Author response:

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

Reviewer #1 (Public Review):

Summary:

This study presents careful biochemical experiments to understand the relationship between LRRK2 GTP hydrolysis parameters and LRRK2 kinase activity. The authors report that incubation of LRRK2 with ATP increases the KM for GTP and decreases the kcat. From this, they suppose an autophosphorylation process is responsible for enzyme inhibition. LRRK2 T1343A showed no change, consistent with it needing to be phosphorylated to explain the changes in G-domain properties. The authors propose that phosphorylation of T1343 inhibits kinase activity and influences monomer-dimer transitions.

Strengths:

The strengths of the work are the very careful biochemical analyses and the interesting result for wild-type LRRK2.

Weaknesses:

A major unexplained weakness is why the mutant T1343A starts out with so much lower activity--it should be the same as wild-type, non-phosphorylated protein. Also, if a monomer-dimer transition is involved, it should be either all or nothing. Other approaches would add confidence to the findings.

We thank the reviewer for these suggestions. We are aware that the T1343A has generally a lower activity compared to the wild type. Therefore, we would like to emphasize that this mutant is the only one not showing an increase in Km values after ATP treatment. Other mutants, also having lower kcat values like T1503A, still show this characteristic change in Km. Our favored explanation for the lower kcat of T1343A is that this mutation lays within a critical region, the so-called ploop, of the Roc domain and is very likely structurally not neutral. Concerning the dimer-monomer transition, we are convinced that there is more than one factor involved in this equilibrium. Most likely, including, but not limited to other LRRK2 domains (e.g. the WD40 domain), binding of co-factors (e.g. Rab29/Rab32 or 14-3-3) and membrane binding. Consistently, also with stapled peptides targeting the Roc or Cor domains we were not able to shift the equilibrium completely to the monomer (Helton et al., ACS Chem Biol. 2021, 16:2326-2338; Pathak et al. ACS Chem Neurosci. 2023, 14(11):1971-1980) We have addressed these points in a revised version of the manuscript.

Reviewer #2 (Public Review):

This study addresses the catalytic activity of a Ras-like ROC GTPase domain of LRRK2 kinase, a Ser/Thr kinase linked to Parkinson's disease (PD). The enzyme is associated with gain-of-function variants that hyper-phosphorylate substrate Rab GTPases. However, the link between the regulatory ROC domain and activation of the kinase domain is not well understood. It is within this context that the authors detail the kinetics of the ROC GTPase domain of pathogenic variants of LRRK2, in comparison to the WT enzyme. Their data suggest that LRRK2 kinase activity negatively regulates the ROC GTPase activity and that PD variants of LRRK2 have differential effects on the Km and catalytic efficiency of GTP hydrolysis. Based on mutagenesis, kinetics, and biophysical experiments, the authors suggest a model in which autophosphorylation shifts the equilibrium toward monomeric LRRK2 (locked GTP state of ROC). The authors further conclude that T1343 is a crucial regulatory site, located in the P-loop of the ROC domain, which is necessary for the negative feedback mechanism. Unfortunately, the data do not support this hypothesis, and further experiments are required to confirm this model for the regulation of LRRK2 activity.

Specific comments are below:

• Although a couple of papers are cited, the rationale for focusing on the T1343 site is not evident to readers. It should be clarified that this locus, and perhaps other similar loci in the wider ROCO family, are likely important for direct interactions with the GTP molecule.

To clarify this point: We, have not only have focused on this specific locus, but instead systematically mutated all known auto-phosphorylation sites with the RocCOR domain (see. supplemental information). Furthermore, it has been shown that this site, at least in the RCKW (Roc to WD40) construct, is quantitatively phosphorylated (Deniston et al., Nature 2020, 588:344-349). We are aware that the T1343 residue is located within the p-loop and that this can impact nucleotide binding capacities (see response to reviewer 1).

We have clarified and addressed these points in a revised version of the manuscript.

• Similar to the above, readers are kept in the dark about auto-phosphorylation and its effects on the monomer/dimer equilibrium. This is a critical aspect of this manuscript and a major conceptual finding that the authors are making from their data. However, the idea that auto-phosphorylation is (likely) to shift the monomer/dimer equilibrium toward monomer, thereby inactivating the enzyme, is not presented until page 6, AFTER describing much of their kinetics data. This is very confusing to readers, as it is difficult to understand the meaning of the data without a conceptual framework. If the model for the LRRK2 function is that dimerization is necessary for the phosphorylation of substrates, then this idea should be presented early in the introduction, and perhaps also in the abstract. If there are caveats, then they should be discussed before data are presented. A clear literature trail and the current accepted (or consensus) mechanism for LRRK2 activity is necessary to better understand the context for these data.

We agree on the reviewer’s opinion. We have revised the introduction accordingly and added a paragraph on page 3 starting from line 27.

• Following on the above concepts, I find it interesting that the authors mention monomeric cytosolic states, and kinase-active oligomers (dimers??), with citations. Again here, it would be useful to be more precise. Are dimers (oligomers?) only formed at the membrane? That would suggest mechanisms involving lipid or membrane-attached protein interactions. Also, what do the authors mean by oligomers? Are there more than dimers found localized to the membrane?

There are multiple studies that have shown that LRRK2 is mainly monomeric in the cytosol while it forms mainly dimeric or higher oligomeric states at membrane (James et al., Biophys. J. 2012, 102, L41–L43; Berger et al., Biochemistry, 2010, 49, 5511–5523). However, we agree with the reviewer that it remains to be determined if the dimeric form is the most active state at the membrane, or a higher oligomeric state. Espescially since a recent study shows that LRRK2 can form active tetramers only when bound to Rab29 (Zhu et al., bioRxiv, 2022, DOI: 10.1101/2022.04.26.489605). We have clarified these points in the introduction of the revised version of the manuscript (page 3, line 27ff).

• Fig 5 is a key part of their findings, regarding the auto-phosphorylation induced monomer formation of LRRK2. From these two bar graphs, the authors state unequivocally that the 'monomer/dimer equilibrium is abolished', and therefore, that the underlying mechanism might be increased monomerization (through maintenance of a GTP-locked state). My view is that the authors should temper these conclusions with caveats. One is that there are still plenty of dimers in the auto-phosphorylated WT, and also in the T1343A mutant. Why is that the case? Can the authors explain why only perhaps a 10% shift is sufficient? Secondly, the T1343A mutant appears to have fewer overall dimers to begin with, so it appears to readers that 'abolition' is mainly due to different levels prior to ATP treatment at 30 deg. I feel these various issues need to be clarified in a revised manuscript, with additional supporting data. Finally, on a minor note, I presume that there are no statistically significant differences between the two sets of bar graphs on the right panel. It would be wise to place 'n.s.' above the graphs for readers, and in the figure legend, so readers are not confused.

Starting with the monomer-dimer equilibrium we are convinced that there is more than the phosphorylation of T1343 (see response to reviewer 1). Therefore a 10% shift in our assay most likely underestimate the effect seen in cells. Consistently, the T1343A mutants show a similar increase in Rab10 phosphorylation assay as the G2019S mutant. This thus shows that the identified feedback mechanism plays an important role in a cellular context. We have addressed this point in the revised manuscript on page 6, line 8ff. As long as the significance indicators in the bar charts are concerned, we agree with reviewer. In order not to overload the figure, we finally decided to include all pairwise comparisons (post-hoc tests) in the supplement.

• Figure 6B, Westerns of phosphorylation, the lanes are not identified and it is unclear what these data mean.

We apologize for this mistake and have added the correct labeling in the revised version of the manuscript.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

The author studies a family of models for heritable epigenetic information, with a focus on enumerating and classifying different possible architectures. The key aspects of the paper are:

• Enumerate all 'heritable' architectures for up-to 4 constituents.

• A study of whether permanent ("genetic") or transient ("epigenetic") perturbations lead to heritable changes

• Enumerated the connectivity of the "sequence space" formed by these heritable architectures

• Incorporating stochasticity, the authors explore stability to noise (transient perturbations)

• A connection is made with experimental results on C elegans.

The study is timely, as there is a renewed interest in the last decade in non-genetic, heritable heterogeneity (e.g., from single-cell transcriptomics). Consequently, there is a need for a theoretical understanding of the constraints on such systems. There are some excellent aspects of this study: for instance, the attention paid to how one architecture "mutates" into another. Unfortunately, the manuscript as a whole does not succeed in formalising nor addressing any particular open questions in the field. Aside from issues in presentation and modelling choices (detailed below), it would benefit greatly from a more systematic approach rather than the vignettes presented.

Despite being foundational, this work was systematic in that (1) for the simple architectures modeled using ordinary differential equations (ODEs) with continuity assumptions, parameters that support steady states were systematically determined for each architecture and then every architecture was explored using genetic changes exhaustively, although epigenetic perturbations were not examined exhaustively because of their innumerable variety; and (2) for the more realistic modeling of architectures as Entity-Sensor-Property systems, the behavior of systems with respect to architecture as well as parameter space that lead to particular behaviors (persistence, heritable epigenetic change, etc.) was systematically explored. A more extensive exploration of parameter space that also includes the many ways that the interaction between any two entities/nodes could be specified using an equation is a potentially ever-expanding challenge that is beyond the scope of any single paper.

Specific aspects that remain to be addressed include the application of multiple notions of heritability to real networks of arbitrary size, considering different types of equations for change of each entity/node, and classifying different behavioral regimes for different sets of parameters.

The key contribution of the paper is an articulation of the crucial questions to ask of any regulatory architecture in living systems rather than the addressing of any question that a field has recognized as ‘open’. Specifically, through the exhaustive listing of small regulatory architectures that can be heritable and the systematic analysis of arbitrary Entity-Sensor-Property systems that more realistically capture regulatory architectures in living systems, this work points the way to constrain inferences after experiments on real living systems. Currently, most experimental biologists engaged in reductionist approaches and some systems biologists examining the function or prevalence of network motifs do not explicitly constrain their models for heritability or persistence. It is hoped that this paper will raise awareness in both communities and lead to more constrained models that minimize biases introduced by incomplete knowledge of the network, which is always the case when analyzing living systems.

Terminology

The author introduces a terminology for networks of interacting species in terms of "entities" and "sensors" -- the former being nodes of a graph, and the latter being those nodes that receive inputs from other nodes. In the language of directed graphs, "entities" would seem to correspond to vertices, and "sensors" those vertices with positive indegree and outdegree. Unfortunately, the added benefit of redefining accepted terminology from the study of graphs and networks is not clear.

The Entities-Sensors-Property (ESP) framework is based on underlying biology and not graph theory, making an ESP system not entirely equivalent to a network or graph, which is much less constrained. The terms ‘entity’, ‘sensor’, and ‘property’ were defined and justified in a previous paper (Jose, J R. Soc. Interface, 2020). While nodes of a network can be parsed arbitrarily and the relationship between them can also be arbitrary, entities and sensors are molecules or collections of molecules that are constrained such that the sensors respond to changes in particular properties of other entities and/or sensors. When considered as digraphs, sensors can be seen as vertices with positive indegree and outdegree. The ESP framework can be applied across any scale of organization in living systems and this specific way of parsing interactions also discretizes all changes in the values of any property of any entity. In short, ESP systems are networks, but not all networks are ESP systems. Therefore, the results of network theory that remain applicable for ESP systems need further investigation.

The key utility of the ESP framework is that it is aligned with the development of mechanistic models for the functions of living systems while being consistent with heredity. In contrast, widely analyzed networks like protein-interaction networks, signaling networks, gene regulatory networks, etc., are not always constrained using these principles.

Model

The model seems to suddenly change from Figure 4 onwards. While the results presented here have at least some attempt at classification or statistical rigour (i.e. Fig 4 D), there are suddenly three values associated with each entity ("property step, active fraction, and number"). Furthermore, the system suddenly appears to be stochastic. The reader is left unsure what has happened, especially after having made the effort to deduce the model as it was in Figs 1 through 3. No respite is to be found in the SI, either, where this new stochastic model should have been described in sufficient detail to allow one to reproduce the simulation.

The Supplementary Information section titled ‘Simulation of simple ESP systems’ provides the requested detailed information and revisions to the writing provide the biologically grounded justification for parsing interacting regulators as ESP systems.

Perturbations

Inspired especially by experimental manipulations such as RNAi or mutagenesis, the author studies whether such perturbations can lead to a heritable change in network output. While this is naturally the case for permanent changes (such as mutagenesis), the author gives convincing examples of cases in which transient perturbations lead to heritable changes. Presumably, this is due the the underlying multistability of many networks, in which a perturbation can pop the system from one attractor to another.

Unfortunately, there appears to be no attempt at a systematic study of outcomes, nor a classification of when a particular behaviour is to be expected. Instead, there is a long and difficult-to-read description of numerical results that appear to have been sampled at random (in terms of both the architecture and parameter regime chosen). The main result here appears to be that "genetic" (permanent) and "epigenetic" (transient) perturbations can differ from each other -- and that architectures that share a response to genetic perturbation need not behave the same under an epigenetic one. This is neither surprising (in which case even illustrative evidence would have sufficed) nor is it explored with statistical or combinatorial rigour (e.g. how easy is it to mistake one architecture for another? What fraction share a response to a particular perturbation?)

As an additional comment, many of the results here are presented as depending on the topology of the network. However, each network is specified by many kinetic constants, and there is no attempt to consider the robustness of results to changes in parameters.

The systematic study of all arbitrary regulatory architectures is beyond the scope of this paper and, indeed, beyond the scope of any one paper. Nevertheless 225,000 arbitrary Entity-Sensor-Property systems were systematically explored and collections of parameters that lead to different behaviors provided (e.g., 78,285 are heritable). These ESP systems more closely mimic regulation in living systems than the coupled ODE-based specification of change in a regulatory architecture.

The example questions raised here are not only difficult to answer, but subjective and present a moving target for future studies. One, ‘how easy is it to mistake one architecture for another?’. Mistaking one architecture for another clearly depends on the number of different types of experiments one can perform on an architecture and the resolution with which changes in entities can be measured to find distinguishing features. Two, ‘What fraction share a response to a particular perturbation?’. ‘Sharing a response’ also depends on the resolution of the measurement after perturbation.

DNA analogy

At two points, the author makes a comparison between genetic information (i.e. DNA) and epigenetic information as determined by these heritable regulatory architectures. The two claims the author makes are that (i) heritable architectures are capable of transmitting "more heritable information" than genetic sequences, and (ii) that, unlike DNA, the connectivity (in the sense of mutations) between heritable architectures is sparse and uneven (i.e. some architectures are better connected than others).

In both cases, the claim is somewhat tenuous -- in essence, it seems an unfair comparison to consider the basic epigenetic unit to be an "entity" (e.g., an entire transcription factor gene product, or an organelle), while the basic genetic unit is taken to be a single base-pair. The situation is somewhat different if the relevant comparison was the typical size of a gene (e.g., 1 kb).

Considering every base being the unit of stored information in the DNA sequence results in the maximal possible storage capacity of a genome of given length. Any other equivalence between entity and units within the genome (e.g., 1 kb gene) will only reduce the information stored in the genome.

Nevertheless, the claim was modified to say that the information content of an ESP system can [italics added] be more extensive than the information content of the genome. This accounts for the possibility of an organism that has an inordinately large genome such that maximal information that can be stored in a particular genome sequence exceeds that stored in a particular configuration of all the contents in a cell.

I thank the reviewer for providing further explanation of this misunderstanding in the second round of review, which helps draw future readers to the sections in the paper that discusses this important point (also see response to Recommendations for the authors).

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I thank the author for their efforts in replying to the comments. I have updated my review accordingly; in particular, I have:

(1) Removed my complaint that Heritability is nowhere defined

(2) Removed issues with the presentation of the ODE model in the supplementary information.

I thank the reviewer for raising these issues and acknowledging the improvements made.

However, given that the manuscript is broadly unchanged from the initial one, many of my prior comments remain justified. Some key points:

(1) The manuscript continues to be difficult to read, for the same reasons as I mentioned when reviewing the paper previously.

(2) The utility of the "ESP" formalism is still unclear.

• As the author notes, continuous ODEs are of course an idealisation of a system with discrete copy number.

• However, discussing this is standard fare in any textbook dealing with chemical dynamics and stochastic processes -- see, for instance, the standard textbook by van Kampen.

• This seems little reason to reject ODEs and implement a poorly defined formalism/simulation scheme.

(3) The author claims that many questions raised are "beyond the scope of this study". Indeed, answering all of these questions are beyond the scope of any one study. However, as I initially wrote, the paper would be much stronger if it focused on a particular problem rather than the many vignettes depicted.

The broad scope of this foundational paper necessitates addressing many issues, which may make it a difficult read for some readers. I hope that future work where each paper focuses on one of the aspects raised here will enable the extensive treatment of limited scope as suggested by the reviewer.

The utility of ODEs is much appreciated and was indeed a computationally efficient way of exploring the vast space of regulatory architectures. As stated in the response to the public reviews, the Entity-Sensors-Property framework provides a biologically grounded way of parsing interacting regulators. This approach is aligned with the development of mechanistic models for the functions of living systems while being consistent with heredity. In contrast, widely analyzed networks like protein-interaction networks, signaling networks, gene regulatory networks, etc., are not always constrained using these principles.

On a final note, on the subject of the comparison with DNA:

Perhaps I have misunderstood something. I simply meant that comparing the "maximal information" with 4 HRAs (12.45 bits) is certainly more than the "maximal information" with 4 basepairs (8 bits), but definitely less than the "maximal information" for four 1-kb genes (4^(4000) combinations, so 8000 bits...)

Perhaps the author means that the growth in information of HRAs is faster than exponential. If so, that should be shown and then remarked on.

For this reason, I maintain my comment that the comparison is tenuous.

This issue was addressed once in the results section and again in the discussion section.

The results section states that “The combinatorial growth in the numbers of HRAs with the number of interactors can thus provide vastly more capacity for storing information in larger HRAs compared to that afforded by the proportional growth in longer genomes.”

The discussion section states that “Despite imposing heritability, regulated non-isomorphic directed graphs soon become much more numerous than unregulated non-isomorphic directed graphs as the number of interactors increase (125 vs. 5604 for 4 interactors, Table 1). With just 10 interactors, there are >3x1020 unregulated non-isomorphic directed graphs [60] and HRAs are expected to be more numerous. This tremendous variety highlights the vast amount of information that a complex regulatory architecture can represent and the large number of changes that are possible despite sparsity of the change matrix (Fig. 3).”

Thus, indeed as the reviewer surmises, the combinatorial explosion in information of HRAs with increases in interacting entities is faster than the proportional growth in information of genome sequence with increases in length.

In summary, I thank the reviewers and editors for their help in improving the paper and would like to make the current manuscript the Version of Record.

---

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

Public Reviews:

Reviewer #1 (Public Review):

The author studies a family of models for heritable epigenetic information, with a focus on enumerating and classifying different possible architectures. The key aspects of the paper are:

• Enumerate all 'heritable' architectures for up to 4 constituents.

• A study of whether permanent ("genetic") or transient ("epigenetic") perturbations lead to heritable changes.

• Enumerated the connectivity of the "sequence space" formed by these heritable architectures.

-Incorporating stochasticity, the authors explore stability to noise (transient perturbations). - A connection is made with experimental results on C elegans.

The study is timely, as there has been a renewed interest in the last decade in nongenetic, heritable heterogeneity (e.g., from single-cell transcriptomics). Consequently, there is a need for a theoretical understanding of the constraints on such systems. There are some excellent aspects of this study: for instance:

• The attention paid to how one architecture "mutates" into another, establishing the analogue of a "sequence space" for network motifs (Fig 3).

• The distinction is drawn between permanent ("genetic") and transient ("epigenetic") perturbations that can lead to heritable changes.

• The interplay between development, generational timescales, and physiological time (as in Fig. 5).

I thank the reviewer for highlighting these aspects of the work.

The manuscript would be very interesting if it focused on explaining and expanding these results. Unfortunately, as a whole, it does not succeed in formalising nor addressing any particular open questions in the field. Aside from issues in presentation and modelling choices (detailed below), it would benefit greatly from a more systematic approach rather than the vignettes presented.

This first paper is foundational and therefore cannot be expected to solve all aspects of the problem of heredity. The work was nevertheless systematic in that (1) for the simple architectures modeled using ordinary differential equations (ODEs) with continuity assumptions, parameters that support steady states were systematically determined for each architecture and then every architecture was explored using genetic changes exhaustively, although epigenetic perturbations were not examined exhaustively because of their wide variety; and (2) for the more realistic modeling of architectures as Entity-Sensor-Property systems, the behavior of systems with respect to architecture as well as parameter space that lead to particular behaviors (persistence, heritable epigenetic change, etc.) was systematically explored. A more extensive exploration of parameter space that also includes the many ways that the interaction between any two entities/nodes could be specified using an equation is a potentially ever-expanding challenge that is beyond the scope of any single paper (see response to additional comments below).

Specific aspects that remain to be addressed include the application of multiple notions of heritability to real networks of arbitrary size, considering different types of equations for change of each entity/node, and classifying different behavioral regimes for different sets of parameters. As is evident from this list of combinatorial possibilities, the space to be explored is vast and beyond the scope of this foundational paper.

The key contribution of the paper is an articulation of the crucial questions to ask of any regulatory architecture in living systems rather than the addressing of any question that a field has recognized as ‘open’. Specifically, through the exhaustive listing for small regulatory architectures that can be heritable and the systematic analysis of arbitrary Entity-Sensor-Property systems that more realistically capture regulatory architectures in living systems, this work points the way to constrain inferences after experiments on real living systems. Currently, most experimental biologists engaged in reductionist approaches and some systems biologists examining the function or prevalence of network motifs do not explicitly constrain their models for heritability or persistence. It is hoped that this paper will raise awareness in both communities and lead to more constrained models that minimize biases introduced by incomplete knowledge of the network, which is always the case when analyzing living systems.

Terminology

The author introduces a terminology for networks of interacting species in terms of "entities" and "sensors" -- the former being nodes of a graph, and the latter being those nodes that receive inputs from other nodes. In the language of directed graphs, "entities" would seem to correspond to vertices, and "sensors" those vertices with positive indegree and outdegree. Unfortunately, the added benefit of redefining accepted terminology from the study of graphs and networks is not clear.

The Entities-Sensors-Property (ESP) framework is based on underlying biology and not graph theory, making an ESP system not entirely equivalent to a network or graph, which is much less constrained. The terms ‘entity’, ‘sensor’, and ‘property’ were defined and justified in a previous paper (Jose, J R. Soc. Interface, 2020). While nodes of a network can be parsed arbitrarily and the relationship between them can also be arbitrary, entities and sensors are molecules or collections of molecules that are constrained such that the sensors respond to changes in particular properties of other entities and/or sensors. When considered as digraphs, sensors can be seen as vertices with positive indegree and outdegree. The ESP framework can be applied across any scale of organization in living systems and this specific way of parsing interactions also discretizes all changes in the values of any property of any entity. In short, ESP systems are networks, but not all networks are ESP systems. Therefore, the results of network theory that remain applicable for ESP systems need further investigation. This justification is now repeated in the paper.

The key utility of the ESP framework is that it is aligned with the development of mechanistic models for the functions of living systems while being consistent with heredity. In contrast, widely analyzed networks like protein-interaction networks, signaling networks, gene regulatory networks, etc., are not always constrained using these principles. In addition, the language of digraphs where sensors can be seen as vertices with positive indegree and outdegree has been also added to aid readers who are familiar with graph theory.

Heritability

The primary goal of the paper is to analyse the properties of those networks that constitute "heritable regulatory architectures". The definition of heritability is not clearly stated anywhere in the paper, but it appears to be that the steady-state of the network must have a non-zero expression of every entity. As this is the heart of the paper, it would be good to have the definition of heritable laid out clearly in either the main text or the SI.

I have now defined the term as used in this paper early, which is indeed as surmised by the reviewer simply the preservation of the architecture and non-zero levels of all entities. I have also highlighted additional notions of heredity that are possible, which will be the focus of future work. These can range from precise reproduction of the concentration and the localization of every entity to a subset of the entities being reproduced with some error while the rest keep varying from generation to generation (as illustrated in Fig. 2 of Jose, BioEssays, 2018). Importantly, it is currently unclear which of these possibilities reflects heredity in real living systems.

Model

As described in the supplementary, but not in the main text, the author first chooses to endow these networks with simple linear dynamics; something like $\partial_t \vec{x} = A x - T x$, where the vector $x$ is the expression level of each entity, $A$ has the structure of the adjacency matrix of the directed graph, and $T$ is a diagonal matrix with positive entries that determines the degradation or dilution rate of each entity. From a readability standpoint, it would greatly aid the reader if the long list of equations in the SI were replaced with the simple rule that takes one from a network diagram to a set of ODEs.

I have abridged the description by eliminating the steady state expression for every HRA as suggested and simply pointed to the earlier version of the paper for those readers who might prefer the explicit derivations of these simple expressions. An overview is now provided for going from any network diagram to a set of ODEs.

The implementation of negative regulation is manifestly unphysical if the "entities" represent the expression level of, say, gene products. For instance, in regulatory network E, the value of the variable z can go negative (for instance, if the system starts with z= and y=0, and x > 0).

Negative values for any entity were avoided in simulations by explicitly setting all such values to zero. This constraint has been added as a note in the section describing the equations for the change of each node/entity in each regulatory network. Specifically, the levels of each entity/sensor was set to zero during any time step when the computed value for that entity/sensor was less than zero. This bounding of the function allows for any approach to zero while avoiding negative values. I apologize for the omission of this constraint from the supplemental material in the last submission. This constraint was used in all the simulations and therefore this change does not affect any of the results presented. In this way, it is ensured that the presence of negative regulation does not lead to negative values.

Formally, the promotion or inhibition of an entity or sensor can be modeled using any function that is either increasing (for promotion) or decreasing (for inhibition). This diversity of possibilities is one of the challenges that prevents exhaustive exploration of all functions. In fact, the use of ODEs after assuming a continuous function is an idealization that facilitates understanding of general principles but is not in keeping with the discreteness of entities or step changes in their values (amount, localization, etc.) observed in living systems. Other commonly used continuous functions include Hill functions for the rate of production of y given as xn/(k + xn) for x activating y, which increases to ~1 as x increases, or given as k/(k + xn) for x inhibiting y, which decreases to ~0 as x increases. Increasing values of ‘n’ result in steeper sigmoidal curves. In reality, levels of all entities/sensors are expected to be discretized by measurement in living systems and the form of the function for any regulation needs empirical measurement in vivo (see response to comment below).

The model seems to suddenly change from Figure 4 onwards. While the results presented here have at least some attempt at classification or statistical rigour (i.e. Fig 4 D), there are suddenly three values associated with each entity ("property step, active fraction, and number"). Furthermore, the system suddenly appears to be stochastic. The reader is left unsure of what has happened, especially after having made the effort to deduce the model as it was in Figs 1 through 3. No respite is to be found in the SI, either, where this new stochastic model should have been described in sufficient detail to allow one to reproduce the simulation.

While ODEs are easier to simulate and understand, they are less realistic as explained above. I have now added more explanation justifying the need for the subsequent simulation of Entity-Sensor-Property systems. I have also expanded the information provided for each aspect of the model (previously outlined in Fig. 4A and detailed within the code) in a Supplementary Information section titled ‘Simulation of simple ESP systems’.

Perturbations

Inspired especially by experimental manipulations such as RNAi or mutagenesis, the author studies whether such perturbations can lead to a heritable change in network output. While this is naturally the case for permanent changes (such as mutagenesis), the author gives convincing examples of cases in which transient perturbations lead to heritable changes. Presumably, this is due the the underlying mutlistability of many networks, in which a perturbation can pop the system from one attractor to another.

Unfortunately, there appears to be no attempt at a systematic study of outcomes, nor a classification of when a particular behaviour is to be expected. Instead, there is a long and difficult-to-read description of numerical results that appear to have been sampled at random (in terms of both the architecture and parameter regime chosen). The main result here appears to be that "genetic" (permanent) and "epigenetic" (transient) perturbations can differ from each other -- and that architectures that share a response to genetic perturbation need not behave the same under an epigenetic one. This is neither surprising (in which case even illustrative evidence would have sufficed) nor is it explored with statistical or combinatorial rigour (e.g. how easy is it to mistake one architecture for another? What fraction share a response to a particular perturbation?)

The systematic study of all arbitrary regulatory architectures is beyond the scope of this paper and, as stated earlier, beyond the scope of any one paper. Nevertheless 225,000 arbitrary Entity-Sensor-Property systems were systematically explored and collections of parameters that lead to particular behaviors provided (e.g., 78,285 are heritable). These ESP systems more closely mimic regulation in living systems than the coupled ODE-based specification of change in a regulatory architecture.

The example questions raised here are not only difficult to answer, but subjective and present a moving target for future studies. One, ‘how easy is it to mistake one architecture for another?’. Mistaking one architecture for another clearly depends on the number of different types of experiments one can perform on an architecture and the resolution with which changes in entities can be measured to find distinguishing features. Two, ‘What fraction share a response to a particular perturbation?’. ‘Sharing a response’ also depends on the resolution of the measurement of entities after perturbation.

As an additional comment, many of the results here are presented as depending on the topology of the network. However, each network is specified by many kinetic constants, and there is no attempt to consider the robustness of results to changes in parameters.

The interpretations presented are conservative determinations of heritability based on the topology of the architecture. In other words, architectures that can be heritable for some set of parameters. Of course, parameter sets can be found that make any regulatory architecture not heritable. As stated earlier, exploring all parameters for even one architecture is beyond the scope of a single study because of the infinitely many ways that the interaction between any two entities can be specified.

DNA analogy

At two points, the author makes a comparison between genetic information (i.e. DNA) and epigenetic information as determined by these heritable regulatory architectures. The two claims the author makes are that (i) heritable architectures are capable of transmitting "more heritable information" than genetic sequences, and (ii) that, unlike DNA, the connectivity (in the sense of mutations) between heritable architectures is sparse and uneven (i.e. some architectures are better connected than others).

In both cases, the claim is somewhat tenuous -- in essence, it seems an unfair comparison to consider the basic epigenetic unit to be an "entity" (e.g., an entire transcription factor gene product, or an organelle), while the basic genetic unit is taken to be a single base-pair. The situation is somewhat different if the relevant comparison was the typical size of a gene (e.g., 1 kb).

Considering every base being the unit of stored information in the DNA sequence results in the maximal possible storage capacity of a genome of given length. Any other equivalence between entity and units within the genome (e.g., 1 kb gene) will only reduce the information stored in the genome.

Nevertheless, the claim has been modified to say that the information content of an ESP system can [italics added] be more extensive than the information content of the genome. This accounts for the possibility of an organism that has an inordinately large genome such that maximal information that can be stored in a particular genome sequence exceeds that stored in a particular configuration of all the contents in a cell.

Reviewer #2 (Public Review):

Summary:

This manuscript uses an interesting abstraction of epigenetic inheritance systems as partially stable states in biological networks. This follows on previous review/commentary articles by the author. Most of the molecular epigenetic inheritance literature in multicellular organisms implies some kind of templating or copying mechanisms (DNA or histone methylation, small RNA amplification) and does not focus on stability from a systems biology perspective. By contrast, theoretical and experimental work on the stability of biological networks has focused on unicellular systems (bacteria), and neglects development. The larger part of the present manuscript (Figures 1-4) deals with such networks that could exist in bacteria. The author classifies and simulates networks of interacting entities, and (unsurprisingly) concludes that positive feedback is important for stability. This part is an interesting exercise but would need to be assessed by another reviewer for comprehensiveness and for originality in the systems biology literature. There is much literature on "epigenetic" memory in networks, with several stable states and I do not see here anything strikingly new.

The key utility of the initial part of the paper is the exhaustive enumeration of all small heritable regulatory architectures. The implications for the abundance of ‘network motifs’ and more generally any part of a network proposed to perform a particular function is that all such parts need to be compatible with heredity. This principle is generally not followed in the literature, resulting in incomplete networks being interpreted as having motifs or modules with autonomous function. Therefore, while the need for positive feedback for stability is indeed obvious, it is not consistently applied by all. For example, the famous synthetic circuit ‘the repressilator’ (Elowitz and Leibler, “A synthetic oscillatory network of transcriptional regulators”, Nature, 2000), which is presented as an example of ‘rational network design’, has three transcription factors that all sequentially inhibit the production of another transcription factor in turn forming a feedback loop of inhibitory interactions. Therefore, the contributions of the factors that promote the expression of each entity is unknown and yet essential for heritability. The comprehensive listing of the heritable regulatory architectures that are simple provide the basis for true synthetic biology where the contributing factors for observed behavior of the network are explicitly considered only after constraining for heredity. Using this principle, the minimal autonomous architecture that can implement the repressilator is the HRA ‘Z’ (Fig. 1).

An interesting part is then to discuss such networks in the framework of a multicellular organism rather than dividing unicellular organisms, and Figure 5 includes development in the picture. Finally, Figure 6 makes a model of the feedback loops in small RNA inheritance in C. elegans to explain differences in the length of inheritance of silencing in different contexts and for different genes and their sensitivity to perturbations. The proposed model for the memory length is distinct from a previously published model by Karin et al. (ref 49).

I thank the reviewer for appreciating this aspect of the paper.

Strengths:

A key strength of the manuscript is to reflect on conditions for epigenetic inheritance and its variable duration from the perspective of network stability.

I thank the reviewer for appreciating the importance of the overall topic.

Weaknesses:

• I found confusing the distinction between the architecture of the network and the state in which it is. Many network components (proteins and RNAs) are coded in the genome, so a node may not disappear forever.

I have added language to clarify the many states of a network versus its architecture (also illustrated in Fig. 4 for ESP systems). Even loss of expression below a threshold can lead to permanent loss if there is not sufficient noise to induce re-expression. For example, consider the simple case of a transcription factor that binds to its own promoter, requiring 10 molecules for the activation of the promoter and thus production of more of the same transcription factor. If an epigenetic change (e.g., RNA interference) reduces the levels to fewer than 10 molecules and if the noise in the system never results in the numbers of the transcription factor increasing beyond 10, the transcription factor has been effectively lost permanently. In this way, reduction of a regulator can lead to permanent change despite the presence of the DNA. Many papers in the field of RNA silencing in C. elegans have provided strong experimental evidence to support this assertion.

• From the Supplementary methods, the relationship between two nodes seems to be all in the form of dx/dt = Kxy . Y, which is just one way to model biological reactions. The generality of the results on network architectures that are heritable and robust/sensitive to change is unclear. Other interactions can have sigmoidal effects, for example. Is there no systems biology study that has addressed (meta)stability of networks before in a more general manner?

Indeed, the relationship between any two entities can in principle be modeled using any function. Extensive exploration of the behavior of any regulatory architecture – even the simplest ones – require simplifications. For example, early work by Stuart Kauffman explored Boolean networks (see ref. 10 in the paper for history and extensive explanations). However, allowing all possible ways of specifying the interactions between components of a network makes analysis both a computational and conceptual challenge.

• Why is auto-regulation neglected? As this is a clear cause of metastable states that can be inherited, I was surprised not to find this among the networks.

Auto-regulation in the sense of some molecule/entity ultimately leading to the production of more of itself is present in every heritable regulatory architecture. Specifically, all auto-regulatory loops rely on a sequence of interactions between two or more kinds of molecules. For example, a transcription factor (TF) binding to the promoter of its own gene sequence, resulting in the production of more TF protein is a positive feedback loop that relies on many interacting factors (transcription, translation, nuclear import, etc.) and can be considered as ‘auto-regulation’ as it is sometimes referred to in the literature. In this sense, every HRA (A through Z) includes ‘auto-regulation’ or more appropriately positive feedback loops. For example, in the HRA ‘A’, x ‘auto-regulates’ itself via y.

• I did not understand the point of using the term "entity-sensor-property". Are they the same networks as above, now simulated in a computer environment step by step (thus allowing delays)?

Please see response to the other reviewer regarding the need for the Entity-SensorProperty framework and how it is distinct from generic networks. Briefly, the ODE-based simple networks, while easy to analyze, are not realistic because of the assumptions of continuity. In contrast ESP systems are more realistic with measurement discretizing changes in property values as is expected in real living systems.

• The final part applies the network modeling framework from above to small RNA inheritance in C. elegans. Given the positive feedback, what requires explanation is how fast the system STOPs small RNA inheritance. A previous model (Karin et al., ref. 49) builds on the fact that factors involved in inheritance are in finite quantity hence the different small RNAs "compete" for amplification and those targeting a given gene may eventually become extinct.

The present model relies on a simple positive feedback that in principle can be modulated, and this modulation remains outside the model. A possibility is to add negative regulation by factors such as HERI-1, that are known to limit the duration of the silencing.

The duration of silencing differs between genes. To explain this, the author introduces again outside the model the possibility of piRNAs acting on the mRNA, which may provide a difference in the stability of the system for different transcripts. At the end, I do not understand the point of modeling the positive feedback.

The previous model (Karin et al., Cell Systems, 2023) can describe populations of genes that are undergoing RNA silencing but cannot explain the dynamics of silencing particular genes. Furthermore, this model also cannot explain cases of effectively permanent silencing of genes that have been reported (e.g., Devanapally et al., Nature Communications, 2021 and Shukla et al., Current Biology, 2021). Finally, the observations of susceptibility to, recovery from, and even resistance to trans silencing (e.g., Fig. 5a in Devanapally et al., Nature Communications, 2021) require an explanation that includes modulation of the HRDE-1-dependent positive feedback loop that maintains silencing across generations.

The specific qualitative predictions regarding the relationship between piRNA-mediated regulation genome-wide and HRDE-1-dependent silencing of a particular gene across generations could guide the discovery of potential regulators of heritable RNA silencing. The equations (4) and (5) in the paper for the extent of modulation needed for heritable epigenetic change provide specific quantitative predictions that can be tested experimentally in the future. I have also revised the title of the section to read ‘Tuning of positive feedback loops acting across generations can explain the dynamics of heritable RNA silencing in C. elegans’ to emphasize the above points.

• From the initial analysis of abstract networks that do not rely on templating, I expected a discussion of possible examples from non-templated systems and was a little surprised by the end of the manuscript on small RNAs.

The heritability of any entity relies on regulatory interactions regardless of whether a templated mechanism is also used or not. For example, DNA replication relies on the interactions between numerous regulators, with only the sequence being determined by the template DNA. The field of small RNA-mediated silencing facilitates analysis of epigenetic changes at single-gene resolution (Chey and Jose, Trends in Genetics, 2022). It is therefore likely to continue to provide insights into heritable epigenetic changes and how they can be modulated. Unfortunately, there are currently no known cases of epigenetic inheritance where the role of any templated mechanism has been conclusively excluded. Future research will improve our understanding of epigenetic states and their modulation in terms of changes in positive feedback loops as proposed in this study and potentially lead to the discovery of such mechanisms that act entirely independent of any template-dependent entity.

Recommendations for the authors:

I thank the reviewers for their specific suggestions to improve the paper.

Reviewer #1 (Recommendations For The Authors):

The paper has many long paragraphs that attempt to explain results, make illustrations, and give intuition. Unfortunately, these are difficult to read. It would aid the reader greatly if these were, say, converted into cartoons (even if only in the SI), or made more accessible in some other way.

I agree with the importance of making the material accessible to readers in multiple ways. I have now added a figure with schematics in the SI titled ‘Illustrations of key concepts’ (new Fig. S2), which collects concepts that are relevant throughout the paper and might aid some readers.

The bulk of the supplementary is currently a collection of elementary mathematics results: to whit, pages 26 to 33 of the combined manuscript carry no more information than a quick description of the general model and the diagrams in Fig 1. Similarly, pages 34 to 39 (non-zero dilution rate), and pages 39 through 58 (response to permanent changes) each express a trivial mathematical point that is more than sufficiently made with one illustrative example.

I agree with the reviewer and have condensed these pages as suggested. I have added a pointer to the earlier version as containing further details for the readers who might prefer the explicit listing of these equations.

Overall, the paper appears to be a collection of numerical results obtained from different models, united by uncertain terminology that is not fully defined in this paper. The most promising aspects of the paper lie either in (a) combinatorially complete enumeration of all regulatory architectures, or (b) relating experimental manipulations in C. elegans to possible underlying regulatory architectures. Focusing on one or the other might improve the readability of the paper.

The two sections of the paper are complementary and when presented together help with the integration of concepts rather than the siloed pursuit of theory versus experimental analysis. When this work was presented at meetings before submission, it was clear that different researchers appreciated different aspects. This divergence is also apparent in the two reviews, with each reviewer appreciating different aspects. I have repeated the definitions and justifications from the earlier paper (Jose, J R Soc Interface, 2020) to provide a more fluid transition between the two complementary sections of the paper. Knowing both sides could aid in the development of models that are not only consistent with measurable quantities (e.g., anything that can be considered an entity) but are also logically constrained (e.g., entities matched with sensors while avoiding any entities that do not have a source of production – i.e., avoiding nodes with indegree = 0).

However, having said that many results of these types are well-known in models of regulatory networks, and it is unclear what precisely warrants the new framework that the author is proposing. Indeed, it would be good to understand in what way the framework here is novel, and how it is distinguished from prior studies of regulatory networks.

The key novelty of the work is the consideration of heritability for any regulation. With the explicit definition of the heritability for a regulatory architecture and the acknowledgement that there can be more than one notion of heredity, this paper now sets the foundation for examining many real networks in this light. I hope that the added justifications for the current framework in the revised paper strengthen these arguments. Future literature reviews on networks in general and how they address heritability or persistence will better define the prevalence of these considerations. Currently, most experimental biologists engaged in reductionist approaches and some systems biologists examining the function or prevalence of network motifs do not explicitly constrain their models for heritability or persistence. It is hoped that this work will raise awareness in both communities and lead to more constrained models that acknowledge incomplete knowledge of the network, which is always the case when analyzing living systems.

Reviewer #2 (Recommendations For The Authors):

Minor points/clarity

• page 1 line 57: "transgenerational waveforms that preserve form and function" is unclear.

This phrase was expanded upon in a previous paper (Jose, BioEssays, 2020). I have now added more explanation in this paper for completeness. The section now reads ‘For example, the localization and activity of many kinds of molecules are recreated in successive generations during comparable stages [1-3]. These recurring patterns can change throughout development such that following the levels and/or localizations of each kind of molecule over time traces waveforms that return in phase with the similarity of form and function across generations [2].’

• page 7 line 3-6: the sentence has an ambiguous structure.

I have now edited this long sentence to read as follows: ‘For systematic analysis, architectures that could persist for ~50 generations without even a transient loss of any entity/sensor were considered HRAs. Each HRA was perturbed (loss-of-function or gain-of-function) after five different time intervals since the start of the simulation (i.e., phases). The response of each HRA to such perturbations were compared with that of the unperturbed HRA.’

• page 9 lines 25-27: the sentence is convoluted: are you defining epigenetic inheritance?

I have simplified this sentence describing prior work by others (Karin et al., Cell Systems, 2023) and moved a clause to the subsequent sentence. This section now reads: ‘Recent considerations of competition for regulatory resources in populations of genes that are being silenced suggest explanations for some observations on RNA silencing in C. elegans [49]. Specifically, based on Little’s law of queueing, with a pool of M genes silenced for an average duration of T, new silenced genes arise at a rate  that is given by M = T’. I have also provided more context by preceding this section with: ‘Although the release of shared regulators upon loss of piRNA-mediated regulation in animals lacking PRG-1 could be adequate to explain enhanced HRDE-1-dependent transgenerational silencing initiated by dsRNA in prg-1(-) animals, such a competition model alone cannot explain the observed alternatives of susceptibility, recovery and resistance (Fig. 6A).’

• page 13 lines 51-53. This last sentence of the discussion is ambiguous/unclear.

I have now rephrased this sentence to read: ‘This pathway for increasing complexity through interactions since before the origin of life suggests that when making synthetic life, any form of high-density information storage that interacts with heritable regulatory architectures can act as the ‘genome’ analogous to DNA.’

• Figure 2: the letters in the nodes are hard to read; the difference between full and dotted lines in the graphs also.

I have enlarged the nodes and widened the gap in the dotted lines to make them clearer. I have also similarly edited Fig. 1 and Fig. S3 to Fig. S9.

Author Response

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

eLife assessment

This important study convincingly shows that the less common D-serine stereoisomer is transported in the kidney by the neutral amino acid transporter ASCT2 and that it is a noncanonical substrate for sodium-coupled monocarboxylate transporter SMCTs. With a multihierarchical approach, this important study further shows that Ischemia-Reperfusion Injury in the kidney causes a specific increment in renal reabsorption carried out, in part, by ASCT2.

Public Reviews:

Reviewer #1 (Public Review):

Most amino acids are stereoisomers in the L-enantiomer, but natural D-serine has also been detected in mammals and its levels shown to be connected to a number of different pathologies. Here, the authors convincingly show that D-serine is transported in the kidney by the neutral amino acid transporter ASCT2 and as a non-canonical substrate for the sodium-coupled monocarboxylate transporter SMCTs. Although both transport D-serine, this important study further shows in a mouse model for acute kidney injury that ASCT2 has the dominant role.

Strengths:

The paper combines proteomics, animal models, ex vivo transport analyses, and in vitro transport assays using purified components. The exhaustive methods employed provide compelling evidence that both transporters can translocate D-serine in the kidney.

Weakness:

In the model for acute kidney injury, the SMCTs proteins were not showing a significant change in expression levels and were rather analysed based on other, circumstantial evidence. Although its clear SMCTs can transport D-serine its physiological role is less obvious compared to ASCT2.

We greatly value the reviewer's efforts and feedback in reviewing our manuscript. We acknowledge the reviewer's observation that the changes indicated by our proteomic results are not markedly pronounced. To reinforce our findings, we have incorporated an analysis of gene alterations at the single-cell level (snRNA-seq) from the publicly accessible IRI mouse model data (Figure supplement 7). The snRNA-seq data align with our proteomic data in terms of the general trend of gene/protein alterations, but reveal more substantial changes in both ASCT2 and SMCTs. These discrepancies might stem from the different quantification methods used, suggesting a possible underestimation in our label-free proteomic quantification. The differences we see between the functional changes in transporters and their quantification in proteomics can be explained by the unique challenges posed by membrane proteins. Post-translational modifications and the complex nature of multiple transmembrane domains often impact the accurate measurement of these proteins in proteomic studies. This complexity can lead to a mismatch between the actual functional changes occurring in the transporters and their perceived abundance or alterations as detected by proteomic methods (Figure 4A) (Schey KL et al. Biochemistry 2015, doi: 10.1021/bi301604j). However, this label-free quantitative proteomics approach is well-suited for our study, given its screening efficiency, compatibility with animal models, and the absence of a labeling requirement. We may consider incorporating alternative quantitative proteomic methods in future for a more thorough comparison. We have included these considerations in lines 351-356 of the revised manuscript.

Manuscript lines 351-356

“When evaluating the extent of gene/protein alterations between the control and IRI conditions, we observed that the gene alterations of both Asct2 and Smcts, as revealed by snRNAsequencing, are more pronounced than the protein alteration ratios obtained from proteomics. This discrepancy may stem from difficulty in the quantification method, especially for membrane transport proteins in label-free quantitative proteomics.”

Regarding the roles of ASCT2 and SMCTs in renal D-serine transport, snRNA-seq showed that ASCT2 expression in the controls is less than 10% of the cell population. We suggest that ASCT2 contributes to D-serine reabsorption because of its high affinity and SMCTs (SMCT1 and SMCT2) would play a role in D-serine reabsorption in the cells without ASCT2 expression. In addition, we included other factors (the turnover rate and the presence of local canonical substrates) that may determine the capability of D-serine reabsorption. We have included this suggestion in the Discussion lines 386-404.

Manuscript lines 386-404

“Kinetics analysis of D-serine transport revealed the high affinity by ASCT2 (Km 167 µM) (Foster et al., 2016) and low affinity by SMCT1 (Km 3.39 mM; Figure 5E). In addition to transport affinity, the expression levels and co-localization of multiple transporters within the same cells are critical for elucidating the physiological roles of transporters or transport systems (Sakaguchi et al., 2024). In our proteome data, the chromatogram intensities of Smct1 (2.9 x 109 AU) and Smct2 (1.6 x 108 AU) were significantly higher than that of Asct2 (1.5 x 107 AU) in control mice (Table 1: abundance in Sham). While direct intensity comparisons between different proteins in mass spectrometry analyses are not precise, they can provide a general indication of relative protein amounts. This finding aligns with the snRNA-seq data, where Asct2 expression was found to be minimal, present in less than 10% of cell populations under both control and IRI conditions, suggesting that many cells do not express Asct2. Conversely, Smct1 and Smct2 show high and ubiquitous expression in control conditions, but their levels are markedly reduced in IRI conditions (Figure supplement 7). Our ex vivo assays demonstrate that both ASCT2 and SMCTs mediate D-serine transport (Figure 7B). Consequently, Asct2 may contribute to D-serine reabsorption due to its high affinity, whereas Smcts, owing to their abundance, particularly in cells lacking Asct2, likely play a significant role in D-serine reabsorption. Moreover, factors such as transport turnover rate (Kcat) and the presence of local canonical substrates are also vital in defining the overall contribution of Dserine transport systems.”

Reviewer #2 (Public Review):

Summary:

The manuscript "A multi-hierarchical approach reveals D-1 serine as a hidden substrate of sodium-coupled monocarboxylate transporters" by Wiriyasermkul et al. is a resubmission of a manuscript, which focused first on the proteomic analysis of apical membrane isolated from mouse kidney with early Ischemia-Reperfusion Injury (IRI), a well-known acute kidney injury (AKI) model. In the second part, the transport of D-serine by Asct2, Smct1, and Smct2 has been characterized in detail in different model systems, such as transfected cells and proteoliposomes.

Strengths:

A major problem with the first submission was the explanation of the link between the two parts of the manuscript: it was not very clear why the focus on Asct2, Smct1, and Smct2 was a consequence of the proteomic analysis. In the present version of the manuscript, the authors have focused on the expression of membrane transporters in the proteome analysis, thus making the reason for studying Asct2, Smct1, and Smct2 transporters more clear. In addition, the authors used 2D-HPLC to measure plasma and urinary enantiomers of 20 amino acids in plasma and urine samples from sham and Ischemia-Reperfusion Injury (IRI) mice. The results of this analysis demonstrated the value of D-serine as a potential marker of renal injury. These changes have greatly improved the manuscript and made it more convincing.

We deeply appreciate the reviewer’s comments on the manuscript. We have responded to the recommendations one by one in the later section.

Reviewer #3 (Public Review):

Summary:

The main objective of this work has been to delve into the mechanisms underlying the increment of D-serine in serum, as a marker of renal injury.

Strengths:

With a multi-hierarchical approach, the work shows that Ischemia-Reperfusion Injury in the kidney causes a specific increment in renal reabsorption of D-serine that, at least in part, is due to the increased expression of the apical transporter ASCT2. In this way, the authors revealed that SMCT1 also transports D-serine.

The experimental approach and the identification of D-serine as a new substrate for SMCT1 merit publication in Elife.

