterized in cell-based assays (Chen et al., Cells 2019, 8(11), 1413; Zhang et al., J Med Chem. 2022 Feb 24;65(4):3420-3433.) Similar characterization has been published for SNX631-6 and its equipotent analog SNX631, which showed drastic antitumor activity when  used in vivo at the same dosing regimen as in this paper (Li et al., J Clin Invest. 2024;134(10):e176709). The comparison of the pharmacokinetics data obtained in the present study and in vitro activity of SNX631-6 in a cell-based assay suggests that the tissue concentrations of this drug should have provided substantial inhibition of Cdk8/19. Unfortunately, there are no known phosphorylation substrates specific for Cdk8/19 that can be used as pharmacodynamic markers. The widely used STAT1 phosphorylation at S727 is exerted not only by CDK8/19 but also by other kinases and shows variable response to CDK8/19 inhibition (Chen et al., Cells 2019, 8(11), 1413). In the revised MS, we have added a Western blot with pSTAT1 S727 staining of WT, 8KO, 19KO and iDKO testes. Cdk8/19 knockout did not decrease and apparently even increased the level of pSTAT1 S727, which demonstrates that this marker of CDK8/19 activity it is not suitable for our tissue type. While the evidence that Cdk8/19 kinase inhibition in the testes after in vivo drug treatment does not match the phenotype of iDKO is admittedly indirect, the same result has been obtained in the cell culture studies with Sertoli cells, where the inhibitor concentration (1 µM Senexin B) was much higher than needed for the maximal Cdk8/19 inhibition.

Finally, I did not find any legends to supplementary figures anywhere.

We apologize for not including legends for supplementary figures, and will correct that in the next version of the manuscript.

Additionally, we addressed the question about the sufficiency of the lipid supply for steroidogenesis in testes. There was a possibility that steroidogenesis is impossible due to the lack of cholesterol input, but OilRed staining revealed that the situation is the opposite: lipid content in iDKO testes is significantly higher than in WT testes. We added corresponding text to the article and the supplementary Fig. S6.

Author response:

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

Public Reviews:

Reviewer #1 (Public review):

Summary: 

This manuscript (Baron, Oviedo et al., 2024) builds on a previous study from the Wiseman lab (Perea, Baron et al., 2023) and describes the identification of novel nucleoside mimetics that activate the HRI branch of the ISR and drive mitochondrial elongation. The authors develop an image processing and analysis pipeline to quantify the effects of these compounds on mitochondrial networks and show that these HRI activators mitigate ionomycin-driven mitochondrial fragmentation. They then show that these compounds rescue mitochondrial morphology defects in patient-derived MFN2 mutant cell lines. 

Strengths: 

The identification of new ISR modulators opens new avenues for biological discovery surrounding the interplay between mitochondrial form/function and the ISR, a topic that is of broad interest. It also reinforces the possibility that such compounds might represent new potential therapeutics for certain mitochondrial disorders. The development of a quantitative image analysis pipeline is valuable and has the potential to extract the subtle effects of various treatments on mitochondrial morphology. 

We thank the reviewer for the positive feedback on our manuscript. We address all of the reviewer’s valuable concerns in the revised submission, as highlighted below. 

Weaknesses: 

I have three main concerns.

First, support for the selectivity of compounds 0357 and 3610 acting downstream of HRI comes from using knockdown ISR kinase cell lines and measuring the fluorescence of ATF4-mApple (Figure 1G and 1H). However, the selectivity of these compounds acting through HRI is not shown for mitochondrial morphology. Is mitochondrial elongation blocked in HRI knockdown cells treated with the compounds? While the ISRIB treatment does block mitochondrial elongation, ISRIB acts downstream of all ISR kinases and doesn't necessarily define selectivity for the HRI branch of the ISR. Additionally, are the effects of these compounds on ATF4 production and mitochondrial elongation blocked in a non-phosphorylatable eIF2alpha mutant? 

We thank the reviewer for highlighting this point. As indicated by the reviewer, we show that compounddependent increases in mitochondrial elongation are blocked by co-treatment with ISRIB, indicating that this effect can be attributed to ISR activation. We prefer the use of this highly selective pharmacologic approach to block ISR activation, as opposed to the MEF^A/A cells, as the use of pharmacologic approaches provide more temporal control over ISR inhibition and can prevent the type of chronic disruption to mitochondria associated with these types of genetic perturbations. However, the reviewer is correct that ISRIB blocks downstream of all ISR kinases, meaning that we cannot explicitly demonstrate that 0357 and 3610 induce mitochondrial elongation downstream of HRI-dependent ISR activation using this tool. Thus, to address this point, we have clarified the discussion of these results to make it clear that our results show that our compounds induce mitochondrial elongation downstream of the ISR, omitting the direct implications of HRI in this phenotype. 

This point of selectivity/specificity of the compounds gets at a semantic stumbling block I encountered in the text where it was often stated "stress-independent activation" of ISR kinases. Nucleoside mimetics are likely a very biologically active class of molecules and are likely driving some level of cell stress independent of a classical ISR, UPR, heat-shock response, or oxidative stress response. 

A major challenge in defining stress-independent activation of stress-responsive signaling pathways is the fact that the activation of these pathways is often used as a primary marker of cellular stress. While this can be overcome by transcriptome-wide profiling (e.g., RNAseq), the reviewer is correct that our focused profiling of select stress-responsive signaling pathways is insufficient to claim the stress-independent activation of the ISR by our prioritized compounds. To address this, we removed this terminology from the revised submission.  

Second, it is difficult for me to interpret the data for the quantification of mitochondrial morphology. In the legend for Figure 2, it is stated that "The number of individual measurements for each condition are shown above." Are the individual measurements the number of total cells quantified? If not, how many total cells were analyzed? If the individual measurements are distinct mitochondrial structures that could be quantified why are the n's for each parameter (bounding box, ellipsoid principal axis, and sphericity) so different? Does this mean that for some mitochondria certain parameters were not included in the analysis? For me, it seems more intuitive that each mitochondrial unit should have all three parameters associated with it, but if this isn't the case it needs to be more carefully described why. 

The number of individual measurements refers to the number of 3D segmentations generated using the “surfaces’ module in Imaris. As the reviewer noted, we expect each surface segmentation to represent a single “mitochondrial unit.” We have now further clarified this in the figure legend. 

Regarding differences in sample size for each group, we used an outlier test (i.e., ROUT outlier test in PRISM 10) to remove apparent outliers in our data. Often, these outliers result from errors in the automatic quantification that inaccurately merge two mitochondria into one large segmentation. This explains the discrepancy in the number of measurements made for each experimental group. We have made this point more clear in the Materials and Methods section of the revised manuscript.  

Third, the impact of these compounds on the physiological function of mitochondria in the MFN2.D414V mutants needs to be measured. Sharma et al., 2021 showed a clear deficit in mitochondrial OCR in MFN2.D414V cells which, if rescued by these compounds, would strengthen the argument that pharmacological ISR kinase activation is a strategy for targeting the functional consequences of the dysregulation of mitochondrial form.

In this manuscript, we demonstrate that pharmacologic activation of the ISR using 0357 and 3610 rescue mitochondrial morphology in patient fibroblasts expressing the disease-associated MFN2^D414V mutant. The reviewer is correct that there are other mitochondrial phenotypes linked to the expression of this mutant. We are currently pursuing this question with more potent ISR activating compounds developed in our laboratory identified using the HTS screening platform described in this manuscript. However, this work, which builds on the studies described herein, uses other ISR activating compounds, which we feel would be best described in subsequent manuscripts that can fully define the activity of these new compounds.  

Reviewer #2 (Public review): 

Summary. 

Mitochondrial dysfunction is associated with a wide spectrum of genetic and age-related diseases. Healthy mitochondria form a dynamic reticular network and constantly fuse, divide, and move. In contrast, dysfunctional mitochondria have altered dynamic properties resulting in fragmentation of the network and more static mitochondria. It has recently been reported that different types of mitochondrial stress or dysfunction activate kinases that control the integrated stress response, including HRI, PERK, and GCN2. Kinase activity results in decreased global translation and increased transcription of stress response genes via ATF4, including genes that encode mitochondrial protein chaperones and proteases (HSP70 and LON). In addition, the ISR kinases regulate other mitochondrial functions including mitochondrial morphology, phospholipid composition, inner membrane organization, and respiratory chain activity. Increased mitochondrial connectivity may be a protective mechanism that could be initiated by pharmacological activation of ISR kinases, as was recently demonstrated for GCN2. 

A small molecule screening platform was used to identify nucleoside mimetic compounds that activate HRI. These compounds promote mitochondrial elongation and protect against acute mitochondrial fragmentation induced by a calcium ionophore. Mitochondrial connectivity is also increased in patient cells with a dominant mutation in MFN2 by treatment with the compounds.

Strengths: 

(1) The screen leverages a well-characterized reporter of the ISR: translation of ATF4-FLuc is activated in response to ER stress or mitochondrial stress. Nucleoside mimetic compounds were screened for activation of the reporter, which resulted in the identification of nine hits. The two most efficacious dose-response tests were chosen for further analysis (0357 and 3610). The authors clearly state that the compounds have low potency. These compounds were specific to the ISR and did not activate the unfolded protein response or the heat shock response. Kinases activated in the ISR were systematically depleted by CRISPRi revealing that the compounds activate HRI.

(2) The status of the mitochondrial network was assessed with an Imaris analysis pipeline and attributes such as length, sphericity, and ellipsoid principal axis length were quantified. The characteristics of the mitochondrial network in cells treated with the compounds were consistent with increased connectivity. Rigorous controls were included. These changes were attenuated with pharmacological inhibition of the ISR. 

(3) Treatment of cells with the calcium ionophore results in rapid mitochondrial fragmentation. This was diminished by pre-treatment with 0357 or 3610 and control treatment with thapsigargin and halofuginone 

(4) Pathogenic mutations in MFN2 result in the neurodegenerative disease Charcot-Marie-Tooth Syndrome Type 2A (CMT2A). Patient cells that express Mfn2-D414V possess fragmented mitochondrial networks and treatment with 0357 or 3610 increased mitochondrial connectivity in these cells.

We appreciate the reviewer’s positive response to these aspects of our manuscript. We address the reviewer’s valuable comments in the revised submission as highlighted below. 

Weaknesses: 

The weakness is the limited analysis of cellular changes following treatment with the compounds. 

(1) Unclear how 0357 or 3610 alter other aspects of cellular physiology. While this would be satisfying to know, it may be that the authors determined that broad, unbiased experiments such as RNAseq or proteomic analysis are not justified due to the limited translational potential of these specific compounds.

The reviewer is correct. The low potency of 0357 and 3610 limit the translational potential for these compounds. However, building on the work described herein, we recently identified more potent HRI activating compounds with higher translational potential. Using RNAseq profiling, we found that these compounds show transcriptomewide selectivity for the ISR and can promote adaptive remodeling of mitochondrial morphology and function in cellular models of multiple other diseases. These compounds will be further described in subsequent studies that expand on the efforts outlined here demonstrating the potential for pharmacologic HRI activators to promote adaptive mitochondrial remodeling.   

(2) There are many changes in Mfn2-D414V patient cells including reduced respiratory capacity, reduced mtDNA copy number, and fewer mitochondrial-ER contact sites. These experiments are relatively narrow in scope and quantifying more than mitochondrial structure would reveal if the compounds improve mitochondrial function, as is predicted by their model.

In this manuscript, we demonstrate that pharmacologic activation of the ISR using 0357 and 3610 rescue mitochondrial morphology in patient fibroblasts expressing the disease-associated MFN2^D414V mutant. The reviewer is correct that there are other mitochondrial phenotypes linked to the expression of this mutant. We are currently pursuing this question with more potent ISR activating compounds developed in our laboratory using the HTS screening platform described in this manuscript. However, this work, which builds on the studies described herein, uses other ISR activating compounds, which we feel would be best described in subsequent manuscripts that can fully define the activity of these new compounds.  

Reviewer #3 (Public review):

Summary: 

Mitochondrial injury activates eiF2α kinases - PERK, GCN2, HRI, and PKR - which collectively regulate the Integrated Stress Response (ISR) to preserve mitochondrial function and integrity. Previous work has demonstrated that stress-induced and pharmacologic stress-independent ISR activation promotes adaptive mitochondrial elongation via the PERK and GCN2 kinases, respectively. Here, the authors demonstrate that pharmacologic ISR inducers of HRI and GCN2 enhance mitochondrial elongation and suppress mitochondrial fragmentation in two disease models, illustrating the therapeutic potential of pharmacologic ISR activators. Specifically, the authors first used an innovative ISR translational reporter to screen for nucleoside mimetic compounds that induce ISR signaling and identified two compounds, 0357 and 3610, that preferentially activate HRI. Using a mitochondrial-targeted GFP MEF cell line, the authors next determined that these compounds (as well as the GCN2 activator, halofuginone) enhance mitochondrial elongation in an ISR-dependent manner. Moreover, pretreatment of MEFs with these ISR kinase activators suppressed pathological mitochondrial fragmentation caused by a calcium ionophore. Finally, pharmacologic HRI and GCN2 activation were found to preserve mitochondrial morphology in human fibroblasts expressing a pathologic variant in MFN2, a defect that leads to mitochondrial fragmentation and is a cause of Charcot Marie Tooth Type 2A disease. 