The manuscript also supports that increased expression of ASCT2, even together with the parallel decreased expression of SMCT1, in renal proximal tubules underlies the increased reabsorption of D-serine responsible for the increment of this enantiomer in serum in a murine model of Ischemia-Reperfusion Injury.

Weaknesses:

Remains to be clarified whether ASCT2 has substantial stereospecificity in favor of D- versus L-serine to sustain a ~10-fold decrease in the ratio D-serine/L-serine in the urine of mice under Ischemia-Reperfusion Injury (IRI).

It is not clear how the increment in the expression of ASCT2, in parallel with the decreased expression of SMCT1, results in increased renal reabsorption of D-serine in IRI.

We thoughtfully appreciate the reviewer’s comment on the manuscript. Considering the alteration of D-/L-serine ratios, there are several factors including protein expression levels at both apical and basolateral sides, properties of the transporters (e.g. transport affinities, substrate stereoselectivities), and the expression of DAAO (D-amino acid oxidase) which selectively degrades D-amino acids. Moreover, the mechanism becomes more complicated when the transport systems of L- and D-enantiomers are different and have distinct stereoselectivities as in the case of serine. Future studies are required to complete the mechanism. However, we would like to explore the mechanism based on the current knowledge.

From this study, we identified ASCT2 and SMCTs (SMCT1 and SMCT2) as D-serine transport systems. We showed that SMCT1 prefers D-serine. Although we did not analyze ASCT2 stereoselectivity, based on the previous studies, ASCT2 recognizes both D- and Lserine with high affinities and slightly prefers L-enantiomer (Km of 18.4 µM for L-serine in oocyte expression system (Utsunomiya-Tate et al. J Biol Chem 1996) and 167 µM for Dserine in oocyte expression system (Foster et al. Plos ONE 2016), and the IC50 of 0.7 mM for L-serine and 4.9 mM for D-serine (in HEK293 expression systems, Foster et al. PLOS ONE 2016). The proteomics showed an increase of ASCT2 (1.6-fold increase) and a decrease of SMCTs (1.7-fold decrease in SMCT1, and 1.3-fold decrease in SMCT2) in IRI conditions. The table below summarizes D-serine transport by ASCT2 and SMCTs.

In the case of L-serine, ASCT2 and B0ATs (in particular B0AT3) have been revealed as L-serine transport systems in the kidneys (Bröer et al. Physiol Rev 2008; Singer et al. J Biol Chem 2009). Proteomics showed that B0ATs have higher expression levels than ASCT2 supporting the idea that B0ATs are the main L-serine transport system (Table S1: Abundance of B0AT1 = 1.34E+09, B0AT3 = 2.13E+08, ASCT2 = 1.46E+07). In IRI conditions, B0AT3 decreased 1.8 fold and B0AT1 decreased 1.1 fold. From these results, we included the contribution of B0ATs in L-serine transport in Author response table 1.

Author response table 1.

Taken together, we suggest that high ratios of D-/L-serine in IRI conditions are a combinational result of 1) increase of D-serine reabsorption by ASCT2 enhancement and SMCTs reduction and 2) decrease of L-serine reabsorption by B0ATs. We have included this suggestion in the Discussion lines 438-451.

Manuscript lines 438-451

“The enantiomeric profiles of serine revealed distinct plasma D/L-serine ratio, with low rations in the normal control but elevated ratios in IRI, despite the weak stereoselectivity of ASCT2 (Figure 1B). This observation suggested differential renal handling of D-serine compared to L-serine. While we identified SMCTs as a D-serine transport system, it has been reported that L-serine reabsorption is mediated by B0AT3 (Singer et al., 2009). We propose that the alterations in plasma and urinary D/L-serine ratios are the combined outcomes of: 1) transport systems for L-serine, and 2) transport systems for D-serine. In normal kidneys, the low plasma D/L-serine ratios could result from the efficient reabsorption of L-serine by B0AT3, coupled with the DAAO activity that degrades intracellular D-serine reabsorbed by SMCTs. In IRI conditions, our enantiomeric amino acid profiling revealed low plasma L-serine and high urinary L-serine (Figure supplements 1B, 2B). Additionally, the proteomic analysis indicated a reduction in B0AT3 levels (4h IRI/sham = 0.56 fold; 8h IRI/sham = 0.65 fold; Table S1). These observations suggest that the low L-serine reabsorption in IRI is a result of B0AT3 reduction.”

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

This is a thorough study that was reviewed previously under the old system. I think the authors have strengthened their findings and have no further suggestions.

We appreciate reviewer 1 for his/her effort and comments, which greatly contributed to improving this manuscript.

Reviewer #2 (Recommendations For The Authors):

The experiments seem to me to have been well performed and the data are readily available.

Weaknesses:

More than weakness I would speak of discussion points: I have a few suggestions that may help to make the paper more accessible to a general audience.

(1) In the Introduction, when the authors introduce the term "micromolecules", it would be beneficial to provide a precise definition or clarification of what they mean by this term. Adding a brief explanation may help the reader to better understand the context.

Following the reviewer’s comment, we have included the explanation of the micromolecule and membrane transport proteins in lines 41-43.

Manuscript lines 41-43

“Membrane transport proteins function to transport micromolecules such as nutrients, ions, and metabolites across membranes, thereby playing a pivotal role in the regulation of micromolecular homeostasis.”

(2) In line 91, I suggest specifying that this is a renal IRI model.

Following the reviewer’s comment, we have added the information that it is a renal IRI model of AKI (lines 90-92).

Manuscript lines 90-92

“We applied 2D-HPLC to quantify the plasma and urinary enantiomers of 20 amino acids of renal ischemia-reperfusion injury (IRI) mice, a model of AKI and AKI-to-CKD transition (Sasabe et al., 2014; Fu et al., 2018).”

(3) Lines 167-168 state that Asct2 is localised to the apical side of the renal proximal tubules. Is there any expression of Asct2 in other nephron segments?

To our knowledge, there is no report of ASCT2 expression in other nephron segments. Our immunofluorescent data of the ASCT2 staining in the whole kidney at the low magnification and another region of Figure 3 (below) as well as immunohistochemistry from Human Protein Atlas (update: Jun 9th, 2023) did not show a strong signal of ASCT2 expression in other regions besides the proximal tubules. Thus, we conclude that ASCT2 is mainly expressed in proximal tubules, but not in other nephron regions.

Author response image 1.

(4) Lines 225-226: Have the authors expressed the candidate genes in HEK293 cells with ASCT2 knockdown?

This experiment was done by expressing the candidate genes in the presence of endogenous ASCT2. We have added the information in lines 225-227 to emphasize this process.

Manuscript lines 225-227

“Based on this finding, we utilized cell growth determination assay as the screening method even in the presence of endogenous ASCT2 expression. HEK293 cells were transfected with human candidate genes without ASCT2 knockdown.”

(5) Lines 254-255: why was D-serine transport enhanced by ASCT2 knockdown in FlpInTRSMCT1 or 2 cells?

We appreciate the reviewer to point out this data. We apologize for causing the confusion in the text. The total amount of D-serine uptake in the cells did not enhance but the net uptake (uptake subtracted from the background) was increased. This enhancement is a result of the lower background by ASCT2 knockdown. We have revised the texts and explained this result in more detail (lines 256-258).

Manuscript lines 256-258

“In the cells with ASCT2 knockdown, the background level was lower, thereby enhancing the D-[3H]serine transport contributed by both SMCT1 and SMCT2 (the net uptake after subtracted with background) (Figure 5C).”

(6) Line 265: The low affinity of SMCT1 for D-serine alone makes it an unlikely transporter for urinary D-serine.

We admitted the reviewer’s concern about the low affinity of SMCT1. However, Km at mM range is widely accepted for several low-affinity amino acid transporters such as proton-coupled amino acid transporter PAT1 (Km = 2 – 5 mM; Miyauchi et al. Biochem J 2010), cationic amino acid transporter CAT2A (Km = 3 – 4 mM; Closs et al. Biochem 1997), and large-neutral amino acid transporter LAT4 (Km = 17 mM; Bodoy et al. J Biol Chem 2005). In the kidneys, many compounds are well-known to be reabsorbed by the low-affinity but high-capacity (high-expression) transporters. Similarly, D-serine was reported to be reabsorbed by the low-affinity transporter (Kragh-Hansen and Sheikh, J Physiol 1984; Shimomura et al. BBA 1988; Silbernagl et al. Am J Physiol Renal Physiol 1999). Moreover, amino acid profile showed urinary D-serine in the range of 100 – 200 µM (Figure supplement 2). This concentration range could drive SMCT1 function (Figure 5). Combined with the high and ubiquitous expression of SMCT1, we propose that SMCT1 is a low-affinity but highcapacity D-serine transporter in the kidneys.

snRNA-seq is a method that can directly compare the expression levels between different genes within the same cells. From Figure supplement 7, expression of SMCT1 is much more abundant than ASCT2. ASCT2 was presented in less than 10% of cell population. It is possible that 90% of the cells that do not express ASCT2 use SMCT1 to reabsorb Dserine.

We have revised the Discussion regarding this comment (lines 386-404).

Manuscript lines 386-404

“Kinetics analysis of D-serine transport revealed the high affinity by ASCT2 (Km 167 µM) (Foster et al., 2016) and low affinity by SMCT1 (Km 3.39 mM; Figure 5E). In addition to transport affinity, the expression levels and co-localization of multiple transporters within the same cells are critical for elucidating the physiological roles of transporters or transport systems (Sakaguchi et al., 2024). In our proteome data, the chromatogram intensities of Smct1 (2.9 x 109 AU) and Smct2 (1.6 x 108 AU) were significantly higher than that of Asct2 (1.5 x 107 AU) in the control mice (Table 1: abundance in Sham). While direct intensity comparisons between different proteins in mass spectrometry analyses are not precise, they can provide a general indication of relative protein amounts. This finding aligns with the snRNA-seq data, where Asct2 expression was found to be minimal, present in less than 10% of cell populations under both control and IRI conditions, suggesting that many cells do not express Asct2. Conversely, Smct1 and Smct2 show high and ubiquitous expression in control conditions, but their levels are markedly reduced in IRI conditions (Figure supplement 7). Our ex vivo assays demonstrate that both ASCT2 and SMCTs mediate D-serine transport (Figure 7B). Consequently, Asct2 may contribute to D-serine reabsorption due to its high affinity, whereas Smcts, owing to their abundance, particularly in cells lacking Asct2, likely play a significant role in D-serine reabsorption. Moreover, factors such as transport turnover rate (Kcat) and the presence of local canonical substrates are also vital in defining the overall contribution of Dserine transport systems.”

(7) Line 316: The authors state that there is a high tubular D-serine reabsorption in IRI and in line 424 there is an inactivation of DAAO during the pathology. This suggests that there is a reabsorption of D-serine mediated by a transport system in the basolateral membrane domain of proximal tubular cells. Do the authors have any information about this transporter?

We agree with the reviewer that transporters at the basolateral membrane are important to complete the D-serine reabsorption in the kidney, and have included this issue in the original manuscript. We stated that transport systems at the basolateral side are necessary to be analyzed in order to complete the picture of D-serine transport systems in the kidney (lines 481-483 of the revised manuscript). However, we did not have any strong candidates for basolateral D-serine transport systems. Because we analyzed the proteome of BBMV, which concentrates on the apical membrane proteins, the analysis did not detect several transporters at the basolateral side.

(8) In lines 462-463, the authors state: "It is suggested that PAT1 is less active at the apical membrane where the luminal pH is neutral". However, the pH of urine in the proximal tubules is normally acidic due to the high activity of NH3. I suggest rewording this sentence.

Thank you for your comment. Proximal tubule (PT) is the first and the main region to maintain acid-base homeostasis in the kidney. In PT cells, NH3 secretes H+ to titrate luminal HCO3- and creates CO2, which is absorbed into PT cells and produces "new intracellular HCO3-", which is subsequently reabsorbed into the blood. Although ion fluxes in PT is to maintain the pH homeostasis, the pH regulation in both luminal and intracellular PT cells is highly dynamic. We totally agree with the reviewer and to follow that, we have revised the text by emphasizing the pH around PT segments, rather than the final urine pH, and leaving the discussion open for the possibility of PAT1 function in PT of normal kidneys (lines 474481).

Manuscript lines 474-481

“PAT1, a low-affinity proton-coupled amino acid transporter (Km in mM range), has been found at both sub-apical membranes of the S1 segment and inside of the epithelia (The Human Protein Atlas: https://www.proteinatlas.org; updated on Dec 7th, 2022) (Sagné et al., 2001; Vanslambrouck et al., 2010). PAT1 exhibits optimum function at pH 5 - 6 but very low activity at pH 7 (Miyauchi et al., 2005; Bröer, 2008b). Future research is required to address the significance of PAT1 on D-serine transport in the proximal tubule segments where pH regulation is known to be highly dynamic (Boron, 2006; Nakanishi et al., 2012; Bouchard and Mehta, 2022; Imenez Silva and Mohebbi, 2022).”

Reviewer #3 (Recommendations For The Authors):

The authors proposed that the increased expression of ASCT2, even together with the decreased expression of SMCT1/2, causes the increased renal reabsorption of D-serine that occurs in IRI. In the discussion, the main argument to sustain this hypothesis is the higher apparent affinity for D-serine of ASCT2 (<200 uM Km) versus SMCT1 (3.4 mM Km). In the Discussion section (page 18- 1st complete paragraph), the authors indicate that the Mass Spec intensities of SMCT1 and 2 are two and one order of magnitude higher respectively than that of ASCT2. This suggests that SMCT1 is clearly more expressed than ASCT2 in control conditions. IRI increments ASCT2 protein expression in brush-border membrane vesicle from kidney 1.6 folds and decreases that of SMCT1 0.6 folds. How this fold changes, even taking into account the lower Km of ASCT2 versus SMCT1 would explain the dramatic changes in the D-/L-serine ratios in plasma and urine in IRI? The authors might discuss whether other transport characteristics, even unknown (e.g., a higher turnover rate of ASCT2 vs SMCT1), would also contribute to the higher D-serine reabsorption in IRI.

SMCT1 shows some enantiomer selectivity for D- vs L-serine. At 50 uM concentration the transport is almost double for D. vs L-serine, but is ASCT2 stereoselective between the two enantiomers of serine? Some of the authors of this manuscript showed in a previous paper that the basolateral transporter Asc1 also participates in the accumulation of D-serine in serum caused by renal tubular damage. (Serum D-serine accumulation after proximal renal tubular damage involves neutral amino acid transporter Asc-1. Suzuki M et al. Sci Rep. 2019 Nov 13;9(1):16705 (PMID: 31723194)). Asc1 shows no stereoselectivity between L- and D-serine. Can the authors discuss possible mechanisms resulting in increased renal reabsorption of Dserine than L-serine in IRI with the participation of transporters with modest stereoselectivity for D- vs L-serine?

We appreciate the reviewer’s comments on the degree of protein alteration in proteomics, the functional contributions of ASCT2 and SMCTs, and the alteration of D/L ratios. We have included the possibilities of the technical concerns and the discussion on the roles of ASCT2 and SMCTs as follows.

• Regarding the expression levels, proteomics and snRNA-seq showed the same tendency that ASCT2 increase and SMCTs decrease in IRI conditions. However, the degrees of alterations are more contrast in snRNA-seq. This may be due to the difference in quantification methods and probably points out the underestimated quantification of membrane transport proteins in label-free proteomics. The accuracy of protein quantifications in the label-free proteomics are often impacted by the presence of post-translational modifications and multiple trans-membrane domains like in the case of the membrane transport proteins (Schey KL et al. Biochemistry 2015, doi: 10.1021/bi301604j). Alternative methods of quantitative proteomics may be added in the future for a more thorough comparison. We have added this issue in lines 351-356 of the revised version.

Manuscript lines 351-356

“When evaluating the extent of gene/protein alterations between the control and IRI conditions, we observed that the gene alterations of both Asct2 and Smcts, as revealed by snRNA-sequencing, are more pronounced than the protein alteration ratios obtained from proteomics. This discrepancy may stem from difficulty in the quantification method, especially for membrane transport proteins in label-free quantitative proteomics.”

• For the functional contributions of ASCT2 and SMCTs in the kidney, we admitted the reviewer’s concern about the low affinity of SMCT1. Following the reviewer’s comment, we have included other factors besides transport affinities, e.g. expression levels and turnover rates of the transporters. From the results of both proteomics and snRNA-seq, ASCT2 expression is significantly lower than SMCTs in the normal conditions. snRNA-seq showed that ASCT2 was presented in less than 10% of the cell population (Figure supplement 7). We propose that most of the cells that do not express ASCT2 may use SMCT1 to reabsorb D-serine. This topic was included in the revised manuscript lines 386-404.

Manuscript lines 386-404

“Kinetics analysis of D-serine transport revealed the high affinity by ASCT2 (Km 167 µM) (Foster et al., 2016) and low affinity by SMCT1 (Km 3.39 mM; Figure 5E). In addition to transport affinity, the expression levels and co-localization of multiple transporters within the same cells are critical for elucidating the physiological roles of transporters or transport systems (Sakaguchi et al., 2024). In our proteome data, the chromatogram intensities of Smct1 (2.9 x 109 AU) and Smct2 (1.6 x 108 AU) were significantly higher than that of Asct2 (1.5 x 107 AU) in the control mice (Table 1: abundance in Sham). While direct intensity comparisons between different proteins in mass spectrometry analyses are not precise, they can provide a general indication of relative protein amounts. This finding aligns with the snRNA-seq data, where Asct2 expression was found to be minimal, present in less than 10% of cell populations under both control and IRI conditions, suggesting that many cells do not express Asct2. Conversely, Smct1 and Smct2 show high and ubiquitous expression in control conditions, but their levels are markedly reduced in IRI conditions (Figure supplement 7). Our ex vivo assays demonstrate that both ASCT2 and SMCTs mediate D-serine transport (Figure 7B). Consequently, Asct2 may contribute to D-serine reabsorption due to its high affinity, whereas Smcts, owing to their abundance, particularly in cells lacking Asct2, likely play a significant role in D-serine reabsorption. Moreover, factors such as transport turnover rate (Kcat) and the presence of local canonical substrates are also vital in defining the overall contribution of D-serine transport systems.”

• As for the dramatic alterations of D/L-serine ratios juxtaposed with minimal changes in ASCT2 and SMCTs expression level, we cautiously refrain from drawing a definitive conclusion regarding the entire mechanism. This caution is grounded in the scientific understanding of a comprehensive elucidation of both L-serine transport systems and D-serine transport systems at both apical and basolateral membranes. Nevertheless, we would like to suggest a mechanism at the apical membrane based on the current knowledge.

For D-serine transport systems, we found ASCT2 and SMCTs contributions in this study. Meanwhile, L-serine was previously reported to be mediated mainly by the neutral amino acid transporters B0AT3 (in particular B0AT3; Bröer et al. Physiol Rev 2008; Singer et al. J Biol Chem 2009). Hence, the mechanism behind the alterations of D/L-serine ratios should include B0AT3 functions as well. In IRI conditions, B0AT3 decreased 1.8 fold. We suggest that high ratios of D-/L-serine in IRI conditions are a combined outcome of 1) increase of D-serine reabsorption by ASCT2 enhancement and SMCTs reduction, and 2) decrease of L-serine reabsorption by B0AT3. We have included this suggestion in the Discussion lines 438-451.

Manuscript lines 438-451

“The enantiomeric profiles of serine revealed distinct plasma D/L-serine ratios, with low ratios in the normal control but elevated ratios in IRI, despite the weak stereoselectivity of ASCT2 (Figure 1B). This observation suggested the differential renal handling of D-serine compared to L-serine. While we identified SMCTs as a Dserine transport system, it has been reported that L-serine reabsorption is mediated by B0AT3 (Singer et al., 2009). We propose that the alterations in plasma and urinary D/Lserine ratios are the combined outcomes of: 1) transport systems for L-serine, and 2) transport systems for D-serine. In normal kidneys, the low plasma D/L-serine ratios could result from the efficient reabsorption of L-serine by B0AT3, coupled with the DAAO activity that degrades intracellular D-serine reabsorbed by SMCTs. In IRI conditions, our enantiomeric amino acid profiling revealed low plasma L-serine and high urinary L-serine (Figure supplements 1B, 2B). Additionally, the proteomics analysis indicated a reduction in B0AT3 levels (4h IRI/sham = 0.56 fold; 8h IRI/sham = 0.65 fold; Table S1). These observations suggest that the low L-serine reabsorption in IRI is a result of B0AT3 reduction.”

• In the case of Asc-1, it was reported to be a D-serine transporter in the brain (Rosenberg et al. J Neurosci 2013). Suzuki et al. 2019 showed the increase of Asc-1 in cisplatin-induced tubular injury. Notably, the mRNA of Asc-1 is predominantly found in Henle’s loop, distal tubules, and collecting ducts but not in proximal tubules, and its protein expression level is dramatically low in the kidney (Human Protein Atlas: update on Jun 19, 2023). Furthermore, in this study, Asc-1 expression was not detected in the brush border membrane proteome. Consequently, we have decided not to include Asc-1 in the Discussion of this study, which primarily focuses on the proximal tubules.

Author Response

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

Reviewer #1 (Recommendations For The Authors):

(1) More explanation/description of Fig 3C and 3D would be helpful for readers, including the color code of 3D and black lines shown in both panels.

We have added more description to the legend of Figure 3, and we have used the same color code as in Figure 2, which we now specifically note in the figure legend as well.

(2) Differences between cranial and trunk NCC could be experimentally shown or discussed. Fig 4C shows some differences between these two populations, but in situ, results using Dlc1/Sp5/Pak3 probes in the trunk region may be informative, like Fig 5 supplement 2 for cranial NCCs.

This is an important point. The focus of our study was on cranial neural crest cells, and the single cell sequencing data is therefore truly reflective of only cranial neural crest cells. We have not functionally tested for the roles of Dlc1/Sp5/Pak3 in trunk neural crest cells, however, based on the expression and loss-of-function phenotypes of Sp5 or Pak3 knockout mice, we predict they individually may not play a significant role. It remains plausible that Dlc1 could play an important role in the delamination of trunk neural crest cells, but we have not tested that definitively. Nonetheless, Sabbir et al 2010 showed in a gene trap mouse mutant that Dlc1 is expressed in trunk neural crest cells. Regarding the similarities and differences between cranial and trunk neural crest cells as noted by the reviewer with respect to Figure 4, it’s important to recognize the temporal differences illustrated in Figure 4. Neural crest cell delamination proceeds in a progressive wave from anterior to posterior, but also that the analysis was designed to quantify cell cycle status before and during neural crest cell delamination. We have compared cranial and trunk neural crest cells in more detail in the discussion and also speculate what might happen in the trunk based on what we know from other species.

(3) Discussion can be added about the potential functions of Dlc1 for NCC migration and/or differentiation based on available info from KO mice.

We have added specific details regarding the published Dlc1 knockout mouse phenotype to the discussion, particularly with respect to the craniofacial anomalies which included frontonasal prominence and pharyngeal arch hyperplasia, and defects in neural tube closure and heart development. Although the study didn’t investigate the mechanisms underpinning the Dlc1 knockout phenotype, the craniofacial morphological anomalies would be consistent with a deficit in neural crest cell delamination reducing the number of migrating neural crest cells, as we observed in our Dlc1 knockdown experiments.

Reviewer #2 (Recommendations For The Authors):

The authors used the (Tg(Wnt1-cre)11Rth Tg(Wnt1-GAL4)11Rth/J) line but work from the Bush lab (see Lewis et al., 2013) has demonstrated fully penetrant abnormal phenotypes that affect the midbrain neuroepithelium, increased CyclinD1 expression and overt cell proliferation as measured by BrdU incorporation. The authors should explain why they used this mouse line instead of the Wnt1-Cre2 mice (129S4-Tg(Wnt1-cre)1Sor/J) in the Jackson Laboratory (which lacks the phenotypic effects of the original Wnt1-Cre line), or a "Cre-only" control, or at a minimum explain the steps they took to ensure there were no confounding effects on their study, especially since cell proliferation was a major outcome measure.

This is an important point, and we thank the reviewer for raising it. Yes, it has been reported that the original Wnt1Cre mice exhibit a midbrain phenotype (Ace et al. 2013). However, it has also been noted that Wnt1Cre2 can exhibit recombination in the male germline leading to ubiquitous recombination (Dinsmore et al., 2022). Therefore, to avoid any potential for bias, we used an equal number of cells derived from the Wnt1 and F10N transgenic line embryos in our scRNA-seq, and this included multiple non-Cre embryos. Our scRNA-seq analysis was therefore not dependent upon Wnt1-Cre, but also because we used whole heads not fluorescence sorted cells. However, Wnt1-Cre lineage tracing was advantageous from a computational perspective to help define cells that were premigratory and migratory in concert with Mef2c-lacZ ¬based on their expression of YFP, LacZ or both. We note these specifics more clearly in the methods.

The Results section (line 122) states that scRNA-seq was performed on dissociated cranial tissues but the Methods section (lines 583-584) implies that whole E8.5 mouse embryos were dissociated. Which was dissociated, whole embryos or just cranial tissues? Obviously, the latter would be a better strategy to enrich for cranial neural crest, but the authors also examine the trunk neural crest. This should be clarified in the text.

We apologize that some of the details regarding the tissue isolation were confusing and we have clarified this in the methods and the text. For the record, after isolating E8.5 embryos, we then dissected the head from those embryos, and performed scRNA-seq on dissociated cranial tissues. As the reviewer correctly noted, this approach strategically enriches for cranial neural crest cells.

The authors do not justify why they chose a knockdown strategy, which has its limitations including its systemic injection into the amniotic cavity, its likely global and more variable effects, and its need to be conducted in culture. Why the authors did not instead use a Wnt1-Cre-mediated deletion of Dlc1, which would have been "cleaner" and more specific to the neural crest, is not clear (maybe so they could specifically target different Dcl1 isoforms?). Also, the authors use Sox10 as a marker to count neural crest cells, but Sox10 may only label a subset of neural crest cells and thus some unaffected lineages may not have been counted. The authors should mention what is known about the regulation of Dcl1 by Sox10 in the neural crest. Although the data are persuasive, a second marker for counting neural crest cells following knockdown would make the analysis more robust. Can the authors explain why they did not simply use the Mef2c-F10N-LacZ line and count LacZ-positive cells (if fluorescence signal was required for the quantification workflow, then could they have used an anti-beta Galactosidase antibody to label cells)?

We thank the reviewer for raising these important considerations. It has previously been noted that although Wnt1-Cre is the gold standard for conditional deletion analyses in neural crest cell development, especially migration and differentiation, it is not a good tool for functional studies of the specification and delamination of neural crest cells due to the timing of Wnt1 expression and Cre activation and excision (see Barriga et al., 2015). Therefore, we chose a knockdown strategy instead, and also because it allows us to more rapidly evaluate gene function. We agree that there are limitations to the approach with respect to variability, however, this is outweighed by the ability to repeatedly perform the knockdown at multiple and more relevant temporal stages such as E7.5 (which is prior to the onset of Wnt1-Cre activity), as well as target different isoforms, and also treat large numbers of embryos for quantitative analyses. The advantage of using Sox10 as a marker for counting neural crest cells is that at the time of analysis, cranial neural crest cells are still migrating towards the frontonasal prominences and pharyngeal arches, and the overwhelming majority of these cells are Sox10 positive. Moreover, we can therefore assay every Dlc1 knockdown embryo for Sox10 expression and count the number of migrating neural crest cells. The limitation of using the Mef2c-F10N-LacZ line is that this transgenic line is maintained as a heterozygote, and thus only half the embryos in a litter could reasonably be expected to be lacZ+. But combining Sox10 and Mef2c-F10N-LacZ fluorescent immunostaining for similar analyses in the future is a great idea.

Reviewer #3 (Recommendations For The Authors):

The putative intermediate cells differentially express mRNAs for genes involved in cell adhesion, polarity, and protrusion relative to bona fide premigratory cells (Fig. 2E). This is persuasive evidence, but only differentially expressed genes are shown. Discussing those markers that have not yet changed, e.g. Cdh1 or Zo1 (?), would be instructive and help to clarify the order of events.

We thank the author for this suggestion and we have provided more detail about adherens junction and tight junctions. Cdh1 is not expressed, and although Myh9 and Myh10 are expressed, we did not detect any significant changes. ZO1 is a tight junction protein encoded by the gene Tjp1, which along with other tight junctions protein encoding genes, is downregulated in intermediate NCCs as shown in the Figure 2E.

It is unclear whether the two putative intermediate state clusters differ other than their stage of the cell cycle. Based on the trajectory analysis in Fig. 3C-D, the authors state that these two populations form simultaneously and independently but then merge into a single population. However, without further differential expression, it seems more plausible that they represent a single population that is temporarily bifurcated due to cell cycle asynchrony.

We have addressed the cell cycle question in the discussion by noting that while it is possible the transition states represent a single population that is temporarily bifurcated due to cell cycle asynchrony, if this were true, then we should expect S phase inhibition to eliminate both transition state groups. Instead, our trajectory analyses suggest that the transition states are initially independent, and furthermore, S phase inhibition did not affect delamination of the other population of neural crest cells.

The authors do not present an in-depth comparison of these neural crest intermediate states to previously reported cancer intermediate states. This analysis would reveal how similar the signatures are and thus how extrapolatable these and future findings in delaminating neural crest are to different types of cancer.

We have also added more detail to the discussion to address the potential for similarities and differences in neural crest intermediate states compared to previously reported cancer intermediate states. The challenge, however, is that none of the cancer intermediate states have been characterized at a molecular level. Nonetheless, with the limited molecular markers available, we have not identified any similarities so far, but our datasets are now available for comparison with future cancer EMP datasets.

The reduction in SOX10+ cells may be in part or wholly attributable to inhibition of proliferation AFTER delamination. Showing that there are premigratory NCCs in G2/M at ~E8.0 would bolster the argument that this population is present from the earliest stages.

The presence of premigratory neural crest cells in G2/M is shown by the scRNA-seq data and cell cycle staining data in the neural plate border.

Lines 248-249: The pseudo-time analysis in Fig 3C/D does indicate that the two most mature cell clusters (pharyngeal arch and frontonasal mesenchyme) may arise from common or similar migratory progenitors. However, given the decades of controversy about fate restriction of neural crest cells, the statement that "EMT intermediate NCC and their immediate lineages are not fate restricted to any specific cranial NCC derivative at this timepoint" should be toned down so as to not give the impression that they have identified common progenitors of ectomesenchyme and neuro/glial/pigment derivatives.

We appreciate this comment, because as the reviewer noted, there has been considerable literature and debate about the fate restriction and plasticity of neural crest cells, and indeed we did not intend to imply we have identified common progenitors of ectomesenchyme and neuro/glial/pigment derivatives. That can only be truly functionally demonstrated by clonal lineage tracing analyses. Rather, we interpret our pseudo-time analyses to indicate that irrespective of cell cycle status at the time of delamination, these two populations come together with equivalent mesenchymal and migratory properties, but in the absence of fate determination in the collective of cells. This does not mean that individual cells are common progenitors of both ectomesenchyme and neuro/glial/pigment derivatives. The nuance is important, and we address this more carefully in the text.

Lines 320-321: "...this overlap in expression was notably not observed in older embryos in areas where EMT had concluded". It is unclear whether the markers no longer overlap in older embryos (i.e. segregate to distinct populations) or are simply no longer expressed.

The data in Figure 5 demonstrates the dynamic and overlapping expression of Dlc1, Sp5 and Pak3 in the different clusters of cells as they transition from being neuroepithelial to mesenchymal. In contrast to Sp5 and Pak3, Dlc1 is not expressed by premigratory neural crest cells but is expressed at high levels in all EMT intermediate stage neural crest cells. Later as Dlc1 continues to be expressed in migrating neural crest cells, Pak3 and Sp5 are downregulated. But the absence of overlapping expression in the dorsolateral neural plate at the conclusion of EMT coincides with their downregulation in that territory.

In the final results section on Dlc1, the previously published mutant mouse lines are referenced as having "craniofacial malformation phenotypes". The lack of detail given on what those malformations are (assuming descriptions are available) makes the argument that they may be related to insufficient delamination less persuasive. The degree of knockdown correlates so well with the percentage reduction in migratory neural crest (Fig. 6) that one would imagine a null mutant to have a very severe phenotype.

The inference from the reviewer is correct and indeed Dlc1 null mutant mice do have a severe phenotype. We have added more specific details regarding the craniofacial and other phenotypes of the Dlc1 mutant mice to the discussion. Of note the frontonasal prominences and the pharyngeal arches are hypoplastic in E10.5 Dlc1 mutant embryos, which would be consistent with a neural crest cell deficit. Although a deficit in neural crest cells can be caused my multiple distinct mechanisms, our Dlc1 knockdown analyses suggest that the phenotype is due to an effect on neural crest cell delamination which diminishes the number of migrating neural crest cells.

Use the same y-axis for Fig. 4C/D

This has been corrected.

Fig. 6C: Please note in the panel which gene is being measured by qPCR

This has been corrected to denoted Dlc1.

Lines 108-117: More concise language would be appropriate here.

As requested, we were more succinct in our language and have shortened this section.

The SABER-FISH images are very dim. I realize the importance of not saturating the pixels, but the colors are difficult to make out.

We thank the reviewer for pointing this out and have endeavored to make the SABER-FISH images brighter and easier to see.

Author Response

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

Reviewer #1 (Public Review):

In this manuscript, the authors report a molecular mechanism for recruiting syntaixn 17 (Syn17) to the closed autophagosomes through the charge interaction between enriched PI4P and the C-terminal region of Syn17. How to precisely control the location and conformation of proteins is critical for maintaining autophagic flux. Particularly, the recruitment of Syn17 to autophagosomes remains unclear. In this paper, the author describes a simple lipid-protein interaction model beyond previous studies focusing on protein-protein interactions. This represents conceptual advances.

We would like to thank Reviewer #1 for the positive evaluation of our study.

Reviewer #2 (Public Review):

Summary:

Syntaxin17 (STX17) is a SNARE protein that is recruited to mature (i.e., closed) autophagosomes, but not to immature (i.e., unclosed) ones, and mediates the autophagosome-lysosome fusion. How STX17 recognizes the mature autophagosome is an unresolved interesting question in the autophagy field. Shinoda and colleagues set out to answer this question by focusing on the C-terminal domain of STX17 and found that PI4P is a strong candidate that causes the STX17 recruitment to the autophasome.

Strengths:

The main findings are: 1) Rich positive charges in the C-terminal domain of STX17 are sufficient for the recruitment to the mature autophagosome; 2) Fluorescence charge sensors of different strengths suggest that autophagic membranes have negative charges and the charge increases as they mature; 3) Among a battery of fluorescence biosensors, only PI4P-binding biosensors distribute to the mature autophagosome; 4) STX17 bound to isolated autophagosomes is released by treatment with Sac1 phosphatase; 5) By dynamic molecular simulation, STX17 TM is shown to be inserted to a membrane containing PI4P but not to a membrane without it. These results indicate that PI4P is a strong candidate that STX17 binds to in the autophagosome.

We would like to thank Reviewer #2 for pointing out these strengths.

Weaknesses:

• It was not answered whether PI4P is crucial for the STX17 recruitment in cells because manipulation of the PI4P content in autophagic membranes was not successful for unknown reasons.

As we explained in the initial submission, we tried to deplete PI4P in autophagosomes by multiple methods but did not succeed. In this revised manuscript, we added the result of an experiment using the PI 4-kinase inhibitor NC03 (Figure 4―figure supplement 1), which shows no significant effect on the autophagosomal PI4P level and STX17 recruitment.

Author response image 1.

The PI 4-kinase inhibitor NC03 failed to suppress autophagosomal PI4P accumulation and STX17 recruitment. HEK293T cells stably expressing mRuby3–STX17TM (A) or mRuby3–CERT(PHD) (B) and Halotag-LC3 were cultured in starvation medium for 1 h and then treated with and without 10 μM NC03 for 10 min. Representative confocal images are shown. STX17TM- or CERT(PHD)-positive rates of LC3 structures per cell (n > 30 cells) are shown in the graphs. Solid horizontal lines indicate medians, boxes indicate the interquartile ranges (25th to 75th percentiles), and whiskers indicate the 5th to 95th percentiles. Differences were statistically analyzed by Welch’s t-test. Scale bars, 10 μm (main), 1 μm (inset).

• The molecular simulation study did not show whether PI4P is necessary for the STX17 TM insertion or whether other negatively charged lipids can play a similar role.

As the reviewer suggested, we performed the molecular dynamics simulation using membranes with phosphatidylinositol, a negatively charged lipid. STX17 TM approached the PI-containing membrane but was not inserted into the membrane within a time scale of 100 ns in simulations of all five structures. This data suggests that PI4P, which is more negatively charged than PI, is required for STX17 insertion. Thus, we have included these data in Figure 5E and F and added the following text to Lines 242–244. “Moreover, if the membrane contained phosphatidylinositol (PI) instead of PI4P, STX17 approached the PI-containing membrane but was not inserted into the membrane (Figure 5E, F, Video 3)."

Author response image 2.

(E) An example of a time series of simulated results of STX17TM insertion into a membrane consisting of 70% phosphatidylcholine (PC), 20% phosphatidylethanolamine (PE), and 10% phosphatidylinositol (PI). STX17TM is shown in blue. Phosphorus in PC, PE and PI are indicated by yellow, cyan, and orange, respectively. Short-tailed lipids are represented as green sticks. The time evolution series are shown in Video 3. (F) Time evolution of the z-coordinate of the center of mass (z_cm) of the transmembrane helices of STX17TM in the case of membranes with PI. Five independent simulation results are represented by solid lines of different colors. The gray dashed lines indicate the locations of the lipid heads. A scale bar indicates 5 nm.

• The question that the authors posed in the beginning, i.e., why is STX17 recruited to the mature (closed) autophagosome but not to immature autophagic membranes, was not answered. The authors speculate that the seemingly gradual increase of negative charges in autophagic membranes is caused by an increase in PI4P. However, this was not supported by the PI4P fluorescence biosensor experiment that showed their distribution to the mature autophagosome only. Here, there are at least two possibilities: 1) The increase of negative charges in immature autophagic membranes is derived from PI4P. However the fluorescence biosensors do not bind there for some reason; for example, they are not sensitive enough to recognize PI4P until it reaches a certain level, or simply, their binding does not occur in a quantitative manner. 2) The negative charge in immature membranes is not derived from PI4P, and PI4P is generated abundantly only after autophagosomes are closed. In either case, it is not easy to explain why STX17 is recruited to the mature autophagosome only. For the first scenario, it is not clear how the PI4P synthesis is regulated so that it reaches a sufficient level only after the membrane closure. In the second case, the mechanism that produces PI4P only after the autophagosome closure needs to be elucidated (so, in this case, the question of the temporal regulation issue remains the same).

We thank the reviewers for pointing this out. While the probe for weakly negative charges (1K8Q) labeled both immature and mature autophagosomes, the probes for intermediate charges (5K4Q and 3K6Q) and PI4P labeled only mature autophagosomes (Figure 2F, Figure 2–figure supplement 1B). Thus, we think that the autophagosomal membrane rapidly and drastically becomes negatively charged, and at the same time, PI4P is enriched. Although immature membranes may have weak negative charges, we did not examine which lipids contribute to the negative charges. Thus, we have added the following sentences to the Discussion part.

“Our data of the 1K8Q probe suggest that immature autophagosomal membranes may also have slight negative charges (Figure 2E). Although the source of the negative charge of immature autophagosomes is currently unknown, it may be derived from low levels of PI4P, which is undetectable by the PI4P probes and/or other negatively charged lipids such as PI and PS (Schmitt et al., EMBO Rep, 2022).” (Lines 279–283) “In any case, it would be important to elucidate how PI 4-kinase activity or PI4P synthesis is upregulated during autophagosome maturation.” (Lines 302–303)

Reviewer #3 (Public Review):

Summary:

In this study, the authors set out to address the question of how the SNARE protein Syntaxin 17 senses autophagosome maturation by being recruited to autophagosomal membranes only once autophagosome formation and sealing is complete. The authors discover that the C-terminal region of Syntaxin 17 is essential for its sensing mechanism that involves two transmembrane domains and a positively charged region. The authors discover that the lipid PI4P is highly enriched in mature autophagosomes and that electrostatic interaction with Syntaxin 17's positively charged region with PI4P drives recruitment specifically to mature autophagosomes. The temporal basis for PI4P enrichment and Syntaxin 17 recruitment to ensure that unsealed autophagosomes do not fuse with lysosomes is a very interesting and important discovery. Overall, the data are clear and convincing, with the study providing important mechanistic insights that will be of broad interest to the autophagy field, and also to cell biologists interested in phosphoinositide lipid biology. The author's discovery also provides an opportunity for future research in which Syntaxin 17's c-terminal region could be used to target factors of interest to mature autophagosomes.

Strengths:

The study combines clear and convincing cell biology data with in vitro approaches to show how Syntaxin 17 is recruited to mature autophagosomes. The authors take a methodical approach to narrow down the critical regions within Syntaxin 17 required for recruitment and use a variety of biosensors to show that PI4P is enriched on mature autophagosomes.

We would like to thank Reviewer #3 for the positive comments.

Weaknesses:

There are no major weaknesses, overall the work is highly convincing. It would have been beneficial if the authors could have shown whether altering PI4P levels would affect Syntaxin 17 recruitment. However, this is understandably a challenging experiment to undertake and the authors outlined their various attempts to tackle this question.

We thank Reviewer #3 for pointing this out. Please see our above response to Reviewer #2 (Public Review).

In addition, clear statements within the figure legends on the number of independent experimental repeats that were conducted for experiments that were quantitated are not currently present in the manuscript.

As pointed out by Reviewer #3, we have added the number of independent experimental repeats in the figure legends.

Reviewer #1 (Recommendations For The Authors):

This paper is well written and all experiments were conducted with a high standard. Several minor issues should be addressed before final publication.

(1) To further confirm the charge interaction, a charge screening experiment should be performed for Fig. 2A.

We have asked Reviewer #1 through the editor what this experiment meant and understood that it was to see the effects of high salt concentrations. We monitored the association of GFP-STX17TM with liposomes in the presence or absence of 1 M NaCl and found that it was blocked in a high ionic buffer. This data supports the electrostatic interaction of STX17 with membranes. We have included this data in Figure 2B and added the following sentences to Lines 124–126.

“The association of STX17TM with PI4P-containing membranes was abolished in the presence of 1 M NaCl (Figure 2B). These data suggest that STX17 can be recruited to negatively charged membranes via electrostatic interaction independent of the specific lipid species.”

Author response image 3.

GFP–STX17TM translated in vitro was incubated with rhodamine-labeled liposomes containing 70% PC, 20% PE and 10% PI4P in the presence of 1 M NaCl or 1.2 M sucrose. GFP intensities of liposomes were quantified and shown as in Figure 1C (n > 30).

(2) The authors claim that "Autophagosomes become negatively charged during maturation", based on experiments using membrane charge probes. Since it's mainly about the membrane, it's better to refine the claim to "The membrane of autophasosomes becomes...", which would be more precise and close to the topic of this paper.

We would like to thank the reviewer for pointing this out. This point is valid. As recommended, we have collected the phrases “Autophagosomes become negatively charged during maturation” to “The membrane of autophagosomes becomes negatively charged during maturation” (Line 72, 118, 262, 969 (title of Figure2), 1068 (title of Figure2–figure supplyment1)).

(3) The authors should add more discussion regarding the "specificity" for recruiting Syn17 through the charge interaction. Particularly, how Syn17 could be maintained before the closure of autophagosomes? For the MD simulations in Fig. 5, the current results don't add much to the manuscript. The cell biology experiments have demonstrated the conclusion. The authors could try to find more details about the insertion by analyzing the simulation movies. Do membrane packing defects play a role during the insertion process? A similar analysis was conducted for alpha-synuclein (https://pubmed.ncbi.nlm.nih.gov/33437978/).

Regarding the mechanism of STX17 maintenance in the cytosol, we do not think that other molecules, such as chaperones, are essential because purified recombinant mGFP-STX17TM used in this study is soluble. However, it does not rule out such a mechanism, which would be a future study.

In the paper by Liu et al. (PMID: 33437978), small liposomes with diameters of 25–50 nm are used. Therefore, there are packing defects in the highly curved membranes, to which alpha-synuclein helices are inserted in a curvature-dependent manner. On the other hand, autophagosomes are much larger (~1 um in diameter) and almost flat for STX17 molecules, so we think it is unlikely that STX17 recognizes the packing defect.

Reviewer #2 (Recommendations For The Authors):

• The two (and other) possibilities with regards to the interpretation of the negative charge/PI4P result in autophagic membranes are hoped to be discussed.

As mentioned above, we have added the following sentences to the Discussion section. “Our data of the 1K8Q probe suggest that immature autophagosomal membranes may also have slight negative charges (Figure 2E). Although the source of the negative charge of immature autophagosomes is currently unknown, it may be derived from low levels of PI4P, which is undetectable by the PI4P probes and/or other negatively charged lipids such as PI and PS (Schmitt et al., EMBO Rep, 2022).” (Lines 279–283)

“In any case, it would be important to elucidate how PI 4-kinase activity or PI4P synthesis is upregulated during autophagosome maturation.” (Lines 302–303)

• Fluorescence biosensors are convenient to give an overview of the intracellular distribution of various lipids, but some of them show false-negative results. For example, evectin-2-PH for PS binds to endosomes but not to the plasma membrane, even though the latter contains abundant PS. With regards to PI4P, some biosensors illuminate both the Golgi and autophagosome, while others do not appear to bind the Golgi. Moreover, fluorescence biosensors for PI(3,5)P2 and PI(3,4)P2, which are also candidates for the STX17 insertion issue, are less reliable than others (e.g., those for PI3P and PI(4,5)P2). These problems need to be considered.

We agree with Reviewer #2 that fluorescence biosensors are not perfect for detecting specific lipids. Based on the Reviewer’s suggestion, we have included a comment on this in the Discussion section as follows (Lines 265–268).

“Given the possibility that fluorescence lipid probes may give false-negative results, a more comprehensive biochemical analysis, such as lipidomics analysis of mature autophagosomes, would be imperative to elucidate the potential involvement of other negatively charged lipids.”

• A negative control for the PI4P biosensor, i.e., a mutant lacking the PI4P binding ability, is better to be tested to confirm the presence of PI4P in autophagosomes.

We would like to thank the Reviewer for this comment. We conducted the suggested experiment and confirmed that the CERT(PHD)(W33A) mutant, which is deficient for PI4P binding (Sugiki et al., JBC. 2012), was diffusely present in the cytosol and did not localize to STX17-positive autophagosomes. This data supports our conclusion that PI4P is indeed present in autophagosomes. We have included this data in Figure 3–figure supplement 2A and explained it in the text (Lines 164–166).

Author response image 4.

Mouse embryonic fibroblasts (MEFs) stably expressing GFP–CERT(PHD)(W33A) and mRuby3–STX17TM were cultured in starvation medium for 1 h. Bars indicate 10 μm (main images) and 1 μm (insets).

• As a control to the molecular dynamic simulation study, STX17 TM insertion into a membrane containing other negative charge lipids, especially PI, needs to be tested. PI is a negative charge lipid that is likely to exist in autophagic membranes (as suggested by the authors' past study).

We thank the reviewers for this suggestion. As mentioned above (Reviewer #2, Public Review), we performed the molecular dynamics simulation using membranes containing PI and added the results in Figure 5E and F and Video 3.

• If the putative role of PI4P could be shown in the cellular context, the authors' conclusion would be much strengthened. I wonder if overexpression of PI4P fluorescence biosensors, especially those that appear to bind to the autophagosome almost exclusively, may suppress the recruitment of STX17 there.

We would like to thank the Reviewer for asking this question. In MEFs stably overexpressing PI4P probes driven by the CMV promoter, STX17 recruitment was not affected. Thus, simple overexpression of PI4P probes does not appear to be effective in masking PI4P in autophagosomes.

Another idea is to use an appropriate molecule (e.g., WIPI2, ATG5) and to recruit Sac1 to autophagic membranes by using the FRB-FKBP system or the like. I hope these and other possibilities will be tested to confirm the importance of PI4P in the temporal regulation of STX17 recruitment.

We tried the FRB-FKBP system using the phosphatase domain of yeast Sac1 fused to FKBP and LC3 fused to FRB, but unfortunately, this system failed to deplete PI4P from the autophagosomal membrane.

Reviewer #3 (Recommendations For The Authors):

A few areas for suggested improvement are:

(1) It would be helpful if the authors could clarify for all figures how many independent experiments were conducted for all experiments, particularly those that have quantitation and statistical analyses.

As pointed out by Reviewer #3, we have added the number of independent experimental repeats in the figure legends.

The authors made several attempts to modulate PI4P levels on autophagosomes although understandably this proved to be challenging. A couple of suggestions are provided to address this area:

(2) Given the reported role of GABARAPs in PI4K2a recruitment and PI4P production on autophagosomes, as well as autophagosome-lysosome fusion (Nguyen et al (2016) J Cell Biol) it would be worthwhile to assess whether GABARAP TKO cells have reduced PI4P and reduced Stx17 recruitment

According to the Reviewer’s suggestion, we examined the localization of STX17 TM and the PI4P probe CERT(PHD) in ATG8 family (LC3/GABARAP) hexa KO HeLa cells that were established by the Lazarou lab (Nguyen et al., JCB 2016). As in WT cells, STX17 TM and CERT(PHD) were still colocalized with each other in hexa KO cells, suggesting that neither STX17 recruitment nor PI4P enrichment depends on ATG8 family proteins (note: the size of autophagosomes in HeLa cells is smaller than in MEFs, making it difficult to observe autophagosomes as ring-shaped structures). We have included this result in Figure 3–figure supplement 2(F) and explained it in the text (Lines 194–196, 198).

Author response image 5.

(F) WT and ATG8 hexa KO HeLa cells stably expressing GFP–STX17TM and transiently expressing mRuby3–CERT(PHD) were cultured in starvation medium. Bars indicate 10 μm (main images) and 1 μm (insets).

(3) Can the authors try fusing Sac1 to one of the PI4P probes (CERT(PHD)) that were used, or alternatively to the c-terminus of Syntaxin 17? This approach would help to recruit Sac1 only to mature autophagosomes and could therefore prevent the autophagosome formation defect observed when fused to LC3B that targeted Sac1 to autophagosomes as they were forming. Understandably, this approach might seem a bit counterintuitive since the phosphatase is removing PI4P which is what is recruiting it but it could be a viable approach to keep PI4P levels low enough on mature autophagosomes so that Syntaxin 17 is no longer recruited. A Sac1 phosphatase mutant might be needed as a control.

We would like to thank the Reviewer for these suggestions. We tried the phosphatase domain of yeast Sac1 or human SAC1 fused with STX17TM, but unfortunately, these fusion proteins did not deplete PI4P from autophagosomes.

Author Response

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

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

To resolve and further test the claim that TBI did not induce cell proliferation:

How many brains did they analyse? Sample sizes must be provided in Figure S1.