Strengths: 

This well-written manuscript has several notable strengths, including the demonstration of the potential therapeutic benefit of ISR modulation. New chemical entities with which to further interrogate this stress response pathway are also reported. In addition, the authors used an elegant screen to isolate compounds that selectively activate the ISR and identify which of the four kinases was responsible for activation. Special attention was also paid to a thorough evaluation of the effect of their compounds on other stress response pathways (i.e. the UPR, and heat and oxidative stress responses), thereby minimizing the potential for off-target effects. The implementation of automated image analysis rather than manual scoring to quantify mitochondrial elongation is not only practical but also adds to the scientific rigor, as does the complementary use of both the calcium ionophore and MFN2 models to enhance confidence and the broad therapeutic potential for pharmacology ISR manipulation. 

We thank the reviewer for their positive response to our manuscript. We address the reviewer’s remaining concerns as outlined below. 

Weaknesses: 

The only minor concerns are with regard to effects on cell health and the timing of pharmacological administration. 

The two compounds described in this manuscript were found to not induce any overt toxicity over a 24 h period in cell culture models. In the revised manuscript, we show data showing that treatment with increasing doses of either 0357 or 3610 do not significantly reduce cellular viability in HEK293 cells (Fig. S1G). 

With regards to treatments, we include all of the relevant information for the timing and dosage of compound treatment in the revised manuscript. 

Recommendations for the authors:

Reviewer #1 (Recommendations for Authors)

(1) Figure S1 "B. ATF4-Gluc activity" -> Fluc, The number of replicates is not consistently stated for each experiment. p-values are not given for D and F. 

We have changed the legend for Fig. S1B to ATF4-FLuc. We show individual replicates for all experiments for all panels described in this figure, except panels C and G, in the revised Figure S1. We explicitly state the number of replicates in panel C and G in the accompanying figure legend. We have repeated the qPCR described in panels C,F and statistics are included in the revised manuscript.

(2) Figure 2 - no p-values for BtdCPU.

Yes. We found that BtdCPU-dependent increases in mitochondrial fragmentation (described in Fig. 2A-D) were not significant when analyzing all the data included in these figures by Brown-Forsythe and Welch ANOVA test. However, the DMSO and BtdCPU conditions were significantly different when directly compared using a Welch’s t-test (p<0.005). Since the statistics in this manuscript are being analyzed by ANOVA, we decided not to include a significance marker for BtdCPU, as it was not observed in this more stringent test and is not the main focus of this manuscript.  

(3) Figure S4 (Supplement to Figure 5) -> Supplement to Figure 4. 

We have corrected this error in the revised manuscript. 

(4) Error in references - duplicated 24 and 46, duplicated 10 and 11.

This is now corrected in the revised submission.

Reviewer #2 (Recommendations for the authors): 

I would love to see an assessment of mitochondrial function and mtDNA in the D414 cells following treatment. 

As indicated above, we are continuing to probe the impact of more potent HRI activating compounds in patientderived cell models expressing disease-relevant MFN2 mutants. Initial experiments suggest that this approach can mitigate additional pathologies beyond deficient elongation in these cells, although we are continuing to pursue these results with our improved HRI activating compounds. We are excited by these results, but feel that they are best suited for a follow-up manuscript describing these new HRI activators.   

Reviewer #3 (Recommendations for the authors):

The only suggestion to broaden the manuscript's impact might be to perform a basic assessment of the impact of pharmaceutical ISR activation on cell viability. Though mitochondrial elongation is often considered a surrogate for mitochondrial health, whether mitochondrial elongation improves cell viability (or not) would be informative. Similarly, the authors did not address the time-dependent effects of the ISR modulators, choosing to focus on the acute rather more chronic outcomes. Finally, does simultaneous (rather than pre-) treatment with an activator and the ionomycin produce similar effects on mitochondrial morphology, especially since therapeutics are typically administered post-injury?

We now include cell viability experiments showing that the two HRI activators discussed in this manuscript, 0357 and 3610, do not significantly reduce viability in HEK293 cells. This work is included in the revised manuscript (see Fig. S1G). 

With respect to acute vs chronic outcomes of ISR activation. As highlighted by the reviewer, we primarily focus this work on defining the impact of acute ISR treatment on mitochondrial morphology. As discussed above, we now show that our prioritized ISR activating compounds 0357 and 3610 do not significantly impact cellular viability over a 24 h timecourse. However, as suggested by the reviewer, additional studies on the potential implications of chronic pharmacologic ISR activation on mitochondrial biology remains to be further explored.

We are continuing to address this in subsequent studies using more potent ISR kinase activating compounds established in our lab. However, we would like to highlight that detrimental phenotypes linked to chronic ISR kinase activation in cell culture does not preclude the translational potential for this approach, as in vivo PK/PD of these compounds can be controlled to prevent complications arising from chronic pathway activity. We previously demonstrated the potential for controlling compound activity through its PK/PD in our establishment of highly selective activators of other stress-responsive signaling pathways such as the IRE1/XBP1s arm of the UPR (e.g., Madhavan et al (2022) Nat Comm).   

We appreciate the reviewer’s comments regarding the timing of compound treatment in them ionomycin experiment. Ionomycin works extremely quick to induce fragmentation (minutes), which would be prior to activation of the ISR induced by these compounds (hours). Thus, co-treatment would lead to fragmentation. It is an interesting question to ask if co-treatment with ISR activators could rescue this fragmentation as the pathway is activated, but we did not explicitly address this question in this manuscript. However, we do show that pharmacologic GCN2 or HRI activators can rescue mitochondrial morphology in patient fibroblasts expressing a MFN2 mutant, where mitochondria are fragmented, indicating that our approach can restore mitochondrial morphology in this context. We feel that these results, in combination with others described in our manuscript, demonstrate the potential for this approach to mitigate pathologic mitochondrial fragmentation associated with different conditions.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This work uses transgenic reporter lines to isolate entpd5a+ cells representing classical osteoblasts in the head and non-classical (osterix-) notochordal sheath cells. The authors also include entpd5a- cells, col2a1a+ cells to represent the closely associated cartilage cells. In a combination of ATAC and RNA-Seq analysis, the genome-wide transcriptomic and chromatin status of each cell population is characterized, validating their methodology and providing fundamental insights into the nature of each cell type, especially the less well-studied notochordal sheath cells. Using these data, the authors then turn to a thorough and convincing analysis of the regulatory regions that control the expression of the entpd5a gene in each cell population. Determination of transcriptional activities in developing zebrafish, again combined with ATAC data and expression data of putative regulators, results in a compelling and detailed picture of the regulatory mechanisms governing the expression of this crucial gene.

Strengths:

The major strength of this paper is the clever combination of RNA-Seq and ATAC analysis, further combined with functional transcriptional analysis of the regulatory elements of one crucial gene. This results in a very compelling story.

Weaknesses:

No major weaknesses were identified, except for all the follow-up experiments that one can think of, but that would be outside of the scope of this paper.

Reviewer #2 (Public Review):

Summary:

Complementary to mammalian models, zebrafish has emerged as a powerful system to study vertebrate development and to serve as a go-to model for many human disorders. All vertebrates share the ancestral capacity to form a skeleton. Teleost fish models have been a key model to understand the foundations of skeletal development and plasticity, pairing with more classical work in amniotes such as the chicken and mouse. However, the genetic foundation of the diversity of skeletal programs in teleosts has been hampered by mapping similarities from amniotes back and not objectively establishing more ancestral states. This is most obvious in systematic, objective analysis of transcriptional regulation and tissue specification in differentiated skeletal tissues. Thus, the molecular events regulating bone-producing cells in teleosts have remained largely elusive. In this study, Petratou et al. leverage spatial experimental delineation of specific skeletal tissues -- that they term 'classical' vs 'non-classical' osteoblasts -- with associated cartilage of the endo/peri-chondrial skeleton and inter-segmental regions of the forming spine during development of the zebrafish, to delineate molecular specification of these cells by current chromatin and transcriptome analysis. The authors further show functional evidence of the utility of these datasets to identify functional enhancer regions delineating entp5 expression in 'classical' or 'non-classical' osteoblast populations. By integration with paired RNA-seq, they delineate broad patterns of transcriptional regulation of these populations as well as specific details of regional regulation via predictive binding sites within ATACseq profiles. Overall the paper was very well written and provides an essential contribution to the field that will provide a foundation to promote modeling of skeletal development and disease in an evolutionary and developmentally informed manner.

Strengths:

Taken together, this study provides a comprehensive resource of ATAC-seq and RNA-seq data that will be very useful for a wide variety of researchers studying skeletal development and bone pathologies. The authors show specificity in the different skeletal lineages and show the utility of the broad datasets for defining regulatory control of gene regulation in these different lineages, providing a foundation for hypothesis testing of not only agents of skeletal change in evolution but also function of genes and variations of unknown significance as it pertains to disease modeling in zebrafish. The paper is excellently written, integrating a complex history and experimental analysis into a useful and coherent whole. The terminology of 'classical' and 'non-classical' will be useful for the community in discussing the biology of skeletal lineages and their regulation.

Weaknesses:

Two items arose that were not critical weaknesses but areas for extending the description of methods and integration into the existing data on the role of non-classical osteoblasts and establishment/canalization of this lineage of skeletal cells.

(1) In reading the text it was unclear how specific the authors' experimental dissection of the head/trunk was in isolating different entp5a osteoblast populations. Obviously, this was successful given the specificity in DEG of results, however, analysis of contaminating cells/lineages in each population would be useful - e.g. using specific marker genes to assess. The text uses terms such as 'specific to' and 'enriched in' without seemingly grounded meaning of the accuracy of these comments. Is it really specific - e.g. not seen in one or other dataset - or is there some experimental variation in this?

We thank the reviewer for pointing this out. Given that the separation from head and trunk is done manually, there will be some experimental variability. We have used anatomical hallmarks (cleithrum and swim bladder), and therefore would expect the variability to be small. Regarding classical osteoblasts contaminating trunk tissue, head removal was consistently performed using the aforementioned anatomical hallmarks in a manner that ensures that the cleithrum does not remain in the trunk tissue.  In order to alleviate concerns regarding trunk cell populations contaminating cranial populations, and to further clarify our strategy, we add the following statement to the Materials and Methods section: “The procedure does not allow for a complete separation of notochordal non-classical osteoblasts from cranial classical osteoblasts, as the notochord extends into the cranium. However, the amount of sheath cells in that portion of the notochord is negligible, compared both to the number of classical (cranial) osteoblasts in head samples, and to notochord cells isolated in trunk samples.”

(2) Further, it would be valuable to discuss NSC-specific genes such as calymmin (Peskin 2020) which has species and lineage-specific regulation of non-classical osteoblasts likely being a key mechanistic node for ratcheting centra-specific patterning of the spine in teleost fishes. What are dynamics observed in this gene in datasets between the different populations, especially when compared with paralogues - are there obvious cis-regulatory changes that correlate with the co-option of this gene in the early regulation of non-classical osteoblasts? The addition of this analysis/discussion would anchor discussions of the differential between different osteoblasts lineages in the paper.

This is an interesting concept and idea, that we will consider in a possible revision or, if requiring substantial additional efforts, in a possible new research line. An excellent starting point for further studies using our datasets.

Reviewer #3 (Public Review):

Summary:

This study characterizes classical and nonclassical osteoblasts as both types were analyzed independently (integrated ATAC-seq and RNAseq). It was found that gene expression in classical and nonclassical osteoblasts is not regulated in the same way. In classical osteoblasts, Dlx family factors seem to play an important role, while Hox family factors are involved in the regulation of spinal ossification by nonclassical osteoblasts. In the second part of the study, the authors focus on the promoter structure of entpd5a. Through the identification of enhancers, they reveal complex modes of regulation of the gene. The authors suggest candidate transcription factors that likely act on the identified enhancer elements. All the results taken together provide comprehensive new insights into the process of bone development, and point to spatio-temporally regulated promoter/enhancer interactions taking place at the entpd5a locus.

Strengths:

The authors have succeeded in justifying a sound and consistent buildup of their experiments, and meaningfully integrating the results into the design of each of their follow-up experiments. The data are solid, insightfully presented, and the conclusion valid. This makes this manuscript of great value and interest to those studying (fundamental) skeletal biology.

Weaknesses:

The study is solidly constructed, the manuscript is clearly written and the discussion is meaningful - I see no real weaknesses.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Minor issues that may need to be addressed or detailed:

Supplementary Figures 1I-J, text page 4, line 24: "photoconversion and imaging": this needs some more detailed description: green fluorescent cells should be actively expressing Kaede, but only if there is a delay between photoconversion and imaging. What is the reason that Supplementary Figure 1F shows mainly green fluorescent cells, contrary to 1G-J?