As per reviewer’s suggestion, we removed one of the unsupported claims shown in Figure S1. The original Figure S1 is shown below with the sample number added.

Author response image 1.

The authors could either improve the TBI method or the detection of cells in S-phase, mitosis or cycling. They could use PCNA-GFP or BrdU, EdU or FUCCI instead and at least provide evidence that they can detect cells in S-phase in intact brains. Timing is critical (ie cell cycle is longer than in larvae) so multiple time points should be tested. Or they could use pH3 but test more time points and rather large sample sizes. If they are not able to provide any evidence, then their lack of evidence is no evidence. The authors should consider removing pH3 and PCNA-GFP related claims instead.

We have removed pH3 and PCNA-GFP related results and claims.

Other unsupported claims:

Figure 2A-C is not very clear what they are showing, but it is not evidence of astrocyte hypertrophy. It does not have cellular resolution and does not show the cell size, membranes, nor number

(1) We have avoided the term “hypertrophy” and changed the description throughout the text to “astrocyte swelling”.

(2) Images in the resolution of Figure 2E and 2F were able to show the enlarged soma of astrocytes, suggesting swelling.

What is the point of using RedStinger in Figure 2?

We used RedStinger to label the astrocyte nuclei.

Figure S5 is not convincing, as anti-Pvr does not look localised to specific cells. Instead, it looks like uniform background. If they really think the antibody is localised, they should do double stainings with cell type specific markers. If the antibody does not work, then remove the data and the claim. They could test with RNAi knock-down in specific cell types and qRT-PCR which cells express pvr instead.

We have removed the claim that “Pvr is predominantly expressed in astrocytes” and changed the description to “Immunostainings using the anti-Pvr antibodies revealed that endogenous Pvr expression is low in the control brains, yet significantly enhanced upon TBI. Reducing Pvr expression, but not Pvr overexpression, in astrocytes blocked the TBI-induced increase of Pvr expression (Figure S5)”.

Figure S6: it is unclear what they are trying to show, but these data do not demonstrate that astrocytes do not engulf debris after TBI, as there isn't sufficient cellular resolution to make such claim. Firstly, they analyse one single cell per treatment. Secondly, the cell projections are not visible in these images, and therefore engulfment cannot be seen. The authors could remove the claim or visualise whether astrocytes phagocytose debris or not either using clones or with TEM.

We agree with the reviewer that our images do not have the resolution to make this claim. We have removed Figure S6 and corresponding text description.

On statistics:

The statistical analysis needs revising as it is wrong in multiple places, eg Fig.1F,G,H; Figure 2D. They only use Student t-tests. These can only be used when data are continuous, distributed uniformly and only two samples are compared; if more than 2 samples, distributed uniformly, then use One-Way ANOVA and multiple comparisons tests. If data are categorical, use Chi-Square.

We have double checked and compared the experimental group to the control separately using the Student t-tests throughout the study.

Other points for improvement:

Figure 2E,F: what are GFP puncta and how are they counted?

I. Each GFP puncta looks like a little circle, likely representing a functional or dysfunctional structure. The biology of the GFP puncta is currently unkonwn.

II. We used the ImageJ to quantify the GFP puncta:

(1) Image- type-8 bits

(2) Process-subtract background (Rolling ball radio:10)

(3) Image-Adjust-Threshold-Apply

(4) Analyze-Measure-set measurements-choose “area” “limit to threshold”-OK

(5) Count the puncta number in the choosing area.

(6) Get the number of puncta per square micron.

All genotypes must be provided (including for MARCM clones), currently they are not.

We have shown the full genotype in the corresponding legend.

Figure 7O,P indicate on figure that these are RNAi

We have revised the labels to RNAi in Figure 7O,P.

Reviewer #2 (Recommendations For The Authors):

Several typos are present in the text.

We have read the manuscript carefully and corrected typos throughout.

Author Response

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

eLife assessment

This study presents valuable findings on the roles of the axon growth regulator Sema7a in the formation of peripheral sensory circuits in the lateral line system of zebrafish. The evidence supporting the claims of the authors is solid, although further work directly testing the roles of different sema7a isoforms would strengthen the analysis. The work will be of interest to developmental neuroscientists studying circuit formation.

Public Reviews:

Reviewer #1 (Public Review):

In this work, Dasguta et al. have dissected the role of Sema7a in fine tuning of a sensory microcircuit in the posterior lateral line organ of zebrafish. They attempt to also outline the different roles of a secreted verses membrane-bound form of Sema7a in this process. Using genetic perturbations and axonal network analysis, the authors show that loss of both Sema7a isoforms causes abnormal axon terminal structure with more bare terminals and fewer loops in contact with presynaptic sensory hair cells. Further, they show that loss of Sema7a causes decreased number and size of both the pre- and post-synapse. Finally, they show that overexpression of the secreted form of Sema7a specifically can elicit axon terminal outgrowth to an ectopic Sema7a expressing cell. Together, the analysis of Sema7a loss of function and overexpression on axon arbor structure is fairly thorough and revealed a novel role for Sema7a in axon terminal structure. However, the connection between different isoforms of Sema7a and the axon arborization needs to be substantiated. Furthermore, an autocrine role for Sema7a on the presynaptic cell is not ruled out as a contributing factor to the synaptic and axon structure phenotypes.

Finally, critical controls are absent from the overexpression paradigm.

Comments: Thank you for your valuable comments. We have analyzed the hair cell scRNA transcriptome data of zebrafish neuromasts from published works and have not identified known expression of receptors of the Sema7A protein, particularly PlexinC1 and Integrin β1 molecules (reference 4 and 15) in hair cells. This result suggests that the Sema7A protein molecule, either secreted or membrane-bound, does not possess its cognate receptor to elicit an autocrine function on the hair cells. Moreover, the GPI-anchored Sema7A lacks a cytosolic domain. So it is unlikely that Sema7A signaling directly induces the formation of presynaptic ribbons. We propose that the decrease in average number and area of synaptic aggregates likely reflects decreased stability of the synaptic structures owing to lack of contact between the sensory axons and the hair cells, which has been identified in zebrafish neuromasts (reference 38).

Thank you for pointing missing critical control experiments. Additional control experiments (lines 333-346) with a new figure (Figure 5) have been added.

These issues weaken the claims made by the authors including the statement that they have identified differential roles for the GPI-anchored verses secreted forms of Sema7a on synapse formation and as a chemoattractant for axon arborization respectively.

Comments: We have rephrased our statement and argue in lines 428-430 that our experiments “suggest a potential mechanism for hair cell innervation in which a local Sema7Asec diffusive cue likely consolidates the sensory arbors at the hair cell cluster and the membrane-anchored Sema7A-GPI molecule guides microcircuit topology and synapse assembly.”

The manuscript itself would benefit from the inclusion of details in the text to help the reader interpret the figures, tools, data, and analysis.

Comments: We have made significant revisions to the text and figures to improve clarity and consistency of the manuscript.

Reviewer #2 (Public Review):

In this work, Dasgupta et al. investigates the role of Sema7a in the formation of peripheral sensory circuit in the lateral line system of zebrafish. They show that Sema7a protein is present during neuromast maturation and localized, in part, to the base of hair cells (HCs). This would be consistent with pre-synaptic Sema7a mediating formation and/or stabilization of the synapse. They use sema7a loss-of-function strain to show that lateral line sensory terminals display abnormal arborization. They provide highly quantitative analysis of the lateral line terminal arborization to show that a number of specific topological parameters are affected in mutants. Next, they ectopically express a secreted form of Sema7a to show that lateral line terminals can be ectopically attracted to the source. Finally, they also demonstrate that the synaptic assembly is impaired in the sema7a mutant. Overall, the data are of high quality and properly controlled. The availability of Sema7a antibody is a big plus, as it allows to address the endogenous protein localization as well to show the signal absence in the sema7a mutant. The quantification of the arbor topology should be useful to people in the field who are looking at the lateral line as well as other axonal terminals. I think some results are overinterpreted though. The authors state: "Our findings demonstrate that Sema7A functions both as a juxtracrine and as a secreted cue to pattern neural circuitry during sensory organ development." However, they have not actually demonstrated which isoform functions in HCs (also see comments below).

Comments: Thank you for making this point. To investigate the presence of both sema7a transcripts in the hair cells of the lateral-line neuromasts, we used the Tg(myo6b:actb1EGFP) transgenic fish to capture the labeled hair cells by fluorescence-activated cell sorting (FACS) and isolated total RNA. Using transcript specific DNA oligonucleotide primers, we have identified the presence of both sema7a transcript variants in the hair cell of the neuromast. Even though we have not developed transcript specific knockout animals, we speculate that the presence of both transcript variants in the hair cell implies that they function in distinct fashion. We have changed our interpretation in lines 32-34 to “Our findings propose that Sema7A likely functions both as a juxtracrine and as a secreted cue to pattern neural circuitry during sensory organ development.”

In future we will utilize the CRISPR/Cas9 technique to target the unique C-terminal domain of the GPI-anchored sema7a transcript variant. We believe that this will only perturb the formation of the full-length Sema7A protein and help us determine the role of the membrane-bound Sema7AGPI molecule as well as the Sema7Asec in sensory arborization and synaptic assembly.

In addition, they have to be careful in interpreting their topology analysis, as they cannot separate individual axons. Thus, such analysis can generate artifacts. They can perform additional experiments to address these issues or adjust their interpretations.

Comments: Thank you for this insightful comment. In a previous eLife publication from our laboratory, we utilized the serial blockface scanning electron micrograph (SBFSEM) technique to characterize the connectome of the neuromast microcircuit where patterns of innervation of all the individual axons can be delineated in five-days-old larvae (reference 8). However, the collective behavior of all the sensory axons that build the innervation network remained enigmatic, especially in a living animal during development. In this paper we addressed how the sensory-axon collective behaves around the clustered hair cells and build the innervation network in living animals during diverse developmental stages. Our analyses have not only identified how the axons associates with the hair cell cluster as the organ matures, but also discovered distinct topological features in the arbor network that emerges during organ maturation, which may influence assembly of postsynaptic aggregates (lines 384-403, Figure 6G-I). We believe that our quantitative approach to capture collective axonal behaviors and their topological attributes during circuit formation have highlighted the importance of understanding network assembly during sensory organ development.

Reviewer #3 (Public Review):

Summary:

This study demonstrates that the axon guidance molecule Sema7a patterns the innervation of hair cells in the neuromasts of the zebrafish lateral line, as revealed by quantifying gain- and loss-of function effects on the three-dimensional topology of sensory axon arbors over developmental time. Alternative splicing can produce either a diffusible or membrane-bound form of Sema7a, which is increasingly localized to the basolateral pole of hair cells as they develop (Figure 1). In sema7a mutant zebrafish, sensory axon arbors still grow to the neuromast, but they do not form the same arborization patterns as in controls, with many arbors overextending, curving less, and forming fewer loops even as they lengthen (Figure 2,3). These phenotypes only become significant later in development, indicating that Sema7a functions to pattern local microcircuitry, not the gross wiring pattern. Further, upon ectopic expression of the diffusible form of Sema7a, sensory axons grow towards the Sema7a source (Figure 4). The data also show changes in the synapses that form when mutant terminals contact hair cells, evidenced by significantly smaller pre- and post-synaptic punctae (Figure 5). Finally, by replotting single cell RNA-sequencing data (Figure 6), the authors show that several other potential cues are also produced by hair cells and might explain why the sema7a phenotype does not reflect a change in growth towards the neuromast. In summary, the data strongly indicate that Sema7a plays a role in shaping connectivity within the neuromast.

Strengths:

The main strength of this study is the sophisticated analysis that was used to demonstrate fine-level effects on connectivity. Rather than asking "did the axon reach its target?", the authors asked "how does the axon behave within the target?". This type of deep analysis is much more powerful than what is typical for the field and should be done more often. The breadth of analysis is also impressive, in that axon arborization patterns and synaptic connectivity were examined at 3 stages of development and in three-dimensions.

Weaknesses:

The main weakness is that the data do not cleanly distinguish between activities for the secreted and membrane-bound forms of Sema7a, which the authors speculate may influence axon growth and synapse formation respectively. The authors do not overstate the claims, but it would have been nice to see some additional experimentation along these lines, such as the effects of overexpressing the membrane-bound form,

Comments: We have accepted this useful suggestion. In lines 333-346 and in Figure 5 we have demonstrated the impact of overexpressing the membrane-bound transcript variant on arborization pattern of the sensory axons.

Some analysis of the distance over which the "diffusible" form of Sema7a might act (many secreted ligands are not in fact all that diffusible), or

Comments: We have reported this in lines 311-317 and in Figure 4F,G.

Some live-imaging of axons before they reach the target (predicted to be the same in control and mutants) and then within the target (predicted to be different).

Comments: We have accepted this useful suggestion. We demonstrate the dynamics of the sensory arbors that are attracted to an ectopic Sema7Asec source in lines 325-332, Figure 4I,J; Figure 4—figure supplement 2A, and Videos 13-16.

Clearly, although the gain-of-function studies show that Sema7a can act at a distance, other cues are sufficient. Although the lack of a phenotype could be due to compensation, it is also possible that Sema7a does not actually act in a diffusible manner within its natural context. Overall, the data support the authors' carefully worded conclusions. While certain ideas are put forward as possibilities, the authors recognize that more work is needed. The main shortcoming is that the study does not actually distinguish between the effects of the two forms of Sema7a, which are predicted but not actually shown to be either diffusible or membrane linked (the membrane linkage can be cleaved). Although the study starts by presenting the splice forms, there is no description of when and where each splice form is transcribed.

Comments: We have utilized the HCR™ RNA-FISH Technology to generate transcript specific probes. To generate transcript-specific HCR probes to distinctly detect the sema7aGPI (NM_001328508) and the sema7asec (NM_001114885) transcripts, Molecular Instruments could design only 11 probes against the sema7aGPI transcript and only one probe against the sema7asec transcript (personal correspondence with Mike Liu, PhD, Head of Operations and Product Development Lead Molecular Instruments, Inc.). The HCR probe against the sema7aGPI transcript showed a very faint signal. Unfortunately, the HCR probe against the sema7asec transcript failed to detect the presence of any transcript. For robust detection of transcripts, the protocol demands a minimum of 20 probes. We believe that the very low number of probes against our transcripts is the primary reason for the absence of a signal.

We therefore utilized fluorescence-activated cell sorting (FACS) to capture the labeled hair cells and isolated total RNA to perform RT-PCR using transcript specific DNA oligonucleotide primers. We identified the presence of both the secreted and the membrane-bound transcripts at four-days-old neuromasts (lines 80-84, Figure 1B-D).

Additionally, since the mutants are predicted to disrupt both forms, it is a bit difficult to disentangle the synaptic phenotype from the earlier changes in circuit topology - perhaps the change at the level of the synapse is secondary to the change in topology.

Comments: Thank you for the insightful suggestion. We have analyzed the relationship between the sensory arbor network topology and the distribution of postsynaptic structures (lines 384-403, Figure 6G-I). We identified that the distribution of the postsynaptic aggregates is closely associated with the topological attributes of the sensory circuit. We further clarify the potential origin of disrupted synaptic assemblies in sema7a-/- mutants in lines 380-382 and lines 417-420.

Further, the authors do not provide any data supporting the idea that the membrane bound form of Sema7a acts only locally. Without these kinds of data, the authors are unable to attribute activities to either form.

Comments: We have accepted this useful suggestion and have prepared the Figure 5 with the necessary details.

The main impact on the field will be the nature of the analysis. The field of axon guidance benefits from this kind of robust quantification of growing axon trajectories, versus their ability to actually reach a target. This study highlights the value of more careful analysis and as a result, makes the point that circuit assembly is not just a matter of painting out paths using chemoattractants and repellants, but is also about how axons respond to local cues. The study also points to the likely importance of alternative splice forms and to the complex functions that can be achieved using different forms of the same ligand.

Reviewer #4 (Public Review):

Summary:

The work by Dasgupta et al identifies Sema7a as a novel guidance molecule in hair cell sensory systems. The authors use the both genetic and imaging power of the zebrafish lateralline system for their research. Based on expression data and immunohistochemistry experiments, the authors demonstrate that Sema7a is present in lateral line hair cells. The authors then examine a sema7a mutant. In this mutant, Sema7a proteins levels are nearly eliminated. Importantly, the authors show that when Sema7a is absent, afferent terminals show aberrant projections and fewer contacts with hair cells. Lastly the authors show that ectopic expression of the secreted form of Sema7a is sufficient to recruit aberrant terminals to non-hair cell targets. The sema7a innervation defects are well quantified. Overall, the paper is extremely well written and easy to follow.

Strengths:

(1) The axon guidance phenotypes in sema7a mutants are novel, striking and thoroughly quantified.

(2) By combining both loss of function sema7a mutants and ectopic expression of the secreted form of Sema7a the authors demonstrate the Sema7a is both necessary and sufficient to guide sensory axons

Weaknesses:

(1) Control. There should be an uninjected heatshock control to ensure that heatshock itself does not cause sensory afferents to form aberrant arbors. This control would help support the hypothesis that exogenously expressed Sema7a (via a heatshock driven promoter) is sufficient to attract afferent arbors.

Comments: Thank you for the suggestion. We have added the uninjected heatshock control experiment in Figure 5 and described experimental details in the text, lines 343-345.

(2) Synapse labeling. The numbers obtained for postsynaptic labeling in controls do not match up with the published literature - they are quite low. Although there are clear differences in postsynaptic counts between sema7a mutants and controls, it is worrying that the numbers are so low in controls. In addition, the authors do not stain for complete synapses (pre- and post-synapses together). This staining is critical to understand how Sema7a impacts synapse formation.

Comments: Thank you for raising this issue. We believe the low average numbers of the postsynaptic punctae in control neuromasts arise from lack of formation of postsynaptic aggregates beneath the immature hair cells, which are abundant in early stages of neuromast maturation. We have performed exhaustive analysis on the formation of pre- and postsynaptic structures and have identified how their distribution changes along neuromast development in control larvae. We have further analyzed how such distribution is perturbed in the sema7a-/- mutants. We do not think analyzing the complete synapse structure will add much to our understanding of how Sema7A influence synapse formation and maintenance.

(3) Hair cell counts. The authors need to provide quantification of hair cell counts per neuromast in mutant and control animals. If the counts are different, certain quantification may need to be normalized.

Comments: We have added the raw data with the hair cell counts in both control and sema7a-/- mutants across developmental stages. The homozygous sema7a-/- mutants have slightly less hair cells and we have normalized all our topological analyses by the corresponding hair cell numbers for each neuromast in each experiment (lines 669-675).

(4) Developmental delay. It is possible that loss of Sema7a simply delays development. The latest stage examined was 4 dpf, an age that is not quite mature in control animals. The authors could look at a later age, such as 6 dpf to see if the phenotypes persist or recover.

Comments: The homozygous sema7a-/- mutants are unviable and die at 6 dpf. We therefore restricted our analysis till 4 dpf. The association of the sensory arbors with the clustered hair cells gradually decreases as the neuromasts mature from 2 dpf to 4dpf in the sema7a-/- mutants (lines 174-176, Figure 2I). Moreover, in the sema7a-/- mutants the sensory axons throw long projections that keep getting farther away from the clustered hair cells as the neuromast matures from 2 dpf to 4 dpf (lines 166-168, Figure 2H; Figure 2—figure supplement 1K,L). These observations suggest that if the phenotypes in the sema7a-/- mutants were due to developmental delays, then we should have seen a recovery of disrupted arborization patterns over time. But instead, we observe a further deterioration of the arborization patterns and other architectural assemblies. These findings confirm that the observed phenotypes in the sema7a-/- mutants are not due to delayed development of the larvae, but a specific outcome for the loss of Sema7A protein.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Major concerns:

Issue 1: One of the most interesting conclusions in this manuscript is the function of the GPIanchored vs. secreted form of Sema7a in axon structure and synapse formation. In lines 357360 of the discussion (for example) the authors state that they have shown that the GPIanchored form of Sema7a is responsible for contact-mediated synapse formation while the secreted form functions as a chemoattractant for axon arbor structure. "We have discovered dual modes of Sema7A function in vivo: the chemoattractive diffusible form is sufficient to guide the sensory arbors toward their target, whereas the membrane-attached form likely participates in sculpting accurate neural circuitry to facilitate contact-mediated formation and maintenance of synapses." However, the data do not support this conclusion. Specifically, no analysis is done showing unique expression of either isoform in hair cells and no functional analysis is done to conclusively determine which isoform is important for either phenotype.

Comments: We have shown that both sema7a transcripts are expressed in the hair cells of four-day-old neuromasts (lines 78-84, Figure 1C,D). Ectopic expression of the sema7asec transcript variant robustly attracts the lateral-line sensory arbors toward itself, whereas ectopic expression of the sema7aGPI variant fails to impart sensory guidance from a distance, suggesting that the membrane-bound form likely participates in contact-mediated neural guidance. These experiments decisively show, for the first time in zebrafish, the dual modes of Sema7A function in vivo. However, we agree that the sema7aGPI transcript-specific knockout animal would be essential to conclusively prove that the membrane-attached form is primarily involved in forming accurate neural circuitry and contact-mediated formation and maintenance of synapses. Hence, we have very carefully stated in lines 427-428 that “the membrane-attached form likely participates in sculpting accurate neural circuitry to facilitate contact-mediated formation and maintenance of synapses”. We will follow up on this suggestion in our upcoming manuscript that will incorporate transcript-specific genetic ablations.

Though the authors present RT-PCR analysis of sema7a isoforms, it is not interpretable. The second reverse primer will also recognize the full-length transcript (from what I can gather) so it does not simply show the presence of the secreted form. Is there a unique 3'UTR for the short transcript that can be used? Additionally, for the GPI-anchored version can you use a forward primer that is not present in the short isoform? This would shed some light on the respective levels of both transcripts.

Comments: The C-termini of the two transcript variants are distinct and we have designed distinct primers that will selectively bind to each transcript (lines 503-511). Since, we have not performed quantitative polymerase chain reaction (qPCR), relative levels of each transcript are hard to determine.

Alternatively, and perhaps of more use, in situ hybridization using unique probes for each isoform would allow you to determine which are actually present in hair cells.

Comments: We have tried this approach and explained the point earlier (refer to lines 203212 of this response letter).

To decisively state that these isoforms have unique functions in axon terminal structure and synapse formation, other experiments are also essential. For example, RNA-mediated rescue analyses using both isoforms would tell you which can rescue the axonal structure and synapse size/number phenotypes. Overexpression of the GPI-anchored form, like the secreted form in Figure 4, would allow you to determine if only the secreted form can cause abnormal axon extension phenotypes. Expression of both forms in hair cells (using a myo6b promotor for example) would allow assessment of their role in presynapse formation.

Comments: We have ectopically expressed the sema7aGPI transcript variant near the sensory arbor network and observed that Sema7A-GPI fails to impart sensory axon guidance from a distance.

Thank you for suggesting the rescue experiments. We are in the process of generating CRISPR/Cas9-mediated transcript-specific knockout animals. We are currently preparing another manuscript that incorporates the above-mentioned rescue experiments to dissect the role of each transcript in regulating arbor topology and synapse formation.

For the overexpression experiments, expression of mKate alone (with and without heat shock) is also a critical control to include.

Comments: We have incorporated two control experiments: (1) larvae injected with hsp70:sema7asec-mKate2 plasmid that were not heat shocked and (2) Uninjected larvae that were heatshocked. We think these two controls are sufficient to demonstrate that the abnormal arborization patterns are not artifacts generated due to plasmid injection and heatshocking.

Issue 2: A second concern is the lack of data showing support cell and hair cell formation and function is unaffected. Analysis of support and hair cell number with loss of Sema7a as well as simple analyses of mechanotransduction (FM4-64) would help alleviate concerns that phenotypes are due to disrupted neuromast formation and basic hair cell function rather than a specific role for Sema7a in this process.

Comments: We have measured the hair cell numbers in both control and sema7a-/- mutants across developmental stages. We have added this to our submitted raw data.

We have utilized the styryl fluorophore FM4-64 to test the mechanotransduction function of the hair cells in sema7a-/- mutants. We have detailed our finding in lines 137141 and in Figure 2—figure supplement 1C,D.

Expression analysis of Sema7a receptors would also help strengthen the argument for a specific effect on lateral line afferent axons.

Comments: Thank you for this suggestion. Currently, we do not possess an RNA transcriptome dataset for the lateral line ganglion. This deficit limits a systematic screen for lateral-line sensory neuronal gene expressions either through antibody stains or via HCRmediated in situ techniques. In future we plan to develop an RNA transcriptome for the lateral-line ganglion and identify potential binding partners for Sema7A.

Issue 3: The manuscript could also be improved to include more detail in some areas and less in others. In general, each section has a fairly long lead up but lacks important experimental details that would help the reader interpret the data. For example:

Figure 1: What is the label for the lateral line axons? Is it a specific transgenic? The legend states that 3 asterisks indicate p<0.0001. What about the other asterisk combinations?

Comments: We have clarified these issues in lines 118-121 and in lines 906-907.

Figure 2: For the network analysis, are the traces for all axons that branch to innervate the neuromast?

Comments: Yes, we have traced the entire arbor containing all the axons that branched from the lateral line nerve and extended toward the clustered hair cells. The three-dimensional traces depict a skeletonized representation of the arbor network.

Can the tracing method distinguish individual axons?

Comments: No, our goal is to understand how the axon-collective behave around the clustered hair cells during development.

How do you know where an end is versus continued looping?

Comments: We have categorically defined the topological attributes in lines 187-191 and in Figure 3A.

Also, are all neuromasts similarly affected or is there a divergence based on which organ you are imaging? What neuromast was imaged in this and other figures?

Comments: Yes, all the neuromasts in the trunk and tail regions were affected similarly by the sema7a mutation. We did not observe any region-specific phenotypic outcome. We consistently imaged the trunk neuromasts, particularly the second, third, and fourth neuromasts.

Discussion: The short discussion failed to put these findings into context or to discuss how this unique topological arrangement of axon terminals impacts function.

Comments: We have added a new segment, lines 432-448, in the discussion section which mentions the potential role of the topological features in arranging the distribution pattern of the postsynaptic densities and thereby potentially influencing the network’s ability to gather sensory inputs through properly placed postsynaptic aggregates.

Can you speculate on how the looping structure may alter number of synaptic contacts per axon for instance? For this, it would be useful to know if normally the synapses form on loops versus bare terminals.

Comments: Thank you for this insightful suggestion. We have performed detailed analysis, as mentioned in lines 384-397, to characterize the distribution of the postsynaptic densities between the two topological attributes.

Does this looping facilitate single axons contacting more hair cells of the same polarity? Would that be beneficial?

Comments: Looping behaviors indeed facilitate the contact between the axons and the hair cells. As we have observed, the primary topological attribute that the sensory arbor network underneath the clustered hair cells adopts is a loop. The bare terminals are predominantly projected transverse to the clustered hair cells and lack contact with them. Whether a single axon, being part of a loop, preferentially contacts hair cells of same polarity is yet to be determined. We can address this question by mosaic labeling a single axon in the arbor network and determine its association with the hair cells. We intend to do these experiments in our upcoming manuscript.

Minor concerns:

(1) For the stacked charts quantifying topological features, I found interpreting them challenging. Is it possible to put these into overlapping histograms or line graphs to better compare wild type to mutant directly?

Comments: Thank you for your suggestion. We tried several ways to represent our data and found that the stacked charts optimally signify our analysis and depict the characteristic phenological differences between the control and the sema7a-/- mutants.

(2) There are numerous strong statements throughout not directly supported by the data, e.g. lines 110-113; 206-208; 357-360 and others. These should be tempered.

Comments: For lines 110-113, we have updated this section with new experiments and the new segment is represented in lines 115-126.

For lines 206-208, we have updated the statement to “This result suggests that the stereotypical circuit topology observed in the mature organ may emerge through transition of individual arbors from forming bare terminals to forming closed loops encircling topological holes” in lines 225-227.

Reviewer #2 (Recommendations For The Authors):

The authors should be careful about making any assumptions which form of sema7a is active in NMs. Their RT-PCR demonstrates presence of both isoforms in a whole animal; however, whether they are similarly present in HCs is not investigated here.

Comments: We have addressed this concern and have updated the manuscript with new experiments, detailed in lines 78-84.

Also, there is an issue of translation and trafficking to the membrane with subsequent secretion. An important experiment that would address this question is expressing two sema7a isoforms in mutant HCs and asking whether this can suppress the mutant phenotype.

Comments: Thank you for suggesting the rescue experiments. We are in the process of generating CRISPR/Cas9-mediated transcript-specific knockout animals. We are currently preparing another manuscript that incorporates the above-mentioned rescue experiments to dissect the role of each transcript in regulating arbor topology and synapse formation.

Presumably, sema7a is trafficked to the membrane during HC maturation. This is consistent with the authors' observation that sema7a localization is changing as NM mature. However, actin-sema7a co-labeling does not actually show whether sema7a is on the membrane. Labeling HCs with a membrane marker (transgene) would be much more convincing. Alternatively, can the authors show sema7a localization actually correlates with the presence of sensory axon terminals? They already have immunos that label both. Thus, this should be pretty straightforward.

Comments: Thank you for these suggestions. We have addressed these issues in lines 112114, and in lines 119-126.

Figure 2 should have a control panel, so the reduced sema7a staining can be compared to the control side-by-side.

Comments: We have depicted Sema7A staining in control neuromasts in multiple images, including Figure 1E, Figure 1H, and in Figure 2—figure supplement 1B. We have kept the control panel in the supplementary figure due to space restrictions in Figure 2.

Arborization topology: While I appreciate the very careful characterization of the topology for wild-type and mutant NMs, I think it would be much more informative to mark individual axons and then analyze their topology. The main reason is that the authors cannot really distinguish whether some aspects of topology they describe are really due to the densely packed overlapping terminals of multiple axons or these are really characteristic, higher order organization of individual axons. Because of this, they cannot be certain what is really happening with sema7a mutant terminals. Related to the point above. While it is clear that the overall topology is abnormal in the mutant, the authors should be careful in concluding that sema7a regulates specific aspects of it. The overall structure is probably highly interconnected perturbing one parameter would likely affect all the others.

Comments: Thank you for this comment. In a previous eLife publication from our laboratory, we utilized the serial blockface scanning electron micrograph (SBFSEM) technique to characterize the connectome of the neuromast microcircuit where patterns of innervation of all the individual axons can be delineated in five-days-old larvae (reference number 8). However, the collective behavior of all the sensory axons that build the innervation network remained enigmatic, especially in a living animal during development. In this paper we addressed how the sensory axon-collective behave around the clustered hair cells and build the innervation network in living animals during diverse developmental stages. Our analyses have not only identified how the axon-collective associates itself with the hair cell cluster as the organ matures, but also discovered distinct topological features in the arbor network that emerges during organ maturation, which may influence assembly of postsynaptic aggregates (lines 384-403, Figure 6G-I). We believe that our quantitative approach to capture collective axonal behaviors and their topological attributes during circuit formation have highlighted the importance of understanding network assembly during sensory organ development.

Experiments with the secreted sema7a isoform would be much more informative if they were compared/contrasted to the GPI anchored isoform.

Comments: We added a new section, lines 338-351, and a new Figure 5 to address this issue.

The phenotype of ectopic projections in sema7a overexpression experiments is pretty dramatic, especially given the fact that these were performed in wild-type animals. Does this mean that the phenotype would be even more dramatic in sema7a mutants, as they have more bare axon terminals according to the authors' analysis. Have the authors attempted this type of experiments?

Comments: That is an interesting suggestion. We have not tested that yet. Our guess is that in the sema7a-/- mutants, the abundant bare terminals will be far more sensitive to an ectopic source of Sema7A. But even in the sema7a-/- mutants, other chemotropic cues are still functional, which may impart certain restrictions on how many bare terminals are allowed to leave the neuromast region.

Reviewer #3 (Recommendations For The Authors):

(1) No raw data are shown, such that it is difficult to assess variability across animals or within animals, just the overall trends within the whole dataset. Raw data need to be shown for every measurement, at least in supplemental figures. It would also be useful to reliably show control next to mutant in the same plot, as it is a bit hard to compare across panels, which occurs in several figures.

Comments: We have uploaded all the raw data related to each experiment.

(2) Given the focus on the two possible forms of Sema7a, the authors should use HCR or another form of reliable in situ hybridization to show the spatiotemporal pattern of expression of each isoform.

Comments: We have utilized the HCR™ RNA-FISH Technology to generate transcript specific probes. To generate transcript-specific HCR probes to distinctly detect the sema7aGPI (NM_001328508) and the sema7asec (NM_001114885) transcripts, Molecular Instruments could design only 11 probes against the sema7aGPI transcript and only one probe against the sema7asec transcript (personal correspondence with Mike Liu, PhD, Head of Operations and Product Development Lead Molecular Instruments, Inc.). The HCR probe against the sema7aGPI transcript showed a very faint signal. Unfortunately, the HCR probe against the sema7asec transcript failed to detect the presence of any transcript. For robust detection of transcripts, the protocol demands a minimum of 20 probes. We believe that the very low number of probes against our transcripts is the primary reason for the lack of a signal.

(3) The authors should explain the criteria used to select the 22 embryos used to analyze the effects of expressing diffusible Sema7a.

Comments: We have explained this in lines 291-292. We identified 22 mosaic sema7asecmKate2 integration events, in which a single mosaic ectopic integration had occurred near the network of sensory arbors, from a total of almost 100 integrations. We rejected events where the sema7asec-mKate2 integration occurred either farther away from the sensory arbor network or had happened in multiple neighboring cells.

(4) Although arbors were imaged in live embryos, time is never presented as a variable, so I cannot tell whether axon topology was changing as the images were collected. This needs to be clarified.

Comments: We imaged the trunk neuromasts of both control and sema7a-/- mutant live zebrsfish larvae at 2, 3, and 4 dpf. We imaged the control and the sema7a-/- mutants of each developmental stage in parallel, within a span of two hours, and repeated these experiments multiple times to gather almost a hundred larvae from each genotype. Even though the sensory arbor network is dynamic, we believe imaging both the genotypes in parallel and within a span of two hours, and averaging almost a hundred larvae from each genotype minimize the temporal variability observed in the arbor architecture.

(5) Ideally, the authors should use CRISPR/cas-9 to create a mutation in the C-terminus that would prevent production of the GPI-anchored form and not of the diffusible form. I understand if this is too much work to do in a short time, and would be satisfied with another experiment that could distinguish roles for at least one isoform more clearly. For instance, it would be interesting to see an analysis of how far an axon can be from a source to detect diffusible Sema7a (live imaging would be ideal for this) and then to show that the effect is different when the membrane bound form is expressed.

Comments: Thank you for this comment. We are currently working in generating transcript specific knockout animals.

We have added live timelapse video microscopy data in lines 330-337, Figure 4H-J, Figure 4—figure supplement 2, Video15,16.

We have added a new segment analyzing the membrane-bound transcript variant in lines 338-351.

Reviewer #4 (Recommendations For The Authors):

Feedback to authors

Overall, this is a very important and novel study. Currently the manuscript does need revision.

Major concerns:

(1) Controls. For the ectoptic expression of Sema7a, injection of a construct expressing Sema7a under a heatshock promoter is used to drive ectopic expression. No heatshock (injected) animal are used as a control. In many systems heatshock can impact neuron morphology. And heatshock proteins are required for normal neurite and synapse formation. Please examine sensory axons in uninjected wildtype animals with heatshock.

Comments: We have added this control experiment in a new segment, explained in detail in lines 348-350 and Figure 5.

(2) Synapse staining - regarding Figure 5 and related supplement

Understanding whether guidance defects ultimately impact synapse formation is an important aspect of this paper. Therefore, is necessary to have accurate measurements of the number of complete synapses, and the overall numbers of pre- and postsynaptic components. Currently the data plotted in Figure 5 is extensive, but the way the data is laid out, the relevant comparisons are challenging to make. Perhaps include this quantification in the supplement, and move the data from the supplement to the main figure? The quantifications in the supplement are easier to follow and easier to compare between genotypes.

Comments: We have performed exhaustive analysis on the formation of pre- and postsynaptic structures and have identified how their distribution changes along neuromast development in control larvae. We have further analyzed how such distribution is perturbed in the sema7a-/- mutants. We believe that showing only the average numbers will not reveal the changes in the distribution of the synaptic structures during development and across genotypes.

Looking at the data itself, there seems to be some discrepancies with the synaptic counts compared to published work. While the CTBP numbers seem in order, the Maguk numbers do not. In both mutant and control there are many hair cells without any Maguk puncta/aggregates-leading to 0.75-1 postsynapses per hair cell (Figure 5 supplement H-I). Typically, the numbers should be more comparable to what was obtained for CTBP, 3-4 puncta per cells (Figure 5 supplement B-C), especially by 3-4 dpf. 3-4 CTPB or Maguk puncta per cell is based on previously published immunostaining and EM work.

The Maguk immunostaining, especially at early stages (2-3 dpf) is challenging. To compound a challenging immunostain, around 2019 Neuromab began to outsource the purification of their Maguk antibody. After this outsourcing our lab was no longer able to get reliable label with the Maguk antibody from Neuromab.

Millipore sells the same monoclonal antibody and it works well: https://www.emdmillipore.com/US/en/product/Anti-pan-MAGUK-Antibody-clone-K2886,MM_NF-MABN72

I would recommend this source.

Comments: Thank you for suggesting the new MAGUK antibody. We have utilized this new MAGUK antibody from Millipore and added a new segment in lines 389-408. In future publication we will utilize this antibody to capture the postsynaptic densities in the sensory arbors.

The discrepancies in the postsynaptic punctae number in our control larvae may arise due to the reliability of the Neuromab MAGUK antibody. We have utilized this same antibody to stain the sema7a-/- mutants and have observed a significant decrease in MAGUK punctae number and area. On grounds of keeping parity between the control and the sema7a-/- mutants, we have decided to keep our experimental results in the manuscript.

In addition to a more accurate Maguk label, a combined pre- and post-synaptic label is essential to understand whether synapses pair properly in the sema7a mutants. This can be accomplished using subtype specific antibodies using goat anti-mouse IgG1/Maguk and goat anti-mouse IgG2a/CTBP secondaries.

Comments: Thank you for suggesting this. We are preparing another manuscript in which we will utilize this technique along with other suggestions to tease apart the role of distinct transcript variants in regulating neural guidance and synapse formation.

(3) Does sema7a lesion impact the number of hair cells per neuromast? If hair cell numbers are reduced several of the quantifications could be impacted.

Comments: We have added the raw data with the hair cell counts in both control and sema7a-/- mutants across developmental stages. The homozygous sema7a-/- mutants have slightly less hair cells and we have normalized all our topological analyses by the corresponding hair cell numbers for each neuromast in each experiment (lines 669-675).

(4) Could innervation just be developmentally delayed in sema7a mutants? At 4 dpf the sensory system is just starting to come online and could still be in the process of refinement. Did you look at slightly older ages, after the sensory system is functional behaviorally, for example, 6 dpf? Do the cores phenotypes (synapse defects and excess arbors) persist at 6 dpf in the sema7a mutants?

Comments: The homozygous sema7a-/- mutants are unviable and start to die at 6 dpf. We therefore restricted our analysis until 4 dpf. The association of the sensory arbors with the clustered hair cells gradually decreases as the neuromasts mature from 2 dpf to 4dpf in the sema7a-/- mutants (lines 174-176, Figure 2I). Moreover, in the sema7a-/- mutants the sensory axons throw long projections that keep getting farther away from the clustered hair cells as the neuromast matures from 2 dpf to 4 dpf (lines 166-168, Figure 2H; Figure 2—figure supplement 1K,L). These observations suggests that if the phenotypes in the sema7a-/- mutants were due to developmental delays, then we should have seen a recovery of disrupted arborization patterns over time. But instead, we observe a further deterioration of the arborization patterns and other architectural assemblies. These findings confirm that the observed phenotypes in the sema7a-/- mutants are not due to delayed development of the larvae, but a specific outcome for the loss of Sema7A protein.

Minor comments to address:

Results

Page 4 lines 89-91. For the readers, explain why you examined levels in Sema7a in rostral and caudal hair cells. Also, this sentence is, in general, a little bit misleading-initially reading that there is no difference in Sema7a at 1.5-4 dpf.

Comments: In lines 44-48, we explain that the hair cells in the neuromast contain mechanoreceptive hair cells of opposing polarities that help them detect water currents from opposing directions. In lines 93-106, we tested whether the Sema7A level varies between the two polarities. We observed that the Sema7A level is similar between the two polarities of hair cells, but the average Sema7A intensity increases significantly over the developmental period of 2 dpf to 4 dpf in both rostrally and caudally polarized hair cells.

Page 10-11 Lines 263-270. What was the frequency of these 2 outcomes- out of the 22 cases with ectopic expression?

Comments: We have explained this in lines 291-292. We identified 22 mosaic sema7asecmKate2 integration events, in which a single mosaic ectopic integration had occurred near the network of sensory arbors, from a total of almost 100 integrations. We rejected events where the sema7asec-mKate2 integration occurred either farther away from the sensory arbor network or had happened in multiple neighboring cells.

Discussion

Page 14 Lines 359-360. There is not enough evidence provided in this work to suggest that the membrane attached form of Sema7a is playing a role. Both the secreted and membrane form are gone in the sema7a mutants. If the membrane attached form was specifically lesioned, and resulted in a phenotype, then there would be sufficient evidence. Currently there is strong evidence for a distinct role for the secreted form. Although the authors qualify the outlined statement with the word 'likely', stating this possibility in the discussion take-home is misleading.

Comments: In future we will utilize the CRISPR/Cas9 technique to target the unique Cterminal domain of the GPI-anchored sema7a transcript variant. We believe that this will only perturb the formation of the full-length Sema7A protein and help us differentiate between the roles of the membrane-bound Sema7AGPI molecule and the secreted Sema7Asec in sensory arborization and synaptic assembly.

It might be interesting in either the intro or discussion to reference the role Sema3F in axon guidance in the mouse auditory epithelium. https://elifesciences.org/articles/07830

Comments: We have added this reference in lines 61-64.

Figures

Please indicate on one of your Figures where the mutation is (roughly) in the sema7a mutant (in addition to stating it in the results).

Comments: We have added this information in Figure 2—figure supplement 1A.

Either state or indicate in a Figure where the epitope used to make the Sema7a antibody-to show that the antibody is predicted to recognize both isoforms.

Comments: We have stated the details of the epitope in lines 528-529.

Figure 2-S1 what is the scale in panel A, is it different between mutant and wildtype?

Comments: We have updated the images. New images are depicted in Figure 2—figure supplement 1A.

Methods

What were the methods used to quantify synapse number and area?

Comments: We have added a new section in lines 702-708 to explain the measurement techniques.

Author Response

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

eLife assessment

This article presents important results describing how the gathering, integration, and broadcasting of information in the brain changes when consciousness is lost either through anesthesia or injury. They provide convincing evidence to support their conclusions, although the paper relies on a single analysis tool (partial information decomposition) and could benefit from a clearer explication of its conceptual basis, methodology, and results. The work will be of interest to both neuroscientists and clinicians interested in fundamental and clinical aspects of consciousness.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this paper, Luppi et al., apply the recently developed integrated information decomposition to the question how the architecture of information processing changes when consciousness is lost. They explore fMRI data from two different populations: healthy volunteers undergoing reversible anesthesia, as well as from patients who have long-term disorders of consciousness. They show that, in both populations, synergistic integration of information is disrupted in common ways. These results are interpreted in the context of the SAPHIRE model (recently proposed by this same group), that describes information processing in the brain as being composed of several distinct steps: 1) gatekeeping (where gateway regions introduce sensory information to the global synergistic workspace where 2) it is integrated or "processed" before 3) by broadcast back to to the brain.

I think that this paper is an excellent addition to the literature on information theory in neuroscience, and consciousness science specifically. The writing is clear, the figures are informative, and the authors do a good job of engaging with existing literature. While I do have some questions about the interpretations of the various information-theoretic measures, all in all, I think this is a significant piece of science that I am glad to see added to the literature.

One specific question I have is that I am still a little unsure about what "synergy" really is in this context. From the methods, it is defined as that part of the joint mutual information that is greater than the maximum marginal mutual information. While this is a perfectly fine mathematical measure, it is not clear to me what that means for a squishy organ like the brain. What should these results mean to a neuro-biologist or clinician?

Right now the discussion is very high level, equating synergy to "information processing" or "integrated information", but it might be helpful for readers not steeped in multivariate information theory to have some kind of toy model that gets worked out in detail. On page 15, the logical XOR is presented in the context of the single-target PID, but 1) the XOR is discrete, while the data analyzed here are continuous BOLD signals w/ Gaussian assumptions and 2) the XOR gate is a single-target system, while the power of the Phi-ID approach is the multi-target generality. Is there a Gaussian analog of the single-target XOR gate that could be presented? Or some multi-target, Gaussian toy model with enough synergy to be interesting? I think this would go a long way to making this work more accessible to the kind of interdisciplinary readership that this kind of article with inevitably attract.

We appreciate this observation. We now clarify that:

“redundancy between two units occurs when their future spontaneous evolution is predicted equally well by the past of either unit. Synergy instead occurs when considering the two units together increases the mutual information between the units’ past and their future – suggesting that the future of each is shaped by its interactions with the other. At the microscale (e.g., for spiking neurons) this phenomenon has been suggested as reflecting “information modification” 36,40,47. Synergy can also be viewed as reflecting the joint contribution of parts of the system to the whole, that is not driven by common input48.”

In the Methods, we have also added the following example to provide additional intuition about synergy in the case of continuous rather than discrete variables:

“As another example for the case of Gaussian variables (as employed here), consider a 2-node coupled autoregressive process with two parameters: a noise correlation c and a coupling parameter a. As c increases, the system is flooded by “common noise”, making the system increasingly redundant because the common noise “swamps” the signal of each node. As a increases, each node has a stronger influence both on the other and on the system as a whole, and we expect synergy to increase. Therefore, synergy reflects the joint contribution of parts of the system to the whole that is not driven by common noise. This has been demonstrated through computational modelling (Mediano et al 2019 Entropy).”

See below for the relevant parts of Figures 1 and 2 from Mediano et al (2019 Entropy), where Psi refers to the total synergy in the system.

Author response image 1.

Strengths

The authors have a very strong collection of datasets with which to explore their topic of interest. By comparing fMRI scans from patients with disorders of consciousness, healthy resting state, and various stages of propofol anesthesia, the authors have a very robust sample of the various ways consciousness can be perturbed, or lost. Consequently, it is difficult to imagine that the observed effects are merely a quirk of some biophysical effect of propofol specifically, or a particular consequence of long-term brain injury, but do in fact reflect some global property related to consciousness. The data and analyses themselves are well-described, have been previously validated, and are generally strong. I have no reason to doubt the technical validity of the presented results.

The discussion and interpretation of these results is also very nice, bringing together ideas from the two leading neurocognitive theories of consciousness (Global Workspace and Integrated Information Theory) in a way that feels natural. The SAPHIRE model seems plausible and amenable to future research. The authors discuss this in the paper, but I think that future work on less radical interventions (e.g. movie watching, cognitive tasks, etc) could be very helpful in refining the SAPHIRE approach.

Finally, the analogy between the PID terms and the information provided by each eye redundantly, uniquely, and synergistically is superb. I will definitely be referencing this intuition pump in future discussions of multivariate information sharing.

We are very grateful for these positive comments, and for the feedback on our eye metaphor.

Weaknesses

I have some concerns about the way "information processing" is used in this study. The data analyzed, fMRI BOLD data is extremely coarse, both in spatial and temporal terms. I am not sure I am convinced that this is the natural scale at which to talk about information "processing" or "integration" in the brain. In contrast to measures like sample entropy or Lempel-Ziv complexity (which just describe the statistics of BOLD activity), synergy and Phi are presented here as quasi-causal measures: as if they "cause" or "represent" phenomenological consciousness. While the theoretical arguments linking integration to consciousness are compelling, is this is right data set to explore them in? For example, the work by Newman, Beggs, and Sherril (nee Faber), synergy is associated with "computation" performed in individual neurons: the information about the future state of a target neuron that is only accessible when knowing both inputs (analogous to the synergy in computing the sum of two dice). Whether one thinks that this is a good approach neural computation or not, it fits within the commonly accepted causal model of neural spiking activity: neurons receive inputs from multiple upstream neurons, integrate those inputs and change their firing behavior accordingly.

In contrast, here, we are looking at BOLD data, which is a proxy measure for gross-scale regional neural activity, which itself is a coarse-graining of millions of individual neurons to a uni-dimensional spectrum that runs from "inactive to active." It feels as though a lot of inferences are being made from very coarse data.

We appreciate the opportunity to clarify this point. It is not our intention to claim that Phi-R and synergy, as measured at the level of regional BOLD signals, represent a direct cause of consciousness, or are identical to it. Rather, our work is intended to use these measures similarly to the use of sample entropy and LZC for BOLD signals: as theoretically grounded macroscale indicators, whose empirical relationship to consciousness may reveal the relevant underlying phenomena. In other words, while our results do show that BOLD-derived Phi-R tracks the loss and recovery of consciousness, we do not claim that they are the cause of it: only that an empirical relationship exists, which is in line with what we might expect on theoretical grounds. We have now clarified this in the Limitations section of our revised manuscript, as well as revising our language accordingly in the rest of the manuscript.

We also clarify that the meaning of “information processing” that we adopt pertains to “intrinsic” information that is present in the system’s spontaneous dynamics, rather than extrinsic information about a task:

“Information decomposition can be applied to neural data from different scales, from electrophysiology to functional MRI, with or without reference to behaviour 34. When behavioural data are taken into account, information decomposition can shed light on the processing of “extrinsic” information, understood as the translation of sensory signals into behavioural choices across neurons or regions 41,43,45,47. However, information decomposition can also be applied to investigate the “intrinsic” information that is present in the brain’s spontaneous dynamics in the absence of any tasks, in the same vein as resting-state “functional connectivity” and methods from statistical causal inference such as Granger causality 49. In this context, information processing should be understood in terms of the dynamics of information: where and how information is stored, transferred, and modified 34.”

References:

(1) Newman, E. L., Varley, T. F., Parakkattu, V. K., Sherrill, S. P. & Beggs, J. M. Revealing the Dynamics of Neural Information Processing with Multivariate Information Decomposition. Entropy 24, 930 (2022).

Reviewer #2 (Public Review):

The authors analysed functional MRI recordings of brain activity at rest, using state-of-the-art methods that reveal the diverse ways in which the information can be integrated in the brain. In this way, they found brain areas that act as (synergistic) gateways for the 'global workspace', where conscious access to information or cognition would occur, and brain areas that serve as (redundant) broadcasters from the global workspace to the rest of the brain. The results are compelling and consisting with the already assumed role of several networks and areas within the Global Neuronal Workspace framework. Thus, in a way, this work comes to stress the role of synergy and redundancy as complementary information processing modes, which fulfill different roles in the big context of information integration.