In our experiments, we could see new Kaede expression under the control of the entpd5a promoter region within 1.5 hours of photoconversion, as shown in Suppl. Figure 1E-H, suggesting that this time window was sufficient for protein generation. The reason for Suppl. Fig 1F showing more green fluorescence we believe relates to the high rate of transcriptional activity at that stage, in the entirety of the notochord progenitor cells. In addition, this is an effect which we attribute to the relatively small number of cells producing red fluorescence at that stage, due to photoconversion of only a few Kaede+ cells at the 15 somites stage (Suppl. Fig. 1E). Therefore, the masking effect of the green fluorescence by the red is not as significant as in G and H, where the red fluorescence resulting from photoconversion right after imaging at 18s and 21s, respectively, significantly overlaps with new green fluorescence. This can be seen in the image as the presence of orange fluorescence in G and H, instead of the clear red shown in E, I and J.

In addition to this, we would like to point out that in Suppl. Fig. 1I, J the reason that green fluorescence is only detected in the ventral region of the notochord, is because the promoter of entpd5a only remains active in the ventral-most sheath cells at that stage. This is stated in the results section of the main text, first subsection, paragraph 3. The reason for this very interesting, strictly localised expression pattern remains unclear.

Somewhat intriguing: green fluorescence in Figure 1B, C (osx:GAL4FF) and Supplementary Figure 1C (entpd5a:GAL4FF) in the CNS? Would that be an artefact of the GAL4FF/UAS:GFP system?

We are confident that the fluorescence pointed out by the reviewer is not an artefact of the GAL4FF/UAS system, for a few reasons. Firstly, osx (Sp7) has been shown to be expressed and to function in the nervous system in mice (Park et al, BBRC, 2011; Elbaz et al, Neuron, 2023). Secondly, not only osx, but also entpd5a can be readily detected in a subset of cranial and spinal neurons in early development using the entpd5a:GAL4FF; UAS:GFP transgenic line (Suppl. Fig 1C). Finally, when establishing transgenic lines with the entpd5a(1.1):GFP construct, expression was almost invariably present in diverse elements of the nervous system, but not in bone (data not shown). This led us to hypothesise that the minimal promoter of entpd5a (and possibly also that of osx) is activated by transcription factors active in the nervous system, and this effect is likely controlled by the surrounding enhancers, but also the genome location. It is unclear at present what the endogenous neural expression of the two genes is like, and we did not further investigate this in this study, as the focus was on the skeleton.

Figure 2: What exactly is "Corrected Total Cell Fluorescence"? Is it green + red fluorescence?

We thank the reviewer for pointing out the absence of more information on this. Corrected total cell fluorescence does not correspond to green+ red fluorescence, rather it is calculated as follows for a single channel:

CTCF = Integrated Density – (Area of selected cell X Mean fluorescence of background readings)

More details can be found in the following website: https://theolb.readthedocs.io/en/latest/imaging/measuring-cell-fluorescence-using-imagej.html

We have edited the Materials and Methods section under “Imaging and image analysis” to include the aforementioned information.

Page 11, line 34: The authors may have missed the recently published "Raman et al., Biomolecules 2024 Vol. 14; doi:10.3390/biom14020139" describing RNA-Seq in 4 dpf osterix+ osteoblasts.

We thank the reviewer for drawing our attention to the Raman et al publication. The reference has now been added in the manuscript.

Figure 5A and B: use a higher resolution version to make the numbers and gene names more readable. Figures 5C and 6A could also use a larger font for the text and numbers.

High resolution files are now included with the revised manuscript, which should significantly help in making figures more easily readable. Although we agree with the reviewer that larger fonts would improve readability, due to the nature of the graphs (very small spaces in some cases, where the numbers would have to fit) this would not be easy to achieve. However, we believe that this issue will be resolved with the availability of higher resolution files. If readability remains a concern, we would be happy to attempt re-organising the graphs to allow for larger fonts.

Reviewer #2 (Recommendations For The Authors):

I suggest no further experiments, but do suggest that a few points be clarified.

In the Discussion, the text "the less evolved osteoblasts of fish and amphibians..." is not accurate. These cells are not less evolved as they represent an independent lineage to tetrapods that have evolved with different stresses for a similar time. However, as teleost fishes and amphibians share characteristics and all share a common ancestor, these signatures represent a putative ancestral state of skeletal differentiation not seen in amniotes, including humans.

We thank the reviewer for pointing out the unfortunate phrasing. The text has now been modified as follows: “Specifically, the osteoblasts of teleost fish and amphibians, whose characteristics are putatively closer to a more ancestral state of skeletal differentiation compared to amniotes, appear to share gene expression with chondrocytes”.

The title could potentially be shortened to reach a broader audience by removing the initial clause of 'integration of ATAC and RNA seq' as this is a commonly performed analysis - "Chromatin and transcriptomic signature in classical and non-classical zebrafish osteoblasts indicate mechanisms of ancestral skeletal differentiation" is more descriptive of the findings and not focused on the method.

We have discussed this internally, but would prefer to retain the current title. The reason is (1) because we would like to see our methodology and datasets be used as platform for further studies, and the current title, in our opinion, facilitates this. In regards to replacing “mechanisms of entpd5a regulation” with “mechanisms of ancestral skeletal differentiation”, we think this does not give an accurate description of our work, which is primarily focused on elucidating entpd5a promoter dynamics.

All datasets should be made available as soon as possible for use in the field.

The datasets (raw and processed) are available on the GEO database. The corresponding accession numbers can be found in our data availability statement.

Minor comments:

(1) Figure 1A. The labels are missing for grey and light blue structures.

These structures are together making up the “notochord sheath”, which is comprised of the basal lamina (grey), the medial layer of fibrillar collagen (light blue) and the outer layer of loosely arranged matrix (lighter blue). We modified the figure legend to indicate that the three layers all correspond to the notochord sheath.

(2) Figure 2A. The constructs in the lower part of the panel are not discussed in the legend and seem out of place in terms of data type and analysis.

We would argue that indicating which non-coding regions and which ATAC peaks were responsible for driving GFP expression in each construct aids in a better understanding of our results. We thank the reviewer for pointing out the lack of mention of these constructs in the figure legend. This issue has now been resolved.

(3) Be wary of red/green color combinations, especially in the figures where these are juxtaposed with each other.

We apologise for the use of red/green colour. Although it is not possible for this manuscript to change the colour patterns, we will make sure to avoid the use of these colours in conjunction in the future.

(4) The use of fish as a term should be classified as teleost fish, as authors are not addressing non-teleost basal ray-finned fishes or the fact that tetrapods are within bony fishes overall.

This is well spotted, we have now remedied this by editing the manuscript. Where the term “fish” was used, we now state “teleost fish”.

(5) Age information is missing in several Figures (e.g. 1D and 2C).

In some of the figures space constrains did not allow for including the stage on the figure itself. However, we have made sure that in those cases the stage is incorporated in the figure legend.

(6) The resolution of several Figures (e.g. Figure 5 and Supplementary Figure 3) is low.

We address this issue by providing high resolution figures with the revised manuscript.

(7) In the sentence (top page before Discussion) "The same conclusion was reached upon isolation from these three..", it was unclear what 'upon isolation' referred to.

We agree with the reviewer that this phrasing is unclear. To enhance clarity, the manuscript now reads as follows: “The same conclusion was reached upon isolation of the DEGs highlighted by our RNA-seq results, from the three aforementioned groups of genes associated with ATAC peaks for each cell population.”

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This study investigated the phosphoryl transfer mechanism of the enzyme adenylate kinase, using SCC-DFTB quantum mechanical/molecular mechanical (QM/MM) simulations, along with kinetic studies exploring the temperature and pH dependence of the enzyme's activity, as well as the effects of various active site mutants. Based on a broad free energy landscape near the transition state, the authors proposed the existence of wide transition states (TS), characterized by the transferring phosphoryl group adopting a meta-phosphate-like geometry with asymmetric bond distances to the nucleophilic and leaving oxygens. In support of this finding, kinetic experiments were conducted with Ca2+ ions at different temperatures and pH, which revealed a reduced entropy of activation and unique pH-dependence of the catalyzed reaction.

Strengths:

A combined application of simulation and experiments is a strength.

Weaknesses:

The conclusion that the enzyme-catalyzed reaction involves a wide transition state is not sufficiently clarified with some concerns about the determined free energy profiles compared to the experimental estimate. (See Recommendations for the authors.)

Reviewer #2 (Public Review):

Summary:

The authors report results of QM/MM simulations and kinetic measurements for the phosphoryl-transfer step in adenylate kinase. The main assertion of the paper is that a wide transition state ensemble is a key concept in enzyme catalysis as a strategy to circumvent entropic barriers. This assertion is based on observation of a "structurally wide" set of energetically equivalent configurations that lie along the reaction coordinate in QM/MM simulations, together with kinetic measurements that suggest a decrease of the entropy of activation.

Thank you for your feedback. The reviewer’s questions are answered below, hoping to clarify them.

Strengths:

The study combines theoretical calculations and supporting experiments.

Weaknesses:

The current paper hypothesizes a "wide" transition state ensemble as a catalytic strategy and key concept in enzyme catalysis. Overall, it is not clear the degree to which this hypothesis is fully supported by the data. The reasons are as follows:

(1) Enzyme catalysis reflects a rate enhancement with respect to a baseline reaction in solution. In order to assert that something is part of a catalytic strategy of an enzyme, it would be necessary to demonstrate from simulations that the activation entropy for the baseline reaction is indeed greater and the transition state ensemble less "wide". Alternatively stated, when indicating there is a "wide transition state ensemble" for the enzyme system - one needs to indicate that is with respect to the non-enzymatic reaction. However, these simulations were not performed and the comparisons not demonstrated. The authors state "This chemical step would take about 7000 years without the enzyme" making it impossible to measure; nonetheless, the simulations of the nonenzymatic reaction would be fairly straight forward to perform in order to demonstrate this key concept that is central to the paper. Rather, the authors examine the reaction in the absence of a catalytically important Mg ion.

Thank you for your thoughtful feedback. QM/MM calculations for uncatalysed phosphoryl-transfer reactions involving either diphosphates or triphosphates have been well documented in the literature showing narrow and symmetric TSE (Klan et al., JACS 2006, 128 (47) 15310-15323; Cui Wang et al., JPCB 2015, 119(9), 3720-3726). We added these references to the revised manuscript.

(2) The observation of a "wide conformational ensemble" is not a quantitative measure of entropy. In order to make a meaningful computational prediction of the entropic contribution to the activation free energy, one would need to perform free energy simulations over a range of temperatures (for the enzymatic and non-enzymatic systems). Such simulations were not performed, and the entropy of activation was thus not quantified by the computational predictions. The authors instead use a wider TS ensemble as a proxy for larger entropy, and miss an opportunity to compare directly to the experimental measurements.

Although we share the reviewers desire to quantify entropies from QM/MM simulations, we agree with discussions in the literature that calculating quantitative entropies from performing QM/MM simulations at different temperatures is not reliable. We therefore felt strongly to stay with a qualitative assessment of entropy differences from our simulations. As the reviewer highlighted, our study combines theoretical calculations and experiments. The entropy of activation is well estimated by the experiments from the experimental accuracy of these temperature-dependent changes in rate constants for the chemical step.  Our computational results agree well with the experimental results and were further validated in 2 rounds of reviews/revisions by additional different free energy calculation methods (MSMD and US), plus committor analysis.

Reviewer #3 (Public Review):

Summary:

By conducting QM/MM free energy simulations, the authors aimed to characterize the mechanism and transition state for the phosphoryl transfer in adenylate kinase. The qualitative reliability of the QM/MM results has been supported by several interesting experimental kinetic studies. However, the interpretation of the QM/MM results is not well supported by the current calculations.

Thank you for your feedback. We appreciate the recognition of our experimental validation but understand your concern about the interpretation of our QM/MM results. To address this, we answer the specific questions below and added clearer explanations of the computational approach, including its limitations. We also better aligned the QM/MM results with both experimental data and theoretical expectations to strengthen the overall interpretation.

Strengths:

The QM/MM free energy simulations have been carefully conducted. The accuracy of the semi-empirical QM/MM results was further supported by DFT/MM calculations, as well as qualitatively by several experimental studies.

Weaknesses:

(1) One key issue is the definition of the transition state ensemble. The authors appear to define this by simply considering structures that lie within a given free energy range from the barrier. However, this is not the rigorous definition of transition state ensemble, which should be defined in terms of committor distribution. This is not simply an issue of semantics, since only a rigorous definition allows a fair comparison between different cases - such as the transition state in an enzyme vs in solution, or with and without the metal ion. For a chemical reaction in a complex environment, it is also possible that many other variables (in addition to the breaking and forming P-O bonds) should be considered when one measures the diversity in the conformational ensemble.