In addition, to prove that the identified high-order interactions are relevant to the phenomenon of consciousness, the same analysis was performed in subjects under anesthesia or with disorders of consciousness (DOC), showing that indeed the loss of consciousness is associated with a deficient integration of information within the gateway regions.

However, there is something confusing in the redundancy and synergy matrices shown in Figure 2. These are pair-wise matrices, where the PID was applied to identify high-order interactions between pairs of brain regions. I understand that synergy and redundancy are assessed in the way the brain areas integrate information in time, but it is still a little contradictory to speak about high-order in pairs of areas. When talking about a "synergistic core", one expects that all or most of the areas belonging to that core are simultaneously involved in some (synergistic) information processing, and I do not see this being assessed with the currently presented methodology. Similarly, if redundancy is assessed only in pairs of areas, it may be due to simple correlations between them, so it is not a high-order interaction. Perhaps it is a matter of language, or about the expectations that the word 'synergy' evokes, so a clarification about this issue is needed. Moreover, as the rest of the work is based on these 'pair-wise' redundancy and synergy matrices, it becomes a significative issue.

We are grateful for the opportunity to clarify this point. We should highlight that PhiID is in fact assessing four variables: the past of region X, the past of region B, the future of region X, and the future of region Y. Since X and Y each feature both in the past and in the future, we can re-conceptualise the PhiID outputs as reflecting the temporal evolution of how X and Y jointly convey information: the persistent redundancy that we consider corresponds to information that is always present in both X and Y; whereas the persistent synergy is information that X and Y always convey synergistically. In contrast, information transfer would correspond to the phenomenon whereby information was conveyed by one variable in the past, and by the other in the future (see Luppi et al., 2024 TICS; and Mediano et al., 2021 arXiv for more thorough discussions on this point). We have now added this clarification in our Introduction and Results, as well as adding the new Figure 2 to clarify the meaning of PhiID terms.

We would also like to clarify that all the edges that we identify as significantly changing are indeed simultaneously involved in the difference between consciousness and unconsciousness. This is because the Network-Based Statistic differs from other ways of identifying edges that are significantly different between two groups or conditions, because it does not consider edges in isolation, but only as part of a single connected component.

Reviewer #3 (Public Review):

The work proposes a model of neural information processing based on a 'synergistic global workspace,' which processes information in three principal steps: a gatekeeping step (information gathering), an information integration step, and finally, a broadcasting step. The authors determined the synergistic global workspace based on previous work and extended the role of its elements using 100 fMRI recordings of the resting state of healthy participants of the HCP. The authors then applied network analysis and two different measures of information integration to examine changes in reduced states of consciousness (such as anesthesia and after-coma disorders of consciousness). They provided an interpretation of the results in terms of the proposed model of brain information processing, which could be helpful to be implemented in other states of consciousness and related to perturbative approaches. Overall, I found the manuscript to be well-organized, and the results are interesting and could be informative for a broad range of literature, suggesting interesting new ideas for the field to explore. However, there are some points that the authors could clarify to strengthen the paper. Key points include:

(1) The work strongly relies on the identification of the regions belonging to the synergistic global workspace, which was primarily proposed and computed in a previous paper by the authors. It would be great if this computation could be included in a more explicit way in this manuscript to make it self-contained. Maybe include some table or figure being explicit in the Gradient of redundancy-to-synergy relative importance results and procedure.

We have now added the new Supplementary Figure 1 to clarify how the synergistic workspace is identified, as per Luppi et al (2022 Nature Neuroscience).

(2) It would be beneficial if the authors could provide further explanation regarding the differences in the procedure for selecting the workspace and its role within the proposed architecture. For instance, why does one case uses the strength of the nodes while the other case uses the participation coefficient? It would be interesting to explore what would happen if the workspace was defined directly using the participation coefficient instead of the strength. Additionally, what impact would it have on the procedure if a different selection of modules was used? For example, instead of using the RSN, other criteria, such as modularity algorithms, PCA, Hidden Markov Models, Variational Autoencoders, etc., could be considered. The main point of my question is that, probably, the RSN are quite redundant networks and other methods, as PCA generates independent networks. It would be helpful if the authors could offer some comments on their intuition regarding these points without necessarily requiring additional computations.

We appreciate the opportunity to clarify this point. Our rationale for the procedure used to identify the workspace is to find regions where synergy is especially prominent. This is due to the close mathematical relationship between synergistic information and integration of information (see also Luppi et al., 2024 TICS), which we view as the core function of the global workspace. This identification is based on the strength ranking, as per Luppi et al (2022 Nature Neuroscience), which demonstrated that regions where synergy predominates (i.e., our proposed workspace) are also involved with high-level cognitive functions and anatomically coincide with transmodal association cortices at the confluence of multiple information streams. This is what we should expect of a global workspace, which is why we use the strength of synergistic interactions to identify it, rather than the participation coefficient. Subsequently, to discern broadcasters from gateways within the synergistic workspace, we seek to encapsulate the meaning of a “broadcaster” in information terms. We argue that this corresponds with making the same information available to multiple modules. Sameness of information corresponds to redundancy, and multiplicity of modules can be reflected in the network-theoretic notion of participation coefficient. Thus, a broadcaster is a region in the synergistic workspace (i.e., a region with strong synergistic interactions) that in addition has a high participation coefficient for its redundant interactions.

Pertaining specifically to the use of resting-state networks as modules, indeed our own (Luppi et al., 2022 Nature Neuroscience) and others’ research has shown that each RSN entertains primarily redundant interactions among its constituent regions. This is not surprising, since RSNs are functionally defined: their constituent elements need to process the same information (e.g., pertaining to a visual task in case of the visual network). We used the RSNs as our definition of modules, because they are widely understood to reflect the intrinsic organisation of brain activity into functional units; for example, Smith et al., (2009 PNAS) and Cole et al (2014 Neuron) both showed that RSNs reflect task-related co-activation of regions, whether directly quantified from fMRI in individuals performing multiple tasks, or inferred from meta-analysis of the neuroimaging literature. This is the aspect of a “module” that matters from the global workspace perspective: modules are units with distinct function, and RSNs capture this well. This is therefore why we use the RSNs as modules when defining the participation coefficient: they provide an a-priori division into units with functionally distinct roles.

Nonetheless, we also note that RSN organisation is robustly recovered using many different methods, including seed-based correlation from specific regions-of-interest, or Independent Components Analysis, or community detection on the network of inter-regional correlations - demonstrating that they are not merely a function of the specific method used to identify them. In fact, we show significant correlation between participation coefficient defined in terms of RSNs, and in terms of modules identified in a purely data-driven manner from Louvain consensus clustering (Figure S4).

(3) The authors acknowledged the potential relevance of perturbative approaches in terms of PCI and quantification of consciousness. It would be valuable if the authors could also discuss perturbative approaches in relation to inducing transitions between brain states. In other words, since the authors investigate disorders of consciousness where interventions could provide insights into treatment, as suggested by computational and experimental works, it would be interesting to explore the relationship between the synergistic workspace and its modifications from this perspective as well.

We thank the Reviewer for bringing this up: we now cite several studies that in recent years have applied perturbative approaches to induce transitions between states of consciousness.

“The PCI is used as a means of assessing the brain’s current state, but stimulation protocols can also be adopted to directly induce transitions between states of consciousness. In rodents, carbachol administration to frontal cortex awakens rats from sevoflurane anaesthesia120, and optogenetic stimulation was used to identify a role of central thalamus neurons in controlling transitions between states of responsiveness121,122. Additionally, several studies in non-human primates have now shown that electrical stimulation of the central thalamus can reliably induce awakening from anaesthesia, accompanied by the reversal of electrophysiological and fMRI markers of anaesthesia 123–128. Finally, in human patients suffering from disorders of consciousness, stimulation of intra-laminar central thalamic nuclei was reported to induce behavioural improvement 129, and ultrasonic stimulation 130,131 and deep-brain stimulation are among potential therapies being considered for DOC patients 132,133. It will be of considerable interest to determine whether our corrected measure of integrated information and topography of the synergistic workspace also restored by these causal interventions.”

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I would appreciate it if the authors could revisit the figures and make sure that:

(1) All fonts are large enough to be readable for people with visual impairments (for ex. the ranges on the colorbars in Fig. 2 are unreadably small).

Thank you: we have increased font sizes.

(2) The colormaps are scaled to show meaningful differences (Fig. 2A)

We have changed the color scale in Figure 2A and 2B.

Also, the authors may want to revisit the references section: some of the papers that were pre-prints at one point have now been published and should be updated.

Thank you: we have updated our references.

Minor comments:

• In Eqs. 2 and 3, the unique information term uses the bar notation ( | ) that is typically indicative of "conditioned on." Perhaps the authors could use a slash notation (e.g. Unq(X ; Z / Y)) to avoid this ambiguity? My understanding of the Unique information is that it is not necessarily "conditioned on", so much as it is "in the context of".

Indeed, the “|” sign of “conditioning” could be misleading; however, the “/” sign could also be misleading, if interpreted as division. Therefore, we have opted for the “\” sign of “set difference”, in Eq 2 and 3, which is conceptually more appropriate in this context.

• The font on the figures is a little bit small - for readers with poor eyes, it might be helpful to increase the wording size.

We have increased font sizes in the figures where relevant.

• I don't quite understand what is happening in Fig. 2A - perhaps it is a colormap issue, but it seems as though it's just a bit white square? It looks like redundancy is broadly correlated with FC (just based on the look of the adjacency matrices), but I have no real sense of what the synergistic matrix looks like, other than "flat."

We have now changed the color scale in Figure 2.

Reviewer #2 (Recommendations For The Authors):

Besides the issues mentioned in the Public review, I have the following suggestions to improve the manuscript:

• At the end of the introduction, a few lines could be added explaining why the study of DOC patients and subjects under anesthesia will be informative in the context of this work.

By comparing functional brain scans from transient anaesthetic-induced unconsciousness and from the persistent unconsciousness of DOC patients, which arises from brain injury, we can search for common brain changes associated with loss of consciousness – thereby disambiguating what is specific to loss of consciousness.

• On page and in general the first part of Results, it is not evident that you are working with functional connectivity. Many times the word 'connection' is used and sometimes I was wondering whether they were structural or functional. Please clarify. Also, the meaning of 'synergistic connection' or 'redundant connection' could be explained in lay terms.

Thank you for bringing this up. We have now replaced the word “connection” with “interaction” to disambiguate this issue, further adding “functional” where appropriate. We have also provided, in the Introduction, an intuitive explanation of what synergy and redundancy mean int he context of spontaneous fMRI signals.

• Figure 2 needs a lot of improvement. The matrix of synergistic interactions looks completely yellow-ish with some vague areas of white. So everything is above 2. What does it mean?? Pretty uninformative. The matrix of redundant connections looks a lot of black, with some red here and there. So everything is below 0.6. Also, what are the meaning and units of the colorbars?.

We agree: we have increased font sizes, added labels, and changed the color scale in Figure 2. We hope that the new version of Figure 2 will be clearer.

• Caption of Figure 2 mentions "... brain regions identified as belonging to the synergistic global workspace". I didn't get it clear how do you define these areas. Are they just the sum of gateways and broadcasters, or is there another criterion?

Regions belonging to the synergistic workspace are indeed the set comprising gateways and broadcasters; they are the regions that are synergy-dominated, as defined in Luppi et al., 2022 Nature Neuroscience. We have now clarified this in the figure caption.

• In the first lines of page 7, it is said that data from DOC and anesthesia was parcellated in 400 + 54 regions. However, it was said in a manner that made me think it was a different parcellation than the other data. Please make it clear that the parcellation is the same (if it is).

We have now clarified that the 400 cortical regions are from the Schaefer atlas, and 54 subcortical regions from the Tian atlas, as for the other analysis. The only other parcellation that we use is the Schaefer-232, for the robustness analysis. This is also reported in the Methods.

• Figure 3: the labels in the colorbars cannot be read, please make them bigger. Also, the colorbars and colorscales should be centered in white, to make it clear that red is positive and blue is negative. O at least maintain consistency across the panels (I can't tell because of the small numbers).

Thank you: we have increased font sizes, added labels, indicated that white refers to zero (so that red is always an increase, and blue is always a decrease), and changed the color scale in Figure 2.

• The legend of Figure 4 is written in a different style, interpreting the figure rather than describing it. Please describe the figure in the caption, in order to let the read know what they are looking at.

We have endeavoured to rewrite the legend of Figure 4 in a style that is more consistent with the other figures.

• In several parts the 'whole-minus-sum' phi measure is mentioned and it is said that it did not decrease during loss of consciousness. However, I did not see any figure about that nor any conspicuous reference to that in Results text. Where is it?

We apologise for the confusion: this is Figure S3A, in the Supplementary. We have now clarified this in the text.

Reviewer #3 (Recommendations For The Authors):

(1) In the same direction, regarding Fig. 2, in my opinion, it does not effectively aid in understanding the selection of regions as more synergistic or redundant. In panels A) and B), the color scales could be improved to better distinguish regions in the matrices (panel A) is saturated at the upper limit, while panel B) is saturated at the lower limit). Additionally, I suggest indicating in the panels what is being measured with the color scales.

Thank you: we have increased font sizes, added labels, and changed the color scale in Figure 2.

(2) When investigating the synergistic core of human consciousness and interpreting the results of changes in information integration measures in terms of the proposed framework, did the authors consider the synergistic workspace computed in HCP data? If the answer is positive, it would be helpful for the authors to be more explicit about it and elaborate on any differences that may be found, as well as the potential impact on interpretation.

This is correct: the synergistic workspace, including gateways and broadcasters, are identified from the Human Connectome Project dataset. We now clarify this in the manuscript.

Minors:

(1) I would suggest improving the readability of figures 2 and 3, considering font size (letters and numbers) and color bars (numbers and indicate what is measured with this scale). In Figure 1, the caption defines steps instead stages that are indicated in the figure.

Thank you: we have increased font sizes, added labels, and replaced steps with “stages” in Figure 1.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Qin et al. set out to investigate the role of mechanosensory feedback during swallowing and identify neural circuits that generate ingestion rhythms. They use Drosophila melanogaster swallowing as a model system, focusing their study on the neural mechanisms that control cibarium filling and emptying in vivo. They find that pump frequency is decreased in mutants of three mechanotransduction genes (nompC, piezo, and Tmc), and conclude that mechanosensation mainly contributes to the emptying phase of swallowing. Furthermore, they find that double mutants of nompC and Tmc have more pronounced cibarium pumping defects than either single mutants or Tmc/piezo double mutants. They discover that the expression patterns of nompC and Tmc overlap in two classes of neurons, md-C and md-L neurons. The dendrites of md-C neurons warp the cibarium and project their axons to the subesophageal zone of the brain. Silencing neurons that express both nompC and Tmc leads to severe ingestion defects, with decreased cibarium emptying. Optogenetic activation of the same population of neurons inhibited filling of the cibarium and accelerated cibarium emptying. In the brain, the axons of nompC∩Tmc cell types respond during ingestion of sugar but do not respond when the entire fly head is passively exposed to sucrose. Finally, the authors show that nompC∩Tmc cell types arborize close to the dendrites of motor neurons that are required for swallowing, and that swallowing motor neurons respond to the activation of the entire Tmc-GAL4 pattern.

Strengths:

• The authors rigorously quantify ingestion behavior to convincingly demonstrate the importance of mechanosensory genes in the control of swallowing rhythms and cibarium filling and emptying

• The authors demonstrate that a small population of neurons that express both nompC and Tmc oppositely regulate cibarium emptying and filling when inhibited or activated, respectively

• They provide evidence that the action of multiple mechanotransduction genes may converge in common cell types

Thank you for your insightful and detailed assessment of our work. Your constructive feedback will help to improve our manuscript.

Weaknesses:

• A major weakness of the paper is that the authors use reagents that are expressed in both md-C and md-L but describe the results as though only md-C is manipulated-Severing the labellum will not prevent optogenetic activation of md-L from triggering neural responses downstream of md-L. Optogenetic activation is strong enough to trigger action potentials in the remaining axons. Therefore, Qin et al. do not present convincing evidence that the defects they see in pumping can be specifically attributed to md-C.

Thank you for your comments. This is important point that we did not adequately address in the original preprint. We have obtained imaging and behavioral results that strongly suggest md-C, rather than md-L, are essential for swallowing behavior.

36 hours after the ablation of the labellum, the signals of md-L were hardly observable when GFP expression was driven by the intersection between Tmc-GAL4 & nompC-QF (see F Figure 3—figure supplement 1A). This observation indicates that the axons of md-L likely degenerated after 36 hours, and were unlikely to influence swallowing. Moreover, the projecting pattern of Tmc-GAL4 & nompC-QF>>GFP exhibited no significant changes in the brain post labellum ablation.

Furthermore, even after labellum ablation for 36 hours, flies exhibited responses to light stimulation (see Figure 3—figure supplement 1B-C, Video 5) when ReaChR was expressed in md-C. We thus reasoned that md-C but not md-L, plays a crucial role in the swallowing process.

• GRASP is known to be non-specific and prone to false positives when neurons are in close proximity but not synaptically connected. A positive GRASP signal supports but does not confirm direct synaptic connectivity between md-C/md-L axons and MN11/MN12.

In this study, we employed the nSyb-GRASP, wherein the GRASP is expressed at the presynaptic terminals by fusion with the synaptic marker nSyb. This method demonstrates an enhanced specificity compared to the original GRASP approach.

Additionally, we utilized +/ UAS-nSyb-spGFP1-10, lexAop-CD4-spGFP11 ; + / MN-LexA fruit flies as a negative control to mitigate potential false signals originating from the tool itself (Author response image 1, scale bar = 50μm). Beside the genotype Tmc-Gal4, Tub(FRT. Gal80) / UAS-nSyb-spGFP1-10, lexAop-CD4-spGFP11 ; nompC-QF, QUAS-FLP / MN-LexA fruit flies discussed in this manuscript, we also incorporated genotype Tmc-Gal4, Tub(FRT. Gal80) / lexAop-nSyb-spGFP1-10, UAS-CD4-spGFP11 ; nompC-QF, QUAS-FLP / MN-LexA fruit flies as a reverse control (Author response image 2). Unexpectedly, similar positive signals were observed, indicating that, positive signals may emerge due to close proximity between neurons even with nSyb-GRASP.

Author response image 1

It should be noted that the existence of synaptic projections from motor neurons (MN) to md-C cannot be definitively confirmed at this juncture. At present, we can only posit the potential for synaptic connections between md-C and motor neurons. A more conclusive conclusion may be attainable with the utilization of comprehensive whole-brain connectome data in future studies.

Author response image 2

• As seen in Figure 2—figure supplement 1, the expression pattern of Tmc-GAL4 is broader than md-C alone. Therefore, the functional connectivity the authors observe between Tmc expressing neurons and MN11 and 12 cannot be traced to md-C alone

It is true that the expression pattern of Tmc-GAL4 is broader than that of md-C alone. Our experiments, including those flies expressing TNT in Tmc+ neurons, demonstrated difficulties in emptying (Figure 2A, 2D). Notably, we encountered challenges in finding fly stocks bearing UAS>FRT-STOP-P2X2. Consequently, we opted to utilize Tmc-GAL4 to drive UAS-P2X2 instead. We believe that the results further support our hypothesis on the role of md-C in the observed behavioral change in emptying.

Overall, this work convincingly shows that swallowing and swallowing rhythms are dependent on several mechanosensory genes. Qin et al. also characterize a candidate neuron, md-C, that is likely to provide mechanosensory feedback to pumping motor neurons, but the results they present here are not sufficient to assign this function to md-C alone. This work will have a positive impact on the field by demonstrating the importance of mechanosensory feedback to swallowing rhythms and providing a potential entry point for future investigation of the identity and mechanisms of swallowing central pattern generators.

Reviewer #2 (Public Review):

In this manuscript, the authors describe the role of cibarial mechanosensory neurons in fly ingestion. They demonstrate that pumping of the cibarium is subtly disrupted in mutants for piezo, TMC, and nomp-C. Evidence is presented that these three genes are co-expressed in a set of cibarial mechanosensory neurons named md-C. Silencing of md-C neurons results in disrupted cibarial emptying, while activation promotes faster pumping and/or difficulty filling. GRASP and chemogenetic activation of the md-C neurons is used to argue that they may be directly connected to motor neurons that control cibarial emptying.

The manuscript makes several convincing and useful contributions. First, identifying the md-C neurons and demonstrating their essential role for cibarium emptying provides reagents for further studying this circuit and also demonstrates the important of mechanosensation in driving pumping rhythms in the pharynx. Second, the suggestion that these mechanosensory neurons are directly connected to motor neurons controlling pumping stands in contrast to other sensory circuits identified in fly feeding and is an interesting idea that can be more rigorously tested in the future.

At the same time, there are several shortcomings that limit the scope of the paper and the confidence in some claims. These include:

a) the MN-LexA lines used for GRASP experiments are not characterized in any other way to demonstrate specificity. These were generated for this study using Phack methods, and their expression should be shown to be specific for MN11 and MN12 in order to interpret the GRASP experiments.

Thanks for the suggestion. We have checked the expression pattern of MN-LexA, which is similar to MN-GAL4 used in previous work (Manzo et al., PNAS., 2012, PMID:22474379) . Here is the expression pattern:

Author response image 3

b) There is also insufficient detail for the P2X2 experiment to evaluate its results. Is this an in vivo or ex vivo prep? Is ATP added to the brain, or ingested? If it is ingested, how is ATP coming into contact with md-C neuron if it is not a chemosensory neuron and therefore not exposed to the contents of the cibarium?

The P2X2 experimental preparation was done ex vivo. We immersed the fly in the imaging buffer, as described in the Methods section under Functional Imaging. Following dissection and identification of the subesophageal zone (SEZ) area under fluorescent microscopy, we introduced ATP slowly into the buffer, positioned at a distance from the brain

c) In Figure 3C, the authors claim that ablating the labellum will remove the optogenetic stimulation of the md-L neuron (mechanosensory neuron of the labellum), but this manipulation would presumably leave an intact md-L axon that would still be capable of being optogenetically activated by Chrimson.

Please refer to the corresponding answers for reviewer 1 and Figure 3—figure supplement 1.

d) Average GCaMP traces are not shown for md-C during ingestion, and therefore it is impossible to gauge the dynamics of md-C neuron activation during swallowing. Seeing activation with a similar frequency to pumping would support the suggested role for these neurons, although GCaMP6s may be too slow for these purposes.

Profiling the dynamics of md-C neuron activation during swallowing is crucial for unraveling the operational model of md-C and validating our proposed hypothesis. Unfortunately, our assay faces challenges in detecting probable 6Hz fluorescent changes with GCaMP6s.

In general, we observed an increase of fluorescent signals during swallowing, but movement of alive flies during swallowing influenced the imaging recording, so we could not depict a decent tracing for calcium imaging for md-C neurons. To enhance the robustness of our findings, patching the md-C neurons would be a more convincing approach. As illustrated in Figure 2, the somata of md-C neurons are situated in the cibarium rather than the brain. patching of the md-C neuron somata in flies during ingestion is difficult.

e) The negative result in Figure 4K that is meant to rule out taste stimulation of md-C is not useful without a positive control for pharyngeal taste neuron activation in this same preparation.

We followed methods used in the previous work (Chen et al., Cell Rep., 2019, PMID:31644916), which we believe could confirm that md-C do not respond to sugars.

In addition to the experimental limitations described above, the manuscript could be organized in a way that is easier to read (for example, not jumping back and forth in figure order).

Thanks for your suggestion and the manuscript has been reorganized.

Reviewer #3 (Public Review):

Swallowing is an essential daily activity for survival, and pharyngo-laryngeal sensory function is critical for safe swallowing. In Drosophila, it has been reported that the mechanical property of food (e.g. Viscosity) can modulate swallowing. However, how mechanical expansion of the pharynx or fluid content sense and control swallowing was elusive. Qin et al. showed that a group of pharyngeal mechanosensory neurons, as well as mechanosensory channels (nompC, Tmc, and Piezo), respond to these mechanical forces for regulation of swallowing in Drosophila melanogaster.

Strengths:

There are many reports on the effect of chemical properties of foods on feeding in fruit flies, but only limited studies reported how physical properties of food affect feeding especially pharyngeal mechanosensory neurons. First, they found that mechanosensory mutants, including nompC, Tmc, and Piezo, showed impaired swallowing, mainly the emptying process. Next, they identified cibarium multidendritic mechanosensory neurons (md-C) are responsible for controlling swallowing by regulating motor neuron (MN) 12 and 11, which control filling and emptying, respectively.

Weaknesses:

While the involvement of md-C and mechanosensory channels in controlling swallowing is convincing, it is not yet clear which stimuli activate md-C. Can it be an expansion of cibarium or food viscosity, or both? In addition, if rhythmic and coordinated contraction of muscles 11 and 12 is essential for swallowing, how can simultaneous activation of MN 11 and 12 by md-C achieve this? Finally, previous reports showed that food viscosity mainly affects the filling rather than the emptying process, which seems different from their finding.

We have confirmed that swallowing sucrose water solution activated md-C neurons, while sucrose water solution alone could not (Figure 4J-K). We hypothesized that the viscosity of the food might influence this expansion process.

While we were unable to delineate the activation dynamics of md-C neurons, our proposal posits that these neurons could be activated in a single pump cycle, sequentially stimulating MN12 and MN11. Another possibility is that the activation of md-C neurons acts as a switch, altering the oscillation pattern of the swallowing central pattern generator (CPG) from a resting state to a working state.

In the experiments with w1118 flies fed with MC (methylcellulose) water, we observed that viscosity predominantly affects the filling process rather than the emptying process, consistent with previous findings. This raises an intriguing question. Our investigation into the mutation of mechanosensitive ion channels revealed a significant impact on the emptying process. We believe this is due to the loss of mechanosensation affecting the vibration of swallowing circuits, thereby influencing both the emptying and filling processes. In contrast, viscosity appears to make it more challenging for the fly to fill the cibarium with food, primarily attributable to the inherent properties of the food itself.

Reviewer #4 (Public Review):

A combination of optogenetic behavioral experiments and functional imaging are employed to identify the role of mechanosensory neurons in food swallowing in adult Drosophila. While some of the findings are intriguing and the overall goal of mapping a sensory to motor circuit for this rhythmic movement are admirable, the data presented could be improved.

The circuit proposed (and supported by GRASP contact data) shows these multi-dendritic neurons connecting to pharyngeal motor neurons. This is pretty direct - there is no evidence that they affect the hypothetical central pattern generator - just the execution of its rhythm. The optogenetic activation and inhibition experiments are constitutive, not patterned light, and they seem to disrupt the timing of pumping, not impose a new one. A slight slowing of the rhythm is not consistent with the proposed function.

Motor neurons implicated in patterned motions can be considered effectors of Central Pattern Generators (CPGs)(Marder et al., Curr Biol., 2001, PMID: 11728329; Hurkey et al., Nature., 2023, PMID:37225999). Given our observation of the connection between md-C neurons and motor neurons, it is reasonable to speculate that md-C neurons influence CPGs. Compared to the patterned light (0.1s light on and 0.1s light off) used in our optogenetic experiments, we noted no significant changes in their responses to continuous light stimulation. We think that optogenetic methods may lead to overstimulation of md-C neurons, failing to accurately mimic the expansion of the cibarium during feeding.

Dysfunction in mechanosensitive ion channels or mechanosensory neurons not only disrupts the timing of pumping but also results in decreased intake efficiency (Figure 1E). The water-swallowing rhythm is generally stable in flies, and swallowing is a vital process that may involve redundant ion channels to ensure its stability.

The mechanosensory channel mutants nompC, piezo, and TMC have a range of defects. The role of these channels in swallowing may not be sufficiently specific to support the interpretation presented. Their other defects are not described here and their overall locomotor function is not measured. If the flies have trouble consuming sufficient food throughout their development, how healthy are they at the time of assay? The level of starvation or water deprivation can affect different properties of feeding - meal size and frequency. There is no description of how starvation state was standardized or measured in these experiments.

Defects in mechanosensory channel mutants nompC, piezo, and TMC, have been extensively investigated (Hehlert et al., Trends Neurosci., 2021, PMID:332570000). Mutations in these channels exhibit multifaceted effects, as illustrated in our RNAi experiments (see Figure 2E). Deprivation of water and food was performed in empty fly vials. It's important to note that the duration of starvation determines the fly's willingness to feed but not the pump frequency (Manzo et al., PNAS., 2012, PMID:22474379).

In most cases, female flies were deprived water and food in empty vials for 24 hours because after that most flies would be willing to drink water. The deprivation time is 12 hours for flies with nompC and Tmc mutated or flies with Kir2.1 expressed in md-C neurons, as some of these flies cannot survive 24h deprivation.

The brain is likely to move considerably during swallow, so the GCaMP signal change may be a motion artifact. Sometimes this can be calculated by comparing GCaMP signal to that of a co-expressed fluorescent protein, but there is no mention that this is done here. Therefore, the GCaMP data cannot be interpreted.

We did not co-express a fluorescent protein with GCaMP for md-C. The head of the fly was mounted onto a glass slide, and we did not observe significant signal changes before feeding.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

.>Abstract: I disagree that swallow is the first step of ingestion. The first paragraph also mentions the final checkpoint before food ingestion. Perhaps sufficient to say that swallow is a critical step of ingestion.

Indeed, it is not rigorous enough to say “first step”. This has been replaced by “early step”.

Introduction:

Line 59: "Silence" should be "Silencing"

This has been replaced.

Results:

Lines 91-92: I am not clear about what this means. 20% of nompC and 20% of wild-type flies exhibit incomplete filling? So nompC is not different from wild-type?

Sorry for the mistake. Viscous foods led to incomplete emptying (not incomplete filling), as displayed in Video 4. The swallowing behavior differs between nompC mutants and wild-type flies, as illustrated in Figure 1C, Figure 1—figure supplement 1A-C and video 1&5.

When fed with 1% MC water solution (Figure 1—figure supplement 1E-H). We found that when fed with 1% MC watere solution, Tmc or piezo mutants displayed incomplete emptying, which could constitute a long time proportion of swallowing behavior; while only 20% of nompC flies and 20% of wild-type flies sporadically exhibit incomplete emptying, which is significantly different. Though the percent of flies displaying incomplete pump is similar between nompC mutant and wild-type files, you can find it quite different in video 1 and 5.

Line 94: Should read: “while for foods with certain viscosity, the pump of Tmc or piezo mutants might"

What evidence is there for weakened muscle motion? The phenotypes of all three mutants is quite similar, so concluding that they have roles in initiation versus swallowing strength is not well supported -this would be better moved to the discussion since it is speculative.

Muscles are responsible for pumping the bolus from the mouth to the crop. In the case of Tmc or piezo mutants, as evidenced by incomplete filling for viscous foods (see Video 4), we speculate that the loss of sensory stimuli leads to inadequate muscle contraction. The phenotypes observed in Tmc and piezo mutants are similar yet distinct from those of the wild-type or nompC mutant, as shown in Video 1 and 4. The phrase "due to weakened muscle motion" has been removed for clarity.

Line 146: If md-L neurons are also labeled by this intersection, then you are not able to know whether the axons seen in the brain are from md-L or md-C neurons. Line 148: cutting the labellum is not sufficient to ablate md-L neurons. The projections will still enter the brain and can be activated with optogenetics, even after severing the processes that reside in the labellum.

Please refer to the responses for reviewer #1 (Public Review):” A major weakness of the paper…” and Figure 4.

Line 162: If the fly head alone is in saline, do you know that the sucrose enters the esophagus? The more relevant question here is whether the md-C neurons respond to mechanical force. If you could artificially inflate the cibarium with air and see the md-C neurons respond that would be a more convincing result. So far you only know that these are activated during ingestion, but have not shown that they are activated specifically by filling or emptying. In addition, you are not only imaging md-C (md-L is also labeled). This caveat should be mentioned.

We followed the methods outlined in the previous work (Chen et al., Cell Rep., 2019, PMID:31644916), which suggested that md-C neurons do not respond to sugars. While we aimed to mechanically stimulate md-C neurons, detecting signal changes during different steps of swallowing is challenging. This aspect could be further investigated in subsequent research with the application of adequate patch recording or two-photon microscopy (TPM).

Figure 3: It is not clear what the pie charts in Figure 3 A refer to. What are the three different rows, and what does blue versus red indicate?

Figure 3A illustrates three distinct states driven by CsChrimson light stimulation of md-C neurons, with the proportions of flies exhibiting each state. During light activation, flies may display difficulty in filling, incomplete filling, or a normal range of pumping. The blue and red bars represent the proportions of flies showing the corresponding state, as indicated by the black line.

Figure 4: Where are the example traces for J? The comparison in K should be average dF/F before ingestion compared with average dF/F during ingestion. Comparing the in vitro response to sucrose to the in vivo response during ingestion is not a useful comparison.

Please refer to the answers for reviewer #2 question d).

Reviewer #2 (Recommendations For The Authors):

Suggested experiments that would address some of my concerns listed in the public review include:

a) high resolution SEZ images of MN-LexA lines crossed to LexAop-GFP to demonstrate their specificity

b) more detail on the P2X2 experiment. It is hard to make suggestions beyond that without first seeing the details.

c) presenting average GCaMP traces for all calcium imaging results

d) to rule out taste stimulation of md-C (Figure 4K) I would suggest performing more extensive calcium imaging experiments with different stimuli. For example, sugar, water, and increasing concentrations of a neutral osmolyte (e.g. PEG) to suppress the water response. I think that this is more feasible than trying to get an in vitro taste prep to be convincing.

Please refer to the responses for public review of reviewer #2.

Reviewer #3 (Recommendations For The Authors):

Below I list my suggestions as well as criticisms.

(1) It would be excellent if the authors could demonstrate whether varying levels of food viscosity affect md-C activation.

That is a good point, and could be studied in future work.

(2) It is not clear whether an intersectional approach using TMC-GAL4 and nompC-QF abolishes labelling of the labellar multidendritic neurons. If this is the case, please show labellar multidendritic neurons in TMC-GAL4 only flies and flies using the intersectional approach. Along with this question, I am concerned that labellum-removed flies could be used for feeding assay.

Intersectional labelling using TMC-GAL4 and nompC-QF could not abolish labelling of the labellar multidendritic neurons (Author response image 4). Labellum-removed flies could be used for feeding assay (Figure 3—figure supplement 1B-C, video 5), but once LSO or cibarium of fly was damaged, swallowing behavior would be affected. Removing labellum should be very careful.

Author response image 4

(3) Please provide the detailed methods for GRASP and include proper control.

Please refer to the responses for public review of reviewer #1.

(4) The authors hypothesized that md-C sequentially activates MN11 and 12. Is the time gap between applying ATP on md-C and activation of MN11 or MN12 different? Please refer to the responses for public review of reviewer #3. The time gap between applying ATP on md-C and activation of MN11 or MN12 didn’t show significant differences, and we think the reason is that the ex vivo conditions could not completely mimic in vivo process.

I found the manuscript includes many errors, which need to be corrected.

(1) The reference formatting needs to be rechecked, for example, lines 37, 42, and 43.

(2) Line 44-46: There is some misunderstanding. The role of pharyngeal mechanosensory neurons is not known compared with chemosensory neurons.

(3) Line 49: Please specify which type of quality of food. Chemical or physical?

(4) Line 80 and Figure 1B-D Authors need to put filling and emptying time data in the main figure rather than in the supplementary figure. Otherwise, please cite the relevant figures in the text(S1A-C).

(5) Line 84-85; Is "the mutant animals" indicating only nompC? Please specify it.

(6) Figure 1a: It is hard to determine the difference between the series of images. And also label filling and emptying under the time.

(7) S1E-H: It is unclear what "Time proportion of incomplete pump" means. Please define it.

(8) Please reorganize the figures to follow the order of the text, for example, figures 2 and 4

(9) Figure 4A. There is mislabelling in Figure 4A. It is supposed to be phalloidin not nc82.

(10) Figure 4K: It does not match the figure legend and main text.

(11) Figure 4D and G: Please indicate ATP application time point.

Thanks for your correction and all the points mentioned were revised.

Reviewer #4 (Recommendations For The Authors):

The figures need improvement. 1A has tiny circles showing pharynx and any differences are unclear.

The expression pattern of some of these drivers (Supplement) seems quite broad. The tmc nompC intersection image in Figure 1F is nice but the cibarium images are hard to interpret: does this one show muscle expression? What are "brain" motor neurons? Where are the labellar multi-dendritic neurons?

Tmc nompC intersection image show no expression in muscles. Somata of motor neurons 12 or 11 situated at SEZ area of brain, while somata of md-C neurons are in the cibarium. Image of md-L neurons was posted in response for reviewer #3 (Recommendations For The Authors):

Why do the assays alternate between swallowing food and swallowing water?

Thank for your suggestion, figure 1A has been zoomed-in. The Tmc nompC intersection image in Figure 2F displayed the position of md-C neurons in a ventral perspective, and muscles were not labelled. We stained muscles in cibarium by phalloidin and the image is illustrated in Figure 4A, while we didn’t find overlap between md-C neurons and muscles. Image of md-L neurons were posted as Author response image 4.

In the majority of our experiments, we employed water to test swallowing behavior, while we used methylcellulose water solution to test swallowing behavior of mechanoreceptor mutants, and sucrose solution for flies with md-C neurons expressing GCaMP since they hardly drank water when their head capsules were open.

How starved or water-deprived were the flies?

One day prior to the behavioral assays, flies were transferred to empty vials (without water or food) for 24 hours for water deprivation. Flies who could not survive 24h deprivation would be deprived for 12h.

How exactly was the pumping frequency (shown in Fig 1B) measured? There is no description in the methods at all. If the pump frequency is scored by changes in blue food intensity (arbitrary units?), this seems very subjective and maybe image angle dependent. What was camera frame rate? Can it capture this pumping speed adequately? Given the wealth of more quantitative methods for measuring food intake (eg. CAFE, flyPAD), it seems that better data could be obtained.

How was the total volume of the cibarium measured? What do the pie charts in Figure 3A represent?

The pump frequency was computed as the number of pumps divided by the time scale, following the methodology outlined in Manzo et al., 2012. Swallowing curves were plotted using the inverse of the blue food intensity in the cibarium. In this representation, ascending lines signify filling, while descending lines indicate emptying (see Figure 2D, 3B). We maintain objectivity in our approach since, during the recording of swallowing behavior, the fly was fixed, and we exclusively used data for analysis when the Region of Interest (ROI) was in the cibarium. This ensures that the intensity values accurately reflect the filling and emptying processes. Furthermore, we conducted manual frame-by-frame checks of pump frequency, and the results align with those generated by the time series analyzer V3 of ImageJ.

For the assessment of total volume of ingestion, we referred the methods of CAFE, utilizing a measurable glass capillary. We then calculated the ingestion rate (nL/s) by dividing the total volume of ingestion by the feeding time.

The changes seem small, in spite of the claim of statistical significance.

The observed stability in pump frequency within a given genotype underscores the significance of even seemingly small changes, which is statistically significant. We speculate that the stability in swallowing frequency suggests the existence of a redundant mechanism to ensure the robustness of the process. Disruption of one channel might potentially be partially compensated for by others, highlighting the vital nature of the swallowing mechanism.

How is this change in pump frequency consistent with defects in one aspect of the cycle - either ingestion (activation) or expulsion (inhibition)?

Please refer to Figure 2, 3. Both filling and emptying process were affects, while inhibition mainly influences emptying time (Figure 1—figure supplement 1).

for the authors:

Line 48: extensively

Line 62 - undiscovered.

Line 107, 463: multi

Line 124: What is "dysphagia?" This is an unusual word and should be defined.

Line 446: severe

Line 466: in the cibarium or not?

Thanks for your correction and all the places mentioned were revised.

Author Response

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

Reply to comments:

(1) It was not clear why the phylogenetic analysis included non-validated GPCRs that clustered with the validated peptidergic receptors. Would restricting the phylogenetic analyses only to confirmed peptidergic GPCRs alter the topology of the tree and subsequent conclusions of independent expansion?

Thank you for this comment. In general, phylogenetic analyses become more robust if a larger diversity and fuller complement of sequences are included. With very sparse sampling, sequences that are homologous but not orthologous may be misleadingly grouped together, because intermediate sequences have been left out. For tree building, we thus did not want to focus only on experimentally validated receptors but also on all receptors that are phylogenetically related to the validated receptors. Only this approach can ensure a comprehensive exploration of the relationship of peptidergic receptors. The broader phylogenetic approach was also essential to identify orthologs to the experimentally validated Nematostella receptors across other cnidarian species.

(2) Clearly, other neuropeptide signaling systems in cnidarians remain to be discovered but this paper represents a huge step forward.

We appreciate this assessment of the paper. We agree that many systems remain to be discovered. Our paper will also help with the identification of further receptors both in Nematostella as well as other cnidarian species. Please note that we have made specific receptor-ligand predictions for several cnidarian species based on our phylogenetic analysis. Our phylogenies could also help prioritize the study of the remaining orphan Nematostella GPCRs.

(3) There are limitations in what can be interpreted from single cell transcriptomic data but the data nevertheless provide the foundations for future studies involving i). detailed anatomical analysis of neuropeptide and neuropeptide receptor expression in N. vectensis using mRNA in situ hybridization and/or immunohistochemical methods and ii). functional analysis of the physiological/behavioral roles of neuropeptide signaling systems in N. vectensis

We fully agree with this comment. The analysis of the available single-cell sequence resources clearly represents only the first step of anatomical and functional analyses. Our aim was to place the identified peptide-receptor interactions into a whole-organism context with cell type resolution, to highlight the potential complexity of peptidergic signaling in this organism and to facilitate the exploration and conceptualisation of our biochemical screen.

Comments to authors

(1) In future, when preparing manuscripts, please use page and line numbers; it makes the task below for reviewers much easier!

We appreciate the suggestion and will do this for future manuscripts.

(2) In the abstract the term "extensively wired" is used. In the context of neuropeptide mediated volume transmission this may not be an appropriate term to use because use of the word "wired" is likely to be associated with point-to-point type classical synaptic transmission; "extensively connected" would be better.

Thank you for this comment. We have changed the text in the abstract to “extensively connected”.

(3) Introduction: Please change "seven-transmembrane proteins and show a slower evolutionary rate than proneuropeptide..." to "seven-transmembrane proteins that show a slower evolutionary rate than proneuropeptide..."

Changed.

(4) Under the section "Creation of a Nematostella neuropeptide library, what is meant by "our regular expressions"? This needs to be rephrased to make it clearer what is meant.

We have now rephrased the relevant sentence to make our approach clearer.

“This predicted secretome was filtered with regular expressions to detect sequences with the repetitive dibasic cleavage sites (K and R in any combination) and amidation sites, using a custom script from a previous publication (Thiel et al., 2021).”

and later:

“Based on the MS data, we included the additional, non-dibasic N-terminal cleavage sites into our script that uses regular expressions to search for repetitive cleavage sites (Thiel et al., 2024) and re-screened the predicted secretome.”

(5) Under the section "Creation of a Nematostella neuropeptide library" the phrase "differ in the length of their N-terminus" needs to be changed to "differ in the length of their N-terminal region". The N-terminus is, as its name implies, one end of the peptide/protein so it can't have a length as such.

Changed.

(6) Under the section "Analysis of metazoan class A GPCRs and selection of N. vectensis neuropeptide-receptor candidates",

Change:

"For a more detailed analysis, we then reduced our sampled species to the cnidarian, the bilaterian with experimentally confirmed GPCRs and Petromyzon marinus, and the two placozoan species (Figure 2B)."

To

"For a more detailed analysis, we then reduced our sampled species to cnidarians, bilaterians with experimentally confirmed GPCRs and Petromyzon marinus, and two placozoan species (Figure 2B)."

Changed.

(7) Under the section "Analysis of metazoan class A GPCRs and selection of N. vectensis neuropeptide-receptor candidates" - change "We re-run" to "We re-ran"

Changed.

(8) Throughout the paper reference is made to a variety of neuropeptides that have or are predicted to have an N-terminal pyroglutmate. However, these are referred to without indicating this post-translational modification e.g. QGRFamide.

This should be corrected throughout the paper, in the text, and figures. Two abbreviations for pyroglutamate are used in the literature:

pQ, which shows that the encoded amino acid is Q (Glutamine)

pE, which shows that the post-translationally modified amino-acid is glutamate (E)

In the neuropeptide field, pQ seems to be more widely used than pE, so our recommendation would be to use pQ.

In the revised version we now write pyroQ whenever we refer to the actual peptide. We now only use the peptide name without indicating this modification when we refer to the precursor of these peptides.

(9) The title for Figure 5 is rather short and vague. A title like "Tissue-specific expression of neuropeptide precursors and receptors in Nematostella" seems more appropriate

We appreciate the reviewer's input, and we have made the change accordingly. The revised figure legend now reads: “Tissue-specific expression of neuropeptide precursors and receptors (GPCRs) in N. vectensis.”

(10) All of the figures in the paper have been saved in bitmap format (e.g. tiff), which means that the resolution of the figures may end up being poor in the published article. All of the figures in this paper should be saved in vector format (e.g. eps) so that there is no loss of resolution when the size of the file/figure is reduced.

We have now uploaded all figures in vector format (.eps or .pdf) to prevent any loss of resolution.

(11) In Figure 3 - supplement 2 - the neuropeptides are referred to here as PRGamides and GPRGamides. Some consistency is needed here. And in Figure B, the G of one of the GPRGamides is not shown in black.

Thank you for spotting this mistake. We now give the correct peptide sequence in parenthesis as "GPRGamide". We also highlighted the missing GPRGamide in the figure.

Author Response

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

Reviewer #1 (Public Review):

Summary:

In the manuscript by Chen et al. entitled, "The retina uncouples glycolysis and oxidative phosphorylation via Cori-, Cahill-, and mini-Krebs-cycle", the authors look to provide insight on retinal metabolism and substrate utilization by using a murine explant model with various pharmacological treatments in conjunction with metabolomics. The authors conclude that photoreceptors, a specific cell within the explant, which also includes retinal pigment epithelium (RPE) and many other types of cells, are able to uncouple glycolytic and Krebs-cycle metabolism via three different pathways: 1) the mini-Krebs-cycle, fueled by glutamine and branched-chain amino acids; 2) the alanine-generating Cahill-cycle; and 3) the lactate-releasing Cori-cycle. While intriguing if determined to be true, these cell-specific conclusions are called into question due to the ex vivo experimental setup with the inclusion of RPE, the fact that the treatments were not cell-specific nor targeted at an enzyme specific to a certain cell within the retina, and no stable isotope tracing nor mitochondrial function assays were performed. Hence, without significant cell-specific methods and future experimentation, the primary claims are not supported.

Strengths:

This study attempts to improve on the issues that have limited the results obtained from previous ex vivo retinal explant studies by culturing in the presence of the RPE, which is a major player in the outer retinal metabolic microenvironment. Additionally, the study utilizes multiple pharmacologic methods to define retinal metabolism and substrate utilization.

Weaknesses:

A major weakness of this study is the lack of in vivo supporting data. Explant cultures remove the retina from its dual blood supply. Typically, retinal explant cultures are done without RPE. However, the authors included RPE in the majority of experimental conditions herein. However, it is unclear if the metabolomics samples included the RPE or not. The inclusion of the RPE, which is metabolically active and can be altered by the treatments investigated herein, further confounds the claims made regarding the neuroretina. Considering the pharmacologic treatments utilized with the explant cultures are not cell-specific and/or have significant off-target effects, it is difficult to ascertain that the metabolic changes are secondary to the effects on photoreceptors alone, which the authors claim. Additionally, the explants are taken at a very early age when photoreceptors are known to still be maturing. No mention or data is presented on how these metabolic changes are altered in retinal explants after photoreceptors have fully matured. Likewise, significant assumptions are made based on a single metabolomics experiment with no stable isotope tracing to support the pathways suggested. While the authors use immunofluorescence to support their claims at multiple points, demonstrating the presence of certain enzymes in the photoreceptors, many of these enzymes are present throughout the retina and likely the RPE. Finally, the claims presented here are in direction contradiction to recent in vivo studies that used cell-specific methods when examining retinal metabolism. No discussion of this difference in results is attempted. Response: We agree with the reviewer that in vivo studies could be very interesting indeed. However, technologically it will be extremely difficult to (repeatedly/continuously) sample the retina of an experimental animal and to combine this with an interventional study, with a subsequent metabolomic analysis. We do not currently have access to such technology nor are we aware of any other lab in the world capable of doing such studies. Moreover, virtually all prior studies on retinal metabolism have been done on explanted retina without RPE. This includes the seminal studies by Otto Warburg in the 1920s. As opposed to this, our retinal samples for also all the metabolomic analyses included the RPE, except for the no RPE condition that was used as a comparator for the earlier investigations.

We note that our metabolomic analysis was done for all five experimental conditions where each condition included at least five independent samples (each derived from different animals).

The reviewer is correct to say that our organotypic explant cultures are early post-natal, with explantation performed at post-natal day 9 and culturing until day 15. Since our retinal explant system has been validated extremely well over more than three decades of pertinent research (see for instance: Caffe et al., Curr Eye Res. 8:1083-92, 1989), we are confident that photoreceptors mature in vitro in ways that are very similar to the in vivo situation. As far as studies in adult retina (i.e. three months or older) are concerned, this is indeed an important question that will be addressed in future studies. Studies employing stable isotope labelling may also be very informative and are planned for the future, also in order to properly determine fluxes. This will likely require an extension to our NMR hardware with an 15N channel probe, something that we plan on implementing in the future.

We are aware that a number of questions relating to retinal metabolism are controversial and that the use of other methodology or experimental systems may lead to alternative interpretations. We have now included citations of other studies that use, for example, conditional and/or inducible knock-outs or in vivo blood sampling (e.g. Wang et al., IOVS 38:48-55, 1997; Yu et al., Invest Ophthalmol Vis Sci. 46:4728-33, 2005; Swarup et al., Am J Physiol Cell Physiol. 316:C121-C133, 2019; Daniele et al., FASEB Journal 36:e22428, 2022) and discuss the pros and cons of such approaches (e.g. in Lines 376-384; 454-472).

Reviewer #2 (Public Review):

Summary:

The authors aim to learn about retinal cell-specific metabolic pathways, which could substantially improve the way retinal diseases are understood and treated. They culture ex vivo mouse retinas for 6 days with 2 - 4 days of various drug treatments targeting different metabolic pathways or by removing the RPE/choroid tissue from the neural retina. They then look at photoreceptor survival, stain for various metabolic enzymes, and quantify a broad panel of metabolites. While this is an important question to address, the results are not sufficient to support the conclusions.

Strengths:

The questions the authors are exploring at extremely valuable and I commend the authors and working to learn more about retina metabolism. The different sensitivity of the cones to various drugs is interesting and may suggest key differences between rods and cones. The authors also provide a thoughtful discussion of various metabolic pathways in the context of previous publications.

Weaknesses:

As the authors point out, ex vivo culture models allow for control over multiple aspects of the environment (such as drug delivery) not available in vivo. Ex vivo cultures can provide good hints as to what pathways are available between interacting tissues. However, there are many limitations to ex vivo cultures, including shifting to a very artificial culture media condition that is extremely different than the native environment of the retina. It is well appreciated that cells have flexible metabolism and will adapt to the conditions provided. Therefore, observations of metabolic responses obtained under culture conditions need to be interpreted with caution, they indicate what the tissue is doing under those specific conditions (which include cells adapting and dying).

Chen et al use pharmacological interventions to the impact of various metabolic pathways on photoreceptor survival and "long term" metabolic changes. The dose and timing of these drug treatments are not examined though. It is also hard to know how these drugs penetrate the tissue and it needs to be validated that the intended targets are being accurately hit. These relatively long-term treatments should be causing numerous downstream changes to metabolism, cell function, and survival, which makes looking at a snapshot of metabolite levels hard to interpret. It would be more valuable to look at multiple time points after drug treatment, especially easy time points (closer to 1 hr). The authors use metabolite ratios to make conclusions about pathway activity. It would be more valuable to directly measure pathway activity by looking a metabolite production rates in the media and/or with metabolic tracers again in time scales closer to minutes and hours instead of days.

It is not clear from the text if the ex vivo samples with RPE/choroid intact are analyzed for metabolomics with the RPE/choroid still intact or if this is removed. If it is not removed, the comparison to the retina without RPE/choroid needs to be re-interpreted for the contribution of metabolites from the added tissue. The composition of the tissue is different and cannot be disentangled from the changes to the neural retina specifically.

While the data is interesting and may give insights into some rod and cone-specific metabolic susceptibility, more work is needed to validate these conclusions. Given the limitations of the model the authors have over-interpreted their findings and the conclusions are not supported by the results. They need to either dramatically limit the scope of their conclusions or validate these hypotheses with additional models and tools.

Response: We thank the reviewer for the insightful comments and agree that some of our interpretations may have been phrased too determinedly. We have therefore rephrased and toned down our conclusions in many instances in the text, and changed the manuscript title to now read “Retinal metabolism: Evidence for uncoupling of glycolysis and oxidative phosphorylation via Cori-, Cahill-, and mini-Krebs-cycle”.

Nevertheless, when considering the major known metabolic pathways and their possible impact on metabolite patterns after the experimental manipulations used here, we believe our interpretations to be consistent with the data obtained. Conversely, the previously suggested retinal aerobic glycolysis cannot explain most of the data we have obtained. Even further, also a predominant use of the classical “full” Krebs-cycle/OXPHOS would not explain the metabolite patterns found (e.g. alanine, N-acetylaspartate (NAA)). While this does not in itself mean that our interpretations are all correct, they seem plausible in view of the data at hand and will hopefully stimulate further research on retinal energy metabolism using complementary technologies that were not available to us for the purpose of this study.

We comment that our organotypic retinal explant cultures, while they do contain their very own, native RPE, do not comprise the choroidal vasculature (in our explantation procedure the RPE readily detaches from the choroid).

As far as the drugs used on retinal explants are concerned, we note that:

(1) all three compounds used are extremely well validated, with literally thousands of studies and decades of research to their credit (i.e., 1,9-dideoxyforskolin: >270 publications since 1984; Shikonin: >1000 publications since 1977; FCCP: >2800 publications since 1967),

(2) all experimental conditions show clear and differential drug effects, as shown, for instance, by the principal component analysis in Figure1I and the cluster analysis in Figure2A,

(3) the response patterns observed for key metabolites match the anticipated drug effects (e.g. decreased glucose consumption with 1,9-dideoxyforskolin; decreased lactate levels with Shikonin; lactate accumulation with FCCP).

One can therefore be reasonably certain that these drugs did penetrate the explanted retina and that their respective drug targets were hit. Assessing dose-responses would certainly be interesting, however, the aim of this initial study was not pharmacodynamics but a general manipulation of energy metabolism. Moreover, given the extensive validation of these drugs, off-target effects seem not very likely at the concentrations used.

We agree with the reviewer that using a longitudinal, time-series type of analysis could give additional insights. We note that each additional time-point will require retinae from 25 animals and a very resource-intensive and time-consuming metabolomic analysis, together with a significantly more complex multivariate analysis (metabolite, experimental condition, time). This is a completely new undertaking that is simply not feasible as an extension of the present study.

To look at pathway activity in more direct ways is very good idea, to this end we aim to implement in the future an idea put forward by the reviewers, namely 13C-labeling and additionally 15N-labeling and tracing for specific metabolic fuels (e.g. glucose, lactate and anaplerotic amino acids such as glutamate and branched chain amino acids).

The reviewer is of course correct to say that the culture condition is somewhat artificial and that this may have introduced changes in the metabolism. However, as noted above in the first response to reviewer #1, the organotypic retinal culture system, using a defined medium, free of serum and antibiotics, has been extremely well studied and validated for decades (cf. Caffé et al., Curr Eye Res. 8:1083-92, 1989). Importantly, this system allows to maintain retinal viability, histotypic organization, and function over many weeks in culture. Moreover, most previous studies on retinal metabolism have also used explanted retina – acute or cultured – i.e. experimental approaches that are similar to what we have used and that may be liable to their own artefactual changes in metabolism. This includes the seminal, 1920s studies by Otto Warburg, or the 1980s studies by Barry Winkler, the results of which the reviewers do not seem to doubt.

We further agree that studying retinal metabolism in a situation closer to in vivo conditions would be thrilling, however to our knowledge to date there is no retina model that fully mimics the complex interplay of the blood metabolome with metabolic tissue activity. This likely means that for each metabolic condition to study (e.g. hyperglycemia, cachexia, etc.), a fairly large number of animals will need to be sacrificed for the molecular investigation of ex vivo retinal biopsies, which would mean a tremendous animal burden.

We hope the reviewer will appreciate that the revised manuscript now includes numerous improvements, along with new, additional datasets and figures, references to further relevant literature, and – as mentioned above – a more cautious phrasing of our interpretations and conclusions, including a more careful wording for the manuscript title.

Reviewer #3 (Public Review):

Summary:

The neural retina is one of the most energetically active tissues in the body and research into retinal metabolism has a rich history. Prevailing dogma in the field is that the photoreceptors of the neural retina (rods and cones) are heavily reliant on glycolysis, and as oxygen tension at the level of photoreceptors is very low, these specialized sensory neurons carry out aerobic glycolysis, akin to the Warburg effect in cancer cells. It has been found that this unique metabolism changes in many retinal diseases, and targeting retinal metabolism may be a viable treatment strategy. The neural retina is composed of 11 different cell types, and many research groups over the past century have contributed to our current understanding of cell-specific metabolism of retinal cells. More recently, it has been shown in mouse models and co-culture of the mouse neural retina with human RPE cultures that photoreceptors are reliant on the underlying retinal pigment epithelium for supplying nutrients. Chen and colleagues add to this body of work by studying an ex vivo culture of the developing mouse retina that maintained contact with the retinal pigment epithelium. They exposed such ex vivo cultures to small molecule inhibitors of specific metabolic pathways, performing targeted metabolomics on the tissue and staining the tissue with key metabolic enzymes to lay the groundwork for what metabolic pathways may be active in particular cell types of the retina. The authors conclude that rod and cone photoreceptors are reliant on different metabolic pathways to maintain their cell viability - in particular, that rods rely on oxidative phosphorylation and cones rely on glycolysis. Further, their data support multiple mechanisms whereby glycolysis may occur simultaneously with anapleurosis to provide abundant energy to photoreceptors. The data from metabolomics revealed several novel findings in retinal metabolism, including the use of glutamine to fuel the mini-Krebs cycle, the utilization of the Cahill cycle in photoreceptors, and a taurine/hypotaurine shuttle between the underlying retinal pigment epithelium and photoreceptors to transfer reducing equivalents from the RPE to photoreceptors. In addition, this study provides robust quantitative metabolomics datasets that can be compared across experiments and groups. The use of this platform will allow for rapid testing of novel hypotheses regarding the metabolic ecosystem in the neural retina.

Strengths:

The data on differences in the susceptibility of rods and cones to mitochondrial dysfunction versus glycolysis provides novel hypothesis-generating conjectures that can be tested in animal models. The multiple mechanisms that allow anapleurosis and glycolysis to run side-by-side add significant novelty to the field of retinal metabolism, setting the stage for further testing of these hypotheses as well.

Weaknesses:

Almost all of the conclusions from the paper are preliminary, based on data showing enzymes necessary for a metabolic process are present and the metabolites for that process are also present. However, to truly prove whether these processes are happening, C13 labeling or knock-out or over-expression experiments are necessary. Further, while there is good data that RPE cultures in vitro strongly recapitulate RPE phenotypes in vivo, ex vivo neural retina cultures undergo rapid death. Thus, conclusions about metabolism from explants should either be well correlated with existing literature or lead to targeted in vivo studies. This paper currently lacks both.

Response: As mentioned above in the first answers to reviewers #1 and #2, we think of our study as a starting point that may provide novel directions for a whole series of investigations into retinal energy metabolism. Especially the use of novel technologies may in the future allow to decipher the different metabolic phenotypes of the 100+ distinct retinal cell types by in situ spatial metabolomics and lipidomics. Currently, we still have to limit the scope of our studies to only certain aspects of this topic. We thus agree that some of our interpretations need to be formulated more carefully and we have done so in the revised version of our manuscript. We also agree with the reviewer that carbon (13C) labelling and tracing studies will be very informative and will engage in such studies in the future. Besides 13C, we aim to further employ 15N labelled substrates, which is especially suitable to study the destiny of amino acids.

As far as our organotypic retinal explant system is concerned, it is arguably one of the best validated such systems available (see responses to reviewers #1 and #2). While the reviewer is correct to say that the neuroretina without RPE degenerates relatively quickly in vitro, in our system, with the neuroretina and its native RPE cultured together, we can routinely culture the retina for four weeks or more, without major cell loss (Söderpalm et al., IOVS 35:3910-21, 1994; Belhadj et al., JoVE 165, 2020). Thus, our retinal cultures with RPE do not undergo rapid death. Within the time-frame of the present study (6 days in vitro) culturing-induced cell death is minimal and unlikely to influence our analyses. For further, more detailed answers to the reviewers’ questions please see our detailed point-to-point response below.

We agree with the reviewer that eventually in vivo studies will be important to confirm our interpretations. As mentioned in our initial response to reviewer #1, such studies will be very challenging and new technologies may need to be developed before in vivo investigations can deliver the answers to the questions at hand (see answer to question Rev#3.17 below), especially if the cross-play between substrate availability from the blood metabolome and the retinal metabolic pathway activity shall be studied.

Recommendations For The Authors

Reviewer #1 (Recommendations For The Authors):

Rev#1.1. The animals should be screened for and lack rd8.

Response: This is a pertinent question from the reviewer. Ever since we first became aware of the presence of rd8 mutations in certain mouse lines from major vendors (e.g. Charles River, Jackson Labs) in around 2010, we have setup regular screening of all our mouse lines for this Crb1 mutation. Accordingly, the mouse lines used in this study were confirmed to be free of the rd8 / Crb1 mutation. A corresponding remark has now been inserted into the SI materials and methods section (Lines 37-38).

Rev#1.2. GLUT1 looks significantly different from in vivo to in vitro. Recommend co-staining with RHO and cone markers (PNA or CAR) to further delineate where it is being expressed. The in vitro cultures appear to have much shorter outer segments (OS). Considering OS biosynthesis is thought to drive a good deal of metabolic adaptations, how relevant is the in vitro model system to what is truly occurring in vivo?

Response: The GLUT1 staining shown in Figure 1 displays the in vivo situation. Since may not have been entirely clear from the previous figure legend, we have now labelled this as “in vivo retina” and distinguish it from “in vitro” samples in the legend to Figure 1 (Lines 774-778). As far as the comparison of GLUT1 staining in vivo (Figure 1A3) vs. in vitro (Figure S1C3) is concerned, in both situations a strong RPE labelling is clearly visible, with essentially no GLUT1 label within the neuroretina.

Nevertheless, to better delineate the expression of GLUT1 in the outer retina, we have now performed an additional co-staining with rhodopsin (RHO) as rod marker and peanut agglutinin (PNA) as cone marker, as suggested by the reviewer (new supplemental Figure S1). In brief, this co-staining confirms the strong expression of GLUT1 in the RPE, while there is essentially no GLUT1 detectable in rod or cone photoreceptors.

Retinal explants in long-term cultures do indeed have somewhat shorter outer segments compared to same age in vivo counterparts (Caffe et al., Curr Eye Res. 8:1083-1092, 1989). However, in the short-term cultures (6 DIV) and at the age studied here (P15) outer segments have only just started to grow out and are around 10 - 12 µm long, both in vitro and in vivo (cf. LaVail, JCB 58:650-661, 1978). Thus, the metabolism required for outer segment synthesis should be equivalent when in vitro and in vivo situations are compared. For considerations on outer segments in retinal explant cultures see also Rev#3.2 and Rev#3.29.

Rev#1.3. Also, recent publications have shown that GLUT1 is expressed in the neuroretina including rods, cones, and muller glia. Was GLUT1 not appreciated in these cells in your ex vivo samples and if so, why? Likewise, these same studies previously demonstrated GLUT1 resulted in rod degeneration but not cone. The results presented here differ significantly. Why the difference in results and is it secondary to the in vitro vs. in vivo setting? Furthermore, the authors state that they thought the no RPE situation would be similar to the GLUT1 inhibitor experimental condition but instead, they were vastly different. Is this secondary to the fact that GLUT1 is expressed outside the RPE.

Response: We are aware that there is a controversy regarding GLUT1 expression in the neuroretina, please see also our response to question Rev#3.1 below. As far as our immunostaining for GLUT1 on in vivo retina is concerned, we find an unambiguous and very marked expression of GLUT1 in RPE cells, at both basal and apical sides. Compared to the RPE, the neuroretina appears devoid of GLUT1 staining. However, at very high gamma values a faint staining in the neuroretina becomes visible, a staining which from its appearance – processes spanning the entire width of the retina – is most compatible with Müller glia cells. Under normal circumstances we would have dismissed such a faint staining as background and false positive. Given the sometimes very contradicting reports in the literature, we cannot fully exclude a weak expression of GLUT1 also in cells other than the RPE, with Müller glial cells perhaps being the most likely candidate. At any rate, GLUT1 expression in the neuroretina can only be much weaker than in the RPE, making its relevance for overall retinal metabolism unclear.

As far as recent publications studying GLUT1 in the retina are concerned, we know of the study by Daniele et al. (FASEB Journal 36:e22428, 2022), which used a rod-specific, conditional knock-out of GLUT1 and found a relatively slow rod degeneration. We are not aware of a selective GLUT1 knock-out in cones, nor are we aware of conditional GLUT3 knock-outs in the retina. For further discussion of the Daniele et al. study please see Rev#3.13.

The reviewer is right, initially we were thinking that, since GLUT1 was expressed only (predominantly) in RPE, the metabolic response to GLUT1 inhibition should look similar to the no RPE situation. However, this initial hypothesis did not consider a key fact: The RPE builds the blood retinal barrier and the tight-junction coupled RPE cells are a barrier to any larger molecule, including glucose. Removing the barrier by removing the RPE dramatically increases the availability of glucose to the retina, a phenomenon that is likely exacerbated by the expression of the high affinity/high capacity GLUT3 on photoreceptors (cf. Figure S1A). In other words, when the RPE is removed the outer retina is “flooded” with glucose and we believe that this is probably the main factor that explains why the metabolic response to GLUT1 inhibition (1,9-DDF group) is so different from the no RPE condition.

We have now included an additional corresponding explanation in the discussion (Lines 422-429). Furthermore, we have added an entire new subchapter to the discussion to debate the expression of glucose transporters in the outer retina (Lines 454-472).

Rev#1.4. Shikonin's mechanism of action via protein aggregation and lack of specificity for PKM2 vs PKM1 at 4uM is an experimental limitation that needs to be taken into account. All treatments utilized are not cell-specific.

Response: While the reviewer is correct to say that Shikonin may have multiple cellular targets and a diverse range of possible applications as an anti-inflammatory, antimicrobial, or anticancer agent (cf. Guo et al., Pharmacol. Res. 149:104463, 2019), numerous studies support its specificity for PKM2 over PKM1, at concentrations ranging from 1 – 10 µM (Chen et al., Oncogene 30:4297-306, 2011; Zhao et al., Sci. Rep. 8:14517, 2018; Traxler et al., Cell Metab. 34:1248-1263, 2022). We settled for 4 μM as an intermediate concentration, considering its effectiveness and specificity in previous studies. We have now inserted references detailing the specificity and concentration range of Shikonin into the SI Materials and Methods section (Line 62).

The concern that “all treatments” are not cell-specific is debatable. Certainly, any given compound may have off-target effects, yet, since the compounds we used in our study have all been studied for decades (see above, initial response to Reviewer #2), their off-target profile is well established and unlikely to play an important role here. Moreover, in our study the cell specificity does not come from the compounds used but from where their targets are expressed. As shown in Figure 1A and in Figure S1C, Shikonin´s target PKM2 is almost exclusively expressed in photoreceptor inner segments. Hence, it seems very reasonable to expect that the vast majority of the metabolomic changes observed by Shikonin treatment are related to photoreceptors. We note that this assertion would still be true even if there was a low-level expression of PKM2 in other retinal cell types and/or if Shikonin had moderate off-target effects on other enzymes since the bulk of the effect on the quantitative metabolomic dataset would still originate from PKM2 inhibition in photoreceptors.

Rev#1.5. What was the method of cone counting in Figure 1?

Response: Cones were counted per 100 µm of retinal circumference based on an arrestin-3 staining (cone arrestin, CAR).

This information is now included in the SI Materials and Methods section under “Microscopy, cell counting, and statistical analysis” (Lines 99-100).

Rev#1.6. How do you know that FCCP is not altering RPE ox phos, disrupting the outer retinal microenvironment and leading to cell death, and therefore, the effects seen are not photoreceptor-specific but rather downstream from the initial insult in RPE?

Response: We propose that FCCP will be acting on both photoreceptors and RPE cells (and all other retinal cell types) at essentially the same time, over the experimental time-frame. Thus, OXPHOS should be inhibited in all cells simultaneously. However, FCCP will primarily affect cells that actually use OXPHOS to a large extent, while cells relying on other metabolic pathways (e.g. glycolysis) will hardly be affected.

We believe the very strong effect of FCCP, seen exclusively in rod photoreceptors, to be a direct drug effect. While we cannot not fully exclude an indirect effect via the RPE – as proposed by the reviewer – we think this to be unlikely because:

(1) RPE viability was not compromised by FCCP treatment.

(2) If the reviewer´s hypothesis was correct, then also cone photoreceptors should have been affected (e.g. because now the RPE consumes all glucose, leaving nothing for cones). However, cones were essentially unaffected by the FCCP treatment, making a dependence on RPE OXPHOS unlikely. Especially so, because blocking GLUT1 and glucose import on the RPE with 1,9-DDF had only relatively minor effects on rod photoreceptor viability but strongly affected cones. This indicates that the RPE is mainly shuttling glucose through to photoreceptors, especially to cones, and this function does not seem to be impaired by FCCP treatment.

(3) We found that enzymes required for Krebs-cycle and OXPHOS activity (i.e. citrate synthase, fumarase, ATP synthase γ) are predominantly expressed in photoreceptors but virtually absent from RPE (Figure 3D, see also answer to following question).

(4) The density of mitochondria (i.e. the target for FCCP) is far lower in RPE than in photoreceptors, as evidenced also by the COX staining shown in Figure 1A. Hence, photoreceptors are far more likely to be hit by FCCP treatment than RPE cells.

To accommodate the reviewer´s concern, we have now added a further comment into the discussion (Lines 440-442).

Rev#1.7. While Figure 3D is interesting, it offers no significant insight into mechanisms as the enzyme levels are not being compared to control nor is mitochondrial fitness in these conditions being assessed, which would provide greater insight than just showing that these enzymes are present in the inner segments, which are known to be rich in mitochondria. Additionally, stating that the low ATP is secondary to decreased Krebs cycle activity and ox phos based on merely ATP levels is not supported by metabolite levels minus citrate nor ox phos enzyme levels or oxygen consumption. Also, citrate is purported to be decreased in the table in Figure 2 in the no RPE condition; however, Supplemental Figure 2 demonstrates this change is not significant then the same data is presented in Supplemental Figure 3 and it is statistically significant again. Why the difference in data and why is the same data being shown multiple times?

Response: The immunostaining shown in Figure 3D shows the in vivo retina, or in other words the localization of enzymes in the native situation. Since this may not have been obvious in the previous manuscript version, we have added a corresponding comment to the legend of Figure 3 (Line 806). The localization of the Krebs-cycle/OXPHOS enzymes citrate synthase, fumarase, and ATP synthase mainly to photoreceptors, but not (or much less) to RPE, is another piece of evidence supporting the idea that OXPHOS is predominantly performed by photoreceptors (see also answer to previous question Rev#1.6).

The decreased ATP levels (together with citrate, aspartate, NAA) shown in Figure 3 in the no RPE group, are an indication that photoreceptor Krebs-cycle activity may be decreased but not abolished in the absence of RPE. Importantly, GTP levels are not reduced in the no RPE group (Figure 2). Since large amounts of GTP can only by synthesized by either SUCLG-1 in the Krebs-cycle or by NDK-mediated exchange with ATP, the most plausible interpretation is that Krebs-cycle dependent ATP-synthesis was decreased in the no RPE situation, but that the (mini) Krebs-cycle or Cahill-cycle, notably the step from succinyl-CoA to succinate, was running. Since there is no RPE in this group, this strongly suggests important Krebs-cycle/OXPHOS activity in photoreceptors where the majority of the corresponding enzymes are located (see above).

We thank the reviewer for pointing out that the information on group comparisons may not have been presented with sufficient clarity. In the figures mentioned by the reviewer the data is shown and compared in different contexts: the table in Figure 2B and the data in Figure S3 (now renumbered to Figure S5) refer to two-way comparisons of treatment condition to control, to elucidate individual treatment effects. Meanwhile Figure S2 (now supplementary Figure S3) refers to a 5-way comparison for a general overview that puts all five groups in context with each other. These differences in comparisons and normalization to the respective common standards entail the use of different statistical tools, resulting in different p-values. The statistical testing approaches and thresholds are now disclosed in the figure legends, and additionally in the SI Materials and Methods section (Lines 145-155).

Rev#1.8. When were the ex vivo samples taken for metabolomics, and if taken when significant TUNEL staining and cell death have occurred, are the changes in metabolism due to cell death or a true indication of differential metabolism? Furthermore, it is unclear if the metabolomics samples included the RPE or not. Considering these treatments will affect most cells in the retina and the RPE, which is included in the ex vivo samples, it is difficult to ascertain that these changes are secondary to the effects on photoreceptors alone.

Response: The samples for metabolomics included the RPE (except for the no RPE condition) and were taken at the same time as the tissues for histological preparations and TUNEL assays, i.e. they were all taken at post-natal day 15. This has now been clarified in the SI Materials and Methods section (Lines 108-110).

We cannot entirely exclude an effect of ongoing cell death caused by the different drug treatments on the retinal metabolome. However, since in the experimental treatments cell death was still comparatively low (even in the FCCP condition, overall cell death was only around 10% of the total retina), and the metabolomic analysis considered the entire tissue, the impact of cell death per se on the total metabolome will be comparatively minor (≤ 10%, i.e. within the typical error margin of the metabolomic analysis).

As mentioned above, the drug treatments should in principle affect all retinal cells at the same time. However, only cells that express the drug targets (i.e. 1,9-DDF targets GLUT1 in RPE cells, Shikonin targets PKM2 in photoreceptors; cf. Figure 1A) should react to the treatment. Even FCCP, in the paradigm employed, will only affect those cells that rely heavily on OXPHOS. Our data indicates that while this is almost certainly the case for rods; cones, RPE cells, and essentially all of the inner retina, are not affected by FCCP treatment, strongly suggesting that OXPHOS is of minor importance for these cell populations.

Rev#1.9. Why were the FCCP and no RPE groups compared? If they have similar metabolite patterns as noted in Figure 2, would that suggest that FCCP's greatest effect is on the ox phos of RPE and the metabolite patterns are secondary to alterations in RPE metabolism? Also, the increase in citrate and decrease in NAD may be related to effects on RPE mitochondrial metabolism when comparing these groups, and the disruption of RPE metabolism may then result in PARP staining of photoreceptors.

Response: The reason for the pair-wise comparison of the no RPE and FCCP groups initially was indeed the similarity in metabolite patterns. This was now rephrased accordingly in the results section “Photoreceptors use the Krebs-cycle to produce GTP” (Lines 218-219). The interpretation that the reviewer proposed here is interesting, but does not conform with the data analysis of this and other group comparisons.

Instead, the similarity between the metabolic patterns found in the no RPE and FCCP groups further supports the idea that a lack of RPE decreases retinal OXPHOS and increases glycolysis. This interpretation is based on the following observations:

(1) Mitochondrial density in the RPE is far lower than in photoreceptors (see COX staining in Figure 1A), thus quantitatively the metabolite pattern caused by a disruption of OXPHOS (via FCCP treatment) will be dominated by metabolites generated by photoreceptors. For the same reason the depletion of retinal NAD+, and the concomitant increase in photoreceptor PAR accumulation after FCCP treatment, is unlikely to be due to changes in RPE.

(2) Similarly, citrate synthase (CS) was found to be almost exclusively expressed in photoreceptor inner segments, with little expression in RPE (Figure 3D). Hence, the quantitative increase of citrate levels after FCCP treatment can only originate in photoreceptors.

(3) The comparison of the control (with RPE) against the no RPE group suggested an increase in (aerobic) glycolysis in the absence of RPE, evidenced notably by a retinal accumulation of lactate, BCAAs, and glutamate (Figure 3A). The very same metabolite pattern is seen for the FCCP treatment (Figure 1B) indicating a marked upregulation of glycolysis (Figure 6C). The latter observation suggests that photoreceptors, after disruption of OXPHOS switch to an exclusively glycolytic metabolism, which, however, rods cannot sustain (Figure 1C, D).

(4) Glucose consumption and lactate release is increased in the no RPE group vs. control (new Supplementary Figure 4). A similar increase in glucose consumption and lactate production is seen in the FCCP group suggesting that also the no RPE situation disrupts OXPHOS in photoreceptors.

Rev#1.10. The conclusions being reached are difficult to interpret secondary to the experimental procedures and the fact that the treatments are not cell-specific and RPE is included with the neuroretina as well. Likewise, stating FCCP is altering the Krebs cycle in the neuroretina is difficult to believe as there are no changes in the Krebs cycle when compared to the control, which also has RPE.

Response: We agree with the reviewer, that some of the conclusions may have been somewhat speculative. Accordingly, we have toned down our conclusions in several instances in the text, notably in abstract, introduction, and discussion.

When it comes to Krebs cycle intermediates a key limitation of our study is indeed the lack of carbon-tracing and metabolic flux analysis as noted by the reviewers, a limitation that we now highlight more strongly in the discussion of the revised manuscript (Lines 545-549). While it is highly probable that the flux of Krebs cycle intermediates is altered by FCCP, our steady-state data does not show significant changes in the metabolites citrate, fumarate, and succinate. However, our study does show a highly significant decrease in GTP levels, which as explained above, is a key indicator of Krebs cycle activity/inactivity. Moreover, while GTP levels were reduced also in the no RPE group, GTP was still significantly higher in the no RPE group compared to the FCCP treatment. Our interpretation of this finding is that there is Krebs-cycle/OXPHOS activity in the neuroretina, which is abolished by FCCP.

Rev#1.11. Supplemental Figure 4C and D states that GAC inhibition affected only photoreceptors, but GAC is expressed throughout the retina and so the inhibition is altering glutamine-glutamate homeostasis throughout the retina. Clearly, based on histology, one can see that the architecture of the retina, especially at the highest dose, is lost likely because all cells are being affected. So it is not photoreceptor-specific and even at low doses one can see that the inner retina is edematous. Moreover, with such a high amount of TUNEL staining in the ONL, are rods more affected than cones?

Response: In our hands the immunostaining for Glutaminase C (GAC) labelled predominantly cone inner segments, the OPL, and perhaps bipolar cells (Figure S1A). The deleterious effects mentioned by the reviewer are only seen at the highest concentration of the GAC inhibitor compound 968. This concentration (10 µM) is 100-fold higher than the dose that produces a significant loss of cones in the outer retina (0.1 µM). We therefore think that this data points to the extraordinary reliance of cones on glutamine and glutamate. As can be seen from the images (Figure S4C) illustrating the effects of 0.1 and 1 µM Compound 968 treatment, the ONL thickness is not significantly reduced by the GAC inhibitor. This strongly indicates that at these doses the rods are not affected by GAC inhibition.

Rev#1.12. The no RPE vs 1,9 DDF data may be interpreted as preventing glucose transport in the RPE increases BCAA catabolism by the RPE, which has been shown to utilize BCAA in culture systems. To this end, when the RPE is not present, the BCAA is increased as compared to the control with RPE.

Response: Our original interpretation of this data was that after GLUT1 inhibition and a correspondingly reduced retinal glucose uptake, the retina switched to an increasing use of anaplerotic substrates, including BCAAs. This is supported by the concomitant upregulation of the Cahill-cycle product alanine and the mini-Krebs-cycle product N-acetylaspartate (NAA). Yet, we agree with the reviewer that BCAAs could also be consumed by the RPE. We have now changed our conclusion at the end of the results chapter “Reduced retinal glucose uptake promotes anaplerotic metabolism“ to also highlight this possibility (Lines 261-262).

Rev#1.13. It is unclear why so much effort is comparing the no RPE group to the treatment groups and not comparing the control group to the different treatment groups.

Response: Previous studies – including the seminal studies of Otto Warburg from the early 1920s – had always used retina without RPE. This “no RPE” situation is therefore something of a reference for our entire study, which is why we dedicated more effort to its analysis. We have now inserted a corresponding remark into the manuscript (Lines 182-184).

Rev#1.14. The conclusions are significantly overstated especially with regards to rods versus cones as these are not cell-specific treatments. For example, the control vs 1,9 DDF vs FCCP clearly shows that there is mitochondrial dysfunction due to decreased NAD, increased AMP/ATP ratio, decreased Asp but increased Gln, and a compensatory increase in lactate production.

Response: We agree with the reviewer and have tried to phrase our statements in more measured fashion. Notably, we have toned down our statements in the title, abstract, results, discussion, and several of the subchapter headings.

Rev#1.15. While metabolic conclusions are drawn on serine/lactate ratio, this ratio is driven by the drastic changes in lactate and not so much serine in the treatment conditions as it was rather stable. Likewise, substrates beyond glucose have the potential to fuel the TCA cycle and make GTP via SUCLG1, such as fatty acids, other AAs, etc. Therefore, this ratio may not tell the entire story about anaplerotic metabolism. Furthermore, knowing that RPE utilize BCAAs to fuel their TCA cycle, the no RPE condition may simply have increased BCAAs due to lack of metabolism by the RPE, which drives the GTP/BCAA ratio. To state that the neuroretina was utilizing BCAAs for anaplerosis is not well supported based on the current data. Similarly, what is to say that the GTP/lactate ratio in the no RPE situation is not driven by the fact that the RPE is no longer present to act as acceptor of retinal lactate production or that more glucose is reaching the retina since the RPE is not present to accept and utilize that produced. Glucose uptake was not assessed to further address these issues.

Response: We agree with reviewer that metabolite ratios may not tell the full story underlying retinal metabolism however based on the robustness of using quantitative and highly reproducible NMR data, they are an important part of the metabolomics toolbox. The reviewer correctly observed that the changes in lactate levels are more dramatic than in serine. Still, also serine was significantly increased in the no RPE, 1,9-DDF, and Shikonin groups. Together with the lactate changes (same or opposite direction) the resulting serine/lactate ratios display marked alterations.

When it comes to the supply of other potential energy substrates mentioned by the reviewer, i.e. fatty acids or amino acids other than BCAAs, these are only supplied in minimal amounts in the defined, serum free R16 medium (Romijn, Biology of the Cell, 63, 263-268, 1988) and – if used to any important extent – would be rapidly depleted by the retina. Thus, for a culture period of 2 days in vitro between medium changes these energy sources are not available and thus cannot be used by the retina.

Our conclusion that the retina is using anaplerosis is based not only on the observations made in the no RPE group but also on, for instance, the metabolite ratios seen in the 1,9-DDF treatment group. In this group decreased glycolytic activity may correspond to increased serine synthesis and anaplerosis.

As far as glucose uptake is concerned, we have analysed the medium samples at P15 (equivalent to the retina tissue collection time point) and now present data that addresses this question more directly via the consumption of glucose from and release of lactate to the culture medium (New Supplementary Figure 4C, D). This new dataset provides another independent observation showing that:

(1) Glucose consumption/lactate release (i.e. aerobic glycolysis) is high in the no RPE situation but low in the control situation. In other words, retinal aerobic glycolysis is most likely stimulated by the absence of RPE.

(2) 1,9-DDF treatment decreases glucose consumption/lactate release as would be expected from a GLUT1 blocker. Since ATP and GTP production are high nonetheless, this indicates that other substrates (i.e. anaplerosis) were used for retinal energy production, in agreement with the analysis shown in Figure 6C.

(3) The FCCP treatment, which disrupts oxidative ATP-production, increases glucose consumption/lactate release in way similar to the no RPE situation. Yet, the no RPE retina can still generate sizeable amounts of GTP but not ATP. Together, this provides further evidence that neuroretinal OXPHOS is decreased in the absence of RPE.

Rev#1.16. The evidence for the mini-Krebs cycle is intriguing but weak considering it is based on certain enzymes being expressed in the photoreceptors, which had already been shown to be present in other publications, and a single ratio of metabolites that is increased in FCCP. One would expect this ratio to be increased under FCCP regardless. There is no stable isotope tracing with certain fuels to confirm the existence of the mini-Krebs cycle.

Response: We thank the reviewer for this suggestion. We agree that our evidence for the mini-Krebs-cycle (and the Cahill-cycle) may be to some extent circumstantial and additional technologies would help to obtain further supportive data. Still, here we would like to invite the reviewer to a thought experiment where he/she could try and interpret our data without considering the Cahill- or the mini-Krebs-cycle. At least we ourselves, when we engaged into such thought experiments, were unable to explain the data observed without these alternative energy-producing cycles. Most notably, we were unable to explain the strong accumulation of either alanine or N-acetyl-aspartate (NAA) when only considering glycolysis and (full) Krebs-cycle metabolism. Of course, this may still be considered “weak” evidence, and we expect that future studies including complementary technologies will either confirm or expand our interpretation of the existing data set.

The suggestion to perform stable isotope-labelled tracing with potential alternative fuels (e.g. glutamate, glutamine, pyruvate, etc.) is very attractive indeed. While such studies are likely to shed further light on the metabolic pathways proposed, this will entail very extensive experimental work, with multiple different conditions and concentrations and variety of analysis methods that is currently not feasible (e.g. a 1.7 mm NMR probe equipped with a 15N channel) as an extension of the present manuscript. Nevertheless, we will certainly consider this approach for future follow-up studies once such techniques are available and will screen for suited collaboration partners. A corresponding comment on such future possibilities has now been inserted into the discussion (Lines 545-549).

Rev#1.17. The discussion does not mention how this data contradicts a recent in vivo study looking at Glut1 knockout in the retina (Daniele et al. FASEB. 2022) or previous in vivo studies that suggest cones may be less sensitive to changes in glucose levels (Swarup et al. 2019). This is a key oversight.

Response: We thank the reviewer for pointing this out. We now included these studies in the revised discussion in a new subchapter on the expression of glucose transporters in the outer retina (Lines 454-472). For a critical review of the Daniele et al., 2022 study please also see our more detailed response to question Rev#3.13 below.

Rev#1.18. GAC is expressed in more than just cones so making cell-specific statements regarding fuel utilization is not well supported.

Response: Our immunostaining for GAC revealed a strong expression in cone inner segments (Figure S1A3). While this does not exclude (relatively minor) expression in other retinal cell types, cones are likely to be more reliant on GAC activity than other cell types. See also answer above.

Rev#1.19. Suggesting that rods utilize the mini-Krebs cycle based on AAT2 being seen in the inner segments without at least co-staining for RHO or PNA is weak evidence for such a cycle. AAT looks to be expressed in the inner segments of all photoreceptors.

Response: We have taken up this suggestion from the reviewer and now provide an additional co-staining for AAT1 and AAT2 with rhodopsin. Note that in response to a pertinent comment from Reviewer #3 we have changed the abbreviation for aspartate aminotransferase from “AAT” to the more commonly used “AST” throughout the manuscript.

New images showing a co-staining for AST1 and AST2 with rhodopsin now replace the former image set in Figure 7D. In brief, the new images show the expression of both AST1 and AST2 across the retina, with, notably an expression in the inner segments of photoreceptors but not in the outer segments, where rhodopsin is expressed.

Reviewer #3 (Recommendations For The Authors):

Rev#3.1. The staining for the glucose transporters GLUT1 and GLUT3 does not reflect what has previously been published by two different groups that were validated by cell-specific knockout mice. As mentioned by the author GLUT1 and GLUT3 have differences in transport kinetics, which would affect their metabolism. Therefore, the lack of GLUT1 in photoreceptors would suggest that photoreceptor metabolism is not faithfully replicated in this system. This difference from the previous literature should be discussed in the discussion.

Response: As the reviewer pointed out, the expression of GLUT1 in the retina is somewhat controversial, with much older literature showing expression on the RPE, while some more recent studies claim GLUT1 expression in photoreceptors. For a brief discussion of our GLUT1 immunostaining please see also our answer to question Rev#1.3 above.

Although the retinal expression of GLUT1 was besides the focus of our study, we feel we must address this point in more detail: In the brain the generally accepted setup for GLUT1 and GLUT3 expression is that low-affinity GLUT1 (Km = 6.9 mM) is expressed on glial cells, which contact blood vessels, while high-affinity GLUT3 (Km = 1.8 mM) is expressed on neurons (Burant & Bell, Biochemistry 31:10414-20, 1992; Koepsell, Pflügers Archiv 472, 1299–1343, 2020). This setup matches decreasing glucose concentration with increasing transporter affinity, for an efficient transport of glucose from blood vessels, to glial cells, to neurons. In the retina, the cells that contact the choroidal blood vessels are the tight-junction-coupled RPE cells. As shown by us and many others, RPE cells strongly express GLUT1 (cf. Figure 1A-3.). To warrant an efficient glucose transport from the RPE to photoreceptors, photoreceptors must express a glucose transporter with higher glucose affinity than GLUT1. We show that this is indeed the case with photoreceptors expressing GLUT3 (cf. Supplemental Figure 1-5.). While a part teleological explanation does not per se prove that our data is correct, at least our data is plausible. In contrast, the glucose transporter setup sometimes claimed in the literature is biochemically implausible, i.e. for the flow of metabolites (glucose) to go against a gradient of transporter affinities, and we are not aware of an example of such a setup occurring anywhere in nature.

However, at this point we cannot exclude low levels of GLUT1 expression on Müller glia cells or even photoreceptors. This expression could, for instance, be relevant in cases where cells were shuttling excess glucose – perhaps produced through gluconeogenesis – onwards to other retinal cells. Still, GLUT1 expression can only be minor when compared to RPE since a major expression would destroy the glucose affinity gradient (see above) required for efficient glucose shuttling into the energy hungry photoreceptors.

To address this request by the reviewer (and also reviewer #1) we now discuss the question of glucose transporter expression in the outer retina in a new subchapter of the discussion (Lines 454-472).

Rev#3.2. Photoreceptor metabolism and aerobic glycolysis are tied to photoreceptor function, as demonstrated by Dr. Barry Winkler. The authors should provide data or mention (if previously published) about photoreceptor OS growth and function in this system.

Response: The studies of Barry Winkler (e.g. Winkler, J Gen Physiol. 77, 667-692, 1981) confirmed the original work of Otto Warburg and expanded on the idea that the neuroretina was using aerobic glycolysis. Importantly, Winkler used a very similar experimental setup as Warburg has used, namely explanted rat retina without RPE. In light of our data where we compare metabolism of mouse retina with and without RPE – where retina cultured without RPE confirms the data of Warburg and Winkler – it appears most likely that the purported aerobic glycolysis occurs mostly in the absence of RPE but only to a lower extent in the native retina.

Photoreceptor outer segment outgrowth is somewhat slower in the organotypic retinal explant cultures compared to the in vivo situation (cf. Caffe et al., Curr Eye Res. 8:1083-1092, 1989 with LaVail, JCB 58:650-661, 1978; see also answer to reviewer #1). Importantly, organotypic retinal explant cultures and their photoreceptors are fully functional and remain so for extended periods in culture (Haq et al., Bioengineering 10:725, 2023; Tolone et al., IJMS 24:15277, 2023). This information has now been added to the manuscript discussion section, into the new subchapter “The retina as an experimental system for studies into neuronal energy metabolism” (Lines 367-395).

Rev#3.3. It is unclear from the description of the experiment in both the results and methods if 1,9DDF, Shikonin, and FCCP were added to both apical and basal media compartments or one or the other and should be specified. The details of what was on the apical compartment would be helpful, as the model is supposed to allow for only nutrients from the basal compartment (as indicated by the authors themselves). Is the apical compartment just exposed to air? How does this affect survival?

Response: In organotypic retinal explant cultures the RPE rests on the permeable culturing membrane such that the basal side is contact with the membrane and the medium below (far schematic drawing see Figure S1B), while the apical side is covered by a thin film of medium created by the surface tension of water (Caffe et al., Curr Eye Res. 1989; Belhadj et al., JoVE, 2020). This thin liquid film ensures sufficient oxygenation and is an important factor that allows the retinal explant to remain viable for several weeks in culture. If the retinal cultures were submerged by the medium, their viability – especially that of the photoreceptors – would drop dramatically and would typically be below 3-5 days. Therefore, in the retinal organotypic explant cultures used here, the nutrients and the drugs applied do indeed reach the outer retina from the basal side, i.e. similar as they would in vivo.

To address this question from the reviewer, corresponding clarifications have been inserted into the SI Materials and Methods section (Lines 64-66).

Rev#3.4. As the metabolomic data obtained was quantitative, several metabolites discussed should be analyzed in terms of ratios, for example, Glutathione and glutathione disulfide should be reported as a ratio. In addition as ATP, ADP, and AMP were measured, they can used to calculate the energy charge of the tissue.

Response: We thank the reviewer for these suggestions and have created corresponding graphs for GSH / GSSG ratio and energy charge. These new graphs have now been added to the SI datasets, to the new Supplementary Figure 4. To accommodate other requests from the Reviewers, this new Figure also contains additional new datasets on glucose and lactate concentrations (see further comments above and below). Please note that all later SI Figures have been renumbered accordingly.

In brief, the ratios for GSH/GSSG show no significant changes between control and the different experimental groups. Meanwhile, the adenylate energy charge of the retinal tissues show a significant decrease in the energy charge for the Shikonin group and the FCCP group. Note that in the new Supplementary Figure 4A, the dotted lines indicate the energy charge window typical for most healthy cells (0.7 – 0.95).

Rev#3.5. I think a missed opportunity when discussing the possible taurine/hypotaurine shuttle would be the impact on the osmosis of the subretinal space as taurine has been hypothesized as a major osmolyte.

Response: This is another interesting recommendation from the reviewer. To address this point, we have now introduced a corresponding paragraph and references in the discussion of the manuscript (Lines 503-504; 512-514).

Rev#3.6. In Figure 3, the distribution of these enzymes should also be studied under the no RPE condition as the culture treatment took several days for these metabolic changes to occur.

Response: The images shown in Figure 3D are from the in vivo retina. Since this may not have been very clear in the previous manuscript version, we have now added a corresponding explanation to the legend of Figure 3. As far as we can tell, the expression and localization of neuroretinal enzymes does not change in cultured retina, during the culture period (compare Figure 1A with Supplementary Figure S1C). However, when it comes to the metabolite taurine its production (localization) changes dramatically in the no RPE situation where taurine is essentially undetectable by immunostaining (not shown but see metabolite data in Figure 2A, Figure 3A).

Rev#3.7. In Figures 4 and 5, it is unclear why the experimental groups were not compared to the control and requires further explanation. Furthermore, the authors should justify the concentrations of drugs used as the cell death could have risen from toxicity to the drugs and not due to disruption of metabolism.

Response: The reviewer is right, the rationale for these comparisons may not have been laid out with sufficient clarity. In Figure 4 the no RPE and FCCP groups are compared because both groups showed similar metabolite changes towards the control situation. The no RPE to FCCP comparison thus focussed on the details of the – at first seemingly minor – differences between these two groups. This has now been clarified in the corresponding part of the results (Lines 218-219).

In Figure 5A, B we compare the no RPE and 1,9-DDF groups with each other, notably because the data obtained seemingly contradicted our initial expectation that these two groups should show similar metabolite patterns. Also here, we have now inserted an additional explanation for this choice of comparisons (Lines 252-253).

In Figure 5C, D we compare the Shikonin and FCCP groups with each other. The idea behind this comparison was that in the 1st group glycolysis was blocked while in the 2nd group OXPHOS was inhibited, or in other words here were compared what happened when the two opposing ends of energy metabolism were manipulated in opposite directions. This reasoning is now given in the results section (Lines 265-268).

As far as the choice of drugs and concentrations is concerned, we used only compounds that have been extremely well validated through up to five decades of scientific research (see initial response to Reviewer #2 above). We therefore are confident that at the concentrations employed the results obtained stem from drug effects on metabolism and not from generic, off-target toxicity. Then again, as we show, prolonged (i.e. 4 days) block of energy metabolism pathways does cause cell death.

Rev#3.8. In line 203, the authors discuss GTP as being primarily a mitochondrial metabolite, however, photoreceptors would require a localized source of GTP synthesis in the outer segments as part of phototransduction, and therefore GTP in photoreceptors cannot be a mitochondrial-specific reaction in photoreceptors. Furthermore, the authors mentioned NDK as being a possible source of GTP, but they do not show NDK localization despite it being reported in the literature to be localized in the OS.

Response: The question as to the source of GTP in photoreceptor outer segments is indeed highly relevant. For GTP production in mitochondria see the answer to the next question below (Rev#3.9). An early study showed nucleoside-diphosphate kinases (NDK) to be expressed on the rod outer segments of bovine retina (Abdulaev et al., Biochemistry 37:13958-13967, 1998). More recently NDK-A was shown to be strongly expressed in photoreceptor inner segments (Rueda et al., Molecular Vision 22:847-885, 2016). We now refer to both studies in the results section of the manuscript (Line 227-228).

Rev#3.9. In the "Impact on glycolytic activity, serine synthesis pathway, and anaplerotic metabolism" section, the authors claim in the no RPE group glycolytic activity was higher due to a depressed GTP-to-lactate ratio. However, this reviewer is under the impression that GTP production in photoreceptors is not mitochondrial specific, so this ratio doesn't make sense (I could be mistaken, however). A better ratio would have been pyruvate/lactate or glucose/lactate when discussing increased glucose consumption.

Response: We appreciate the reviewers’ comment, yet we do indeed believe we can show that GTP-production in our experimental context is mainly mitochondrial. As explained in the manuscript results section (“Photoreceptors use the Krebs-cycle to produce GTP”), there are essentially only two possibilities for a photoreceptor to produce sizeable amounts of GTP. In the mitochondria via SUCLG1 – i.e. an enzyme highly expressed in photoreceptor inner segments (Figure 5D) – and the cytoplasm via NDK from excess ATP. The claim about the depressed GTP-to-lactate ratio in the no RPE situation takes this into account. Importantly, since in the no RPE situation ATP-levels are significantly lower than GTP, here GTP can only be produced via SUCLG1 and OXPHOS. Moreover, this contrasts with the FCCP group where mitochondrial OXPHOS is disrupted and both ATP and GTP are depleted.

As far as ratios with pyruvate and glucose are concerned, we agree that these could potentially be very interesting to analyse. Unfortunately, in our retinal tissue 1H-NMR spectroscopy- based metabolomics analysis the levels of both pyruvate and glucose were below the detection limits which likely reflects their rapid metabolic turnover (cf. table S1). While this might be attributable to the marked consumption of these metabolites within the tissue, it does not allow for us to calculate the suggested ratios to lactate. Then again, in the supernatant medium which was collected at the same time point as the retina tissue, we can readily detect glucose and lactate levels, for this data please see the new Supplementary Figure 4.

Rev#3.10. Aspartate aminotransferase should be abbreviated as AST, as it is more commonly noted.

Response: In response to this comment from the reviewer, we have changed the abbreviation for aspartate aminotransferase from AAT to AST throughout the manuscript.

Rev#3.11. In the discussion the assumptions of the ex vivo culture systems should be clearly stated. One that was not mentioned, but affects the implications of the data, is that the retinas used in this study are from the developing mouse eye. Another important assumption that was made in this paper was that the changes in retinal metabolism were due to photoreceptors even though the whole neural retina was included.

Response: The reviewer is correct; we have added these two points to the discussion section of the manuscript. Notably, we now included a new subchapter “The retina as an experimental system for studies into neuronal energy metabolism” (Lines 367-395) to present different in vitro and in vivo test systems.

Rev#3.12. Starting at line 347: As the authors know, the RPE has been shown to be highly reliant on mitochondrial function, and disruption of RPE mitochondrial metabolism leads to photoreceptor degeneration (numerous papers have shown this). Furthermore, the lower levels of lactate detected in their explants when RPE was present suggests that lactate is actively transported out of the neural retina by the RPE.

Response: The reviewer is right about lactate being exported from the retina to the blood stream in vivo, or, in our in vitro study, to the culture medium. In the new dataset showing glucose and lactate concentrations in the culture medium (new Supplementary Figure 4C, D), we show that without RPE (no RPE group) and the retina releases more significantly lactate into the medium than control retina with RPE. At the same time the no RPE retina consumes more glucose than control retina.

Rev#3.13. Line 360: Again, in mouse photoreceptors (by bulk RNAseq and scRNAseq), there is no GLUT3 expression (encoded by slc2a3). It was also recently shown by Dr. Nancy Philp's lab that rod photoreceptors express GLUT1, encoded by slc2a1 (PMCID: PMC9438481). The differences reported in this study and previous studies should be discussed.

Response: Although this comment may not make us very popular, we are somewhat sceptical of RNAseq data (especially single cell RNAseq) since the underlying methodology – at the current level of technological development – is notoriously unreliable when it comes to the assessment of low abundance transcripts and suffers from apoor batch reproducibility, compared to NMR based metabolomics. Due to methodological constraints RNAseq have a propensity to display erroneously high or low expression. Moreover, and perhaps even more important, dissociated cells in scRNAseq studies undergo rapid gene expression changes that can significantly falsify the image obtained (Rajala et al., PNAS Nexus 2:1-12, 2023). Finally, it cannot be emphasized enough that mRNA expression profiles DO NOT equate protein expression and there are numerous examples for divergent expression profiles when mRNA and protein is compared.