In the revised manuscript, the authors included committor analysis. However, the discussion of the result is very brief. In particular, if we use the common definition of the transition state ensemble (TSE) as those featuring the committor around 0.5, the reaction coordinate of the TSE would span a much narrower range than those listed in Table 1. This point should be carefully addressed.

The reviewer is right, the TSE is rigorously defined in terms of the committor distribution. We actually calculated the committor distribution for the reaction with and without Mg. We now added the figure showing the committor distribution for both reactions (Figure 3 – supplement 9). We did not include these results before because the committor distribution histogram would need more points to have a more accurate shape, requiring a high computational cost. We followed the reviewer’s suggestion and updated table 1 with the values from the committor distribution analysis.

(2) While the experimental observation that the activation entropy differs significantly with and without the Ca2+ ion is interesting, it is difficult to connect this result with the "wide" transition state ensemble observed in the QM/MM simulations so far. Even without considering the definition of the transition state ensemble mentioned above, it is unlikely that a broader range of P-O distances would explain the substantial difference in the activation entropy measured in the experiment. Since the difference is sufficiently large, it should be possible to compute the value by repeating the free energy simulations at different temperatures, which would lead to a much more direct evaluation of the QM/MM model/result and the interpretation.

See our answer above about this point.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Major comments:

One of the remaining issues with this revision is the assertion of the wide transition states in the presence of Mg2+ ion. When discussing the transition state of phosphoryl transfer reactions, it is important to consider their nature as involving both the cleavage and formation of P-O bonds. While these two events can occur in concert with a single transition state, many studies have shown that one event often precedes the other. Sometimes, there is a slight drop in free energy between the two events, forming a transient intermediate. However, due to its very short lifetime, this intermediate state may not be detectable experimentally. Depending on the sequence of events, the transition state or the transient intermediate may exhibit characteristics of a metaphosphate or phosphorane-like species. Based on the DFTB simulation results presented in the paper, it appears that the reaction forms a metaphosphate-like transition state. In the present reaction, since the two oxygen atoms involved in the reaction are very good leaving groups with similar reactivity, it is not surprising to observe the two events near the TS with very similar relative free energies, and therefore, the free energy profile can be very flat near the TS. This is consistent with the statement that "the transferring phosphate can be much closer to the leaving oxygen than the attacking oxygen and vice versa" on page 9. In my opinion, however, this should not be considered as a wide transition state but rather a consequence of the two events occurring very close to each other along the reaction coordinate. This distinction can be considered a semantic issue, and as long as the authors clearly discuss this issue and clarify the meaning of the TS ensemble, the reviewer is okay with that. In its current form, the statement of the wide TS ensemble may lead to a misleading interpretation of the reaction under study.

An intermediate is clearly defined as a minimum in the free energy landscape. We see no evidence in any of your simulations of a minimum flanked by two transitions states, nor do we see any evidence in our NMR relaxation data or crystal structure ensemble refinement. We report our experimental and computational results, so that the reader can directly interpret the free energy landscapes for this system, avoiding semantics due to language ambiguity.

Second, based on the kinetic study, the free energy of the catalytic reaction is approximately zero. The authors suggest that at pH near 7, the ADP exists as a roughly

50-50 mixture between the singly protonated and fully charged states and consequently, the reaction free energies between the two scenarios cancel each other out. However, this argument is not correct. If [ADP(H)]/[ADP] is close to 1, the two reaction free energies, one with +6 kcal/mol and the other with -6 kcal/mol, imply that the protonation of the products (either ATP or AMP) requires ~12 kcal/mol (i.e., 9 pKa unit shift). Given the symmetric nature of the reaction and the similar pKa values between ATP, ADP versus AMP, such a large shift in the pKa of the product state is not expected, and for the calculated results to be accurate, the pKa shifts in the reactant state versus the product state must be opposite, with a total relative shift of 9 pKa units. This is difficult to understand given the nature of the reaction catalyzed by the adenylate kinase enzyme.

We thank this reviewer for this question, which made us realize that we cannot compare the free energies of our QM/MD simulations with the experimentally determined ADP and ATP/AMP ratios. In the experiment we determine the entire pool of ADP and AMP/ATP bound to the enzyme, but could not distinguish if the protonated and or nonprotonated states are contributing to the measured observed rate constants (Kerns, S. et al.,(2015). In the present study, we now discovered that the nonprotonated forms have a lower activation barrier, but the protonated states also contribute to the overall reaction. Therefore, we removed this paragraph from our discussion.

Minor comments:

The difference in the free energy barrier determined by the MSMD and umbrella sampling is not negligible. Considering that umbrella sampling is commonly used in this type of research, the MSMD method appears to overestimate the barrier by more than 3 kcal/mol. Would the TS geometries obtained by umbrella sampling be comparable to those obtained by MSMD?

This is an excellent suggestion, since the umbrella sampling is the more accurate method. The TSE from both methods are indeed comparable, and we added new figure panels about this results to Fig. 4.

Figure 5 shows that the enthalpy of activation is similar for reactions with and without Ca2+. Do the authors expect the enthalpy of activation to decrease when Ca2+ is replaced by Mg2+ without a significant change in the entropy of activation? Any justification?

In (Kerns, S. et al.,(2015) we had experimentally determined the dependence of the observed rate of the P-transfer on the nature of the divalent metal, with Mg2+ being by far superior to the other divalent metals. We proposed that this majorly is an effect on the enthalpy of activation, that other divalent metals provide poor orbital overlap, in agreement with published work on P-transfer reactions that show selectivity for a specific metal.

Please provide proper citations for SHAKE and WHAM.

The citations were added.

Reviewer #2 (Recommendations For The Authors):

The authors did not really address in the revised manuscript the main points of the previous review, which included examination of non-enzymatic reaction (via simulation, not measurement) and quantification of the connection between the reported wide TS ensemble and the increase in entropy (by additional simulations). The authors should also add reference to the AM1/d-PhoT model of Nam et al. which is now discussed.

QM/MM calculations for uncatazlysed phosphoryl-transfer reactions involving either diphosphates or triphosphates have been well documented in the literature showing narrow and symmetric TSE (Klahn et al., JACS 2006, 128 (47) 15310-15323; Cui Wang et al., JPCB 2015, 119(9), 3720-3726). We added these references to the revised manuscript.

The reference to AM1/d-PhoT model was added.

Reviewer #3 (Recommendations For The Authors):

In the revised ms, the authors indeed addressed many of the points raised in the previous round of review. In addition to the issue of TSE and committor mentioned above, another point that needs to be carefully explained is the very significant difference between umbrella sampling results and those in Fig. 1C - especially for the case without Mg2+ - the difference of more than 20 kcal/mol is not something that can be ignored at a qualitative level.

We thank the reviewer for pointing out that the difference in free energy profiles between umbrella sampling (US) and MSMD, especially in the case without Mg²+ needs to be addressed.

We believe that the key reason for this difference lies in the methodological approaches of these techniques.

Umbrella sampling is an equilibrium enhanced sampling method, that allows for a balanced and thorough exploration of the free energy landscape, the MSMD is a non-equilibrium method and estimation depends of the averaging scheme used and the number of trajectories. In the present work, the free energy was estimated using an exponential average. This averaging scheme has a slow convergence, small variance and may overestimate the free energy barrier, specially if the barrier as seen in the absence of Mg is quite high. This factor could explain the significant difference between umbrella sampling and MSMD combined with Jarzynski’s equality.

We have added new panels to Fig. 4 to compare the TSE from the more accurate umbrella sampling to the MSMD simulations, buttressing the validity of our original findings. We revised the manuscript discuss the differences between the MSMD and the umbrella sampling free energy profiles.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

The work analyzes how centrosomes mature before cell division. A critical aspect is the accumulation of pericentriolar material (PCM) around the centrioles to build competent centrosomes that can organize the mitotic spindle. The present work builds on the idea that the accumulation of PCM is catalyzed either by the centrioles themselves (leading to a constant accumulation rate) or by enzymes activated by the PCM itself (leading to autocatalytic accumulation). These ideas are captured by a previous model derived for PCM accumulation in C. elegans (ref. 8) and are succinctly summarized by Eq. 1. The main addition of the present work is to allow the activated enzymes to diffuse in the cell, so they can also catalyze the accumulation of PCM in other centrosomes (captured by Eqs. 2-4). The authors claim that this helps centrosomes to reach the same size, independent of potential initial mismatches.

A strength of the paper is the simplicity of the equations, which are reduced to the bare minimum and thus allow a detailed inspection of the physical mechanism. One shortcoming of this approach is that all equations assume that the diffusion of molecules is much faster than any of the reactive time scales, although there is no experimental evidence for this.

We appreciate the reviewer’s recognition of the strengths of our work. Indeed, the centrosome growth model incorporates multiple timescales corresponding to various reactions, and existing experimental data do not directly provide diffusion constants for the cytosolic proteins. However, we can estimate these diffusion constants using protein mass, based on the Stokes-Einstein relation, and compare the diffusion timescales with the reaction timescales obtained from FRAP analysis. For example, we estimate that the diffusion timescale for centrosomes separated by 5-10 micrometers is much smaller than the reaction timescales deduced from the FRAP experiments. Specifically, for SPD-5, a scaffold protein with a mass of ~150 kDa, the estimated diffusion constant is ~17 µm²/s, using the Stokes-Einstein relation and a reference diffusion constant of ~30 µm²/s for a 30 kDa GFP protein (reference: Bionumbers book). This results in a diffusion timescale of ~1 second for centrosomes 10 µm apart. In contrast, FRAP recovery timescales for SPD-5 in C. elegans embryos are on the order of several minutes, suggesting that scaffold protein binding reactions are much slower than diffusion. Therefore, a reaction-limited model is appropriate for studying PCM self-assembly during centrosome maturation. We have revised the manuscript to clarify this point and to include a discussion of the diffusion and reaction timescales.

Spatially extended model with diffusion

Both the reviewers have pointed out the importance of considering diffusion effects in centrosome size dynamics, and we agree that this is important to explore. We have developed a spatially extended 3D version of the centrosome growth model, incorporating stochastic reactions and diffusion (see Appendix 4). In this model, the system is divided into small reaction volumes (voxels), where reactions depend on local density, and diffusion is modeled as the transport of monomers/building blocks between voxels.

We find that diffusion can alter the timescales of growth, particularly when the diffusion timescale is comparable to or slower than the reaction timescale, potentially mitigating size inequality by slowing down autocatalysis. However, the main conclusions of the catalytic growth model remain unchanged, showing robust size regulation independent of diffusion constant or centrosome separation (Figure 2—figure supplement 3). Hence, we focused on the effect of subunit diffusion on the autocatalytic growth model. We find that in the presence of diffusion, the size inequality reduces with increasing diffusion timescale, i.e., increasing distance between centrosomes and decreasing diffusion constant (Figure 2—figure supplement 4). However, the lack of robustness in size control in the autocatalyic growth model remains, i.e., the final size difference increases with increasing initial size difference. Notably, in the diffusion-limited regime (very small diffusion or large distances), the growth curve loses its sigmoidal shape, resembling the behavior in the non-autocatalytic limit (Figure 2). These findings are discussed in the revised manuscript.

Another shortcoming of the paper is that it is not clear what species the authors are investigating and how general the model is. There are huge differences in centrosome maturation and the involved proteins between species. However, this is not mentioned in the abstract or introduction. Moreover, in the main body of the paper, the authors mention C. elegans on pages 2 and 3, but refer to Drosophila on page 4, switching back to C. elegans on page 5, and discuss Drosophila on page 6. This is confusing and looks as if they are cherry-picking elements from various species. The original model in ref. 8 was constructed for C. elegans and it is not clear whether the autocatalytic model is more general than that. In any case, a more thorough discussion of experimental evidence would be helpful.

We believe one strength of our approach is its applicability across organisms. Our goal in comparing the theoretical model with experimental data from C. elegans and D.

melanogaster is to demonstrate that the apparent qualitative differences in centrosome growth across species (see e.g., the extent of size scaling discussed in the section “Cytoplasmic pool depletion regulates centrosome size scaling with cell size”) may arise from the same underlying mechanisms in the theoretical model, albeit with different parameter values. We acknowledge differences in regulatory molecules between species, but the core pathways remain conserved see e.g. Raff, Trends in Cell Biology 2019, section: “Molecular Components of the Mitotic Centrosome Scaffold Appear to Have Been Conserved in Evolution from Worms to Humans”. In the revised manuscript, we have expanded the introduction to clarify this point and explain how our theory applies across species. We have also provided a clearer discussion of the experimental systems used throughout the manuscript and the available experimental evidence.

The authors show convincingly that their model compensates for initial size differences in centrosomes and leads to more similar final sizes. These conclusions rely on numerical simulations, but it is not clear how the parameters listed in Table 1 were chosen and whether they are representative of the real situation. Since all presented models have many parameters, a detailed discussion on how the values were picked is indispensable. Without such a discussion, it is not clear how realistic the drawn conclusions are. Some of this could have been alleviated using a linear stability analysis of the ordinary differential equations from which one could have gotten insight into how the physical parameters affect the tendency to produce equal-sized centrosomes.

Following the suggestion of the reviewer, we have revised the manuscript to add references and discussions justifying the choice of the parameter values used for the numerical simulations. These references and parameter choices can be found in Table 1 and Table 2, and are also discussed in relevant figure captions and within the manuscript text.

We thank the reviewer for the excellent suggestion of including linear stability analysis of the ODE models of centrosome growth. We included linear stability analyses of the catalytic and autocatalytic growth models in Appendix 3. Analysis of the catalytic growth model reaffirms the robustness of size equality and the analysis of autocatalytic growth provides an approximate condition of size inequality. We have modified the revised manuscript to discuss these results.

The authors use the fact that their model stabilizes centrosome size to argue that their model is superior to the previously published one, but I think that this conclusion is not necessarily justified by the presented data. The authors claim that "[...] none of the existing quantitative models can account for robustness in centrosome size equality in the presence of positive feedback." (page 1; similar sentence on page 2). This is not shown convincingly. In fact, ref 8. already addresses this problem (see Fig. 5 in ref. 8) to some extent.

The linear stability analysis shown in Fig 5 in ref 8 (Zwicker et al, PNAS, 2014) shows that the solutions are stable around the fixed point and it was inferred from this result that Ostwald ripening can be suppressed by the catalytic activity of the centriole, therefore stabilizing the centrosomes (droplets) against coarsening by Ostwald ripening. But, if size discrepancy arises from the growth process (e.g., due to autocatalysis) the timescale of relaxation for such discrepancy is not clear from the above-mentioned result. We show (in figure 2 - figure supplement 3) that for any appreciable amount of positive feedback, the solution moves very slowly around the fixed point (almost like a line attractor) and cannot reach the fixed point in a biologically relevant timescale. Hence the model in ref 8 does not provide a robust mechanism for size control in the presence of autocatalytic growth. We have added this discussion in the Discussion section.

More importantly, the conclusion seems to largely be based on the analysis shown in Fig. 2A, but the parameters going into this figure are not clear (see the previous paragraph). In particular, the initial size discrepancy of 0.1 µm^3 seems quite large, since it translates to a sphere of a radius of 300 nm. A similarly large initial discrepancy is used on page 3 without any justification. Since the original model itself already showed size stability, a careful quantitative comparison would be necessary.

We thank the reviewer for the valuable suggestions. The parameters used in Fig. 2A are listed in Table 1 with corresponding references, and we used the parameter values from Zwicker et al. (2014) for rate constants and concentrations.

The issue of initial size differences between centrosomes is important, but quantitative data on this are not readily available for C. elegans and Drosophila. Centrosomes may differ initially due to disparities in the amount and incorporation rate of PCM between the mother and daughter centrioles. Based on available images and videos (Cabral et al, Dev. Cell, 2019, DOI: https://doi.org/10.1016/j.devcel.2019.06.004), we estimated an initial radius of ~0.5 μm for centrosomes. Accounting for a 5% radius difference would lead to a volume difference of ~0.1 μm³, which was used in our analysis (Fig. 2A). These differences likely arise from distinct growth conditions of centrosomes containing different centrioles (older mother and newer daughter).

More importantly, we emphasize that the initial size difference does not qualitatively alter the results presented in Figure 2. We agree that a quantitative analysis will further clarify our conclusions, and we have revised the manuscript accordingly. For example, Figure 2—figure supplement 3 provides a detailed analysis of how the final centrosome size depends on initial size differences across various parameter values. Additionally, Appendix 3 now includes analytical estimates of the onset of size inequality as a function of these parameters.

The analysis of the size discrepancy relies on stochastic simulations (e.g., mentioned on pages 2 and 4), but all presented equations are deterministic. It's unclear what assumptions go into these stochastic equations, and how they are analyzed or simulated. Most importantly, the noise strength (presumably linked to the number of components) needs to be mentioned. How is this noise strength determined? What are the arguments for this choice? This is particularly crucial since the authors quote quantitative results (e.g., "a negligible difference in steady-state size (∼ 2% of mean size)" on page 4).

As described in the Methods, we used the exact Gillespie method (Gillespie, JPC, 1977) to simulate the evolution of the stochastic trajectories of the systems, corresponding to the deterministic growth and reaction kinetics outlined in the manuscript. We've expanded the Methods to include further details on the stochastic simulations and refer to Appendix 1, where we describe the chemical master equations governing autocatalytic growth..

The noise strength (fluctuations about the mean size of centrosome) does depend on the total monomer concentration (the pool size), and this may affect size inequality. Similar values of the total monomer concentration were used in the catalytic (0.04 uM) and autocatalytic growth (0.33 uM) simulations. These values for the pool size are similar to previous studies (Zwicker et al, PNAS, 2012) and have been optimized to obtain a good fit with experimental growth curves from C. elegans embryo data.

To present more quantitative results, we have revised our manuscript to add data showing the effect of pool size on centrosome size inequality (Figure 3 - figure supplement 2). We find the size inequality in catalytic growth to increase with decreasing pool size as the origin of this inequality is the stochastic fluctuation in individual centrosome size. The size inequality (ratio of dv/<V>) in the autocatalytic growth does not depend (strongly) on the pool size (dv and <V> both increase similarly with pool size).

Moreover, the two sets of testable predictions that are offered at the end of the paper are not very illuminative: The first set of predictions, namely that the model would anticipate an "increase in centrosome size with increasing enzyme concentration, the ability to modify the shape of the sigmoidal growth curve, and the manipulation of centrosome size scaling patterns by perturbing growth rate constants or enzyme concentrations.", are so general that they apply to all models describing centrosome growth. Consequently, these observations do not set the shared enzyme pool apart and are thus not useful to discriminate between models. The second part of the first set of predictions about shifting "size scaling" is potentially more interesting, although I could not discern whether "size scaling" referred to scaling with cell size, total amount of material, or enzymatic activity at the centrioles. The second prediction is potentially also interesting and could be checked directly by analyzing published data of the original model (see Fig. 5 of ref. 8). It is unclear to me why the authors did not attempt this.

In response to the reviewers' valuable feedback, we have revised the manuscript to include results on potential methods for distinguishing catalytic growth from autocatalytic growth. Since the growth dynamics of a single centrosome do not significantly differ between these two models, it is necessary to experimentally examine the growth dynamics of a centrosome pair under various initial size perturbations. In Figure 3-figure supplement 2, we present theoretical predictions for both catalytic and autocatalytic growth models, illustrating the correlation between initial and final sizes after maturation. The figure demonstrates that the initial size difference and final size difference should be correlated only in the autocatalytic growth and the relative size inequality decreases with increasing subunit pool size in catalytic growth while remains almost unchanged in autocatalytic growth. These predictions can be experimentally examined by inducing varying centrosome sizes at the early stage of maturation for different expression levels of the scaffold former proteins.

A second experimentally testable feature of the catalytic growth model involves sharing of the enzyme between both centrosomes. This could be tested through immunofluorescent staining of the kinase or by constructing a FRET reporter for PLK1 activity, where it can be studied if the active form of the PLK1 is found in the cytoplasm around the centrosomes indicating a shared pool of active enzyme. Additionally, photoactivated localization microscopy could be employed, where fluorescently tagged enzyme can be selectively photoactivated in one centrosome and intensity can be measured at the other centrosome to find the extent of enzyme sharing between the centrosomes.

We also discuss shifts in centrosome size scaling behavior with cell size by varying parameters of the catalytic growth model (Fig 4). While quantitative analysis of size scaling in Drosophila is currently unavailable, such an investigation could enable us to distinguish catalytic growth mode with other models. We have included this point in the Discussion section.

“The second prediction is potentially also interesting …” We assume the reviewer is referencing the scenario in Zwicker et al. (ref 8), where differences in centriole activity lead to unequal centrosome sizes. The data in that study represent a case of centrosome growth with variable centriole activity, resulting in size differences in both autocatalytic and catalytic growth models. This differs from our proposed experiment, where we induce unequal centrosome sizes without modifying centriole activity. We have now revised the text to clarify this distinction.

Taken together, I think the shared enzyme pool is an interesting idea, but the experimental evidence for it is currently lacking. Moreover, the model seems to make little testable predictions that differ from previous models.

We appreciate the reviewer’s interest in the core idea of our work. As mentioned earlier, we have improved the clarity in model predictions in the revised discussion section. Unfortunately, the lack of publicly available experimental data limits our ability to provide more direct experimental evidence. However, we are hopeful that our theoretical model will inspire future experiments to test these model predictions.

Reviewer #2 (Public Review):

Summary:

In this paper, Banerjee & Banerjee argue that a solely autocatalytic assembly model of the centrosome leads to size inequality. The authors instead propose a catalytic growth model with a shared enzyme pool. Using this model, the authors predict that size control is enzyme-mediate and are able to reproduce various experimental results such as centrosome size scaling with cell size and centrosome growth curves in C. elegans.

The paper contains interesting results and is well-written and easy to follow/understand.

We are delighted that the reviewer finds our work interesting, and we appreciate the thoughtful suggestions provided. In response, we have revised the text and figures to incorporate these recommendations. Below, we address each of the reviewer’s comments point by point:

Suggestions:

● In the Introduction, when the authors mention that their "theory is based on recent experiments uncovering the interactions of the molecular components of centrosome assembly" it would be useful to mention what particular interactions these are.

As the reviewer suggested, we have modified the introduction section to add the experimental observations upon which we build our model.

● In the Results and Discussion sections, the authors note various similarities and differences between what is known regarding centrosome formation in C. elegan and Drosophila. It would have been helpful to already make such distinctions in the Introduction (where some phenomena that may be C. elegans specific are implied to hold centrosomes universally). It would also be helpful to include more comments for the possible implications for other systems in which centrosomes have been studied, such as human, Zebrafish, and Xenopus.

We thank the reviewer for this suggestion. We have modified the Introduction to motivate the comparative study of centrosome growth in different organisms and draw relevant connections to centrosome growth in other commonly studied organisms like Zebrafish and Xenopus.

● For Fig 1.C, the two axes are very close to being the same but are not. It makes the graph a little bit more difficult to interpret than if they were actually the same or distinctly different. It would be more useful to have them on the same scale and just have a legend.

We have modified the Figure 1C in the revised manuscript. The plot now shows the growth of a single and a pair of centrosomes both on the same y-axis scale.

● The authors refer to Equation 1 as resulting from an "active liquid-liquid phase separation", but it is unclear what that means in this context because the rheology of the centrosome does not appear to be relevant.

We used the term “active liquid-liquid phase separation” simply to refer to a previous model proposed by Zwicker et al (PNAS, 2014) where the underlying process of growth results from liquid-liquid phase separation. We agree with the reviewer that the rheological property of the centrosome is not very relevant in our discussions and we have thus removed the sentence from the revised manuscript to avoid any confusion.

● The authors reject the non-cooperative limit of Eq 1 because, even though it leads to size control, it does not give sigmoidal dynamics (Figure 2B). While I appreciate that this is just meant to be illustrative, I still find it to be a weak argument because I would guess a number of different minor tweaks to the model might keep size control while inducing sigmoidal dynamics, such as size-dependent addition of loss rates (which could be due to reactions happen on the surface of the centrosome instead of in its bulk, for example). Is my intuition incorrect? Is there an alternative reason to reject such possible modifications?

The reviewer raises an interesting point here. However, we disagree with the idea that minor adjustments to the model can produce sigmoidal growth curves while still maintaining size control. In the absence of an external, time-dependent increase in building block concentration (which would lead to an increasing growth rate), achieving sigmoidal growth requires a positive feedback mechanism in the growth rate. This positive feedback alone could introduce size inequality unless shared equally between the centrosomes, as it is in our model of catalytic growth in a shared enzyme pool. The proposed modification involving size-dependent addition or loss rates due to surface assembly/disassembly may result in unequal sizes precisely because of this positive feedback. A similar example is provided in Appendix 1, where assembly and disassembly across the pericentriolic material volume lead to sigmoidal growth but also generate significant size inequality and lack of robustness in size control.

● While the inset of Figure 3D is visually convincing, it would be good to include a statistical test for completeness.

Following the reviewer’s suggestion, we present a statistical analysis in Figure 3 - Figure supplement 2 in the modified manuscript to enhance clarity. We show that the size difference values are uncorrelated (Pearson’s correlation coefficient ~ 0) with the initial size difference indicating the robustness of the size regulation mechanism.

● The authors note that the pulse in active enzyme in their model is reminiscent of the Polo kinase pulse observed in Drosophila. Can the authors use these published experimental results to more tightly constrain what parameter regime in their model would be relevant for Drosophila? Can the authors make predictions of how this pulse might vary in other systems such as C. elegans?