The Daniele et al. study (FASEB Journal 36:e22428, 2022; PMCID: PMC9438481) used in situ hybridization to study the mRNA expression of GLUT1 (slc2a1) and GLUT3 (slc2a3). In line with our comment just above, the Daniele et al. study may provide for an example of divergence between mRNA and protein expression, since it seemingly showed only minor expression of GLUT1/slc2a1 in the RPE, i.e. precisely in the one cell type that is well-known for its very strong GLUT1 protein expression.

Furthermore, Daniele et al. used a conditional GLUT1 knock-out in photoreceptors induced by repeated Tamoxifen injections. The photoreceptor GLUT1 knock-out led to a relatively mild phenotype with only about 45% of the outer nuclear layer lost over a 4-months time-course. This is in stark contrast with the FCCP or the 1,9-DDF treatment, which would ablate nearly all rod photoreceptors in under one or two weeks, respectively.

As a side note, Tamoxifen is an oestrogen receptor antagonist (with partial agonistic behaviour) with a long history of causing retinal and photoreceptor damage. Notably, oestrogen receptor signalling is important for maintaining photoreceptor viability (Nixon & Simpkins, IOVS 53:4739-47, 2012; Xiong et al., Neuroscience 452:280-294, 2021). Therefore, the relatively minor effects of the conditional GLUT1 KO in photoreceptors found in Daniele et al. may have been confounded by direct tamoxifen photoreceptor toxicity. On a wider level, this possible confounding factor related to the use of Tamoxifen points to general problems associated with certain forms of genetic manipulations.

We now mention the controversy around the expression of glucose transporters in the retina, including the Daniele et al. study in a new subchapter of the discussion on "Expression of glucose transporters in the outer retina” (Lines 454-472).

Rev#3.14. Lines 370-372: FCCP caused a strong cell death phenotype in rods, however under stress rods upregulate the secretion of RdCVF, which leads to cone photoreceptor survival by the upregulation of aerobic glycolysis in cones. The data should be re-interpreted in the context of this previous literature.

Response: We thank the reviewer for this comment; however, we could not find a reference that would state that “…under stress rods upregulate the secretion of RdCVF”. What we did find was a reference stating that similar factors such as thioredoxins (TRX80) are secreted from blood monocytes under stress (Sahaf & Rosén, Antioxid Redox Signal 2:717-26, 2000). However, we consider these cells to be too dissimilar to rod photoreceptors to warrant a corresponding comment. Moreover, the research group who discovered RdCVF originally showed that rod-secreted RdCVF cannot prevent cone degeneration if the corresponding Nxnl1 gene is knocked-out in cones, arguing for a cell-autonomous mechanism of RdCVF -dependent cone protection (Mei et al., Antioxid Redox Signal. 24:909-23, 2016).

Since it is very possible that we may have missed the correct reference(s), we would welcome further guidance by the reviewer.

Rev#3.15. Line 374: 1,9-DDF caused a 90% loss of cones, however, previous studies by Dr. Nancy Philp have shown glucose deprivation in the outer retina affects primarily rod photoreceptors. The differences should be discussed.

Response: We thank the reviewer for directing us to these studies. As mentioned above (Rev#3.13.) the Daniele et al. 2022 study yielded only relatively mild effects for a rod-specific conditional GLUT1 KO on photoreceptor viability. Similarly, in an earlier study (Swarup et al., Am J Physiol Cell Physiol. 316: C121–C133, 2019) the Philp group found that also a GLUT1 KO in the RPE caused only a minor phenotype in the photoreceptor layer. We would argue that if glucose, and by extension aerobic glycolysis, were indeed of major importance for (rod) photoreceptor survival, the degenerative effect of these genetic GLUT1 ablations should have been devastating and should have destroyed most of the outer retina in a matter of days. The fact that this was not seen in both studies is another piece of independent evidence that rod photoreceptors do not rely to any major extent on glycolytic metabolism.

The two studies from the Philp lab (Swarup et al., 2019; Daniele et al., 2022) are now cited in the discussion (Lines 417-419 and 458-460).

Rev#3.16. Line 375: Yes Dr. Claudio Punzo and Dr. Leveillard Thierry along with other groups have shown glycolysis is required to maintain cone survival when under stress, however, the authors should emphasize that it is under stress that this is observed.

Response: In response to this comment we have now specifically extended our corresponding remark in the discussion of the manuscript (Lines 446-447).

Rev#3.17. The section "Cone photoreceptors use the Cahill-cycle". The presence of ALT in photoreceptors was surprising and suggests alternatives to the Cori reaction. However, previous measurements of glucose and lactate from localized in vivo cannulation of animal eyes suggest the majority of glucose taken up by the retina is released back to the blood as lactate. Again, this section should discuss this idea in terms of the previous literature.

Response: Here, we believe the reviewer is referring to studies performed in the late 1990s where, in anaesthetized cats, the lactate concentration in blood samples obtained from choroidal vein cannulation was compared against that in blood samples obtained from femoral arteries (Wang et al., IOVS 38:48-55, 1997). We note that a more relevant in vivo measurement of retinal glucose consumption and lactate production would likely require the simultaneous cannulation of the central retinal artery (CRA) and the central retinal vein (CRV). This would need to be combined with repeated (online) blood sampling, drug applications, and subsequent metabolomic analysis. We are not aware of any in vivo studies where such procedures have been successfully performed and further miniaturization and increased sensitivity of metabolomic analytic equipment will likely be required before such an undertaking may become feasible. Even so, such studies may not be feasible in small rodents (mice, rats) and may instead require larger animal species (e.g. dog, monkey) to overcome limitations in eye and blood sample size.

We have now extended the discussion of our manuscript with a new subchapter on “The retina as an experimental system for studies into neuronal energy metabolism”. Within this new subchapter we now present two different in vivo experimental approaches that addressed retinal energy metabolism (Lines 376-384). Moreover, we now present new data on retinal lactate release to the culture medium, showing, for instance, a strong increase in lactate release in the no RPE condition compared to control (new Supplementary Figure 4).

Rev#3.18. Lines 431-433: The study cited suggested that the mitochondrial AST was detected in other cells, in agreement with the data shown. However, the authors' statements in this section are misleading as they do not take into consideration the contribution of AST from other cell types.

Response: The reviewer is right, we found both AST1 and AST2 to be expressed not only in photoreceptor inner segments but also in the inner retina, especially in the inner plexiform layer (new Figure 6D). Since this might indicate mini-Krebs-cycle activity also in retinal synapses, we have added a corresponding comment to the discussion (Lines 540-543).

Grammatical and wording fixes:

Rev#3.19. Line 98 - "the recycling of the photopigment, retinal."

Response: We have inserted a comma after “photopigment”.

Rev#3.20. Results section and Figure 1 start without providing context for the model system where staining is being done.

Response: We have added this information to the beginning of the results section (Lines 105-106).

Rev#3.21. Supplementary Figure 2 is not mentioned in the main text - there is no context for this figure.

Response: Supplementary Figure 2 was originally referenced in the legend to Figure 2. We now mention supplementary figure 2 (now renumbered to supplementary figure S3) also in the main text, in the results section under “Experimental retinal interventions produce characteristic metabolomic patterns” (Line 148).

Rev#3.22. Volcano plot in Supplementary Figures 3, 5, 6, 7, and 8 don't indicate what Log2(FC) is in reference to.

Response: The log2 fold change (FC) is calculated as follows: log2 (fold change) = log2 (mean metabolite concentration in condition A) - log2 (mean metabolite concentration in condition B) where condition A and condition B are two different experimental groups being compared. This is now explained in the SI Materials and Methods (Lines 145-147) and indicated in abbreviated form in the figure legends. Please note that supplemental figures have now been renumbered due to the insertion of an additional, new Figure.

Rev#3.23. Line 331 - –a“d allowed to analyze the..." ”s incorrect phrasing.

Response: This phrasing was changed.

Rev#3.24. Line 343 "c“cled" ”

Response: This phrasing was changed.

Rev#3.25. Line 446 is misworded.

Response: This phrasing was changed.

Technical questions:

Rev#3.26. At what point after explant was the IHC done in Supplemental Figure 1? If early, but experiments are done later, there's’a chance things are more disorganized at the end of the experiment.

Response: Staining and metabolomics analysis were both done at the end of each experiment, at the same time, at P15. This is now mentioned in the SI materials and methods section (Lines 67, 108-110).

Rev#3.27. FCCP affects plasma membrane permeability, which is particularly critical in neurons that undergo repolarization and depolarization - –ow do we know FCCP on cell death via metabolism? See: https://www.sciencedirect.com/science/article/pii/S2212877813001233

Response: The reviewer is correct, a significant permeabilization of cell membranes in general would likely cause extensive neuronal cell death, unrelated to a disruption of OXPHOS. However, the FCCP concentration used here (5 µM) is at the lower end of what was used in the mentioned Kenwood et al. study (Mol Metab. 3:114-123, 2014) and the effect on cell membrane permeability in tissue culture is likely to be rather small, as opposed to what was seen by Kenwood et al. in cultures of individual cells. This view is supported by the fact that in our FCCP treatments, we did not observe any significant increases of cell death in any retinal cell type (including RPE) other than in rod photoreceptors. Together with the fact that only photoreceptors strongly express Krebs-cycle/OXPHOS related enzymes, this strongly suggests that the FCCP effects seen by us were due to disruption of OXPHOS.

Rev#3.28. Numerous metabolite comparisons are being made throughout the manuscript – what type of multiple hypothesis testing corrections are utilized? Only certain figures mention multiple hypothesis testing (e.g. Figure 6).

Response: In general, in this manuscript we used two different statistical methods: 1) For two-group comparisons, we used an unpaired, two-tailed t-test, which reports a p-value with 95% confidence interval without additional multiple hypothesis testing (e.g. in Figure 2, Suppl. Figures 4, 6, 7, 8). 2) For multiple group comparisons we used a one-way ANOVA analysis with Tukey’s multiple comparisons post-hoc test (except suppl. Figure 9 where Fisher´s LSD post-hoc test was used). The information on which statistical test was used for what dataset is now given in the figure legends and in the SI Material and Methods section.

Rev#3.29. For Figure 3, how do we know that the removal of RPE is causing the metabolite changes due to RPE-PR coupling? How do you rule out the fact that it isn’t just: I – a thicker physical barrier between media and the neural retina that is causing the changes, or II – removal of RPE from PR causes OS shearing and a stress response that alters metabolism?

Response: We believe these concerns can be ruled out: The RPE cells are linked by tight junctions and are not “just a thicker barrier” but a barrier that is almost impermeable for most metabolites unless they are carried by specific transporters. Outer segment shearing via RPE removal would indeed be a concern if we had used adult retina. However, we explanted that retina at P9 when it does not possess any sizeable outer segments yet. As a matter of fact, photoreceptors grow out outer segments only after P9.

Rev#3.30. While 1,9-dideoxyforskolin blocks GLUT1, it is known to have other effects, including on potassium channels. How do we know the effects of 1,9-dideoxyforskolin are specific to GLUT1? Utilizing a GLUT1 KO and showing no additional effects when adding 1,9-dideoxyforskolin would be helpful as a control.

Response: This is a good suggestion from the reviewer. We note that this is technically not easy to achieve as it would require an RPE-specific knock-out that should be inducible at a given experimental time-point, in a quantitative manner. The study by Swarup et al. (see above Rev#3.13.) used an RPE specific knock-out that was, however, not inducible. Moreover, if the corresponding inducible knock-out animals could be generated, then the stochastic nature of the inducing treatment would probably affect only a limited number of cells within a given cell population. In our experimental context, a less than quantitative knock-out would significantly complicate interpretation of results, even to the point that no additional insight might be gained.

Rev#3.31. The analysis in Figure 6, even with attempts to control drug treatments, is highly speculative. One really needs animals with predominately cones vs. predominately rods to do this analysis (e.g. with NRL mice).

Response: The reviewer is right, the analysis shown in Figure 6 was an explorative approach to try and deduce features of rod and cone metabolism. This is now mentioned in the results section (Lines 282-284). Since the experiments were not initially intended to address such questions, by necessity the interpretations remain speculative. The comparison of mouse mutants in which there are either no cones (e.g. cpfl1 mouse) or no rods (e.g. NRL knock-out mouse) may allow to disentangle the metabolic contributions of rods and cones. We appreciate the suggestion from the reviewer and have now inserted a relating suggestion for future studies into the discussion section (Lines 450-452).

Rev#3.32. Overall, much of the paper suggests intriguing pathways, but without C13 tracing or relevant genetic knock-outs, the pathways would have to be speculative rather than definitive.

Response: We agree with the reviewer that further research, including 13C and 15N-tracing studies, will be necessary to evaluate which pathway(s) are used by what retinal cell type under what condition. Still, the high robustness and quantitative nature of the NMR metabolomics data allows us to draw pathway conclusions based on metabolites that are unique to specific pathways/cell types or using ratios. We now relate to the advantages of such carbon-tracing studies in the discussion of the manuscript (Lines 545-549).

Stylistic suggestions:

Rev#3.33. This is a very dense paper to read. It would be helpful for each figure to have a summary diagram of the relevant metabolite changes and how they fit together. Further, for those not metabolism-inclined, defining the mini-Kreb’s, Cahill, and Cori cycles and their brief implications at some point early in the manuscript would be helpful.

Response: We have been thinking a lot about how we could add in the suggested summary diagrams into each figure. Unfortunately, whatever idea we contemplated would have significantly increased the complexity of the figures, while the actual benefit in terms of improved understandability was unclear.

However, we did include the suggestion from the reviewer to present the terms Cori, Cahill-, and mini-Krebs-cycle already in the introduction and we hope that this has improved the understandability of the manuscript overall (Lines 79-92).

Rev#3.34. More discussion about the step-by-step ways that the mini-Kreb’s reaction “uncouples” glycolysis from the Kreb’s cycle would be helpful. What do you mean by “uncouple” in this context?

Response: We thank the reviewer for this suggestion. Uncoupling in this context means that glycolysis and Krebs cycle are not metabolically coupled to each other via pyruvate. Instead both pathways can run independently from each other and in parallel, as long as the Krebs-cycle uses glutamate, BCAAs or other amino acids as fuels. We now also address this point already in the introduction of the manuscript (Lines 87-90).

Conceptual questions:

Rev#3.35. As the proposal that PR undergo heavy amounts of OXPHOS is controversial, it would be helpful for the authors to review the literature on lactate production by the retina and what studies have shown previously about retina use of lactate, specifically lactate making its way into TCA cycle intermediates, suggesting OXPHOS, in PRs.

Response: In response to this question we have added several new references to the introduction and discussion of the manuscript. The question of lactate production (aerobic glycolysis) vs. the use of OXPHOS is now discussed in Lines 77-81, Lines 367-384.

Rev#3.36. Why would cones die more in the no RPE condition? The authors suggest this has something to do with GLUT1 expression on RPE and the transport of glucose to cones. Even if we accept that cones are highly glycolytic, loss of RPE should expose the neural retina to even more glucose in your experimental set-up.

Response: This is a very interesting question from the reviewer. Indeed, loss of the RPE and blood-retinal barrier function should increase photoreceptor access to glucose, even more so if they are expressing high affinity GLUT3. In the discussion (Lines 420-424), we speculate that this may trigger the Crabtree effect, shutting down OXPHOS and causing the cells to exclusively rely on glycolysis. This, however, will likely not yield sufficient ATP to maintain their viability, so that they “starve” to death even in the presence of ample glucose. Since cones require at least twice as much ATP as rods, they may be more sensitive to a Crabtree-dependent shut-down of OXPHOS. However, if this speculation was correct then the question remains why the FCCP treatment, which abolishes OXPHOS more directly, does not cause cone death. Here, we again can only speculate that high glucose may have additional toxic effects on cones that are independent of OXPHOS. We now try to present this reasoning in the discussion (Lines 426-429).

Author Response

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

We thank the reviewers and editors for their comments, as well as for the time dedicated to make useful suggestions that have contributed to improve the manuscript. We have responded to the concerns raised by the reviewers, and after that, we have also responded to the different points highlighted in the Recommendations for the authors:

Reviewer #1

While in vivo injury was used to assess regeneration from subsets of PNS neurons, different in vitro neurite growth or explant assays were used for further assessments. However, the authors did not assess whether the differential "regenerative" responses in vivo could be recapitulated in vitro. Such results will be important in interpreting the results.

We included a supplementary figure evaluating the neurite extension in vitro and updated the text accordingly.

Intriguingly, even in individual groups of PNS neurons, not all neurons regenerate to the same extent. It is known that the distance between the cell body and the lesion site affects neuronal injury responses. It would be interesting to test this in the observed regeneration.

Although it is true that the distance can affect the outcome, here we used a physiological model where all neurons are lesioned at the same point in the nerve. Not only distance is different for motoneurons, but also the microenvironment surrounding their somas and therefore the direct comparison of these neurons with sensory neurons is limited. We extended the discussion on this matter in the new manuscript.

Fig 1: The authors quantified the number of regenerating axons at two different time points. However, the total numbers of neurons/axons in each subset are different. The authors should use these numbers to normalize their regenerative axons.

Figure 1D shows the normalization of data from figure 1C (normalized against the number of control axons in each neuron type). This has been clarified in the text.

Fig 2-5: In explaining differential regeneration of individual groups of neurons, there are at least two possibilities: (1). Each group of neurons has different injury/regenerative responses; (2). The same set of injury/regenerative responses are differentially activated. Some data in this manuscript suggested the latter possibility. But some other data point in the opposite direction. It would be informative for the authors to analyze/discuss this further.

From our point of view, these two options can be considered differential response to injury and could be potentially used for the modulation of regeneration. However, if the second possibility is correct, the regenerative program could be more influenced by the time chosen to study the response. Given the importance of this, we added some discussion about this topic.

Fig 6: Is it possible to assess the regenerative effects of knockdown Med12 after in vivo injury?

It is possible, but it is out of the scope of this work. Here, we aimed to describe the regenerative response and validate our data by testing a potential target for specific regeneration. Future studies will focus on the modulation of this specific regeneration both in vitro and in vivo.

Reviewer #2

It seems that the most intriguing outcome of this paper revolves around the role of Med12 in nerve regeneration. The authors should prioritize this finding. Drawing a conclusion regarding Med12's role in proprioceptor regeneration based solely on this in vitro model may be insufficient. This noteworthy result requires further investigation using more animal models of nerve regeneration.

The main goal of this work was to compare the regenerative responses of different neuron subpopulations. We modulated Med12 to validate our data and the potential of our findings. Unfortunately, investigating in depth the role of Med12 in regeneration is out of the scope of this paper. For this reason, we did not prioritise this finding here. As this finding was striking, we strongly agree that the next step should be studying how it modulates regeneration.

One critique revolves around the authors' examination of only a single time point within the dynamic and continuously evolving process of regeneration/reinnervation. Given that this process is characterized by dynamic changes, some of which may not be directly associated with active axon growth during regeneration, and encompasses a wide range of molecular alterations throughout reinnervation, concentrating solely on a single time point could result in the omission of critical molecular events.

We agree that this is probably the main limitation of this study, as we discussed in the text. We chose 7 days postinjury as a standard time point widely described in literature and to have a correlate with our histological data. Although the main aim was to compare populations, analyzing an additional time point after injury could add valuable information.

Reviewer #3

No concerns were expressed by that reviewer.

Recommendations for the authors:

The authors should assess whether the differential "regenerative" responses in vivo could be recapitulated in vitro.

We included a supplementary figure evaluating the neurite extension in vitro and updated the text accordingly.

Optional:

It will be interesting to test if the distances between the cell body and the lesion site contribute to the observed differences in individual subsets of PNS neurons.

Figure 1D shows the normalization of data from figure 1C (normalized against the number of control axons in each neuron type). This has been clarified in the text.

Fig 2-5: In explaining differential regeneration of individual groups of neurons, there are at least two possibilities: (1). Each group of neurons has different injury/regenerative responses; (2). The same set of injury/regenerative responses are differentially activated. Some data in this manuscript suggested the latter possibility. But some other data point in the opposite direction. At least the authors should discuss these.

From our point of view, these two options can be considered differential response to injury and could be potentially used for the modulation of regeneration. However, if the second possibility is correct, the regenerative program could be more influenced by the time chosen to study the response. Given the importance of this, we added some discussion about this topic.

While the paper is technically well-executed, the conclusions and some of the findings appear to be incomplete and challenging to draw meaningful conclusions from. This manuscript presents some interesting findings, but the title is quite broad and may suggest that the authors have unveiled fundamental mechanisms explaining the varying regenerative abilities of peripheral axons. However, the results do not substantiate such a conclusion. Further comments and suggestions follow.

We eliminated the word “regenerative (response)” from the title, as it could lead to think that all changes seen in these neurons are related only to regeneration. We think that “Neuron-specific RNA-sequencing reveals different responses in peripheral neurons after nerve injury” highlights the differences between neurons that we found without misleading towards thinking that we described regenerative mechanisms in all neurons.

What's notably absent here is the validation of certain genes found with the ribosomes, especially those highlighted in the subsequent figures. The question arises as to whether the changes depicted in the figures align with changes in the DRGs in vivo. Is there concordance between the presence of these genes and their transcriptional changes? It would greatly enhance the study's value if the authors could show evidence of upregulation or downregulation of certain genes over time in tissue sections, utilizing techniques such as in situ hybridization or immunocytochemistry.

We selected some factors that were specifically upregulated in subsets of neurons to corroborated by immunohistochemistry these findings. Changes in the immunofluorescence of P75 in motoneurons and ATF2 in cutaneous mechanoreceptors, were evaluated in controls and animals that received a nerve crush one week before. Supplementary figures with the images have been added.

The authors discovered intriguing distinctions, such as the presence of specific signaling pathways unique to neurons projecting to muscle as opposed to those projecting to the skin. Among these pathways were those associated with receptor tyrosine kinases like VEGF, erbB, and neurotrophin signaling among others. The question now arises: do these pathways play a role in natural peripheral regeneration processes? To answer this, it is imperative to conduct in vivo studies. However, the authors employed an in vitro DRG neurite outgrowth assay to demonstrate that various types of neurons exhibit different responses to the presence of different neurotrophins. This does not reflect what actually happens in vivo. While neurotrophins indeed play a role in neuron survival and axon extension during development, their role in postnatal periods changes over time, and it remains unclear whether they play any role in the natural regenerative processes of the peripheral nerve. Therefore, this experiment may not be directly relevant in this case, especially during the early axon extension period of the regenerating axons. if the authors aim to establish a causal link with neurotrophin signaling, it becomes crucial to conduct in vivo experiments by manipulating the expression of key molecules like the receptors.

It has been widely described that different types of peripheral neurons have a differential expression of Trk receptors, even in the adult, and that these respond differentially to neurotrophins. In our study, we do not stablish a causal relationship between the expression of Trk and neurite extension, but instead we show (as many others) that distinct neurons respond differentially to these neurotrophins. The fact that in vivo studies fail to show a clear effect does not necessarily mean that neurotrophins are not specific. It might mean that their effect is not strong enough to be a useful guide in the complex microenvironment found after an injury. For instance, NGF acts on TrkA (present in some neurons), but in vivo it has been shown to accelerate the clearance of myelin debris in Schwann cells (Li et al., 2020), which could facilitate regeneration of all type of axons, masking any potential specific effect on the subtypes of neurons expressing TrkA. In contrast, in an in vitro setting on neuronal cultures, the specific neuronal effect can be more evident.

Additionally, it's worth noting that another paper utilizing the same methodology and experimental setup (PMID: 29756027, "Translatome Regulation in Neuronal Injury and Axon Regrowth" by Rozenbaum et al.) exists. Are there any significant differences or shared findings with that study?

This study shows the transcriptomic response after an injury 4, 12 and 24 hours after an injury in a very similar experimental setup. They focus on comparing the neuronal vs the glial response to the injury, using a Ribotag line that tags ribosomes from all neurons in the DRG rather than specific neuron subtypes. As the time postinjury (24h vs 7 days) and the cell types studied are different, we could not directly compare our results. We did see an upregulation in both datasets of previously described growth-associated genes (Jun, Atf3, Sox11, Sprr1a, Gal…). We included the article in the references for its relevance in the topic.

It would be helpful for readers to illustrate the finding of the fastest axon regeneration of nociceptors by showing fluorescence micrographs of the nerve samples in addition to the graphs shown in Fig. 1 C/D.

In figure 1B, we show fluorescence micrographs of the nerves 7 days postinjury. As explained in the results, we counted the number of axons at 2 distances from the injury, we did not analyse the fastest axon. This is due to technical reasons: 7 days after the injury the fastest axon has surpassed our evaluation point, which was the further distance that we could assess in our experimental setting in a consistent manner. If the reviewer thinks that we need to include more images from our evaluations (from 9 dpi for example), we could prepare a new figure.

The labeling in Fig. 2B is confusing. Is the CHAT immunoreactivity shown in the last panel illustrated by green or red signals? Is the red signal counterstaining with beta-tubulin?

The labelling was changed in the figure to increase clarity.

The references to the supplementary data throughout the manuscript are confusing. For example, where can the "Supp data 2" be found? (mention on p. 14 in the merged pdf file). Are they referring to the Excel spreadsheets?

We divided the supplementary material in supplementary figures/table (found in the pdf) and supplementary data. Supplementary data refers to excel spreadsheets found outside the pdf file. We hope this will be clearer after the final formatting of the article.

What does the following statement on p. 14 mean?: "The caveat in these analyses was that molecular classification by these approaches may be arbitrary, and not reflective of protein repurposing." This reviewer notes that these databases consider the fact that components participate in different pathways.

Indeed, we aimed to explain that many proteins participate in different pathways, and this is a limitation of the enrichment analysis. We modified the sentence in the text.

First paragraph on p. 15: The PPAR and AMPK pathways have much broader roles, and are not only "related to fatty acid metabolism". This factual inaccuracy should be corrected in the manuscript.

The sentence has been corrected.

The authors should consider showing increased TGF-beta signaling in their neurons after downregulation of Med12 given the previous implication of TGF-beta signaling in axon regeneration.

We tried to demonstrate the effect of our knockdown in TGF-beta pathway by analyzing the expression of typical targets from this pathway by qPCR in our cultures. However, we could not detect any difference. We think that this can have two explanations: (1) as only a few cells upregulate Med12 whereas many cells downregulate it, the effect is masked (presumably only proprioceptors will have a significant difference in this pathway and, thus, it would be very difficult to see the effect), or (2) Med12 is not exerting its effect through this pathway. We added a supplementary figure with these data and discussed it in the manuscript.

It would be helpful to eliminate typos and improve syntax/grammar/style.

We revised the text to improve style.

Author Response

Public Reviews:

Reviewer #1

Strengths:

Overall, the work is novel and moves the field of Alzheimer's disease forward in a significant way. The manuscript reports a novel concept of aberrant activity in VIP interneurons during the early stages of AD thus contributing to dysfunctions of the CA1 microcircuit. This results in the enhancement of the inhibitory tone on the primary cells of CA1. Thus, the disinhibition by VIP interneurons of Principal Cells is dampened. The manuscript was skillfully composed, and the study was of strong scientific rigor featuring well-designed experiments. Necessary controls were present. Both sexes were included.

We express our gratitude to the reviewer for their keen appreciation of our efforts and their enthusiasm for the outcomes of this research.

Limitations:

(1) The authors attributed aberrant circuit activity to the accumulation of "Abeta intracellularly" inside IS-3 cells. That is problematic. 6E10 antibody recognizes amyloid plaques in addition to Amyloid Precursor Protein (APP) as well as the C99 fragment. There are no plaques at the ages 3xTg mice were examined. Thus, the staining shown in Figure 1a is of APP/C99 inside neurons, not abeta accumulations in neurons. At the ages of 3-6 months, 3xTg starts producing abeta oligomers and potentially tau oligomers as well (Takeda et al., 2013 PMID: 23640054; Takeda et al., 2015 PMID: 26458742 and others). Emerging literature suggests that abeta and tau oligomers disrupt circuit function. Thus, a more likely explanation of abeta and tau oligomers disrupting the activity of VIP neurons is plausible.

The Reviewer correctly points out that 3xTg-AD mice typically do not exhibit plaques before 6 months of age, with limited amounts even up to 12 months, particularly in the hippocampus. To the best of our knowledge, the 6E10 antibody binds to an epitope in APP (682-687) that is also present in the Abeta (3-8) peptide. Consequently, 6E10 detects full-length APP, α-APP (soluble alpha-secretase-cleaved APP), and Abeta (LaFerla et al., 2007). Nonetheless, we concur with the Reviewer's observation that the detected signal includes Abeta oligomers and the C99 fragment, which is currently considered an early marker of AD pathology (Takasugi et al., 2023; Tanuma et al., 2023). Studies have demonstrated intracellular accumulation of C99 in 3-month-old 3xTg mice (Lauritzen et al., 2012), and its binding to the Kv7 potassium channel family, which results in inhibiting their activity (Manville and Abbott, 2021). If a similar mechanism operates in IS-3 cells, it could explain the changes in their firing properties observed in our study. Consequently, we will revise the manuscript to include this crucial information in both the Results and Discussion sections.

(2) Authors suggest that their animals do not exhibit loss of synaptic connections and show Figure 3d in support of that suggestion. However, imaging with confocal microscopy of 70micron thick sections would not allow the resolution of pre- and post-synaptic terminals. More sensitive measures such as electron microscopy or array tomography are the appropriate techniques to pursue. It is important for the authors to either remove that data from the manuscript or address the limitations of their technique in the discussion section. There is a possibility of loss of synaptic connections in their mouse model at the ages examined.

We appreciate the Reviewer’s perspective on the techniques used for imaging synaptic connections. While we acknowledge the limitations of confocal microscopy for resolving pre- and post-synaptic structures in thick sections, we respectfully disagree regarding the exclusive suitability of electron microscopy (EM). Our approach involved confocal 3D image acquisition using a 63x objective at 0.2 um lateral resolution and 0.25 Z-step, providing valuable quantitative insights into synaptic bouton density. Despite the challenges posed by thick sections, this method together with automatic analysis allows for careful quantification. Although EM offers unparalleled resolution, it presents challenges in quantification. We will ensure to include the important details regarding image acquisition and analysis in the revised manuscript.

Reviewer #2 (Public Review):

Summary:

The submitted manuscript by Michaud and Francavilla et al., is a very interesting study describing early disruptions in the disinhibitory modulation exerted by VIP+ interneurons in CA1, in a triple transgenic model of Alzheimer's disease. They provide a comprehensive analysis at the cellular, synaptic, network, and behavioral level on how these changes correlate and might be related to behavioral impairments during these early stages of the disease.

Main findings:

3xTg mice show early Aß accumulation in VIP-positive interneurons.

3xTg mice show deficits in a spatially modified version of the novel object recognition test. - 3xTg mice VIP cells present slower action potentials and diminished firing frequency upon current injection.

3xTg mice show diminished spontaneous IPSC frequency with slower kinetics in Oriens / Alveus interneurons.

3xTg mice show increased O/A interneuron activity during specific behavioral conditions.

3xTg mice show decreased pyramidal cell activity during specific behavioral conditions.

Strengths:

This study is very important for understanding the pathophysiology of Alzheimer´s disease and the crucial role of interneurons in the hippocampus in healthy and pathological conditions.

We are thankful to the reviewer for their insightful recognition of our efforts and their enthusiasm for the results of this research.

Weaknesses:

Although results nicely suggest that deficits in VIP physiological properties are related to the differences in network activity, there is no demonstration of causality.

RE: We completely agree with the reviewer's observation regarding the lack of demonstration of causality in our results. Investigating causality in the relationship between deficits in VIP physiological properties and differences in network activity is indeed a crucial aspect of this project. However, achieving this goal will require a significant amount of time and dedicated manipulations in a new mouse model (VIP-Cre-3xTg). We appreciate the importance of this line of investigation and consider it as a priority for our future research endeavors.

Author Response

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

We are grateful to the reviewers for their constructive comments. The following is our point-to-point responses.

Reviewer #1 (Recommendations For The Authors):

Point 1- Abstract: advanced morning peak « opposite » to pdf/pdfr mutants. To my knowledge, the alteration of PDF/PDFR suppresses the morning peak. I am not sure that an advance of the peak is « opposite » to its inhibition?

Mutants with disruptions in CNMa or CNMaR display advanced morning activity, indicating an enhanced state. Mutants with disruptions in Pdf or Pdfr exhibit no morning anticipation, suggesting a promoting role of these genes in morning anticipation. Therefore, our revised version is: “Specific elimination of each from clock neurons revealed that loss of the neuropeptide CNMa in two posterior dorsal clock neurons (DN1ps) or its receptor (CNMaR) caused advanced morning activity, indicating a suppressive role of CNMa-CNMaR on morning anticipation, opposite to the promoting role of PDF-PDFR on morning anticipation.” (Line 43-51)

Point 2- Fig 1K-L: the authors should show the sleep phenotype of the homozygous nAChRbeta2 mutant (if not lethal) for a direct comparison with the FRT/FLP genotype and thus evaluate the efficiency of the system.

We have incorporated sleep profiles of nAChRbeta2 mutant and W1118 into Fig 1K-L. nAChRbeta2 mutants (red) exhibited a sleep amount comparable to that of pan-neural nAChRbeta2 knockout flies (dark red), as shown below.

Author response image 1.

Point 3- Dh31-EGFP-FRT expression patterns look different in figS1 A (or fig1 H) and J. why that?

We re-examined the original data. Both (with R57C10-GAL4 for Fig. S1A, right, S1J, left) are Dh31EGFP.FRT samples displayed below which demonstrated consistent primary expression subsets. Any observed disparities in region "e" could potentially be attributed to variations during dissection.

Author response image 2.

Point 4- The knockdown experiments with the elav-switch (RU486) system (fig S2) do not seem to be as efficient as the HS-FLP system (fig 1H-J). The conclusions on the efficiency should be toned down.

We have revised accordingly: "Near Complete Disruption of Target Genes by GFPi and Flp-out Based cCCTomics" (Line 130): "Knocking out at the adult stage using either hsFLP driven Flp-out (Golic and Lindquist, 1989) (Fig. 1H-1J) or neural (elav-Switch) driven shRNAGFP (Nicholson et al., 2008; Osterwalder et al., 2001) (Fig. S2A-S2I), also resulted in the elimination of most, though not all, GFP signals." (Line 145-149)

Point 5- Fig 2H-J: the LD behavioral phenotype of pdfr pan-neuronal cripsr does not seem to correspond to what is described in the literature for the pdfr mutant (han), see hyun et al 2005 (no morning anticipation and advanced evening peak). I understand that the activity index is lower than controls but fig2H shows a large anticipatory activity that seems really unusual, and no advanced evening peak is observed. I think that the authors should show the CRISPR flies and pdfr mutants together, to better compare the phenotypes.

Thank you for pointing out that the phenotypes of pan-neuronal knockout of PDFR by unmodified Cas9 (Fig. 2H-2I of the previous version) whose morning anticipation still exist (Fig, 2H of the previous manuscript), although the significant decrease of morning anticipation index (Fig 2I of the previous manuscript) and advanced evening activity are not as pronounced as observed in han5304 (Fig. 3C in Hyun et al., 2005).

First, we have separated the activity plots of Fig. 2H of previous manuscript, as shown below. The activity from ZT18 to ZT24 shows a tendency of decreasing from ZT18 to ZT21 and a tendency of increasing from ZT21 to ZT24. The lowest activity before dawn during ZT18 to ZT24 shows at about ZT21, and the activity at ZT18 is comparable to the activity at ZT24. This is significantly different compared to the two control groups, whose activity tends to increase activity from ZT18 to ZT24 with an activity peak at ZT24.

The activity from ZT6 to ZT12 increased much faster in Pdfr knockout flies and get to an activity plateau at about ZT11 compared to two control groups with a slower activity increasing from ZT6 to ZT12 with no activity plateau but an activity peak at ZT12.

Author response image 3.

Second, we have incorporated the phenotype of Pdfr mutants we previously generated (Pdfr-attpKO Deng et al., 2019) with Pdfr pan-neuronal knockout by Cas9.HC. This mutant lacks all seven transmembrane regions of Pdfr (a). The phenotypes are very similar between Pdfr-attpKO flies and Pdfr pan-neuronal knockout flies. In this experimental repeat, we found that a much more obvious advanced evening activity peak is observed both in pan-neuronal knockout flies and Pdfr-attpKO flies.

To further analyze the phenotypes of Pdfr pan-neuronal knockout flies by Cas9.HC, we referred to the literature. The activity pattern at ZT18 to ZT24 (activity tends to decrease from ZT18 to ZT21 and tends to increase from ZT21 to ZT24, with the lowest activity before dawn occurring at about ZT21, and activity at ZT18 comparable to activity at ZT24) is also reported in Pdfr knockout flies such as Fig3C and 3H in Hyun et al., 2005, Fig 2B in Lear et al., 2009, Fig 3B in Zhang et al., 2010, Fig .5A in Guo et al., 2014, and Fig 5B in Goda et al., 2019. Additionally, the less pronounced advanced evening activity peak compared to han5304 (Fig. 3C in Hyun et al., 2005) is also reported in Fig. 2B in Lear et al., 2009, Fig. 3B in Zhang et al., 2010, and Fig. 5B in Goda et al., 2019. We consider that this difference is more likely to be caused by environmental conditions or recording strategies (DAM system vs. video tracing).

Therefore, we revised the text to: “Pan-neuronal knockout of Pdfr resulted in a tendency towards advanced evening activity and weaker morning anticipation compared to control flies (Fig. 2H-2I), which is similar to Pdfr-attpKO flies. These phenotypes were not as pronounced as those reported previously, when han5304 mutants exhibited a more obvious advanced evening peak and no morning anticipation (Hyun et al., 2005)”.

Author response image 4.

Point 6-The authors should provide more information about the DD behavior (power is low, but how about the period of rhythmic flies, which is shortened in pdf (renn et al) and pdfr (hyun et al) mutants).

We have incorporated period data into Fig. 2I. Indeed, conditional knock out of Pdfr by Cas9.HC driven by R57C10-GAL4 shortens the period length, as shown below (previous data), also in Fig. 2I of the revised version.

In the revised Fig. 2I, we tested 45 Pdfr-attpKO flies during DD condition (3 out of 48 flies died during video tracing in DD condition), and only one fly was rhythmic. In contrast, 9 out of 48 Pdfr pan-neuronal knockout flies were rhythmic.

Author response image 5.

Point 7- P15 and fig6. The authors indicate that type II CNMa neurons do not show advanced morning activity as type I do, but Figs 6 I and K seem to show some advance although less important than type I. I am not sure that this supports the claim that type I is the main subset for the control of morning activity. This should be toned down.

We have re-organized Fig. 6 and revised the summary of these results as: “However, Type II neurons-specific CNMa knockout (CNMa ∩ GMR91F02) showed weaker advanced morning activity without advanced morning peak (Fig. 6N), while Type I neurons-specific CNMa knockout did (Fig. 6J), indicating a possibility that these two type I CNMa neurons constitute the main functional subset regulating the morning anticipation activity of fruit fly”. (Line 400-405)

Point 8- Figs 6M and N: is power determined from DD data? if yes, how about the period and arrhythmicity? Please also provide the LD activity profiles for the mutants and rescued pdfr genotypes.

Yes, the power was determined from the DD data. In the new version of the manuscript, we have included the activity plots for the LD phase in supplementary Fig S13, as well as shown below (A, B), and the period and arrhythmicity data for the DD phase in Fig. 6S and Table S7. We have also refined the related description as follows: “Moreover, knocking out Pdfr by GMR51H05, GMR79A11 and CNMa GAL4, which cover type I CNMa neurons, decreased morning anticipation of flies (Fig. 6T, Fig. S13B). However, the decrease in morning anticipation observed in the Pdfr knockout by CNMa-GAL4 was not as pronounced as with the other two drivers. Because the presumptive main subset of functional CNMa is also PDFR-positive, there is a possibility that CNMa secretion is regulated by PDF/PDFR signal”. (Line 413-419)

Author response image 6.

Point 9- Fig 7: does CNMaR affect DD behavior? This should be tested.

We analyzed the CNMaR-/- activity in the dark-dark condition over a span of six days. Results revealed a higher power in CNMaR mutants compared to control flies (Power: 93.5±41.9 (CNMaR-/-, n=48) vs 47.3±31.6 (w1118, n=47); Period: 23.7±0.3 h (CNMaR-/-, n=46) vs 23.7±0.3 h (w1118, n=47); arrhythmic rate 2/48 (CNMaR-/-) vs 0/47 (w1118)). Considering that mutating CNMa had no obvious effect on DD behavior, even if CNMaR affects DD behavior, it cannot be attributed to CNMa signal, we did not further repeat and analyze DD behavior of CNMaR mutant. We believe this raises another question beyond the scope of our current discussion.

Reviewer #2 (Recommendations For The Authors):

Point 1-One major concern is the apparent discrepancies in clock network gene expression using the Flp-Out and split-LexA approaches compared to what is known about the expression of several transmitter and peptide-related genes. For example, it is well established that the 5th-sLNv expresses CHAT (along with a single LNd), yet there appears to be no choline acetyltransferase (ChAT) signal in the 5th-sLNv as assayed by the Split-LexA approach (Fig. 4). This approach also suggests that DH31 is expressed in the s-LNvs, which, as one of the most intensely studied clock neuron are known to express PDF and sNPF, but not DH31. The results also suggest that the sLNvs express ChAT, which they do not. Remarkably PDF is not included in the expression analysis, this peptide is well known to be expressed in only two subgroups of clock neurons, and would therefore be an excellent test case for the expression analysis in Fig. 4. PDF should therefore be added to analysis shown in Fig. 4. Another discrepancy is PdfR, which split LexA suggests is expressed in the Large LNvs but not the small LNvs, the opposite of what has been shown using both reporter expression and physiology. The authors do acknowledge that discrepancies exist between their data and previous work on expression within the clock network (lines 237 and 238). However, the extent of these discrepancies is not made clear and calls into question the accuracy of Flp-Out and Split LexA approaches.

The concerns mentioned above are:

(1) sLNvs express PDF and sNPF but not Dh31;

(2) ChAT presents in 5th-sLNv and one LNd but not in other sLNvs;

(3) PDFR presents in sLNvs but not l-LNvs.

(4) PDF is not included in the analysis.

To verify the accuracy of these intersection analyses, all related to PDF positive neurons (except 5th-sLNv and LNds), we stained PDF and examined the co-localization between PDF-positive LNvs and the respective drivers ChAT-KI-LexA, Pdfr-KI -LexA, Dh31-KI -LexA, and Pdf-KI -LexA.

First, Dh31-KI-LexA labeled four s-LNvs, as shown below (also in Fig. S9A). Therefore, the results of the intersection analysis of Dh31-KI-LexA with Clk856-GAL4 are correct. The difference in the results compared to previous literature is attributed to Dh31-KI-LexA labels different neurons than the previous driver or antibody.

Second, no s-LNv was labeled by ChAT-KI -LexA as shown below. We rechecked our intersection data and found that we analyzed 10 brains of ChAT-KI-LexA∩Clk856-GAL4 while only two brains showed sLNvs positively. To enhance the accuracy of intersection analysis results, we marked all positive signal records when positive subsets were found in less than 1/3 of the total analyzed brains (Table S4).

Third, one l-LNv and at least two s-LNvs were labeled by Pdfr-KI-LexA, as shown below (also in Fig. S9B). Fourth, Pdf-KI-LexA labels all PDF-positive neurons, but the intersection analysis by Pdf-KI-LexA and Clk856-GAL4 only showed scattered signals, as shown below (D, also in Fig. S9C). For these cases, we found some positive signals expected but not observed in our dissection. The possible reason could be the inefficiency of LexAop-FRT-myr::GFP driven by LexA. Therefore, our intersection results must miss some positive signals.

Author response image 7.

Finally, we revised the text to (Line 286-317):

To assess the accuracy of expression profiles using CCT drivers, we compared our dissection results with previous reports. Initially, we confirmed the expression of CCHa1 in two DN1s (Fujiwara et al., 2018), sNFP in four s-LNvs and two LNds(Johard et al., 2009), and Trissin in two LNds (Ma et al., 2021), aligning with previous findings. Additionally, we identified the expression of nAChRα1, nAChRα2, nAChRβ2, GABA-B-R2, CCHa1-R, and Dh31-R in all or subsets of LNvs, consistent with suggestions from studies using ligands or agonists in LNvs (Duhart et al., 2020; Fujiwara et al., 2018; Lelito and Shafer, 2012; Shafer et al., 2008) (Table S4).

Regarding previously reported Nplp1 in two DN1as (Shafer et al., 2006), we found approximately five DN1s positive for Nplp-KI-LexA, indicating a broader expression than previously reported. A similar pattern emerged in our analysis of Dh31-KI-LexA, where four DN1s, four s-LNvs, and two LNds were identified, contrasting with the two DN1s found in immunocytochemical analysis (Goda et al., 2016). Colocalization analysis of Dh31-KI-LexA and anti-PDF revealed labeling of all PDF-positive s-LNvs but not l-LNvs (Fig S9A), suggesting that the differences may arise from the broader labeling of 3' end knock-in LexA drivers or the amplitude effect of the binary expression system. The low protein levels might go undetected in immunocytochemical analysis. This aligns with transcriptome analysis findings showing Nplp1 positive in DN1as, a cluster of CNMa-positive DN1ps, and a cluster of DN3s (Ma et al., 2021), which is more consistent with our dissection.

Despite the well-known expression of PDF in LNvs and PDFR in s-LNvs (Renn et al., 1999; Shafer et al., 2008), we did not observe stable positive signals for both in Flp-out intersection experiments, although both Pdf-KI-LexA and Pdfr-KI-LexA label LNvs as expected (Fig S9B-S9C). We also noted fewer positive neurons in certain clock neuron subsets compared to previous reports, such as NPF in three LNds and some LNvs (Erion et al., 2016; He et al., 2013; Hermann et al., 2012; Johard et al., 2009; Lee et al., 2006) and ChAT in four LNds and the 5th s-LNv (Johard et al., 2009; Duhart et al., 2020) (Table S4). We attribute this limitation to the inefficiency of LexAop-FRT-myr::GFP driven by LexA, acknowledging that our intersection results may miss some positive signals.

Point 2-Related to this, the authors rather inaccurately suggest that the field's understanding of PdfR expression within the clock neuron network is "inconsistent" and "variable" (lines 368-377). This is not accurate. It is true that the first attempts to map PdfR expression with antisera and GAL4s were inaccurate. However, subsequent work by several groups has produced strong convergent evidence that with the exception of the l-LNvs after several days post-eclosion, PdfR is expressed in the Cryptochrome expressing a subset of the clock neuron network. This section of the study should be revised.

We thank the reviewer for pointing this out. As we have already addressed and revised the related part in the RESULTS section (Line 308-317), we have now removed this part from the DISCUSSION section of the revised version.

Point 3-One minor issue that would avoid unnecessary confusion by readers familiar with the circadian literature is the say that activity profiles are plotted in the study. The authors have centered their averaged activity profiles on the 12h of darkness. This is the opposite of the practice of the field, and it leads to some initial confusion in the examination of the morning and evening peak data. The authors may wish to avoid this by centering their activity plots on the 12h light phase, which would put the morning peak on the left and the evening peak on the right. This is the way the field is accustomed to examining locomotor activity profiles.

The centering of averaged activity profiles on the 12 h of darkness is done to highlight the phenotype of advanced morning activity. To prevent any confusion among readers, we have included a sentence in the figure legend explaining the difference in our activity profiles compared to previous literatures: "Activity profiles were centered of the 12 h darkness in all figures with evening activity on the left and morning activity on the right, which is different from general circadian literatures. (Fig. 2H legend)" (Line 957-959))

Point 4-The authors conclude that the loss of PDF and CNMa have opposite effects on the morning peak of locomotor activity (line 392). But they also acknowledge, briefly, that things are not that simple: loss of CNMa causes a phase advance, but loss of PDF causes a loss or reduction in the anticipatory peak. It is still significant to find a peptide transmitter with the clock neuron network that regulates morning activity, but the authors should revise their conclusion regarding the opposing actions of PDF and CNMa, which is not well supported by the data.

We have revised the relevant parts.

ABSTRACT: “Specific elimination of each from clock neurons revealed that loss of the neuropeptide CNMa in two posterior dorsal clock neurons (DN1ps) or its receptor (CNMaR) caused advanced morning activity, indicating a suppressive role of CNMa-CNMaR on morning anticipation, opposite to the promoting role of PDF-PDFR on morning anticipation.” (Line 43-48)

DISCUSSION: “Furthermore, given that the morning anticipation vanishing phenotype of Pdf or Pdfr mutant indicates a promoting role of PDF-PDFR signal, while the enhanced morning anticipation phenotype of CNMa mutant suggests an inhibiting role of CNMa signal, we consider the two signals to be antagonistic.” (Line 492-495)

Point 5-The authors should acknowledge, cite, and incorporate the substantive discussion of CNMa peptide and the DN1p neuronal class in Reinhard et al. 2022 (Front Physiol. 13: 886432).

We have revised the text accordingly and cited this paper: “Type I with two neurons whose branches projecting to the anterior region, as in CNMa∩GMR51H05, CNMa∩Pdfr, and CNMa∩GMR79A11 (Fig. 6E, 5G, 6H), and type II with four neurons branching on the posterior side with few projections to the anterior region, as in CNMa∩GMR91F02 (Fig. 6F). These two types of DN1ps’ subsets were also reported and profound discussed previously (Lamaze et al., 2018; Reinhard et al., 2022)”. (Line 393-397)

Reviewer #3 (Recommendations For The Authors):

Point 1-Throughout the manuscript figure legends (axis, genotypes, etc) are too small to be appreciated. Fig. 1. Panel A. The labels are very difficult to read.

We have attempted to enlarge the font as much as possible in the revised version.

Point 2-Fig. 1. H-J Why is efficiency not mentioned in all the examples?

In the revised manuscript, the results of Fig 1H-1J are discussed in the revised version (Line 145-147). The reason that we did not calculate the exact efficiency is that the GFP intensity is not stable enough which might change during dissection, mounting or intensity of laser in our experimental process. Therefore, in all results related to GFP signal (Fig. 1B-1J, Fig. S1, Fig. S2, Fig. 2B-2D), we relied on qualitative judgment rather than quantitative judgment, unless the GFP signal was easily quantifiable (such as in cases with limited cells or no GFP signal in the experimental group).

Point 3-Fig. 1. Panel L, left (light phase): the statistical comparisons are not clearly indicated (the same happens in Figs 3Q and 3R).

We have now re-arranged Fig. 1L and Fig. 3Q-3R to make the statistical comparisons clear in the new version.

Point 4-Line 792. Could induced be introduced?

Yes, we have now corrected this typo.

Point 5-Fig. S1. Check labels for consistency. GMR57C10 Gal4 driver is most likely R57C10.

We have now revised the labels (Fig. S1).

Point 6-Fig. S2. If the experiments were repeated and several brains were observed, the authors should include the efficiency and the number of flies as reported in Fig. S1.

We have now added the number of flies in Fig. S2 as reported in Fig. S1. As Response to Point 2 mentioned, due to the instability of the GFP signal, we are unable to provide a quantitative efficiency in this context.