Thank you for the insightful suggestion regarding the use of pulse dynamics in experiments to better constrain the model’s parameter regime. In our revised manuscript, we attempted this analysis; however, the data from Wong et al. (EMBO 2022) for Drosophila are presented as normalized intensity in arbitrary units, rather than as quantitative measures of centrosome size or Polo enzyme concentration. This lack of quantitative data limits our ability to benchmark the model beyond capturing qualitative trends. We thus believe that quantitative measurements of centrosome size and enzyme concentration are necessary to achieve a tighter alignment between model predictions and biological data.

We discuss the enzyme dynamics in C. elegans in the revised manuscript. We find the enzyme dynamics corresponding to the fitted growth curves of C. elegans centrosomes are distinctly different from the ones observed in Drosophila. Instead of the pulse-like feature, we find a step-like increase in (cytosolic) active enzyme concentration.

● The authors mention that the shared enzyme pool is likely not diffusion-limited in C. elegans embryos, but this might change in larger embryos such as Drosophila or Xenopus. It would be interesting for the authors to include a more in-depth discussion of when diffusion will or will not matter, and what the consequence of being in a diffusion-limit regime might be.

Both the reviewers have pointed out the importance of considering diffusion effects in centrosome size dynamics, and we agree that this is important to explore. We have developed a spatially extended 3D version of the centrosome growth model, incorporating stochastic reactions and diffusion (see Appendix 4). In this model, the system is divided into small reaction volumes (voxels), where reactions depend on local density, and diffusion is modeled as the transport of monomers/building blocks between voxels.

We find that diffusion can alter the timescales of growth, particularly when the diffusion timescale is comparable to or slower than the reaction timescale, potentially mitigating size inequality by slowing down autocatalysis. However, the main conclusions of the catalytic growth model remain unchanged, showing robust size regulation independent of diffusion constant or centrosome separation (Figure 2—figure supplement 3). Hence, we focused on the effect of subunit diffusion on the autocatalytic growth model. We find that in the presence of diffusion, the size inequality reduces with increasing diffusion timescale, i.e., increasing distance between centrosomes and decreasing diffusion constant (Figure 2—figure supplement 4). However, the lack of robustness in size control in the autocatalyic growth model remains, i.e., the final size difference increases with increasing initial size difference. Notably, in the diffusion-limited regime (very small diffusion or large distances), the growth curve loses its sigmoidal shape, resembling the behavior in the non-autocatalytic limit (Figure 2). These findings are discussed in the revised manuscript.

● The authors state "Firstly, our model posits the sharing of the enzyme between both centrosomes. This hypothesis can potentially be experimentally tested through immunofluorescent staining of the kinase or by constructing FRET reporter of PLK1 activity." I don't understand how such experiments would be helpful for determining if enzymes are shared between the two centrosomes. It would be helpful for the authors to elaborate.

Our results indicate the necessity of the centrosome-activated enzyme to be shared for the robust regulation of centrosome size equality. If a FRET reporter of the active form of the enzyme (e.g., PLK1) can be constructed then the localization of the active form of the enzyme may be determined in the cytosol. We propose this based on reports of studying PLK activities in subcellular compartments using FRET as described in Allen & Zhang, BBRC (2006). Such experiments will be a direct proof of the shared enzyme pool. Following the reviewer’s suggestion, we have modified the description of the FRET based possible experimental test for the shared enzyme pool hypothesis in the revised manuscript.

Additionally, we have added another possible experimental test based on photoactivated localization microscopy (PALM), where tagged enzyme can be selectively photoactivated in one centrosome and intensity measured at the other centrosome to indicate whether the enzyme is shared between the centrosomes.

Recommendations for the authors:

The manuscript needs to clarify better what species the model describes, how alternative models were rejected, and how the parameters were chosen.

In the revised manuscript, we have connect the chemical species in our model to those documented in organisms like Drosophila and C. elegans. This connection is detailed in the main text under the Catalytic Growth Model section and summarized in Table 2. We discuss alternative models and our reasons for excluding them in the first results section on autocatalytic growth, with additional details provided in Appendix 1 and the accompanying supplementary figures. The selection of model parameters is addressed in the main text and methods, with references listed in Table 1. We believe that these revisions, along with our point-by-point responses to reviewer comments, comprehensively address all reviewer concerns.

Reviewer #1 (Recommendations For The Authors):

I think the style and structure of the paper could be improved on at least two accounts:

(1) What's the role of the last section ("Multi-component centrosome model reveals the utility of shared catalysis on centrosome size control.")? It seems to simply add another component, keeping the essential structure of the model untouched. Not surprisingly, the qualitative features of the model are preserved and quantitative features are not discussed anyway.

This model provides a more realistic description of centrosome growth by incorporating the dynamics of the two primary scaffold-forming subunits and their interactions with an enzyme. It is based on the observation that the major interaction pathways among centrosome components are conserved across many organisms (see Raff, Trends in Cell Biology, 2019 and Table 2), typically involving two scaffold-forming proteins and one enzyme that mediates positive feedback between them. These pathways may involve homologous proteins in different species.

This model allows us to validate the experimentally observed spatial spread of the two subunits, Cnn and Spd-2, in Drosophila. Additionally, we used it to investigate the impact of relaxing the assumption of a shared enzyme pool on size control. Although similar insights could be obtained using a single-component model, the two-component model offers a more biologically relevant framework. We have highlighted these points in the revised manuscript to ensure clarity.

(2 ) The very long discussion section is not very helpful. First, it mostly reiterates points already made in the main text. Second, it makes arguments for the choice of modeling (top left column of page 8), which probably should have been made when introducing the model. Third, it introduces new results (lower left column of page 8), which should probably be moved to the main text. Fourth, the interpretation of the model in light of the known biochemistry is useful and should probably be expanded although I think it would be crucial to keep information from different organisms clearly separate (this last point actually holds for the entire manuscript).

We thank the reviewer for the feedback. We have modified the discussion section to focus more on the interpretation of the results, model predictions and future outlook with possible experiments to validate crucial aspects of the model. We have moved most of the justifications to the main text model description.

Here are a few additional minor points:

* page 1: Typo "for for" → "for"

* Page 8: Typo "to to" → "to"

We thank the reviewer for the useful recommendations. We have corrected all the typos in the revised manuscript.

* Why can diffusion be neglected in Eq. 1? This is discussed only very vaguely in the main text (on page 3). Strangely, there is some discussion of this crucial initial step in the discussion section, although the diffusion time of PLK1 is compared to the centrosome growth time there and not the more relevant enzyme-mediate conversion rate or enzyme deactivation rate.

We now discuss the justification of neglecting diffusion while motivating the model. We have added a more detailed discussion in the Methods section. We estimate the timescale of diffusion for the scaffold formers and the enzyme and compare them with the turnover timescales of the respective proteins Spd-2, Cnn and Polo. We find the proteins to diffuse fast compared to their FRAP recovery timescales indicating reaction timescales to be slower than the timescales of diffusion. Nevertheless, following the reviewer’s suggestion, we have also investigated the effect of diffusion on the growth process in Appendix 4.

* Page 3: The comparison k_0^+ ≫ k_1^+ is meaningless without specifying the number of subunits n. I even doubt that this condition is the correct one since even if k_0^+ is two orders of magnitude larger than k_1^+, the autocatalytic term can dominate if there are many subunits.

We thank the reviewer for the insightful comment on the comparison between the growth rates k^+_0 and k^+_1. Indeed, the pool size matters and we have now included a linear stability analysis of the autocatalytic growth equations in Appendix 3 to estimate the condition for size inequality. We have commented on these new findings in the revised manuscript.

* The Eqs. 2-4 are difficult to follow in my mind. For instance, it is not clear why the variables N_av and N_av^E are introduced when they evidently are equivalent to S_1 and E. It would also help to explicitly mention that V_c is the cell volume. Moreover, do these equations contain any centriolar activity? If so, I could not understand what term mediates this. If not, it might be good to mention this explicitly.

Following the reviewer’s suggestion, we have modified the equations 2-4 and added the definition of V_c to enhance clarity in the revised manuscript. The centriole activity is given by k^+ in the catalytic model. We now explicitly mention it.

* Page 4: The observed peak of active enzyme (Fig 3C) is compared to experimental observation of a PLK1 peak at centrosomes in Drosophila (ref. 28). However, if I understand correctly, the peak in the model refers to active enzyme in the entire cell (and the point of the model is that this enzymatic pool is shared everywhere), whereas the experimental measurement quantified the amount of PLK1 at the centrosome (and not the activity of the enzyme). How are the quantity in the model related to the experimental measurements?

The reviewer is correct in pointing out the difference between the quantities calculated from our model and those measured in the experiment by Wong et al. We have clarified this point in the revised manuscript. We hypothesize that if, in future experiments, the active (phosphorylated) polo can be observed by using a possible FRET reporter of activity then the cytosolic pulse can be observed too. We discuss this point in the revised manuscript.

* Page 6: The asymmetry due to differences in centriolar activity is apparently been done for both models (Eq. 1 and Eqs. 2-4), referring to a parameter k_0^+ in both cases. How does this parameter enter in the latter model? More generally, I don't really understand the difference in the two rows in Fig. 5 - is the top row referring to growth driven by centriolar activity while the lower row refers to pure autocatalytic growth? If so, what about the hybrid model where both mechanisms enter? This is particularly relevant, since ref. 8 claims that such a hybrid model explains growth curves of asymmetric centrosomes quantitatively. Along these lines, the analysis of asymmetric growth is quite vague and at most qualitative. Can the models also explain differential growth quantitatively?

We believe the reviewer’s comment on centrosome size asymmetry may stem from a lack of clarity in our initial explanation. In this section, as shown in Figure 5, we compare the full autocatalytic model (where both k_0^+ and k_1^+ are non-zero) with the catalytic model. The confusion might have arisen due to an unclear definition of centriolar activity in the catalytic growth model, which we have clarified in the revised manuscript. Specifically, we use k+ in the catalytic model and k0+ in the autocatalytic model as indicators of centriolar activity.

Our findings quantitatively demonstrate that variations in centriole activity can robustly drive size asymmetry in catalytic growth, independent of initial size differences. However, in autocatalytic growth, increased initial size differences make the system more vulnerable to a loss of regulation, as positive feedback can amplify these differences, ultimately influencing the final size asymmetry. Our results do not contradict Zwicker et al. (ref 8); rather, they complement it. We show that size asymmetry in autocatalytic growth is governed by both centriole activity and positive feedback, highlighting that centriole activity alone cannot robustly regulate centrosome size asymmetry within this framework.

* The code for performing the simulations does not seem to be available

We have now made the main codes available in a GitHub repository. Link: https://github.com/BanerjeeLab/Centrosome_growth_model

Author response:

We thank the reviewers for their constructive comments. While we work on a revision that addresses all points raised, we would already like to point out that both reviewers seem to have misunderstood how we reported the percentages of filament types in our reactions. Because we included all picked images in our calculations (including false positives from the picking, as well as damaged, overlapping or otherwise unsuitable filaments), we may have inadvertently given the impression that these filament preparations are not pure. In fact, the opposite is true: 0N3R PAD12 tau and the mixture of 0N3R:0N4R PAD12 tau assemble into highly pure paired helical filaments with the Alzheimer fold. Discarding images is common practice for high-resolution cryo-EM structure determination. Our reported percentages of discarded images (20-30%) are much lower than in typical cryo-EM studies, which is another reflection of the high quality of these samples. The main impurity lies in smaller fractions (~10%) of single protofilaments with the Alzheimer fold. We will make this clearer in our revised manuscript.

Author response:

(1) discuss the non-native properties of ROCKET and compare CDL binding in native proteins

ROCKET is indeed a non-native protein with exceptional stability, which makes it immune to mutations with subtle effects on structure or dynamics. We would argue that this is an advantage, allowing us to find the features with the most pronounced impact on CDL-mediated stability. The reviewers are right that there certainly are other structural features which impact CDL binding, which cannot be investigated using ROCKET. This is the reason we then apply our findings to GlpG - to translate back to native systems.

The CDL binding site geometry that we tested experimentally was derived by Corey et al (Sci Adv 2022) from large-scale computational analysis of native protein structures. Our data adds some basic rules for flexibility, which helped us to identify GlpG as a potentially CDL-regulated protein. Following the reviewers’ suggestion, we will screen the dataset from Corey et al. for experimentally confirmed examples of CDL-mediated stabilization and analyze whether they conform to the rules derived from analysis of ROCKET. In this way, we may be able to assess how general our findings are.

(2) clarify the limitations of combining MS and nMS

The reviewers correctly point out that there are differences between the MD and MS data: although the binding Site 1 has nearly 100% occupancy in MD, MS shows that ca 50% of the protein is CDL-free and that not all subunits in the tetramer have a CDL bound. Furthermore, MD shows that aromatic residues are important, but this is not tested by MS. Both points relate to the shortcomings of nMS, which requires desolvation, ionization, and detergent stripping to detect protein-lipid complexes. These processes can potentially affect lipid binding, e.g. by leading to loss of lipids that are not tightly bound. As a result, absolute quantitative comparisons between MD and MS are challenging, and contributions from subtle non-electrostatic interactions involving aromatic residues are difficult to detect. For this reason, we use relative changes in lipid interactions between different ROCKET variants to compare MD and MS data. We will discuss these factors in the revision.