Point 7-Fig S4. The fig legend describes panels I-J which are not shown in the current version of the manuscript.

We now have deleted them.

Point 8-Fig 2I. Surprising values for morning anticipation indexes even for controls (0.5 would indicate ¨no anticipation¨; in controls, the expected values would be >>0.5, as most of the activity is concentrated right before the transition. Could the authors explain this unexpected result?

We have revised the description of the calculation in the methods section (Line 612). After calculating the ratio of the last three hours of activity to the total six hours of activity, the results were further subtracted by 0.5. Therefore, the index should be ≤0.5. When the index is equal to 0, it indicates no morning anticipation.

Point 9-Fig 2K/L. The authors mention that not all genes are effectively knocked out with their strategy. Could this be accounted for the specific KD strategy, its duration, or the promotor strength? It is surprising no explanation is provided in the text (page 9 line 179).

In our pursuit of establishing a broadly effective method for gene editing, Fig. 2H-2L and Fig. 2D revealed that previous attempts have fallen short of achieving this objective. The observed inefficiency may be attributed to the intensity of the promoter, resulting in inadequate expression. Alternatively, the insufficient duration of the operation may also contribute to the lack of success. However, in the context of sleep and rhythm research applications, the age of the fruit fly tests is typically fixed, limiting the potential to enhance efficiency by extending the manipulation time. Moreover, increasing the expression level may pose challenges related to cytotoxicity, as reported in previous studies (Port et al., 2014). We refrain from offering specific explanations, as we lack a definitive plan and cannot provide additional robust evidence to support the above speculations. Consequently, in our ongoing efforts, we aim to enhance the efficiency of the tool system while operating within the current constraints.

Point 10-Page 9, line 179. Can the authors include a brief description of the reason for the different modifications? Only one was referenced.

We have revised related part in the manuscript (Line 223-231):

Cas9.M9: We fused a chromatin-modulating peptide (Ding et al., 2019), HMGN1 183 (High mobility group nucleosome binding domain 1), at the N-terminus of Cas9 and HMGB1 184 (High mobility group protein B1) at its C-terminus with GGSGP linker, termed Cas9.M9.

Cas9.M6: We also obtained a modified Cas9.M6 with HMGN1 at the N-terminus and an undefined peptide (UDP) at the C-terminus. (NOTE：UDP was gained by accident)

Cas9.M0: We replaced the STARD linker between Cas9 and NLS in Cas9.HC with GGSGP the linker (Zhao et al., 2016), termed Cas9.M0

Point 11-The authors tested the impact of KO nAChR2 across the different versions of conditional disruption (Fig 1K-L, Fig 2L, Fig 3R). It is surprising they observe a difference in daytime sleep upon knocking down with Cas9.HC (2L) but not with Cas9.M9 (3R) and the reverse is seen for night-time sleep. Could the authors provide an explanation? Efficiency is not the issue at stake, is it?

In Fig. 2K, the day sleep of flies (R57C10-GAL4/UAS-sgRNAnAChRbeta2; UAS-Cas9/+) was significantly decreased compared to flies (R57C10-GAL4/UAS-sgRNAnAChRbeta2; +/+), but not when compared to flies (R57C10-GAL4/+; UAS-Cas9/+). Our criterion for asserting a difference is that the experimental group must show a significant distinction from both control groups. Therefore, we concluded that there was no significant difference between the experimental group and the control groups in Fig. 2K.

Point 12-Fig. 4. Which of the two strategies described in A-B was employed to assemble the expression profile of CCT genes in clock neurons shown in C? This information should be part of the fig legend.

We have now revised the legend as follows: “(A-B) Schematic of intersection strategies used in Clk856 labelled clock neurons dissection, Flp-out strategy (A) and split-LexA strategy (B). The exact strategy used for each gene is annotated in Table S5.”

Point 13-Similarly, how many brains were analyzed to give rise to the table shown in C?

We have now revised the legend of Table S4 to address this concern. As indicated in: “The largest N# for each gene in Table S4 is the brain number analyzed for each gene”.

Point 14-Finally, the sentence ¨The figure is...¨ requires revision.

We have now revised it: “The exact cell number for each subset is annotated in Table S4”.

Point 15-Legend to Table S3. The authors have done an incredible job testing many gRNAs for each gene potentially relevant for communication. However, there is very little information to make the most out of it; for instance, the legend does not inform why many of the targeted genes do not appear to have been tested any further. It would be useful to the reader to discern whether despite being the 3 most efficient gRNAs, they were still not effective in targeting the gene of interest, or whether they showed off-targets, or it was simply a matter of testing the educated guesses. This information would be invaluable for the reader.

First, we designed and generated transgenic UAS-sgRNA fly lines for all these sgRNAs. We randomly selected 14 receptor genes, known for their difficulty in editing based on our experience, to assess the efficiency of our strategy, as depicted in Fig. 3M-3P, Fig. S5, and Fig. S6. We believe these results are representative and indicative of the efficiency of sgRNAs designed using our process and applied with the modified Cas9.

Secondly, we acknowledge your valid concern. While we selected sgRNAs with no predicted off-target effects through various prediction models (outlined in the Methods under C-cCCTomics sgRNA design), we did not conduct whole-genome sequencing. Consequently, we can only assert that the off-target possibility is relatively low. To address potential misleading effects arising from off-target concerns, it is essential to validate these results through mutants, RNAi, or alternative UAS-sgRNAs targeting the same gene.

Point 16-Table S4. Some of the data presented derives from observations made in 1-2 brains for a specific cluster; isn´t it too little to base a decision on whether a certain gene is (or not) expressed? It is surprising since the same CCT line was observed/analysed in more brains for other clusters. Can the authors explain the rationale?

The N# number represents the GFP positive number, and we have revised the legend of Table S4. The largest N# number denotes the total number of brains analyzed for a specific CCT line. It's possible that, due to variations in our dissection or mounting process, some clusters were only observed in 1-2 brains out of the total brains analyzed. To enhance the accuracy of intersection analysis results, we marked all positive signal records when positive subsets were found in less than 1/3 of the total analyzed brains (Table S4).

Point 17-The paragraph describing this data in the results section needs revising (lines 233-243).

We have now revised this. (Line 286-317)

Point 18-While it is customary for authors to attempt to improve the description of the activity patterns by introducing new parameters (i.e. MAPI and EAPI, lines 253-258) it would be interesting to understand the difference between the proposed method and the one already in use (which compares the same parameter, i.e., the slope (defined as ¨the slope of the best-fitting linear regression line over a period of 6 h prior to the transition¨, i.e., Lamaze et al. 2020 and many others). Is there a need to introduce yet another one?

This approach is necessary. The slope defined by Lamaze et al. utilizes data from only 2 time points, which may not accurately capture the pattern within a period before light on or off. Linear regression is not well-suited for a single fly due to the high variability in activity at each time point, making it challenging to fit the model at the individual level. The parameters we have introduced (MAPI and EAPI) in this paper are concise and can be applied at the individual level, effectively reflecting the morning or evening anticipation characteristics of each fly.

As an alternative, the activity plot of a certain fly line could be represented by an average of all flies' activity in one experiment. This would make linear regression easier to fit. However, several independent experiments are required for statistical robustness, necessitating the inclusion of hundreds of flies for each strain in a single analysis.

Point 19-In general, the legends of supplementary figures are a bit too brief. S7 and S8: it is not clear which of the two intersectional strategies were used (it would benefit whoever is interested in replicating the experiments). Legend to Fig S8 should read ¨similar to Fig S7¨.

We have now revised the legend and included “The exact strategy used for each gene is annotated in Table S5” in the legend.

Point 20-The legend in Table S6 should clearly state the genotypes examined. What does the marking in bold refer to?

We have now revised annotation of Table S6. Marking in bold refer to results out of one SD compared to control group.

Point 21-Line 314. The sentence needs revision.

We have revised these sentences.

Point 22-Line 391 (and also in the results section). The authors attempt to describe the CNMa phenotype as the opposite of pdf/pdfr mutant phenotypes. However, no morning anticipation/advanced morning anticipation are not necessarily opposite phenotypes.

We have revised related description.

ABSTRACT: “Specific elimination of each from clock neurons revealed that loss of the neuropeptide CNMa in two posterior dorsal clock neurons (DN1ps) or its receptor (CNMaR) caused advanced morning activity, indicating a suppressive role of CNMa-CNMaR on morning anticipation, opposite to the promoting role of PDF-PDFR on morning anticipation.” (Line 43-48)

DISCUSSION: “Furthermore, given that the morning anticipation vanishing phenotype of Pdf or Pdfr mutant indicates a promoting role of PDF-PDFR signal, while the enhanced morning anticipation phenotype of CNMa mutant suggests an inhibiting role of CNMa signal, we consider the two signals to be antagonistic.” (Line 492-495)

Reference

Deng, B., Li, Q., Liu, X., Cao, Y., Li, B., Qian, Y., Xu, R., Mao, R., Zhou, E., Zhang, W., et al. (2019). Chemoconnectomics: mapping chemical transmission in Drosophila. Neuron 101, 876-893.e874.

Ding, X., Seebeck, T., Feng, Y., Jiang, Y., Davis, G.D., and Chen, F. (2019). Improving CRISPR-Cas9 genome editing efficiency by fusion with chromatin-modulating peptides. Crispr j 2, 51-63.

Duhart, J.M., Herrero, A., de la Cruz, G., Ispizua, J.I., Pírez, N., and Ceriani, M.F. (2020). Circadian Structural Plasticity Drives Remodeling of E Cell Output. Curr Biol 30, 5040-5048.e5045.

Erion, R., King, A.N., Wu, G., Hogenesch, J.B., and Sehgal, A. (2016). Neural clocks and Neuropeptide F/Y regulate circadian gene expression in a peripheral metabolic tissue. eLife 5, e13552.

Fujiwara, Y., Hermann-Luibl, C., Katsura, M., Sekiguchi, M., Ida, T., Helfrich-Förster, C., and Yoshii, T. (2018). The CCHamide1 neuropeptide expressed in the anterior dorsal neuron 1 conveys a circadian signal to the ventral lateral neurons in Drosophila melanogaster. Front Physiol 9, 1276.

Goda, T., Tang, X., Umezaki, Y., Chu, M.L., Kunst, M., Nitabach, M.N.N., and Hamada, F.N. (2016). Drosophila DH31 neuropeptide and PDF receptor regulate night-onset temperature preference. J Neurosci 36, 11739-11754.

Goda, T., Umezaki, Y., Alwattari, F., Seo, H.W., and Hamada, F.N. (2019). Neuropeptides PDF and DH31 hierarchically regulate free-running rhythmicity in Drosophila circadian locomotor activity. Sci Rep 9, 838.

Guo, F., Cerullo, I., Chen, X., and Rosbash, M. (2014). PDF neuron firing phase-shifts key circadian activity neurons in Drosophila. Elife 3.

He, C., Cong, X., Zhang, R., Wu, D., An, C., and Zhao, Z. (2013). Regulation of circadian locomotor rhythm by neuropeptide Y-like system in Drosophila melanogaster. Insect Mol Biol 22, 376-388.

Hermann, C., Yoshii, T., Dusik, V., and Helfrich-Förster, C. (2012). Neuropeptide F immunoreactive clock neurons modify evening locomotor activity and free-running period in Drosophila melanogaster. J Comp Neurol 520, 970-987.

Hyun, S., Lee, Y., Hong, S.T., Bang, S., Paik, D., Kang, J., Shin, J., Lee, J., Jeon, K., Hwang, S., et al. (2005). Drosophila GPCR Han is a receptor for the circadian clock neuropeptide PDF. Neuron 48, 267-278.

Johard, H.A., Yoishii, T., Dircksen, H., Cusumano, P., Rouyer, F., Helfrich-Förster, C., and Nässel, D.R. (2009). Peptidergic clock neurons in Drosophila: ion transport peptide and short neuropeptide F in subsets of dorsal and ventral lateral neurons. J Comp Neurol 516, 59-73.

Lamaze, A., Krätschmer, P., Chen, K.F., Lowe, S., and Jepson, J.E.C. (2018). A Wake-Promoting Circadian Output Circuit in Drosophila. Curr Biol 28, 3098-3105.e3093.

Lear, B.C., Zhang, L., and Allada, R. (2009). The neuropeptide PDF acts directly on evening pacemaker neurons to regulate multiple features of circadian behavior. PLoS Biol 7, e1000154.

Lee, G., Bahn, J.H., and Park, J.H. (2006). Sex- and clock-controlled expression of the neuropeptide F gene in Drosophila. 103, 12580-12585.

Lelito, K.R., and Shafer, O.T. (2012). Reciprocal cholinergic and GABAergic modulation of the small ventrolateral pacemaker neurons of Drosophila's circadian clock neuron network. J Neurophysiol 107, 2096-2108.

Ma, D., Przybylski, D., Abruzzi, K.C., Schlichting, M., Li, Q., Long, X., and Rosbash, M. (2021). A transcriptomic taxonomy of Drosophila circadian neurons around the clock. Elife 10.

Port, F., Chen, H.M., Lee, T., and Bullock, S.L. (2014). Optimized CRISPR/Cas tools for efficient germline and somatic genome engineering in Drosophila. Proc Natl Acad Sci USA 111, E2967-2976.

Reinhard, N., Schubert, F.K., Bertolini, E., Hagedorn, N., Manoli, G., Sekiguchi, M., Yoshii, T., Rieger, D., and Helfrich-Förster, C. (2022). The Neuronal Circuit of the Dorsal Circadian Clock Neurons in Drosophila melanogaster. Front Physiol 13, 886432.

Renn, S.C., Park, J.H., Rosbash, M., Hall, J.C., and Taghert, P.H. (1999). A pdf neuropeptide gene mutation and ablation of PDF neurons each cause severe abnormalities of behavioral circadian rhythms in Drosophila. Cell 99, 791-802.

Shafer, O.T., Helfrich-Förster, C., Renn, S.C., and Taghert, P.H. (2006). Reevaluation of Drosophila melanogaster's neuronal circadian pacemakers reveals new neuronal classes. J Comp Neurol 498, 180-193.

Shafer, O.T., Kim, D.J., Dunbar-Yaffe, R., Nikolaev, V.O., Lohse, M.J., and Taghert, P.H. (2008). Widespread receptivity to neuropeptide PDF throughout the neuronal circadian clock network of Drosophila revealed by real-time cyclic AMP imaging. Neuron 58, 223-237.

Zhang, L., Chung, B.Y., Lear, B.C., Kilman, V.L., Liu, Y., Mahesh, G., Meissner, R.A., Hardin, P.E., and Allada, R. (2010). DN1(p) circadian neurons coordinate acute light and PDF inputs to produce robust daily behavior in Drosophila. Curr Biol 20, 591-599.

Zhao, P., Zhang, Z., Lv, X., Zhao, X., Suehiro, Y., Jiang, Y., Wang, X., Mitani, S., Gong, H., and Xue, D. (2016). One-step homozygosity in precise gene editing by an improved CRISPR/Cas9 system. Cell Res 26, 633-636.

Author Response

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

eLife assessment

This fundamental study provides an unprecedented understanding of the roles of different combinations of NaV channel isoforms in nociceptors' excitability, with relevance for the design of better strategies targeting NaV channels to treat pain. Although the experimental combination of electrophysiological, modeling, imaging, molecular biology, and behavioral data is convincing and supports the major claims of the work, some conclusions need to be strengthened by further evidence or discussion. The work may be of broad interest to scientists working on pain, drug development, neuronal excitability, and ion channels.

Reviewer #1 (Public Review):

Summary:

In this work, Xie, Prescott, and colleagues have reevaluated the role of Nav1.7 in nociceptive sensory neuron excitability. They find that nociceptors can make use of different sodium channel subtypes to reach equivalent excitability. The existence of this degeneracy is critical to understanding neuronal physiology under normal and pathological conditions and could explain why Nav subtype-selective drugs have failed in clinical trials. More concretely, nociceptor repetitive spiking relies on Nav1.8 at DIV0 (and probably under normal conditions in vivo), but on Nav1.7 and Nav1.3 at DIV4-7 (and after inflammation in vivo).

The conclusions of this paper are mostly well supported by data, and these findings should be of broad interest to scientists working on pain, drug development, neuronal excitability, and ion channels.

Strengths:

(1.1) The authors have employed elegant electrophysiology experiments (including specific pharmacology and dynamic clamp) and computational simulations to study the excitability of a subpopulation of DRGs that would very likely match with nociceptors (they take advantage of using transgenic mice to detect Nav1.8-expressing neurons). They make a strong point showing the degeneracy that occurs at the ion channel expression level in nociceptors, adding this new data to previous observations in other neuronal types. They also demonstrate that the different Nav subtypes functionally overlap and are able to interchange their "typical" roles in action potential generation. As Xie, Prescott, and colleagues argue, the functional implications of the degenerate character of nociceptive sensory neuron excitability need to be seriously taken into account regarding drug development and clinical trials with Nav subtype-selective inhibitors.

Weaknesses:

(1.2) The next comments are minor criticisms, as the major conclusions of the paper are well substantiated. Most of the results presented in the article have been obtained from experiments with DRG neuron cultures, and surely there is a greater degree of complexity and heterogeneity about the degeneracy of nociceptors excitability in the "in vivo" condition. Indeed, the authors show in Figures 7 and 8 data that support their hypothesis and an increased Nav1.7's influence on nociceptor excitability after inflammation, but also a higher variability in the nociceptors spiking responses. On the other hand, DRG neurons targeted in this study (YFP (+) after crossing with Nav1.8-Cre mice) are >90% nociceptors, but not all nociceptors express Nav1.8 in vivo. As shown by Li et al., 2016 ("Somatosensory neuron types identified by high-coverage single-cell RNA-sequencing and functional heterogeneity"), there is a high heterogeneity of neuron subtypes within sensory neurons. Therefore, some caution should be taken when translating the results obtained with the DRG neuron cultures to the more complex "in vivo" panorama.

We agree that most but not all Nav1.8+ DRG cells are nociceptors and that not all nociceptors express Nav1.8. We targeted small neurons that also express (or at some point expressed) Nav1.8, thus excluding larger neurons that express Nav1.8. This allowed us to hone in on a relatively homogeneous set of neurons, which is crucial when testing different neurons to compare between conditions (as opposed to testing longitudinally in the same neuron, which is not feasible). We expect all neurons are degenerate but likely on the basis of different ion channel combinations. Indeed, even within small Nav1.8+ neurons, other channels that we did not consider likely contribute to the degenerate regulation (as now better reflected in the revised Discussion).

That said, there are multiple sources of heterogeneity. We suspect that heterogeneity is more increased after inflammation than after axotomy because all DRG neurons experience axotomy when cultured whereas neurons experience inflammation differently in vivo depending on whether their axon innervates the inflamed area (now explained on lines 214-215). This is not so much about whether the insult occurs in vivo or in vitro, but about how homogeneously neurons are affected by the insult. Granted, neurons are indeed more likely to be heterogeneously affected in vivo since conditions are more complex. But our goal in testing PF-71 in behavioral tests (Fig. 8) was to show that changes observed in nociceptor excitability in Figure 7, despite heterogeneity, were predictive of changes in drug efficacy. In short, we establish Nav interchangeability by comparing neurons in culture (Figs 1-6), but we then show that similar Nav shifts can develop in vivo (Fig 7) with implications for drug efficacy (Fig 8). Such results should alert readers to the importance of degeneracy for drug efficacy (which is our main goal) even without a complete picture of nociceptor degeneracy or DRG neuron heterogeneity. Additions to the Discussion (lines 248-259, 304-308) are intended to highlight these considerations.

(1.3) Although the authors have focused their attention on Nav channels, it should be noted that degeneracy concerning other ion channels (such as potassium ion channels) could also impact the nociceptor excitability. The action potential AHP in Figure 1, panel A is very different comparing the DIV0 (blue) and DIV4-7 examples. Indeed, the conductance density values for the AHP current are higher at DIV0 than at DIV7 in the computational model (supplementary table 5). The role of other ion channels in order to obtain equivalent excitability should not be underestimated.

We completely agree. We focused on Nav channels because of our initial observation with TTX and because of industry’s efforts to develop Nav subtype-selective inhibitors, whose likelihood of success is affected by the changes we report. But other channels are presumably changing, especially given observed changes in the AHP shape (now mentioned on lines 304-308). Investigation should be expanded to include these other channels in future studies.

Reviewer #2 (Public Review):

Summary:

The authors have noted in preliminary work that tetrodotoxin (TTX), which inhibits NaV1.7 and several other TTX-sensitive sodium channels, has differential effects on nociceptors, dramatically reducing their excitability under certain conditions but not under others. Partly because of this coincidental observation, the aim of the present work was to re-examine or characterize the role of NaV1.7 in nociceptor excitability and its effects on drug efficacy. The manuscript demonstrates that a NaV1.7-selective inhibitor produces analgesia only when nociceptor excitability is based on NaV1.7. More generally and comprehensively, the results show that nociceptors can achieve equivalent excitability through changes in differential NaV inactivation and NaV expression of different NaV subtypes (NaV 1.3/1.7 and 1.8). This can cause widespread changes in the role of a particular subtype over time. The degenerate nature of nociceptor excitability shows functional implications that make the assignment of pathological changes to a particular NaV subtype difficult or even impossible.

Thus, the analgesic efficacy of NaV1.7- or NaV1.8-selective agents depends essentially on which NaV subtype controls excitability at a given time point. These results explain, at least in part, the poor clinical outcomes with the use of subtype-selective NaV inhibitors and therefore have major implications for the future development of Nav-selective analgesics.

Strengths:

(2.1) The above results are clearly and impressively supported by the experiments and data shown. All methods are described in detail, presumably allow good reproducibility, and were suitable to address the corresponding question. The only exception is the description of the computer model, which should be described in more detail.

We failed to report basic information such as the software, integration method and time step in the original text. This information is now provided on lines 476-477. Notably, the full code is available on ModelDB plus all equations including the values for all gating parameters are provided in Supplementary Table 5 and values for maximal conductance densities for DIV0 and DIV7 models are provided in Supplementary Table 6. Changes in conductance densities to simulate different pharmacological conditions are reported in the relevant figure legends (now shown in red). We did not include model details in the main text to avoid disrupting the flow of the presentation, but all the model details are reported in the Methods, tables and/or figure legends.

(2.2) The results showing that nociceptors can achieve equivalent excitability through changes in differential NaV inactivation and expression of different NaV subtypes are of great importance in the fields of basic and clinical pain research and sodium channel physiology and pharmacology, but also for a broad readership and community. The degenerate nature of nociceptor excitability, which is clearly shown and well supported by data has large functional implications. The results are of great importance because they may explain, at least in part, the poor clinical outcomes with the use of subtype-selective NaV inhibitors and therefore have major implications for the future development of Nav-selective analgesics.

In summary, the authors achieved their overall aim to enlighten the role of NaV1.7 in nociceptor excitability and the effects on drug efficacy. The data support the conclusions, although the clinical implications could be highlighted in a more detailed manner.

Weaknesses:

As mentioned before, the results that nociceptors can achieve equivalent excitability through changes in differential NaV inactivation and NaV expression of different NaV subtypes are impressive. However, there is some "gap" between the DRG culture experiments and acutely dissociated DRGs from mice after CFA injection. In the extensive experiments with cultured DRG neurons, different time points after dissociation were compared. Although it would have been difficult for functional testing to examine additional time points (besides DIV0 and DIV47), at least mRNA and protein levels should have been determined at additional time points (DIV) to examine the time course or whether gene expression (mRNA) or membrane expression (protein) changes slowly and gradually or rapidly and more abruptly.

Characterizing the time course of NaV expression changes is worthwhile but, insofar as such details are not necessary to establish that excitability is degenerate, it was not include in the current study. Furthermore, since mRNA levels do not parallel the functional changes in Nav1.7 (Figure 6A), we do not think it would be helpful to measure mRNA levels at intermediate time points. Measuring protein levels would be more informative, however, as now explained on lines 362-369, neurons were recorded at intermediate time points in initial experiments and showed a lot of variability. Methods that could track fluorescently-tagged NaV channels longitudinally (i.e. at different time points in the same cell) would be well suited for this sort of characterization, but will invariably lead to more questions about membrane trafficking, phosphorylation, etc. We agree that a thorough characterization would be interesting but we think it is best left for a future study.

It would also be interesting to clarify whether the changes that occur in culture (DIV0 vs. DIV47) are accompanied by (pro-)inflammatory changes in gene and protein expression, such as those known for nociceptors after CFA injection. This would better link the following data demonstrating that in acutely dissociated nociceptors after CFA injection, the inflammationinduced increase in NaV1.7 membrane expression enhances the effect of (or more neurons respond to) the NaV1.7 inhibitor PF-71, whereas fewer CFA neurons respond to the NaV1.8 inhibitor PF-24.

These are some of the many good questions that emerge from our results. We are not particularly keen to investigate what happens over several days in culture, since this is not so clinically relevant, but it would be interesting to compare changes induced by nerve injury in vivo (which usually involves neuroinflammatory changes) and changes induced by inflammation. Many previous studies have touched on such issues but we are cautious about interpreting transcriptional changes, and of course all of these changes need to be considered in the context of cellular heterogeneity. It would be interesting to decipher if changes in NaV1.7 and NaV1.8 are directly linked so that an increase in one triggers a decrease in the other, and vice versa. But of course many other channels are also likely to change (as discussed above), and they too warrant attention, which makes the problem quite difficult. We look forward to tackling this in future work.

The results shown explain, at least in part, the poor clinical outcomes with the use of subtypeselective NaV inhibitors and therefore have important implications for the future development of Nav-selective analgesics. However, this point, which is also evident from the title of the manuscript, is discussed only superficially with respect to clinical outcomes. In particular, the promising role of NaV1.7, which plays a role in nociceptor hyperexcitability but not in "normal" neurons, should be discussed in light of clinical results and not just covered with a citation of a review. Which clinical results of NaV1.7-selective drugs can now be better explained and how?

We wish to avoid speculating on which particular clinical results are better explained because our study was not designed for that. Instead, our take-home message (which is well supported; see Discussion on lines 309-321) is that NaV1.7-selective drugs may have a variable clinical effect because nociceptors’ reliance on NaV1.7 is itself variable – much more than past studies would have readers believe. At the end of the results (line 235), which is, we think, what prompted the reviewer’s comment, we point to the Discussion. The corollary is that accounting for degeneracy could help account for variability in drug efficacy, which would of course be beneficial. The challenge (as highlighted in the Abstract, lines 21-22) is that identifying the dominant Nav subtype to predict drug efficacy is difficult. We certainly don’t have all the answers, but we hope our results will point readers in a new direction to help answer such questions.

Another point directly related to the previous one, which should at least be discussed, is that all the data are from rodents, or in this case from mice, and this should explain the clinical data in humans. Even if "impediment to translation" is briefly mentioned in a slightly different context, one could (as mentioned above) discuss in more detail which human clinical data support the existence of "equivalent excitability through different sodium channels" also in humans.

We are not aware of human data that speak directly to nociceptor degeneracy but degeneracy has been observed in diverse species; if anything, human neurons are probably even more degenerate based on progressive expansion of ion channel types, splice variants, etc. over evolution. Of course species differences extend beyond degeneracy and are always a concern for translation, because of a species difference in the drug target itself or because preclinical pain testing fails to capture the most clinically important aspects of pain (which we mention on line 35). Line 39 now reiterates that these explanations for translational difficulties are not mutually exclusive, but that degeneracy deserves greater consideration that is has hitherto received. Indeed, throughout our paper we imply that degeneracy may contribute to the clinical failure of Nav subtype-specific drugs, but those failures are certainly not evidence of degeneracy. In the Discussion (line 320-321), we now cite a recent review article on degeneracy in the context of epilepsy, and point out how parallels might help inform pain research. We wish we had a more direct answer to the reviewer’s request; in the absence of this, we hope our results motivate readers to seek out these answers in future research.

Although speculative, it would be interesting for readers to know whether a treatment regimen based on "time since injury" with NaV1.7 and NaV1.8 inhibitors might offer benefits. Based on the data, could one hypothesize that NaV1.7 inhibitors are more likely to benefit (albeit in the short term) in patients with neuropathic pain with better patient selection (e.g., defined interval between injury and treatment)?

We like that our data prompt this sort of prediction. However, this is potentially complicated since the injury may be subtle, which is to say that the exact timing may not be known. There are scenarios (e.g. postoperative pain) where the timing of the insult is known, but in other cases (e.g. diabetic neuropathy) the disease process is quite insidious, and different neurons might have progressed through different stages depending on how they were exposed to the insult. Our own experiments with CFA are a case in point. Notwithstanding the potential difficulties about gauging the time course, any way of predicting which Nav subtype is dominant could help more strategically choose which drug to use.

Reviewer #3 (Public Review):

Summary:

In this study, the authors used patch-clamp to characterize the implication of various voltagegated Na+ channels in the firing properties of mouse nociceptive sensory neurons. They report that depending on the culture conditions NaV1.3, NaV1.7, and NaV1.8 have distinct contributions to action potential firing and that similar firing patterns can result from distinct relative roles of these channels. The findings may be relevant for the design of better strategies targeting NaV channels to treat pain.

Strengths:

The paper addresses the important issue of understanding, from an interesting perspective, the lack of success of therapeutic strategies targeting NaV channels in the context of pain. Specifically, the authors test the hypothesis that different NaV channels contribute in a plastic manner to action potential firing, which may be the reason why it is difficult to target pain by inhibiting these channels. The experiments seem to have been properly performed and most conclusions are justified. The paper is concisely written and easy to follow.

Weaknesses:

(1) The most critical issue I find in the manuscript is the claim that different combinations of NaV channels result in equivalent excitability. For example, in the Abstract it is stated that: "...we show that nociceptors can achieve equivalent excitability using different combinations of NaV1.3, NaV1.7, and NaV1.8". The gating properties of these channels are not identical, and therefore their contributions to excitability should not be the same. I think that the culprit of this issue is that the authors reach their conclusion from the comparison of the (average) firing rate determined over 1 s current stimulation in distinct conditions. However, this is not the only parameter that determines how sensory neurons convey information. For instance, the time dependence of the instantaneous frequency, the actual firing pattern, may be important too. Moreover, the use of 1 s of current stimulation might not be sufficient to characterize the firing pattern if one wants to obtain conclusions that could translate to clinical settings (i.e., sustained pain). A neuron in which NaV1.7 is the main contributor is expected to have a damping firing pattern due to cumulative channel inactivation, whereas another depending mainly on NaV1.8 is expected to display more sustained firing. This is actually seen in the results of the modelling.

This concern seems to boil down to how equivalent is equivalent? The spike shape or the full inputoutput curve for a DIV0 neuron (Nav1.8-dominant) is never equivalent to what’s seen in a DIV47 neuron (Nav1.7-dominant), but nor are any two DIV0 neurons strictly equivalent, and likewise for any two DIV4-7 neurons. Our point is that DIV0 and DIV4-7 neurons are a far more similar (less discriminable) in their excitability than expected from the qualitative difference in their TTX sensitivity (and from repeated claims in the literature that Nav1.7 is necessary for spike generation in nociceptors). Nav isoforms need not be identical to operate similarly; for instance, Nav1.8 tends to activate at “suprathreshold” voltages, but this depends on the value of threshold; if threshold increases, Nav1.8 can activate at subthreshold voltages (see Fig 5). We have modified lines 155- 175 to help clarify this.

We completely agree that firing rate is not the only way to convey sensory information, and of course injecting current directly into the cell body via a patch pipette is not a natural stimulus. These are all factors to keep in mind when interpreting our data. Nonetheless, our data show that excitability is similar between DIV0 and DIV 4-7, so much so that data from any one neuron (without pharmacological tests or capacitance measurements) would likely not reveal if that cell is DIV0 or DIV4-7; this “indiscriminability” qualifies as “equivalent” for our purposes, and is consistent with phrasing used by other authors studying degeneracy. Notably, not every DIV4-7 neuron exhibits spike height attenuation (see Fig. 1A), likely because of concomitant changes in the AHP that were not captured in our computer model or directly tested in our experiments. This highlights that other channel changes may also contribute to degeneracy and the maintenance of repetitive spiking.

(2) In Fig. 1, is 100 nM TTX sufficient to inhibit all TTX-sensitive NaV currents? More common in literature values to fully inhibit these currents are between 300 to 500 nM. The currents shown as TTX-sensitive in Fig. 1D look very strange (not like the ones at Baseline DIV4-7). It seems that 100 nM TTX was not enough, leading to an underestimation of the amplitude of the TTXsensitive currents.

As now summarized in Supplementary Table 3 (which is newly added), 100 nM TTX is >20x the EC50 for Nav1.3 and Nav1.7 (but is still far below the EC50 for Nav1.8). Based on this, TTXsensitive channels are definitely blocked in our TTX experiments.

(3) Page 8, the authors conclude that "Inflammation caused nociceptors to become much more variable in their reliance of specific NaV subtypes". However, how did the authors ensure that all neurons tested were affected by the CFA model? It could be that the heterogeneity in neuron properties results from distinct levels of effects of CFA.

We agree with the reviewer. We also believe that variable exposure to CFA is the most likely explanation for the heightened variability in TTX-sensitivity reported in Figure 7 (now more clearly explained on lines 214-215). One could try co-injecting a retrograde dye with the CFA to label cells innervating the injection site, but differential spread of the CFA and dye are liable to preclude any good concordance. Alternatively, a pain model involving more widespread (systemic) inflammation might cause a more homogeneous effect. But, our main goal with CFA injections was to show that a Nav1.8®Nav1.7 switch can occur in vivo (and is therefore not unique to culturing), and that demonstration is true even if some neurons do not switch. Subsequent testing in Figure 8 shows that enough neurons switch to have a meaningful effect in terms of the behavioral pharmacology. So, notwithstanding tangential concerns, we think our CFA experiments succeeded in showing that Nav channels can switch in vivo and that this impacts drug efficacy.

Recommendations for the authors:

All reviewers agreed that these results are solid and interesting. However, the reviewers also raised several concerns that should be addressed by the authors to improve the strength of the evidence presented. Revisions considered to be essential include:

(1) Discuss how degeneracy concerning other ion channels (such as potassium ion channels) could also impact nociceptor excitability (reviewer #1). Additionally, the translation of results from DRG neuron cultures to "in vivo" nociceptors should be better discussed.

We have added a new paragraph to the Discussion (line 248-259) to remind readers that despite our focus on Nav channels, other ion channels likely also change (and that these changes involve diverse regulatory mechanisms that require further investigation). Likewise, despite our focus on the changes caused by culturing neurons, we remind readers that subtler, more clinically relevant in vivo perturbations can likewise cause a multitude of changes. We end that paragraph by emphasizing that although accounting for all the contributing components is required to fully understand a degenerate system, meaningful progress can be made by studying a subset of the components. We want to emphasize this because there is some middle ground between focusing on one component at a time (which is the norm) vs. trying to account for everything (which is an infeasible ideal). Additional text on lines 304-308 also addresses related points.

(2) Discuss how different combinations of NaV channels result in equivalent excitability, in the context of the experimental conditions used (see main comment by reviewer #3). It should also be discussed in more detail which human clinical data support the existence of "equivalent excitability through different sodium channels" also in humans (reviewer #2).

Regarding the first part of this comment, reviewer 3 wrote in the public review that “The gating properties of these channels are not identical, and therefore their contributions to excitability should not be the same.” Differences in gating properties are commonly used to argue that different Nav subtypes mediate different phases of the spike, for example, that Nav1.7 initiates the spike whereas Nav1.8 mediates subsequent depolarization because Nav1.7 and Nav1.8 activate at perithreshold and suprathrehold voltages, respectively (see lines 134-135, now shown in red). But such comparison is overly simplistic insofar as it neglects the context in which ion channels operate. For instance, if Nav1.7 is not expressed or fully inactivates, voltage threshold will be less negative, enabling Nav1.8 to contribute to spike initiation; in other words, previously “suprathreshold” voltages become “perithreshold”. Figure 5 is dedicated to explaining this context-sensitivity; specifically, we demonstrate with simulations how Nav1.8 takes over responsibility for initiating a spike when Na1.7 is absent or inactivated. Text on lines 155- 184 has been edited to help clarify this. Regarding the second part of this comment, we are not aware of any direct evidence from human sensory neurons that different sodium channels produce equivalent excitability, but that is certainly what we expect. We suggest that failure of Nav subtype-specific drugs is, at least in part, because of degeneracy, but such failures do not demonstrate degeneracy unless other contributing factors can be excluded (which they can’t). Recognizing degeneracy is difficult, and so variability that might be explained by degeneracy will go unexplained or attributed to other factors unless, by design or serendipity, experiments quantify the effects of degeneracy (as we have attempted to do here). We now cite a recent review article on degeneracy and epilepsy (line 320), which addresses relevant themes that might help inform pain research; for instance, most existing antiseizure medications act on multiple targets whereas more recently developed single-target drugs have proven largely ineffective. This is similar to but better documented than for analgesics. With this in mind, we revised the text to emphasize the circumstantial nature of existing evidence and the need to test more directly for degeneracy (lines 320-323).

(3) Extend the discussion about the poor clinical outcomes with the use of subtype-selective NaV inhibitors. In particular, the promising role of NaV1.7, which plays a role in nociceptor hyperexcitability but not in "normal" neurons, should be discussed in light of clinical results and not just covered with a citation of a review. Which clinical results of NaV1.7-selective drugs can now be better explained and how? (reviewer #2)

As discussed above, we are cautious avoid speculating on which clinical results are attributable to degeneracy. Instead, our take-home message (see Discussion, lines 309-323) is that NaV1.7selective drugs may have a variable clinical effect because nociceptors’ reliance on NaV1.7 is itself variable – much more than past studies would have readers believe. The corollary is that accounting for degeneracy could help account for variability in drug efficacy, which would of course be beneficial. The challenge (as highlighted in the Abstract, lines 21-22) is that identifying the dominant Nav subtype to predict drug efficacy is not trivial. Interpreting clinical data is also complicated by the fact that we are either dealing with genetic mutations (with unclear compensatory changes) or pharmacological results (where NaV1.7-selective drugs have a multitude of problems that might contribute to their lack of efficacy, separate from effects of degeneracy). We have striven to contextualize our results (e.g. last paragraph of results, lines 222-235). We think this is the most we can reasonably say based on the limitations of existing clinical data.

(4) Provide a clearer and more detailed description of the computational model (reviewers #2 and #3).

We added important details on line 476-477 but, in our honest opinion, we think our computational model is thoroughly explained. The issue seems to boil down to whether details are included in the Results vs. being left for the Methods, tables and figure legends. We prefer the latter.

(5) Better clarify the effects of the CFA model, to provide further evidence relating inflammation with nociceptors variability (reviewers #2 and #3)

As explained in response to a specific point by reviewer #3, we believe that variable exposure to CFA explains the heightened variability in TTX-sensitivity reported in Figure 7 (now explained on lines 214-215). One could try co-injecting a retrograde dye with the CFA to label cells innervating the injection site, but differential spread of the inflammation and dye are liable to preclude any good concordance. Alternatively, a pain model involving more widespread (systemic) inflammation might cause a more homogeneous effect. But, our main goal with CFA injections was to show that a Nav1.8®Nav1.7 switch can occur in vivo (and is therefore not unique to culturing); that demonstration holds true even if some neurons do not switch. Subsequent testing (Fig 8) shows that enough neurons switch to drug efficacy assessed behaviorally. This is emphasized with new text on lines 225-227. Overall, we think our CFA experiments succeed in showing that Nav channels can switch in vivo and, despite variability, that this occurs in enough neurons to impact drug efficacy.

(6) Revise the text according to all recommendations raised by the reviewers and listed in the individual reviews.

Detailed responses are provided below for all feedback and changes to the text were made whenever necessary, as identified in our responses.

Reviewer #1 (Recommendations For The Authors):

Minor points/recommendations:

Protein synthesis inhibition by cercosporamide could be the direct cause of a smaller-thanexpected increase in Nav1.7 levels at DIV5. But for Nav1.8, there is a mitigation in the increased levels at DIV5, that only could be explained by several indirect mechanisms, including membrane trafficking and posttranslational modifications (phosphorylation, SUMOylation, etc.) on Nav1.8 or protein regulators of Nav1.8 channels. The authors suggest that "translational regulation is crucial", but also insinuate that other processes (membrane trafficking, etc.) could contribute to the observed outcome. It is difficult to assess the relative importance of these different explanations without knowing the exact mechanisms that are acting here.

We agree. We relied on electrophysiology (and pharmacology) to measure functional changes, but we wanted to verify those data with another method. We expected mRNA levels to parallel the functional changes but, when that did not pan out, we proceeded to look at protein levels. Perhaps we should have stopped there, but by blocking protein translation, we show that there is not enough Nav1.7 protein already available that can be trafficked to the membrane. That does not explain why Nav1.8 levels drop. Our immunohistochemistry could not tease apart membrane expression from overall expression, which limits interpretation. We have enhanced the text to discuss this (lines 200-204), but further experiments are needed. Though admittedly incomplete, our initial finding help set the stage for future experiments on this matter.

Page 15, typo: "contamination from genomic RNA" -> "contamination from genomic DNA" (appears twice).

This has been corrected on lines 420 and 421.

Page 17: I could not find the computer code at ModelDB (http://modeldb.yale.edu/267560). It seems to be an old web link. It should be available at some web repository.

We confirmed that the link works. Entry is password-protected (password = excitability; see line 476). Password protection will be removed once the paper is officially published.

Page 19, reference 36, typo: "Inhibitio of" -> "Inhibition of".

This has been corrected (line 557).

Page 33, typo: "are significantly larger than differences at DIV1" -> "are significantly larger than differences at DIV0".

This has been corrected (line 796).

Page 35, figure 6 legend. The number of experiments (n) is not indicated for panel C data.

N = 3 is now reported (line 828).

Reviewer #2 (Recommendations For The Authors):

p. 3/4 and Data of Fig. 6: It should be commented on why days 1-3 were not investigated. An investigation of the time course (by higher frequency testing) would certainly have an added value because it would be possible to deduce whether the changes develop slowly and gradually, or whether the excitability induced by different NaVs changes suddenly. At least mRNA and protein levels should be determined at additional time points to examine the time course or whether gene expression (mRNA) or membrane expression (protein) changes slowly and gradually or rapidly and more abruptly. It would also be interesting to clarify whether the changes that occur in culture (DIV0 vs. DIV4-7) are accompanied by (pro-)inflammatory changes in gene and protein expression, such as those known for nociceptors after CFA injection. Or is the latter question clear in the literature?

We now explain (lines 362-369) that intermediate time points (DIV1-3) were tested in initial current clamp recordings. Those data showed that TTX-sensitivity stabilized by DIV4 and differed from the TTX-insensitivity observed at DIV0. TTX-sensitivity was mixed at DIV1-3 and crosscell variability complicated interpretation. Subsequent experiments were prioritized to clarify why NaV1.7 is not always critical for nociceptor excitability, contrary to past studies. Our efforts to measure mRNA and protein levels were primarily to validate our electrophysiological findings; we are also interested in deciphering the underlying regulatory processes but this is an entire study on its own. Unfortunately, the existing literature does not help or point to an explanation for the Nav1.7/1.8 shift we observed.

Our evidence that mRNA levels do not parallel functional changes argues against pursuing transcriptional changes in Nav1.7, though transcriptional changes in other factors might be important. Interpretation of immuno quantification would be complicated by the high variability we observed with the physiology at intermediate time points and, furthermore, we cannot resolve surface expression from overall expression based on available antibodies. Methods conducive to longitudinal measurements would be more appropriate (as now mentioned on line 367-369). In short, a lot more work is required to understand the mechanisms involved in the switch, but we think the existing demonstration suffices to show that NaV1.7 and NaV1.8 protein levels vary, with crucial implications for which Nav subtype controls nociceptor excitability, and important implications for drug efficacy. Explaining why and how quickly those protein levels change will be no small feat is best left for a future study.

p. 4 and following: In order to enable the interpretation of the used concentration of PF-24, PF71, and ICA, the respective IC50 should be indicated.

A table (now Supplementary Table 3; line 861) has been added to report EC50 values for all drugs for blocking NaV1.7, NaV1.8 and NaV1.3. The concentrations we used are included on that table for easy comparison.

p. 5, end of the middle paragraph: Here it should be briefly explained -for less familiar readers- why NaV1.1 cannot be causative (ICA inhibits NaV1.1 and 1.3).

We now explain (lines 117-120) that NaV1.1 is expressed almost exclusively in medium-diameter (A-delta) neurons whereas NaV1.3 is known to be upregulated in small-diameter neurons, and so the effect we observe in small neurons is most likely via blockade NaV1.3.

p. 6, lines 4/5: At least once it should read computer model instead of model.

“Computer” has been added the first time we refer to DIV0 or DIV4-7 computer models (lines 138-139)

p. 6: the difference between Fig. 4B and Fig. 4 - Figure suppl. 1 should be mentioned briefly.

We now explain (lines 150-154) that Fig. 4B involves replacing a native channel with a different virtual channel (to demonstrate their interchangeability) whereas and Fig. 4 - Figure supplement 1 involves replacing a native channel with the equivalent virtual channel (as a positive control).

p. 6/7: the text and the conclusions regarding Figure 5 are difficult to follow. Somewhat more detailed explanations of why which data demonstrate or prove something would be helpful.

The text describing Figure 5 (lines 155-175) has been revised to provide more detail.

p. 7, last sentence of the first paragraph: How is this supported by the data? Or should this sentence be better moved to the discussion?

This sentence (now lines 182-184) is designed as a transition. The first half – “a subtype’s contribution shifts rapidly (because of channel inactivation)” – summarizes the immediately preceding data (Figure 5). The second half – “or slowly (because of [changes in conductance density])” – introduces the next section. The text show in square brackets has been revised. We hope this will be clearer based on revisions to the associated text.

p. 7, second paragraph, line 3: Please delete one "at both".

Corrected

p. 7, second paragraph: Please explain why different time points (DIV4-7, DIV5, or DIV7) were used or studied.

Initial electrophysiological experiments determined that TTX sensitivity stabilized by DIV 4 (see response to opening point) and we did not maintain neurons longer than 7 days, and so neurons recorded between DIV4 and 7 were pooled. If non-electrophysiological tests were conducted on a specific day within that range, we report the specific day, but any day within the DIV4-7 range is expected to give comparable results. This is now explained on lines 365-367.

p. 8: the text regarding Fig. 7 should also include the important data (e.g. percentage of neurons showing repetitive spinking) mentioned in the legend.

This text (lines 216-220) has been revised to include the proportion of neurons converted by PF71 and PF-24 and the associated statistical results.

Fig. 1: third panel (TTX-sensitive current...) of D & Fig. 2 subpanel of A (Nav1.8 current...). These panels should be explained or mentioned in the text and/or legends.

We now explain in the figure legends (lines 708-710; 714-715; 736-738) how those currents are found through subtraction.

Fig. 2 - figure supplement 2. One might consider taking Panel A to Fig. 2 so that the comparison to DIV0 is apparent without switching to Suppl. Figs.

We left this unchanged so that Figures 2 and 3 are equivalently organized, with negative control data left to the supplemental figures. Elife formatting makes it easy to reach the supplementary figure from the main figure, so we hope this won’t be an impediment to readers.

Fig. 6 C, middle graph (graph of Nav1.7): Please re-check, whether DIV5 none vs. 24 h and none vs. 120 h are really significantly different with such a low p-value.

We re-checked the statistics and the difference pointed out by the reviewer is significant at p=0.007. We mistakenly reported p<0.001 for all comparisons, and so this p value has been corrected; all the other p values are indeed <0.001. Notably, the data are summarized as median ± quartile because of their non-Gaussian distribution; this is now explained on line 827 (as a reminder to the statement on lines 461-462). Quartiles are more comparable to SD than to SEM (in that quartiles and SD represent the distribution rather than confidence in estimating the mean, like SEM), and so medians can differ very significantly even if quartiles overlap, as in this case.

Reviewer #3 (Recommendations For The Authors):

(1) A critical issue in the manuscript is the use of teleological language. It is likely that this is not the intention, but careful revision of the language should be done to avoid the use of expressions that confer purpose to a biological process. Please, find below a list of statements that I consider require correction.

• In the Abstract, the first sentence: "Nociceptive sensory neurons convey pain signals to the CNS using action potentials". Neurons do not really "use" action potentials, they have no will or purpose to do so. Action potentials are not tools or means to be "used" by neurons. Other examples of misuse of the verb "use" are found in several other sentences:

"...nociceptors can achieve equivalent excitability using different combinations of NaV1.3, NaV1.7, and NaV1.8"

"Flexible use of different NaV subtypes - an example of degeneracy - compromises..."

"Nociceptors can achieve equivalent excitability using different sodium channel subtypes" "...degeneracy - the ability of a biological system to achieve equivalent function using different components..."

"...nociceptors can achieve equivalent excitability using different sodium channel subtypes..."

"Our results show that nociceptors can achieve similar excitability using different NaV channels" "...the spinal dorsal horn circuit can achieve similar output using different synaptic weight combinations..."

"Contrary to the view that certain ion channels are uniquely responsible for certain aspects of neuronal function, neurons use diverse ion channel combinations to achieve similar function" "In summary, our results show that nociceptors can achieve equivalent excitability using different NaV subtypes"

“Use” can mean to put into action (without necessarily implying intention). Based on definitions of the word in various dictionaries, we feel we are well within the realm of normal usage of this term. In trying to achieve a clear and succinct writing style, we have stuck with our original word choice.

• At the end of page 5 and in the legend of Fig. 7, the word "encourage" is not properly used in the sentence "The ability of NaV1.3, NaV1.7 and NaV1.8 to each encourage repetitive spiking is seemingly inconsistent with the common view...". Encouraging is really an action of humans or animals on other humans or animals.

Like for “use”, we verified our usage in various dictionaries and we do not think that most readers will be confused or disturbed by our word choice. We use “encourage” to explain that increasing NaV1.3, NaV1.7 or NaV1.8 can increase the likelihood of repetitive spiking; we avoided “cause” because the probability of repetitive spiking is not raised to 100%, since other factors must always be considered.

• In the Abstract and other places in the manuscript, the word "responsibility" seems to be wrongly employed. It is true that one can say, for instance, on page 4 last paragraph "we sought to identify the NaV subtype responsible for repetitive spiking at each time point". However, to confer channels with the human quality of having "responsibility" for something does not seem appropriate. See also page 8 last paragraph, the first paragraph of the Discussion, and the three paragraphs of page 11.

Again, we must respectfully disagree with the reviewer. We appreciate that this reviewer does not like our writing style but we do not believe that our style violates English norms.

(2) In the first sentence of the Abstract, nociceptive sensory neurons do not convey "pain signals". Pain is a sensation that is generated in the brain.

“Pain” is used as an adjective for “signal” and is used to help identify the type of signal. Nonetheless, since the word count allowed for it, we now refer to “pain-related signals” (line 10).