(3) more detailed investigation of the structure-function relationship in GlpG-CDL complexes

We use the insights from ROCKET to identify a stabilizing CDL site in GlpG and find that CDL binding switches substrate preference from transmembrane to soluble substrates. We do not verify the binding site with mutagenesis in our study, but the MD and MS data are very unambiguous that there is only one site, and its location provides a rationale for how CDL affects substrate binding, which is described in the supplementary data.

We agree that the regulatory effect of CDL on GlpG activity raises a wide range of interesting questions relating to the mechanism of allosteric inhibition, the evolutionary background, and biological implications of E. coli using changes in membrane CDL content to steer GlpG activity. Work in our labs is on-going to investigate this further, including the mutational analysis suggested by the reviewers, but it moves beyond of the scope of the current study. We will discuss our rationale for the absence of mutagenesis data in the revision.

Author response:

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

Reviewer 1:

We thank the reviewer for the time and effort in providing very useful comments and suggestions for our manuscript.

(1) The results do not support the conclusions. The main "selling point" as summarized in the title is that the apoptotic rate of zebrafish motorneurons during development is strikingly low (~2% ) as compared to the much higher estimate (~50%) by previous studies in other systems. The results used to support the conclusion are that only a small percentage (under 2%) of apoptotic cells were found over a large population at a variety of stages 24-120hpf. This is fundamentally flawed logic, as a short-time window measure of percentage cannot represent the percentage in the long term. For example, at any year under 1% of the human population dies, but over 100 years >99% of the starting group will have died. To find the real percentage of motorneurons that died, the motorneurons born at different times must be tracked over the long term or the new motorneuron birth rate must be estimated. A similar argument can be applied to the macrophage results. Here the authors probably want to discuss well-established mechanisms of apoptotic neuron clearance such as by glia and microglia cells.

We chose the time window of 24-120 hpf based on the following two reasons: 1) Previous studies showed that although the time windows of motor neuron death vary in chick (E5-E10), mouse (E11.5-E15.5), rat (E15-E18), and human (11-25 weeks of gestation), the common feature of these time windows is that they are all the developmental periods when motor neurons contact with muscle cells. The contact between zebrafish motor neurons and muscle cells occurs before 72 hpf, which is included in our observation time window of 24-120 hpf. 2) Zebrafish complete hatching during 48-72 hpf, and most organs form before 72 hpf. More importantly, zebrafish start swimming around 72 hpf, indicating that motor neurons are fully functional at 72 hpf. Thus, we are confident that this 24-120 hpf time window covers the time window during which motor neurons undergo programmed cell death during zebrafish early development. We have added this information to the revised manuscript.

We frequently used “early development” in this manuscript to describe our observation. However, we missed “early” in our title. We therefore have added this ket word of “early” in the title in the revised manuscript.

Previous studies in zebrafish have shown that the production of spinal cord motor neurons largely ceases before 48 hpf, and then the motor neurons remain largely constant until adulthood (doi: 10.1016/j.celrep.2015.09.050; 10.1016/j.devcel.2013.04.012; 10.1007/BF00304606; 10.3389/fcell.2021.640414). Our observation time window covers the major motor neuron production process. Therefore, we believe that neurogenesis will not affect our findings and conclusions.

We discussed the engulfment of dead motor neurons by other types of cells in the discussion section.

(2) The transgenic line is perhaps the most meaningful contribution to the field as the work stands. However, the mnx1 promoter is well known for its non-specific activation - while the images suggest the authors' line is good, motor neuron markers should be used to validate the line. This is especially important for assessing this population later as mnx1 may be turned off in mature neurons.

The mnx1 promoter has been widely used to label motor neurons in transgenic zebrafish. Previous studies have shown that most of the cells labeled in the mnx1 transgenic zebrafish are motor neurons. In this study, we observed that the neuronal cells in our sensor zebrafish formed green cell bodies inside of the spinal cord and extended to the muscle region, which is an important morphological feature of the motor neurons.

Reviewer 2:

We thank the reviewer for the time and effort in making very useful comments and suggestions for our manuscript.

The FRET-based programmed cell death biosensor described in this manuscript could be very useful. However, the authors have not considered what is already known about the development and programmed cell death of zebrafish spinal motor neurons, and potential differences between motor neuron populations innervating different types of muscles in different vertebrate models. Without this context, the application of their new biosensor tool does not provide new insights into zebrafish motor neuron programmed cell death. In addition, the authors have not carried out controls to show the efficacy and specificity of their morpholinos. Nor have they described how they counted dying motor neurons, or why they chose the specific developmental time points they addressed. These issues are addressed more specifically below.

(1) Lines 12-13: Previous studies in zebrafish showed death of identified spinal motor neurons.

Line 103: In Figure 2A the cell body in the middle is that of identified motor neuron VaP. VaP death has previously been described in several publications. The cell body on the right of the same panel appears to belong to an interneuron whose axon can be seen extending off to the left in one of the rostrocaudal axon bundles that traverse the spinal cord. Higher-resolution imaging would clarify this.

Lines 163-164: Is this the absolute number of motor neurons that died? How were the counts done? Were all the motor neurons in every segment counted? There are approximately 30 identifiable VaP motor neurons in each embryo and they have previously been reported to die between 24-36 hpf. So this analysis is likely capturing those cells.

Our study examined the overall motor neuron apoptosis rather than a specific type of motor neuron death, so we did not emphasize the death of VaP motor neurons. We agree that the dead motor neurons observed in our manuscript contain VaP motor neurons. However, there were also other types of dead motor neurons observed in our study. The reasons are as follows: 1) VaP primary motor neurons die before 36 hpf, but our study found motor neuron cells died after 36 hpf and even at 84 hpf (revised Figure 4A). 2) The position of the VaP motor neuron is together with that of the CaP motor neuron, that is, at the caudal region of the motor neuron cluster. Although it’s rare, we did observe the death of motor neurons in the rostral region of the motor neuron cluster (revised Figure 2C). 3) There is only one or zero VaP motor neuron in each motor neuron cluster. Although our data showed that usually one motor neuron died in each motor neuron cluster, we did observe that sometimes more than one motor neuron died in the motor neuron cluster (revised Figure 2C). We included this information in the revised discussion.

(2) Lines 82-83: It is published that mnx1 is expressed in at least one type of spinal interneuron derived from the same embryonic domain as motor neurons.

The mnx1 promoter has been widely used to label motor neurons in transgenic zebrafish. Previous studies have shown that most of the cells labeled in the mnx1 transgenic zebrafish are motor neurons. In this study, we observed that the neuronal cells in our sensor zebrafish formed green cell bodies inside of the spinal cord and extended to the muscle region, which is an important morphological feature of the motor neurons.

Furthermore, a few of those green cell bodies turned into blue apoptotic bodies inside the spinal cord and changed to blue axons in the muscle regions at the same time, which strongly suggests that those apoptotic neurons are not interneurons. Although the mnx1 promoter might have labeled some interneurons, this will not affect our major finding that only a small portion of motor neurons died during zebrafish early development.

(3) Lines 161-162: Although this may be the major time window of neurogenesis, there are many more motor neurons in adults than in larvae. Neither of these references describes the increase in motor neuron numbers over this particular time span, so the rationale for this choice is unclear.

Lines 168-171: It is known that later developing motor neurons are still being generated in the spinal cord at this time, suggesting that if there is a period of programmed cell death similar to that described in chick and mouse, it would likely occur later. In addition, most of the chick and mouse studies were performed on limb-innervating motor neurons, rather than the body wall muscle-innervating motor neurons examined here.

Lines 237-238: Especially since new motor neurons are still being generated at this time.

Previous studies have shown that the production of spinal cord motor neurons largely ceases before 48 hpf in zebrafish, and then the motor neurons remain largely constant until the adulthood (doi: 10.1016/j.celrep.2015.09.050; 10.1016/j.devcel.2013.04.012; 10.1007/BF00304606; 10.3389/fcell.2021.640414). Our observation time window covers the major motor neuron production process. Therefore, we believe that neurogenesis will not affect our data and conclusions.

The death of motor neurons in limb-innervating motor neurons has been extensively studied in chicks and rodents, as it is easy to undergo operations such as amputation. However, previous studies have shown this dramatic motor neuron death does not only occur in limb-innervating motor neurons but also occurs in other spinal cord motor neurons (doi: 10.1006/dbio.1999.9413). In our manuscript, we studied the naturally occurring motor neuron death in the whole spinal cord during the early stage of zebrafish development.

(4) Lines 184-187: Previous publications showed that death of VaP is independent of limitations in muscle innervation area, suggesting it is not coupled to muscle-derived neurotrophic factors.

Lines 328-334: There have been many publications describing appropriate morpholino controls. The authors need to describe their controls and show that they know that the genes they were targeting were downregulated.

For the morpholinos, we did not confirm the downregulation of the target genes. These morpholino-related data are a minor part of our manuscript and shall not affect our major findings. We have removed the neurotrophic factors and morpholino-related data in the revised manuscript.

Author response:

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study demonstrates the significant role of secretory leukocyte protease inhibitor (SLPI) in regulating B. burgdorferi-induced periarticular inflammation in mice. They found that SLPI-deficient mice showed significantly higher B. burgdorferi infection burden in ankle joints compared to wild-type controls. This increased infection was accompanied by infiltration of neutrophils and macrophages in periarticular tissues, suggesting SLPI's role in immune regulation. The authors strengthened their findings by demonstrating a direct interaction between SLPI and B. burgdorferi through BASEHIT library screening and FACS analysis. Further investigation of SLPI as a target could lead to valuable clinical applications.

The conclusions of this paper are mostly well supported by data, but two aspects need attention:

(1) Cytokine Analysis:

The serum cytokine/chemokine profile analysis appears without TNF-alpha data. Given TNF-alpha's established role in inflammatory responses, comparing its levels between wild-type and infected B. burgdorferi conditions would provide valuable insight into the inflammatory mechanism.

(2) Sample Size Concerns:

While the authors note limitations in obtaining Lyme disease patient samples, the control group is notably smaller than the patient group. This imbalance should either be addressed by including additional healthy controls or explicitly justified in the methodology section.

We thank the reviewer for the careful review and positive comments.

(1) We did look into the level of TNF-alpha in both WT and SLPI-/- mice with and without B. burgdorferi infection. At serum level, using ELISA, we did not observe any significant difference between all four groups. At gene expression level, using RT-qPCR on the tibiotarsal tissue, we also did not observe any significant differences. Our RT-qPCR result is consistent with the previous microarray study using the whole murine joint tissue (DOI: 10.4049/jimmunol.177.11.7930). The microarray study did not show significant changes in TNF-alpha level in C57BL/6 mice following B. burgdorferi infection. The above data suggest that TNF-alpha does not involve in SLPI-regulated immune responses in the murine tibiotarsal tissue following B. burgdorferi infection. A brief discussion will be added, and the above data will be provided as a supplemental figure in the revised manuscript.

(2) We agree with the reviewer that the control group is smaller than the patient group. Among the archived samples that are available, the number of adult healthy controls are limited. It has been shown that the serum level of SLPI in healthy volunteers is in average about 40 ng/ml  (DOI: 10.3389/fimmu.2019.00664 and 10.1097/00003246-200005000-00003). The median level in the healthy control in our data was 38.92 ng/ml, which is comparable to the previous results. A brief discussion will be added in the revised manuscript.

Reviewer #2 (Public review):

Summary:

This manuscript by Yu and coworkers investigates the potential role of Secretory leukocyte protease inhibitor (SLPI) in Lyme arthritis. They show that, after needle inoculation of the Lyme disease (LD) agent, B. burgdorferi, compared to wild type mice, a SLPI-deficient mouse suffers elevated bacterial burden, joint swelling and inflammation, pro-inflammatory cytokines in the joint, and levels of serum neutrophil elastase (NE). They suggest that SLPI levels of Lyme disease patients are diminished relative to healthy controls. Finally, they find that SLPI may interact directly the B. burgdorferi.

Strengths:

Many of these observations are interesting and the use of SLPI-deficient mice is useful (and has not previously been done).

We appreciate the reviewer’s careful reading and positive comments.

Weaknesses:

(a) The known role of SLPI in dampening inflammation and inflammatory damage by inhibition of NE makes the enhanced inflammation in the joint of B. burgdorferi-infected mice a predicted result;

We agree that the observation of the elevated NE level and the enhanced inflammation is theoretically likely. Indeed, that was the hypothesis that we explored, and often what is theoretically possible does not turn out to occur. In addition, despite the known contribution of neutrophils to the severity of murine Lyme arthritis, the importance of the neutrophil serine proteases and anti-protease has not been specifically studied, and neutrophils secrete many factors. Therefore, our data fill an important gap in the knowledge of murine Lyme arthritis development – and set the stage for the further exploration of this hypothesis in the genesis of human Lyme arthritis.

(b) The potential contribution of the greater bacterial burden to the enhanced inflammation is not addressed;

We agree with the reviewer’s viewpoint that the increased infection burden in the tibiotarsal tissue of the infected SLPI-/- mice could contribute to the enhanced inflammation. A brief discussion of this possibility will be added to the revised manuscript.

(c) The relationship of SLPI binding by B. burgdorferi to the enhanced disease of SLPI-deficient mice is not clear; and

We agree with the reviewer that we have not shown the importance of the SLPI-B. burgdorferi binding in the development of periarticular inflammation. It is an ongoing project in our lab to identify the SLPI binding partner in B. burgdorferi. Our hypothesis is that SLPI could bind and inhibit an unknown B. burgdorferi virulence factor that contributes to murine Lyme arthritis. We will include the above discussion in the revised manuscript.

(d) Several methodological aspects of the study are unclear.

We appreciate the critique and will modify the method session in greater detail in the revised manuscript.

Reviewer #3 (Public review):

Summary:

The authors investigated the role of secretory leukocyte protease inhibitors (SLPI) in developing Lyme disease in mice infected with Borrelia burgdorferi. Using a combination of histological, gene expression, and flow cytometry analyses, they demonstrated significantly higher bacterial burden and elevated neutrophil and macrophage infiltration in SLPI-deficient mouse ankle joints. Furthermore, they also showed direct interaction of SLPI with B. burgdorferi, which likely depletes the local environment of SLPI and causes excessive protease activity. These results overall suggest ankle tissue inflammation in B. burgdorferi-infected mice is driven by unchecked protease activity.

Strengths:

Utilizing a comprehensive suite of techniques, this is the first study showing the importance of anti-protease-protease balance in the development of periarticular joint inflammation in Lyme disease.

We greatly appreciate the reviewer’s careful reading and positive comments.

Weaknesses:

Due to the limited sample availability, the authors investigated the serum level of SLPI in both in Lyme arthritis patients and patients with earlier disease manifestations.

We agree with the reviewer that it would be ideal to have more samples from Lyme arthritis patients. However, among the available archived samples, samples from Lyme arthritis patients are limited. For the samples from patients with single EM, the symptom persisted into 3-4 month after diagnosis, the same timeframe when arthritis is developed. We will add the above discussion in the revised manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations for the authors):

(1) In Figure 2, for histological scoring, do they have similar n numbers?

In panel B, 20 infected WT mice and 19 infected SLPI-/- mice were examined. In panel D, 13 infected WT and SLPI-/- mice were examined. Without infection, WT and SLPI-/- mice do not develop spontaneous arthritis. Due to the slow breeding of the SLPI-/- mice, a small number of uninfected control animals were used.

(2) In Figure 3, for macrophage population analysis, maybe consider implementing Ly6G-negative gating strategy to prevent neutrophil contamination in macrophage population?

We appreciate reviewer’s suggestion. We will analyze the data using the Ly6G-negative gating strategy and provide the result in a supplemental figure. We will compare the results using the two gating strategies in the revised manuscript.

Reviewer #2 (Recommendations for the authors):

(1) The investigators should address the possibility that much of the enhanced inflammatory features of infected SLPI-deficient mice are simply due to the higher bacterial load in the joint.

We agree with the reviewer’s viewpoint that the increased infection burden in the tibiotarsal tissue of the infected SLPI-/- mice could contribute to the enhanced inflammation. A brief discussion of this possibility will be added to the revised manuscript.

(2) Fig. 1. (A) There is no statistically significant difference in the bacterial load in the heart or skin, in contrast to the tibiotarsal joint. It would be of interest to know whether other tissues that are routinely sampled to assess the bacterial load, such as injection site, knee, and bladder, also harbored increased bacterial load in SLPI-deficient mice. (B) Heart and joint burden were measured at "21-28" days. The two time points should be analyzed separately rather than pooled.

(A) We appreciate the reviewer’s suggestion. We agree that looking into the infection load in other tissues is helpful. However, studies into murine Lyme arthritis have been predominantly focused on tibiotarsal tissue, which displays the most consistent and prominent swelling that’s easy to observe and measure. Thus, we focused on the tibiotarsal joint in our study. (B) We collected the heart and joint tissue approximately 3-week post infection within a 3-day window based on the feasibility and logistics of the laboratory. Using “21-28 d”, we meant to describe between 21-24 days post infection. We apologize for the mislabeling and will correct it in the revised manuscript, stating approximately 3 weeks in the results, and defining approximately 3-weeks as between 21-24 days in the methods.

(3) Fig. 2. (A) The same ambiguity as to the days post-infection as cited above in Point 2B exists in this figure. (B) Panel B: Caliper measurements to assess joint swelling should be utilized rather than visual scoring. (In addition, the legend should make clear that the black circles represent mock-infected mice.)

(A) The histology scoring, and histopathology examination were performed at the same time as heart and joint tissue collection, approximately 3 weeks post infection within a 3-day window based on the feasibility and logistics of the laboratory. We apologize for the mislabeling and will correct it in the revised manuscript.  (B) We appreciate the reviewer’s suggestion. However, our extensive experience is that caliper measurement can alter the assessment of swelling by placing pressure on the joints and did not produce consistent results. Double blinded scoring was thus performed. Histopathology examination was performed by an independent pathologist and confirmed the histology score and provided additional measurements.

(4) Fig. 3. (A) See Point 2B. (B) For Panels C-E, uninfected controls are lacking.

We apologize for this omission. Uninfected controls will be provided in the revised manuscript.

(5) Fig. 4. Fig. 4. Some LD subjects were sampled multiple times (5 samples from 3 subjects with Lyme arthritis; 13 samples from 4 subjects with EM), and samples from same individuals apparently are treated as biological replicates in the statistical analysis. In contrast, the 5 healthy controls were each sampled only once.

We agree with the reviewer that the control group is smaller than the patient group. Among the archived samples that are available, the number of adult healthy controls are limited, and sampled once. We used these samples to establish the baseline level of SLPI in the serum. It has been shown that the serum level of SLPI in healthy volunteers is in average about 40 ng/ml  (DOI: 10.3389/fimmu.2019.00664 and 10.1097/00003246-200005000-00003). The median level in the healthy control in our data was 38.92 ng/ml, which is comparable to the previous results. A brief discussion will be added in the revised manuscript.

(6) Fig. 5. (A) Panel A: does binding occur when intact bacteria are used? (B) Panels B, C: Were bacteria probed with PI to indicate binding likely to occur to surface? How many biological replicates were performed for each panel? Is "antibody control" a no SLPI control? What is the blue line?

Actively growing B. burgdorferi were collected and used for binding assays. We do not permeabilize the bacteria for flow cytometry. Thus, all the binding detected occurs to the bacterial surface. Three biological replicates were performed for each panel. The antibody control is no SLPI control. For panel D, the bacteria were stained with Hoechst, which shows the morphology of bacteria. We apologize for the missing information. A complete and detailed description of Figure 5 will be provided in the revised manuscript. 

(7) Sup Fig. 1. (A) Panel A: Was this experiment performed multiple times? I.e., how many biological replicates? (B) Panel B: Strain should be specified.

The binding assay to B. burgdorferi B31A was performed two times. In panel B, B. burgdorferi B31A3 was used. We apologize for the missing information. A complete and detailed description will be provided in the revised manuscript. 

(8) Fig. S2. It is not clear that the condition (20% serum) has any bactericidal activity, so the potential protective activity of SLPI cannot be determined. (Typical serum killing assays in the absence of specific antibody utilized 40% serum.)

In Fig. S2, panel B, the first two bars (without SLPI, with 20% WT anti serum) showed around 40% viability. It indicates that the 20% WT anti serum has bactericidal activity. Serum was collected from B. burgdorferi-infected WT mice at 21 dpi, which should contain polyclonal antibody against B. burgdorferi.

Reviewer #3 (Recommendations for the authors):

It was a pleasure to review! I congratulate the authors on this elegant study. I think the manuscript is very well-written and clearly conveys the research outcomes. I only have minor suggestions to improve the readability of the text.

We greatly appreciate the reviewer’s recognition of our work.

Line 92: Please briefly summarize the key results of the study at the end of the introduction section.

We appreciate the reviewer’s suggestion. A brief summary will be added in the revised manuscript.

Line 108: Why is the inflammation significantly occurred only in ankle joints of SLPI-I mice? Could you please provide a brief explanation?

The inflammation may also happen in other joints the B. burgdorferi infected SLPI-/- mice, which has not been studied. The study into murine Lyme arthritis has been predominantly done in the tibiotarsal tissue, which displays the most prominent swelling that’s easy to observe and measure. Thus, we focused on the tibiotarsal joint in our study.

Line 136: Please also include the gene names in Figure 3.

We apologize for the omission. Gene names will be included in the revised manuscript.

Line 181: Please briefly introduce BASEHIT. Why did you use this tool? What are the benefits?

We appreciate the reviewer’s suggestion. We will provide more background information on BASEHIT in the revised manuscript.

Author response:

We thank the three Reviewers for the extensive evaluation of our work, which was largely positive and constructive. Prompted by their reviews and the many suggestions, we plan to do additional control experiments to add further data in a revised manuscript in order to improve the statistics and quantitation. Furthermore, we plan to expand the discussion. We agree that a more comprehensive mechanistic framework would be welcome but note that the system is a complex multicomponent system which is challenging. We plan to expand the work in future follow-up research.

Author response:

eLife Assessment

This important study reveals a role for IκBα in the regulation of embryonic stem cell pluripotency. The solid data in mouse embryonic stem cells include separation of function mutations in IκBα to dissect its non-canonical role as a chromatin regulator and its canonical function as NF-κB inhibitor. The conclusions could be strengthened by including better markers of differentiation status and additional controls or orthogonal approaches.

We are thankful to the two reviewers and editors for their kind feedback and for highlighting the impact of NF-kB-independent IkBa function in stabilizing naïve pluripotency.

In order to address reviewer’s comments, we will perform further analysis of differentiation trajectories, as well as a deeper comparison of the epigenetic features in our IkBa-KO mESCs with the Serum/LIF and 2i/LIF conditions. Moreover, we recognize that some sentences need to be modified to soften our conclusions in terms of effects on block in the naïve state or the global epigenetic effects, as the reviewers pointed out.

Public Reviews:

Reviewer #1 (Public review):

Summary:

This study probes the role of the NF-κB inhibitor IκBa in the regulation of pluripotency in mouse embyronic stem cells (mESCs). It follows from previous work that identified a chromatin-specific role for IκBa in the regulation of tissue stem cell differentiation. The work presented here shows that a fraction of IκBa specifically associates with chromatin in pluripotent stem cells. Using three Nfkbia-knockout lines, the authors show that IκBa ablation impairs the exit from pluripotency, with embryonic bodies (an in vitro model of mESC multi-lineage differentiation) still expressing high levels of pluripotency markers after sustained exposure to differentiation signals. The maintenance of aberrant pluripotency gene expression under differentiation conditions is accompanied by pluripotency-associated epigenetic profiles of DNA methylation and histone marks. Using elegant separation of function mutants identified in a separate study, the authors generate versions of IκBa that are either impaired in histone/chromatin binding or NF-κB binding. They show that the provision of the WT IκBa, or the NF-κB-binding mutant can rescue the changes in gene expression driven by loss of IκBa, but the chromatin-binding mutant can not. Thus the study identifies a chromatin-specific, NF-κB-independent role of IκBa as a regulator of exit from pluripotency.

Strengths:

The strengths of the manuscript lie in: (a) the use of several orthogonal assays to support the conclusions on the effects of exit from pluripotency; (b) the use of three independent clonal Nfkbia-KO mESC lines (lacking IκBa), which increase confidence in the conclusions; and (c) the use of separation of function mutants to determine the relative contributions of the chromatin-associated and NF-κB-associated IκBa, which would otherwise be very difficult to unpick.

Weaknesses:

In this reviewer's view, the term "differentiation" is used inappropriately in this manuscript. The data showing aberrant expression of pluripotency markers during embryoid body formation are supported by several lines of evidence and are convincing. However, the authors call the phenotype of Nfkbia-KO cells a "differentiation impairment" while the data on differentiation markers are not shown (beyond the fact that H3K4me1, marking poised enhancers, is reduced in genes underlying GO processes associated with differentiation and organ development). Data on differentiation marker expression from the transcriptomic and embryoid body immunofluorescent experiments, for example, should be at hand without the need to conduct many more experiments and would help to support the conclusions of the study or make them more specific. The lack of probing the differentiation versus pluripotency genes may be a missed opportunity in gaining in-depth understanding of the phenotype associated with loss of the chromatin-associated function of IκBa.