(3) I do not see the point of plotting the firing rate as a function of relative stimulus amplitude (normalized to the rheobase, e.g., Fig. 1A bottom panels, Fig. 2B, bottom-right, Fig. 2 Supp2A right, Fig. 3 B bottom panels, etc) instead of as a function of the actual stimulus amplitude. I have the impression that this maneuver hides information. This is equivalent to plotting the current amplitudes as a function of the voltage normalized by the voltage threshold for current activation, which is obviously not done.

This is how the experiments were performed, so it would be impossible to perform the statistical analysis using the absolute amplitudes post-hoc; specifically, stimulus intensities were tested at increments defined relative to rheobase rather than in absolute terms. There are pros and cons to each approach, and both approaches are commonly used. Notably, we report the value of rheobase on the figures so that readers can, with minimal arithmetic, convert to absolute stimulus intensities. No information is hidden by our approach.

(4) On page 4 it is stated that "We show later that similar changes develop in vivo following inflammation with consequences for drug efficacy assessed behaviourally (see Fig. 8), meaning the NaV channel reconfiguration described above is not a trivial epiphenomenon of culturing". However, what happens in culture may have nothing in common with what happens in vivo during inflammation. Thus, the latter data may not serve to answer whether the culture conditions induce artifacts or not. I suggest tuning down this statement by changing "meaning" to "suggesting".

On line 97, we now write “suggesting”.

(5) Page 5, first paragraph, I miss a clear description of the mathematical models. Having to skip to the Methods section to look for the details of the models as the artifices introduced to simulate different conditions is rather inconvenient.

So as not to disrupt the flow of the presentation with methodological details, we only provide a short description of the model in the Results. We have slightly expanded this to point out that the conductance-based model is also single-compartment (line 111). We provide a very thorough description of our model in the Methods, especially considering all the details provided in Supplementary Tables 1, 5 and 6. We also report conductance densities and % changes in figure legends (lines 722, 747-748; now shown in red). This is also true for Figure 3-figure supplement 2 (lines 756-759). We tried very hard to find a good balance that we hope most readers will appreciate.

(6) Page 6, second paragraph, simulations do not serve to "measure" currents.

The sentence been revised to indicate that simulations were used to “infer” currents during different phases of the spike (line 155).

(7) Page 7, regarding the tile of the subsection "Control of changes in NaV subtype expression between DIV0 and DIV4-7", the authors measured the levels of expression, but not really the mechanisms "controlling" them. I suggest writing "changes in NaV subtype expression between DIV0 and DIV4-7"

We have removed “control of” from the section title (line 185)

(8) What was the reason for adding a noise contribution in the model?

We now explain that noise was added to reintroduce the voltage noise that is otherwise missing from simulations (line 474). For instance, in the absence of noise, membrane potential can approach voltage threshold very slowly without triggering a spike, which does not happen under realistically noisy conditions. Of course membrane potential fluctuates noisily because of stochastic channel opening and a multitude of other reasons. This is not a major issue for this study, and so we think our short explanation should suffice.

(9) Please, define the concept of degeneracy upon first mention.

Degeneracy is now succinctly defined in the abstract (line 20).

Author Response

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

Our answer to the final point(s) raised is as follows:

"We thank the reviewer for the comment. We checked our datasets accordingly. Typically, the n of cells showed deviations of maximally 20% from experiment to experiment (e.g. 16-24 cells per experiment). Additionally, experiments were performed using different passages of the cells. Moreover, data were validated at different time-points during the study using newly thawed cell lines."

---

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

Public Reviews:

Reviewer #1 (Public Review):

Bischoff et al present a carefully prepared study on a very interesting and relevant topic: the role of ion channels (here a Ca2+-activated K+ channel BK) in regulating mitochondrial metabolism in breast cancer cells. The potential impact of these and similar observations made in other tumor entities has only begun to be appreciated. That being said, the authors pursue in my view an innovative approach to understanding breast cancer cell metabolism. Considering the following points would further strengthen the manuscript:

We thank reviewer #1 for the overall positive feedback on our study.

Methods:

(1) The authors use an extracellular Ca2+ concentration (2 mM) in their Ringer's solutions that is almost twice as high as the physiologically free Ca2+ concentration (ln 473). Moreover, the free Ca2+ concentration of their pipette solution is not indicated (ln 487).

Indeed, we utilized 2 mM of Ca2+ in the physiologic live-cell imaging buffer. This concentration could actually be a little lower than the total Ca2+ concentration (ranging usually from 2.2 to 2.6 mM) in the body, while the free Ca2+ concentration is typically half as high. Nevertheless, we find multiple studies different from ours, which utilized 2 mM for their live-cell-based experiments. Please check the following studies, which represent only a small selection:

https://doi.org/10.1038/s41598-019-49070-8

https://doi.org/10.1016/j.bpj.2020.08.045

https://doi.org/10.1016/j.redox.2022.102319

However, to ensure that the applied conditions are physiologically relevant, we reperformed experiments using MMTV-PyMT WT and MMTV-PyMT BK-KO cells and compared cytosolic Ca2+ concentrations over time in response to cell stimulation with ATP, either in the presence of 1.0 mM (Author response image 1A) or 2.0 mM extracellular Ca2+ (Author response image1B). The respective graphs are attached in the following for reviewer’s inspection. As expected, we find that the intracellular Ca2+ concentration in MMTV-PyMT WT and BK-KO cells was dependent on the extracellular Ca2+ concentration. Importantly, however, irrespective of the exact Ca2+ concentration applied, we observed a similar difference in basal cytosolic Ca2+ between MMTV-PyMT WT and BK-KO cells (Author response image1C).

Author response image 1.

Cytosolic Ca2+ concentrations over-time in the presence of 1.mM or 2.0 mM extracellular Ca2+.

Concerning the Ca2+ concentration in the patch-pipette – we are very glad that you uncovered an error in our description and apologize for the mistake. Actually, the information the reviewer is referring to was already given in the previous version of the manuscript, but unclear because a comma was shifted (see line 487 in the originally submitted manuscript). The Ca2+ concentration of the patch-pipette was 0.1 mM in the presence of 0.6 mM EGTA, which should (according to Ca-EGTA calculator, https://somapp.ucdmc.ucdavis.edu/pharmacology/bers/maxchelator/CaEGTA-NIST.htm) be equivalent to ~30 nM of free Ca2+ in the patch pipette. We corrected the mistake in the manuscript and thank the reviewer again for spotting this inaccuracy.

(2) Ca2+I measurements: The authors use ATP to elicit intracellular Ca2+ signals. Is this then a physiological stimulus for Ca2+ signaling in breast cancer? What is the rationale for using ATP? Moreover, it would be nice to see calibrated baseline values of Ca2+i.

We thank the reviewer for the comment and suggestion. Importantly, it was demonstrated recently, that all of the utilized cell lines respond to treatment with extracellular ATP with a prominent increase in Ca2+I, most probably indicating the expression of purinergic receptors, which was a prerequisite to observe ATP induced changes in [Ca2+]i.

https://doi.org/10.1038/s41419-022-05329-z,

https://doi.org/10.1093/carcin/bgt493

https://doi.org/10.1038/s41598-018-26459-5

Furthermore, ATP plays a crucial role in the tumor microenvironment, where high rates of cell death occur. Hence, ATP is of pathophysiologic relevance for the utilized cancer cell lines.

https://doi.org/10.1038/s41568-018-0037-0

https://doi.org/10.3390/cells9112496

https://doi.org/10.1002/jcp.30580

Following the suggestions by Reviewer #1 (and #2), we included calibrations of Ca2+cyto and Ca2+mito in the manuscript, by depleting the intracellular Ca2+ stores using Ionomycin in the absence of extracellular Ca2+ (EGTA) to validate the basal difference in Ca2+cyto and Ca2+mito. Additionally, Ca2+cyto was calibrated under basal and inhibitor treated conditions, and values in nM are given in the text (p. 5, lines 185-190, 193-195 and 199-200, in the tracked changes version of the MS). The new data can be found in new Figure S2F – Figure S2J and new Figure S2R – Figure S2V. Moreover, we calculated basal [Ca2+]cyto in the different BKCa pro- and deficient cell lines and under inhibitor treated conditions. We additionally added information about the pathophysiologic relevance of ATP in the tumor microenvironment in lines 175-178 in the tracked changes version of the manuscript.

(3) Membrane potential measurements: It would be nice to see a calibration of the potential measurements; this would allow us to correlate the IV relationship with membrane potential. Without calibration, it is hard to compare unless the identical uptake of the dye is shown. Does paxilline or IbTx also induce depolarization?

We thank the reviewer for the suggestion. Indeed, membrane potential calibrations/ measurements using the membrane potential sensitive dye Dibac4(3) would be interesting, however, technically hardly feasible. The reason is that the principle of the dye is based on different uptake in response to differences in membrane potential, and not ratiometric as for most other dyes/ sensors used. Considering this limitation, we decided to perform membrane potential measurements by patch-clamp analysis. Additionally, we performed these experiments upon inhibition of PM-located BKCa by IBTX. Current-clamp experiments confirmed the difference in basal membrane potential between MMTV-PyMT WT and BK-KO cells (consult new Figure S1C and lines 127-130 in the tracked changes version of the manuscript). Interestingly, IBTX treatment depolarized the PM potential to the BK-KO cell level, which validates that BK activity and PM potential are connected. In addition to this approach, we utilized our recently developed genetically encoded K+ sensors revealing basal differences in [K+]cyto between MMTV-PyMT WT and BK-KO cells. Also this difference between both genotypes was equalized by IBTX as the respective treatment increased [K+]cyto only in WT cells, which most likely explains the cause of PM depolarization (consult lines 130-135 in the tracked changes version of the manuscript and new Figure S1D and Figure S1E).

(4) Mito-potential measurements: Why did the authors use such a long time course and preincubate cells with channel blockers overnight? Why did they not perform paired experiments and record the immediate effect of the BK channel blockers in the mito potential?

We thank the reviewer for the suggestion. We performed TMRM-based experiments with MMTV-PyMT WT cells in response to short-term exposure to paxilline, which did not significantly affect the mitochondrial membrane potential, at least within 15 minutes of treatment (Author response image 2). This indicates, that further downstream processes subsequent to (mito) BKCa inhibition affect the mitochondrial membrane potential(MMP), most probably including remodeling processes of the respiratory chain, mitochondrial ion homeostasis or glycolytic activity, ultimately also delivering reduction equivalents to mitochondria. Our final goal was to validate potential differences between a BKCa pro-and deficient cell model, whereby the latter cells lacked the BKCa channel since its origination. Hence, “long-term” (~12h) BKCa inhibition as performed in our experiments rather reflects the BK-KO cell situation. Taken together with the new experiment (Author response image 2), we can now state that the effect of BK inhibition on the MMP is at least not the consequence of an acute (within minutes) channel blockade.

Author response image 2.

Mitochondrial membrane potential, as measured using TMRM, in response to acute short-term administration of 5µM paxilline, followed by mitochondrial depolarization using FCCP.

(5) MTT assays are also based on mitochondrial function - since modulation of mito function is at the core of this manuscript, an alternative method should be used.

We thank the reviewer for the important comment. We performed additional, immunofluorescence-based experiments using Ki-67 staining to assess cell proliferation rates. The newly added data can be found in the text, lines 409-412 in the tracked changes version of the manuscript and new Figure S6D-F. The results obtained confirm the MTTbased results (Fig.6H-I).

Results:

(1) Fig. 5G: The number of BK "positive" mitoplasts is surprisingly low - how does this affect the interpretation? Did the authors attempt to record mitoBK current in the "whole-mitoplast" mode? How does the mitoBK current density compare with that of the plasma membrane? Is it possible to theoretically predict the number of mitoBK channels per mitochondrion to elicit the observed effects? Can these results be correlated with the immuno-localization of mitoBK channels?

Indeed, the number of BKCa-positive mitoplasts appears low on a first view. However, as these experiments were performed in a mitoplast-attached mode, it is important to keep in mind that only a very small area of the actual mitoplast is investigated with each patch. If no channel was detected in such region, the patch was depicted as “empty”, as presented in Fig.5G, which does, however, not mean that the entire mitochondria was actually BKCa negative. Hence, the density of BKCa in the IMM might be higher than expected from our experiments. Nevertheless, already earlier results using glioblastoma cell lines – considered to be one of the cell lines mostly enriched in mitoBKCa – demonstrated a quite low density of BKCa β4 regulatory subunit in mitochondria – please see figure 2B in the following paper: 10.1371/journal.pone.0068125 – which (based on 1:1 stoichiometry of α and β subunits) also suggests that the density of the alpha subunit of BKCa might be low in this compartment.

Author response image 3.

Author response image 3: Schematic representation of mitoplast attached patch-clamp experiments

Theoretically, density predictions of mitoBK compared to PM localized BKCa would be possible if whole-mitoplast experiments were performed, however, we are unsure what added value this information would actually burst, allowing the pharmacologic modulation of structures originally located within the mitochondrial matrix. Please also consult Author response image 3. According to the most recent models, even if there are other views on this, mitoBKCa is oriented in a way, that the C-terminus with its Ca2+ binding bowl is located within the mitochondrial matrix. Hence, to allow Ca2+ sensitivity experiments of the channel, broken up (by swelling) mitoplasts are required to make the Ca2+ binding bowl accessible for Ca2+ manipulations in the bath solution. This approach does not allow us to compare the channel density to that of the PM.

Finally, to the best of our knowledge, a combination of immunofluorescence with mitoplast patch-clamp experiments is not feasible yet, and would probably be impossible due to the low density of the mitoBKCa as well as the lack of highly sensitive and specific antibodies.

(2) There are also reports about other mitoK channels (e.g. Kv1.3, KCa3.1, KATP) playing an important role in mitochondrial function. Did the authors observe them, too? Can the authors speculate on the relative importance of the different channels? Is it known whether they are expressed organ-/tumor-specifically?

Author response image 4.

Representative single channels different to mitoBKCa detected in MDAMB-453 mitoplasts.

The reviewer is right, other K+ channels have been found in mitochondria and these also play a role in tumor cells. This is also consistent with our data (Fig.5G), where we observed other channels in the mitoplasts of BCCs as well. These all four cell lines tested. According to their conductance and our expectations from literature, these channels may e.g. include mitoIKCa, mitoSKCa, mitoKATP orothers (10.1146/annurev-biophys-092622-094853). As we focused, however, on patches containing a mitoBKCa, we did not further pharmacologically characterize these channels. Two examples of channels we found in these mitoplasts besides BKCa are presented for reviewers’ inspection (Author response image 4). As our manuscript focusses on mitoBKCa, we did not further classify these channels in smaller subgroups according to their conductance, as we feel that a differentiation between BKCa (~210 pS), and channels showing a conductance ≤150pS, or a conductance ≤100 pS is sufficient. Furthermore, this additional information would dilute our story too much making it difficult for the (non-specialist) reader to follow the red thread of the study. We added respective information in the manuscript, however. Please consult lines 365-366 in the tracked changes version of the manuscript.

Reviewer #1 is right, the observed the different K+ channels might of course be organ- or tumor-specific. For example, it has been reported that the expression of K+ channels is different in various cancer cell (lines) (https://doi.org/10.2174/13816128113199990032, 10.1016/j.pharmthera.2021.107874, 10.1038/nrc3635), a fact, which also according to our study might be exploited for pharmacological manipulation, aiming to affect proliferation/apoptosis of cancer cells. Further, a recently published single-cell and spatially resolved atlas of human breast cancer implies that the expression of different K+ channels (such as mitoIKCa, mitoSKCa, mitoKATP) might even differ between cancer- and non-cancer cells within a single tumour (https://doi.org/10.1038/s41588-021-00911-1).

Reviewer #2 (Public Review):

Summary:

The large-conductance Ca2+ activated K+ channel (BK) has been reported to promote breast cancer progression, but it is not clear how. The present study carried out in breast cancer cell lines, concludes that BK located in mitochondria reprograms cells towards the Warburg phenotype, one of the metabolic hallmarks of cancer.

Strengths:

The use of a wide array of modern complementary techniques, including metabolic imaging, respirometry, metabolomics, and electrophysiology. On the whole, experiments are astute and well-designed and appear carefully done. The use of BK knock-out cells to control for the specificity of the pharmacological tools is a major strength. The manuscript is clearly written.

There are many interesting original observations that may give birth to new studies.

Weaknesses:

The main conclusion regarding the role of a BK channel located in mitochondria appears is not sufficiently supported. Other perfectible aspects are the interpretation of co-localization experiments and the calibration of Ca2+ dyes. These points are discussed in more detail in the following paragraphs:

We thank reviewer #2 for the thorough assessment of our study.

(1) May the metabolic effects be ascribed to a BK located in mitochondria? Unfortunately not, at least with the available evidence. While it is clear these cells have a BK in mitochondria (characteristic K+ currents detected in mitoplasts) and it is also well substantiated that the metabolic effects in intact cells are explained by an intracellular BK (paxilline effects absent in the BK KO), it does not follow that both observations are linked. Given that ectopic BKDEC appeared at the surface, a confounding factor is the likely expression of BK in other intracellular locations such as ER, Golgi, endosomes, etc. To their credit, authors acknowledge this limitation several times throughout the text ("...presumably mitoBK...") but not in other important places, particularly in the title and abstract.

We thank the reviewer for this important comment and amended the title and abstract, respectively. The title of the manuscript was changed to “mitoBKCa is functionally expressed in murine and human breast cancer cells and potentially contributes to metabolic reprogramming.” Additionally, we changed appropriate passages in the text, to emphasize that mitoBKCa potentially mediates the metabolic reprogramming, but other intracellular channels could also contribute to these processes.

(2) MitoBK subcellular location. Pearson correlations of 0.6 and about zero were obtained between the locations of mitoGREEN on one side, and mRFP or RFP-GPI on the other (Figs. 1G and S1E). These are nice positive and negative controls. For BK-DECRFP however, the Pearson correlation was about 0.2. What is the Z resolution of apotome imaging? Assuming an optimum optical section of 600 nm, as obtained by a 1.4 NA objective with a confocal, that mitochondria are typically 100 nm in diameter and that BK-DECRFP appears to stain more structures than mitoGREEN, the positive correlation of 0.2 may not reflect colocalization. For instance, it could be that BK-DECRFP is not just in mitochondria but in a close underlying organelle e.g. the ER. Along the same line, why did BK-RFP also give a positive Pearson? Isn´t that unexpected? Considering that BK-DEC was found by patch clamping at the plasma membrane, the subcellular targeting of the channel is suspect. Could it be that the endogenous BK-DEC does actually reside exclusively in mitochondria (a true mitoBK), but overflows to other membranes upon overexpression? Regarding immunodetection of BK in the mitochondrial Percoll preparation (Fig. S5), the absence of NKA demonstrates the absence of plasma membrane contamination but does not inform about contamination by other intracellular membranes.

Indeed, it seems that BKCa-DEC is not an exclusive mitoBKCa, at least not upon (over-/)expression in MCF-7 cells. It is known from literature, that mitochondrial K+ channels are encoded by the nuclear genome, as no obvious gene for a K+ channel is found in the mitochondrial genome. Channel proteins are synthetized by cytosolic ribosomes and likely translocated into mitochondria via the TOM/TIM system. Although some K+ channels possess a mitochondrial targeting sequence at the N-terminus, their import is mostly far from a general mechanism, and this seems also to be true for BK channels. In the case of the K+ channel Kv1.3, an even more complex scenario is hypothesized, as the channel located in the PM could be transferred to mitochondria via mitochondria-associated membranes (MAM) structures of the ER (https://doi.org/10.3390/ijms20030734). Yet, the detailed mechanism for BK shuttling to mitochondria is not fully understood. Possibly, overflow is exactly what is happening, due to very high levels of BK-DEC expression upon transfection. However, that the channel translocates to the IMM upon transfection is not surprising and was also demonstrated for other cell models including HEK293 – see e.g. 10.1038/s41598-021-904653. Unfortunately, transfection efficiency of MCF-7 is quite low compared to HEK293 – hence, quantitative statements from mito-patches upon transfection are difficult.

In order to ensure that the mitochondrial colocalization is not a matter of poor microscope resolution, we reperformed these experiments using confocal imaging on a Zeiss LSM980 with an Airyscan 2 detector, yielding z resolutions of ~ 450 nm. These experiments confirmed the increased colocalization of BKCa-DEC with mitochondria compared to BKCa lacking the DEC exon. Furthermore, this imaging at higher resolution demonstrated, that, unfortunately, colocalization might not be the best analysis, as especially fragmented mitochondria showed a clear MitoGREEN stained matrix, surrounded by red fluorescence derived from BKCaDECRFP present in the IMM (revised Fig. 1G).

To validate the results derived from immunoblotting, we additionally stained the membranes for TMX1, a marker for the ER membrane. This analysis confirmed the high purity of the mitochondrial isolation without ER-membrane contamination after percoll purification, and hence validated the presence of BKCa in the mitochondrial membrane (revised Fig. S5D). The additional information can be found in lines 156-159 in the tracked changes version of the manuscript.

(3) Calibration of fluorescent probes. The conclusion that BK blockers or BK expression affects resting Ca2+ levels should be better supported. Fluorescent sensors and dyes provide signals or ratios that need to be calibrated if comparisons between different cell types or experimental conditions are to be made. This is implicitly acknowledged here when monitoring ER Ca2+, with an elaborate protocol to deplete the organelle in order to achieve a reading at zero Ca2+.

We thank the reviewer for the important comment. Please note that at no point in the manuscript we aim to compare different cell lines concerning their intracellular Ca2+ concentration, but we only compare the same cell lines after the different treatments, as we are aware of this limitation of fluorescent probes. However, to validate the differences in intracellular Ca2+ concentrations, we calibrated the signals derived from Fura-2 and 4mtD3cpV using ionomycin in combination with cellular Ca2+ depletion/ saturation. The newly added data can be found in the text, lines 185-190, 192-195, 199-200, and 228-230 in the tracked changes version of the manuscript, as well as new Figure S2F – Figure S2J and new Figure S2R – Figure S2V

Line 203. "...solely by the expression of BKCa-DECRFP in MCF-7 cells". Granted, the effect of BKCa-DECRFP on the basal FRET ratio appears stronger than that of BK-RFP, but it appears that the latter had some effect. Please provide the statistics of the latter against the control group (after calibration, see above).

Author response image 5.

Dot blot for data shown in Figure 2I.

The reviewer is right, it seems that BKCaRFP may also affect [Ca2+]mito. However, the effect is not significant and shows a p-value of p>0.999 using Kruskal-Wallistest followed by Dunn’s multiple comparison test, due to the non-normally distributed nature of the data. p=0.0002 for ctrl vs. BKCa-DECRFP and 0.0022 for BKCaRFP vs. BKCa-DECRFP, however. We added a scatter dot-blot of the respective data as Author response image 5 for reviewer’s inspection. Additionally, first, even using a more stringent statistical test by only comparing ctrl vs BKCaRFP using Mann-Whitney test, the results are not significant, as the p-value was determined at 0.4467, and second, we performed the requested Ca2+calibration using ionomycin under these conditions, which confirmed the difference between ctrl cells and BKCa-DECRFP expressing cells, but not BKCaRFP expressing ones. Please see Figure S2V.

Reviewer #3 (Public Review):

The original research article, titled "mitoBKCa is functionally expressed in murine and human breast cancer cells and promotes metabolic reprogramming" by Bischof et al, has demonstrated the underlying molecular mechanisms of alterations in the function of Ca2+ activated K+ channel of large conductance (BKCa) in the development and progression of breast cancer. The authors also proposed that targeting mitoBKCa in combination with established anti-cancer approaches, could be considered as a novel treatment strategy in breast cancer treatment.

The paper is clearly written, and the reported results are interesting.

Strengths:

Rigorous biophysical experimental proof in support of the hypothesis.

Weaknesses:

A combinatorial synergistic study is missing.

We thank reviewer #3 for the positive summary of our study. Indeed, we propose that targeting of mitoBKCa in combination with established anti-cancer drugs may represent a novel anti-cancer treatment strategy. Unfortunately, we feel that the manuscript is very condensed already, and that adding respective required experiments and data to support this hypothesis will make the flow of the manuscript more complex or even incomprehensible. As no attempts linking mitoBKCa activity with anti-cancer therapies have been made so far, we removed the respective information from the abstract and only discuss this aspect.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) Statistics: Legends have to contain information about the number of biological replicates (N) and cells analysed (n). Statistics must be calculated with the averages of the replicates.

Author response image 6.

Representative single cell responses of Fura-2 loaded MMTV-PyMT WT cells.

We thank the reviewer for the comment and added the missing details to all figure legends.

We feel that using each cell represents exactly the power of high-resolution live-cell imaging, as there is no better biological replicate than a single separated cell, which is observed by fluorescence microscopy. This analysis is also able to visualize cell-to-cell differences in the microscopy area, similarly to patch-clamp experiments, where each single cell or mitoplast patched is used as a single replicate. Please find a representative dataset derived from fluorescence microscopy of different responses of neighboring single cells in Author response image 6.

(2) Fig. 1G: This is a poor resolution figure, mostly because of its far too small size; in its current form it bears very little information.

We agree with reviewer #1 and reperformed the imaging experiments using high resolution confocal imaging and exchanged the respective images. We feel that this increased the quality of the images significantly. Unfortunately, we were not able to increase the size of the images in the main figure, hence, we added magnifications of the respective images as new Figure S1I.

(3) Fig. 1H: What do the dotted grey lines and the labels stand for?

We believe Reviewer #1 is probably referring to Figure 1G. As indicated in the figure panel and in the text, the grey dotted lines and labels indicate the colocalization scores of mtRFP and RFP-GPI with MitoGREEN, respectively. These data are also shown in Figure S1H, including error bars and statistics. We added additional information in the text to make the meaning of the lines clearer to the reader. Please consult lines 149 – 150 in the tracked changes version of the manuscript.

Reviewer #2 (Recommendations For The Authors):

(1) May the metabolic effects be ascribed to a BK located in mitochondria? Short of a way to tackle BK function and metabolism specifically in mitochondria, the conclusion may best be toned down to "intracellular BK". For the time being the term "mitoBK" appears too ambitious.

We fell that you are right and that our previous overstatement requires adaptation as a clear (100%) attribution of the observed metabolic effects solely to mitoBKCa is not definitely possible. We have therefore amended all relevant passages in the entire MS accordingly.

(2) MitoBK subcellular location. Please address the points raised in the Public Review.

As stated above we addressed the point raised in the public review accordingly (please consult new Figure S1I and revised Figure 1G).

(3) Calibration of fluorescent probes. Please provide calibrations for cytosolic and mitochondrial Ca2+, for example, the standard high Ca2+/ionophore/metabolic inhibition treatment to reach saturation followed by Ca2+ chelation to obtain zero Ca2+.

We thank the reviewer for the comment. As you can see from our response to the public review, we performed the respective experiments, and datasets were added in the manuscript.

(4) Line 203. "...solely by the expression of BKCa-DECRFP in MCF-7 cells". Granted, the effect of BKCa-DECRFP on the basal FRET ratio appears stronger than that of BK-RFP, but it appears that the latter had some effect. Please provide the statistics of the latter against the control group (after calibration, see above).

Please consult our response to the (same) comment in the public review.

(5) Line 228. The statement "Similar results were obtained in MDA-MB-453 cells" is confusing. As shown in Fig.3, pax reduced ECAR and OCR in MMTV-PyMT WT cells. As ibtx was without effect, it is suggested that intracellular BK support metabolism. However, the effect of pax on MDA cells was the opposite. Doesn´t this divergence speak against a universal role of intracellular BKs in promoting metabolism in BCCs? A similar point may be made regarding metabolomics, which showed no effects of pax on lactate and pyruvate in MMTV-PyMT WT cells but stimulation in MDA cells. Perhaps the word "promotes" in the title of the figure should be replaced by something more neutral like "affects" or "alters", as used elsewhere,

We thank the reviewer for pointing out the overstatement regarding intracellular BK functions and changed the title of the figure as suggested.

With regard to the experiments mentioned, we would like to point out the following aspects:

First, the cell lines used strongly differ in their metabolic settings under basal conditions. While both, MMTV-PyMT and MDA-MB-453 cells seem to show similar basal ECAR levels (if BKCa was present), their OCR seems to differ strongly. MMTV-PyMT cells seem to show a basal OCR which is almost at the maximum already, while MDA-MB-453 cells possess a tremendous capacity in their OCR, as observed upon mitochondrial uncoupling using FCCP. Of note, both, ECAR and OCR are indirect metabolic measures. On the one hand, ECAR measures extracellular acidification, which is accomplished by H+ along with lactate secretion. However, lactate secretion is not the only process leading to extracellular acidification, and ECAR may hence measure a variety of H+ releasing processes, including processes of vesicle secretion. On the other hand, OCR is not directly linked to ATP production, as mitochondrial complex IV is consuming O2, ATP, however, is produced by mitochondrial complex V. This becomes even more evident when having a look on OCRs after FCCP treatment – under these conditions, the H+ gradient is destroyed and ATP synthase activity is reduced, OCR, however, increases to the maximum due to increased supply of mitochondrial complex IV with H+.

Second, please note that the LC-MS-based metabolomics derive from a static single time point and not from an over-time “live” read-outs. Moreover, underlying dynamics of the parameters measured can not be assessed. Hence, as an example, increasing levels of pyruvate can e.g. indicate faster generation, or slower subsequent degradation/ metabolization. A clear in-depth statement about what is happening under basal and BKCa inhibitor treated conditions is hence not possible. The only conclusion possible to draw from these experiments is that paxilline treatment differentially affects metabolic pathways in these cells.

Based on these limitations of both methods, we decided to perform our in-depth fluorescence microscopy-based analysis, which provided strong evidence for intracellular BKCa channels on mitochondrial ATP production. Despite opposing effects of BKCa inhibition on OCR in MMTV-PyMT WT and MDA-MB-453 cells, mitochondrial ATP production was reduced, if BKCa-DECRFP was expressed/ intracellular BKCa was functional.

In line with these findings, mitoBKCa was recently described as an uncoupling protein, which could furthermore explain the differential effects of intracellular BKCa inhibition on OCR. https://doi.org/10.1038/s41598-021-90465-3

Minor

(6) Fig. 1C. Average fluorescence intensity in 6 experiments was about 20% higher in BK-KO cells relative to WT. Such a small difference is significant but should not be evident to the eye. The pictures selected for illustration appear to show a much larger difference and therefore may not be representative. If this is the case, please omit them. The same goes for the other representative pictures.

Author response image 7.

: Representative images at different brightnesses.

Please note, that the analysis of the images was done in an unbiased way using a Fiji macro. After analysis, we chose representative images, which were closest to the average.

Furthermore, we must kindly disagree with the reviewer as changes of 20% in fluorescence intensity are indeed evident to the eye (consult Author response image 7). This panels show the same image at different brightness levels with intensity differences of 20%. Hence, we feel, that all the images the reviewer was referring are representative for the values given.

(7) Line 130. The definition of "recent" is of course relative, but 10 years?

We are very glad that you have discovered this “inconsistency", and reworded the respective phrase accordingly.

(8) Line 327. "conductivity" is the property of a medium, "conductance" is the property of a component, such as a channel.

We thank the reviewer for the important comment. We revised the text accordingly.

(9) Various figures. FRET sensor data are expressed as Ratio(FRET/CFP). This is unusual, typically it should be FRET ratio (YFP/CFP), FRET ratio(mTFP/Venus), etc. Please note that the FRET partners differ between sensors.

We acknowledge the comment of the reviewer. It is correct that fluorescent proteins vary widely between the sensors (used). Please note, however, the following: The emission measured from these sensors actually represents FRET, as CFP but not YFP is directly excited. Hence, emission is FRET, not the “intrinsic” fluorescence of the YFP. This is getting more and more important to differentiate, as there are probes existing, which can also be “alternately” excited, i.e. CFP and YFP separately, which will then yield the YFP/CFP ratio (https://doi.org/10.1021/acssensors.8b01599). In case of only CFP excitation, we feel, that the term FRET/CFP is preferable over other labelings such as YFP/CFP.

(10) BK-DEC makes BCCs cells less oxidative. However, BK-DEC was first described in cardiomyocytes, which are among the most oxidative cell types. It would be useful if authors could address this apparent contradiction in the Discussion Section.

That is an exciting point that we addressed as follows in the revised MS:

First, it is important to mention that cardiac myocytes do not show a metabolic Warburg setting and are – under physiologic conditions – maintained in a high O2 environment.

Second, a recent study from our group addressed the question about the role of mitoBKCa in primary cardiac myocytes. Indeed, mitoBKCa was functionally expressed in these cells. Interestingly, under physiologic conditions, the channel did not alter (multiple) cell behaviours nor overall cardiac physiology in a mouse model. However, upon induction of ischemia/ reperfusion injury, a lack of BK increased cardiac susceptibility to cell death resulting in increased infarction size (https://doi.org/10.1161/CIRCULATIONAHA.117.028723). Hence, also in this cell model, BKCa only played a role under oxygen limited conditions/ conditions where mitochondria were not properly functioning. Thus, the results derived from cardiac myocytes support our recent findings in BCCs, as BKCa mediates BCC resistance to hypoxic stress/ makes BCCs more independent from oxidative metabolism.

Parts of this discussion were included in the revised MS. Please consult lines 490-500 in the tracked changes version of the manuscript.

Reviewer #3 (Recommendations For The Authors):

(1) The study is very well designed and most of the computational analyses were done rigorously.

We highly appreciate the positive feedback by reviewer #3.

(2) The authors should discuss the expression of BKCa in different subsets of breast cancer. Authors may also debate on the level of steroid receptors and BKCa expressions.

We thank reviewer #3 for the important suggestion and added the requested information in the discussion, lines 445-447 and 450-454 in the tracked changes version of the manuscript.

(3) In the discussion section, the authors mentioned that the MCF7 cell is the best model to study this hypothesis. Does it imply that triple-negative breast cancer cell lines express lower levels of BKCa? The authors should discuss this.

We thank the reviewer for the interesting comment; we would like to point out that the ERα-positive MCF-7 cell line was used to study experimental overexpression of BKCa at an otherwise low baseline level. This does not imply that BKCa is expressed at lower levels in TNBC cell lines; in fact a recent study showed the opposite, i.e. overexpression of BKCa in TNBC patients (10.1186/s12885-020-07071-1). Consistent with our work, the authors conclude that the channel could even be a new strategy for development of a targeted therapy in TNBC. We also added this information in the discussion, lines 450-454 in the tracked changes version of the manuscript.

(4) The authors propose that combinatorial targeting of mitoBKCa along with known breast cancer chemotherapeutics can open a new horizon in breast cancer treatment. However, the authors did not perform any experiment to show the synergistic effect as mentioned.

As already stated in the public reviews, we feel that the manuscript is very condensed already, and that adding the respective experiments and data will make the flow of the study even more complex. For the moment, we removed all information and statements linking mitoBKCa with anti-cancer treatment strategies from the abstract and only discuss this aspect. We hope that the reviewer agrees with us that an extensive analysis of the functional mitoBKCa status in the context of established breast cancer therapies must be addressed by (our) future studies.

Minor Comments:

There are several typos and grammatical errors that need further attention and rephrasing.

We thank the reviewer for the comment and revised the text accordingly.

Author Response

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

eLife assessment

This paper represents important findings when identifying untargeted metabolomics and its differences between metabolomes of different biological samples. GromovMatcher is the fantasy name for the soft development. The main idea behind it is built on the assumption of featuring and matching complex datasets. Although the manuscript reflects a solid analysis, it remains incomplete for validation with putative non-curated datasets.

We are grateful to the eLife editor for taking the time and effort to assess our manuscript.

We are however unsure of what the editor means by “it remains incomplete for validation with putative non-curated datasets”. As noted by Reviewer 2, manually curated datasets that could be used for validation are scarce. Most publicly available datasets do not contain sufficient information to establish a ground truth matching on which GromovMatcher, M2S, or metabCombiner can be tested. Even in the case where such a ground truth matching can be established, it must be performed by-hand through a manual matching process which is extremely time-consuming and requires very specific expertise. This, in our opinion, only highlights the need for automatic alignment methods such as metabCombiner, M2S or GromovMatcher.

We do agree that the performance of GromovMatcher (and its competitors) needs to be validated further, and we plan to continue validating GromovMatcher as additional data becomes available in EPIC and other cohorts. With that in mind, the lack of publicly available validation data is the reason why we conducted such an extensive simulation study, arguably more comprehensive than previous validations, exploring challenging settings that we believe reflect real-life scenarios (main text “Validation on ground-truth data” and Appendix 3). We would like to stress that this allows us to highlight previously ignored limitations of the previously published methods, metabCombiner and M2S.

We wish to thank the editor and reviewers for their time and efforts in reviewing our manuscript which led to many significant additions to our paper. Namely we:

• Performed an additional sensitivity analysis (Appendix 3) exploring how an imbalance in the number of features or samples between two studies being matched (e.g. the dataset split), affects the quality of matchings found by GromovMatcher, metabCombiner, and M2S.

• Investigated how changing or removing the reference dataset (Appendix 5) in the EPIC study (main text “Application to EPIC data”), affects the results of GromovMatcher.

• Improved alignment matrix visualizations in Fig. 3a for all four methods tested on the validation data, to highlight more clearly which feature matches were correctly identified or missed.

The revised paper is uploaded as the file “main_elife_revision.pdf” where all revisions are highlighted in blue as well as a copy “main_elife_revision_nohighlights.pdf” where revisions are not highlighted.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The authors have implemented the Optimal Transport algorithm in GromovMatcher for comparing LC/MS features from different datasets. This paper gains significance in the proteomics field for performing meta-analysis of LC/MS data.

Strengths:

The main strength is that GromovMatcher achieves significant performance metrics compared to other existing methods. The authors have done extensive comparisons to claim that GromovMatcher performs well.

Weaknesses:

There are two weaknesses.

(1) When the number of features is reduced the precision drops to ~0.8.

We would like to clarify that this drop in precision occurs in the challenging setting where only a small proportion of metabolites are shared between both datasets (e.g., the overlap – or proportion of shared features - was 25% in our simulation study). When two untargeted metabolic datasets share only 25% of their features, this is a challenging setting for any automated matching method as the vast majority 75% of the features in both datasets must remain unmatched.

In such settings, the reviewer correctly observes that the precision of GromovMatcher algorithms (GM and GMT) drops within the range of 0.80 - 0.85 (Figure 3b, top left panel). Such a precision of 0.8 or larger is still competitive compared with the alternative methods MetabCombiner (mC) and M2S whose precisions drop below 0.8 (see main text Fig. 3b, top left panel).

Precision is measured as the number of metabolite pairs correctly matched divided by all matches identified by a method. In other words, even in the challenging setting when the number of shared features (true matches) between both datasets is small (e.g. low 25% overlap), upwards of 80% of the feature matches found by GromovMatcher are correct which is a very encouraging result.

(2) How applicable is the method for other non-human datasets?

We thank the reviewer for raising this question. The crux of the matter concerning the application to animal data revolves around the hypothesis that correlations between metabolites in two different studies are preserved. Theoretically, the metabolome operates under similar principles in humans, governed by an underlying network of biochemical reactions. Consequently, in comparable human populations, the GM hypothesis is likely to hold to some extent.

However, in practice, application to animal data is more complicated. Animal studies tend to have smaller sample sizes and often stem from intervention-driven scenarios, such as mice subjected to specific diets or chemicals. This results in deliberate alterations in metabolic structures which makes finding two comparable animal studies less likely. To investigate the reviewer’s question, we have searched through the two predominant LC-MS dataset repositories (MetaboLights and NIH Metabolomics Workbench) but did not find any pairs of comparable animal studies due to the reasons mentioned above. One potential strategy to navigate this issue could entail regressing the metabolic intensities against the variables that notably differ between the two animal populations and running GM using the residual intensities. This would be an interesting direction for future research and additional validation would be needed to test the robustness of GM in this setting.

Reviewer #2 (Public Review):

Summary:

The goal of untargeted metabolomics is to identify differences between metabolomes of different biological samples. Untargeted metabolomics identifies features with specific mass-to-charge ratio (m/z) and retention time (RT). Matching those to specific metabolites based on the model compounds from databases is laborious and not always possible, which is why methods for comparing samples on the level of unmatched features are crucial.

The main purpose of the GromovMatcher method presented here is to merge and compare untargeted metabolomes from different experiments. These larger datasets could then be used to advance biological analyses, for example, for the identification of metabolic disease markers. The main problem that complicates merging different experiments is m/z and RT vary slightly for the same feature (metabolite).

The main idea behind the GromovMatcher is built on the assumption that if two features match between two datasets (that feature I from dataset 1 matches feature j from dataset 2, and feature k from dataset 1 matches feature l from dataset 2), then the correlations or distances between the two features within each of the datasets (i and k, and j and l) will be similar. The authors then use the Gromov-Wasserstein method to find the best matches matrix from these data.

The variation in m/z between the same features in different experiments is a user-defined value and it is initially set to 0.01 ppm. There is no clear limit for RT deviations, so the method estimates a non-linear deviation (drift) of RT between two studies. GromovMatcher estimates the drift between the two studies and then discards the matching pairs where the drift would deviate significantly from the estimate. It learns the drift from a weighted spline regression.

The authors validate the’performance of their GromovMatcher method by a validation experiment using a dataset of cord blood. They use 20 different splits and compare the GromovMatcher (both its GM and GMT iterations, whereby the GMT version uses the deviation from estimated RT drift to filter the matching matrix) with two other matching methods: M2S and metabCombiner.

The second validation was done using a (scaled and centered) dataset of metabolics from cancer datasets from the EPIC cohort that was manually matched by an expert. This dataset was also used to show that using automatic methods can identify more features that are associated with a particular group of samples than what was found by manual matching. Specifically, the authors identify additional features connected to alcohol consumption.

Strengths:

I see the main strength of this work in its combination of all levels of information (m/z, RT, and higher-order information on correlations between features) and using each of the types of information in a way that is appropriate for the measure. The most innovative aspect is using the Gromov-Wasserstein method to match the features based on distance matrices.

We thank the reviewer for acknowledging this strength of our proposed GromovMatcher method.

The authors of the paper identify two main shortcomings with previously established methods that attempt to match features from different experiments: a) all other methods require fine-tuning of user-defined parameters, and, more importantly, b) do not consider correlations between features. The main strength of the GromovMatcher is that it incorporates the information on distances between the features (in addition to also using m/z and RT).

Weaknesses:

The first, minor, weakness I could identify is that there seem not to be plenty of manually curated datasets that could be used for validation.

We thank the reviewer for raising this issue concerning manually curated validation data.

Manually curated datasets available for validation purposes are indeed scarce. This stems from the laborious nature of matching features across diverse studies, hence the need for automatic matching methods. Our future strategy involves further validation of the GromovMatcher approach as more data becomes accessible in EPIC and other cohorts.

The scarcity of real-life publicly available datasets that can be used for validation purpose is the reason why we conducted an extensive simulation study (main text “Validation on ground-truth data” and Appendix 3). It is notably thorough, arguably more comprehensive than previous validations, utilizes real-life untargeted data, and imitates situations where data originates from distinct untargeted metabolomics studies, complete with realistic noise parameters encompassing RT, mz, and feature intensities. Our validation study comprehensively explores the performance of GromovMatcher, M2S, and metabCombiner, including in challenging realistic settings where there is a nonlinear drift in retention times, varying levels of feature overlaps between studies, normalizations of feature intensities, as well as imbalances in the number of features and samples present in the studies being matched.

The second is also emphasized by the authors in the discussion. Namely, the method as it is set up now can be directly used only to compare two datasets.

This is indeed a limitation that is common to all three methods considered in this paper. However, all these methods, GromovMatcher, M2S, and metabCombiner, can still be used to compare and pool multiple datasets using a multi-step procedure. Namely, this can be done by designating a 'reference' dataset and aligning all studies to it one by one. We take this exact approach in our paper when aligning the CS, HCC, and PC studies of the EPIC data in positive mode (main text “Application to EPIC data”). Namely, the HCC and PC studies are both aligned to the CS study by running GromovMatcher twice, and after obtaining these matchings, our analysis is restricted to those features in HCC and PC that are present in the CS study.

After the reviewer’s comment, we have added an additional sensitivity analysis in Appendix 5, to compare the results produced by GromovMatcher depending on the choice of the reference study. Namely, setting the reference study to either the CS study or the HCC study, GromovMatcher identified 706 and 708 common features respectively, with an overlap of 640 features. This highlights that the choice of reference does matter to some extent. In our original analysis of the EPIC data, choosing CS as the reference was motivated by the fact that CS had the largest sample size (compared to HCC and PC) and a subset of features in HCC and PC were already matched by experts to the CS study which we could use for validation (see Loftfield et al. (2021). J Natl Cancer Inst.).

As mentioned in the discussion section of our manuscript, the recently proposed multimarginal Gromov-Wasserstein algorithm (Beier, F., Beinert, R., & Steidl, G. (2023). Information and Inference) could potentially allow multiple metabolomic studies to be matched using one optimization routine (e.g. without the designation of a ‘reference study’ for matching). We have not explored this possibility in depth yet as fast numerical methods for multimarginal GW are still in their infancy. Also, such multimarginal methods rely on the computation and storage of coupling or matching matrices that are tensors where the number of dimensions is equal to the number of datasets being matched. Therefore, multimarginal methods have large memory costs, which currently precludes their application for the matching of multiple metabolomics datasets.

Reviewer #2 (Recommendations For The Authors):

(1) I was struggling with the representation used in Figure 3a. The gray points overlayed over the green points on a straight line are difficult to visually quantify. I found that my eyes mainly focused on the pattern of the red dots.

Figure 3a has been modified to improve visual clarity. Namely we have consistently reordered the rows and columns of the coupling matrices such that the true positive matches (green points) are spatially separated from the false negative matches (red points). Now the fraction of true positive and false negative matches can be appreciated much more clearly by eye in Figure 3a.

(2) I would also like to add the caveat that I cannot judge whether the authors used the other two methods that they compare with GromovMatcher (the M2S and metabCombiner) optimally. But I also do not see any evidence that they did not. Hopefully one of the other reviewers can address that.

We appreciate the reviewer for highlighting the comparison of our approach GromovMatcher to the other existing methods M2S and MetabCombiner (mC). Both M2S and mC depend on tens of hyperparameters each with a discrete or continuous set of values that must be properly optimized to infer accurate matchings between dataset features. We detail in Appendix 2 how the hyperparameters of the M2S and mC methods are optimally tuned to achieve the best possible performance on the validation ground-truth data. Namely, both in the simulation study and on EPIC data, we grid-search over all important hyperparameters in the M2S and mC methods and choose those parameter combinations that result in the highest F1 score, averaged over 20 random trials. We remark that no such hyperparameter optimization was performed for our GromovMatcher method. As shown in Figures 3 and 4 of the main text, we find that GromovMatcher outperforms M2S and mC even in these cases when the hyperparameters of M2S and mC are tuned to predict optimal feature matchings.

Given the large combinatorial space of hyperparameter choices, we believe we have thoroughly tested the important hyperparameter combinations that users of M2S and mC would be likely to explore in their own research.

(3) Validation

(3a) The first validation is done on a split cord blood dataset. I could not clearly see from the paper how sensitive the result is to the dataset split.

We are grateful for the reviewer’s question and have included new experiments in Appendix 3 which show how the results of GromovMatcher, M2S, and MetabCombiner are affected by the dataset split. In our original manuscript, our validation ground-truth experiment began with an untargeted metabolomic dataset consisting of n = 499 samples and p = 4,712 metabolic features which is split equally into two datasets consisting of an equal number of samples n1 = n2 and an equal number of metabolic features p1 = p2. The features of these equal-sized datasets would then be matched by our method.

Now in Appendix 3 (Figs. 1-3) we show the sensitivity of all three alignment methods (GromovMatcher, M2S, and MetabCombiner) when we vary the fraction of samples in dataset 1 over dataset 2 given by n1/ n2, the overlap in shared features between both datasets, and the fraction of metabolic features in dataset 1 that are not present in dataset 2 which affects the feature sizes of both datasets p1/ p2. We find that all alignment methods are able to maintain a consistent precision and recall score when these three dataset split parameters are varied. GromovMatcher achieves a higher precision and recall than M2S and MetabCombiner for all choices of dataset split, agreeing with the validation experiment results from the main text (see main text Fig. 3). All three methods tested decrease in precision (without dropping in recall) when dataset 1 and dataset 2 contain an equal number of unshared features (e.g. when p1 = p2). Therefore, these sensitivity experiments in Appendix 3 show that our results in the main text are performed in the most challenging setting for the dataset split.

(3b) The second validation was done using a (scaled and centered) dataset of metabolics from cancer datasets from the EPIC cohort that was manually matched by an expert. Here the authors observe that metabCombiner has good precision, but lags in recall. And M2S has a very similar performance to GromovMatcher. The authors explain this by the fact that the drift in RT between the two experiments is mostly linear and thus does not affect the M2S performance. Can the authors find a different validation dataset where the drift in RT is not linear? If yes, it would be interesting to add it to the paper.

We thank the reviewer for raising this question. As mentioned above, curated validation datasets such as the EPIC study analyzed in our paper are very rare and we do not currently have a validation study with a nonlinear retention time drift.

Nevertheless, we performed an additional analysis of simulated data (reported in Appendix 2 – “M2S hyperparameter experiments” and Appendix 2 – Table 1) that demonstrates the decrease in M2S performance when the simulated drift is nonlinear. As presented in Appendix 2 – Table 1, in a low overlap setting with a linear drift which corresponds to the EPIC data, precision and recall were 0.831 and 0.934 respectively, instead of 0.769 and 0.905 in the main analysis where the drift was nonlinear.

Author Response

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

eLife assessment

This important study reports a novel mechanism linking DHODH inhibition-mediated pyrimidine nucleotide depletion to antigen presentation. Alternative means of inducing antigen presentation provide therapeutic opportunities to augment immune checkpoint blockade for cancer treatment. While the solid mechanistic data in vitro are compelling, in vivo assessments of the functional relevance of this mechanism are still incomplete.

Public Reviews:

We thank all Reviewers for their insightful comments and excellent suggestions.

Reviewer #1 (Public Review):

The manuscript by Mullen et al. investigated the gene expression changes in cancer cells treated with the DHODH inhibitor brequinar (BQ), to explore the therapeutic vulnerabilities induced by DHODH inhibition. The study found that BQ treatment causes upregulation of antigen presentation pathway (APP) genes and cell surface MHC class I expression, mechanistically which is mediated by the CDK9/PTEFb pathway triggered by pyrimidine nucleotide depletion.

No comment from authors

The combination of BQ and immune checkpoint therapy demonstrated a synergistic (or additive) anti-cancer effect against xenografted melanoma, suggesting the potential use of BQ and immune checkpoint blockade as a combination therapy in clinical therapeutics.

No comment from authors

The interesting findings in the present study include demonstrating a novel cellular response in cancer cells induced by DHODH inhibition. However, whether the increased antigen presentation by DHODH inhibition actually contributed to the potentiation of the efficacy of immune-check blockade (ICB) is not directly examined is the limitation of the study.

No comment from authors for preceding text, comment addresses the following text

Moreover, the mechanism of the increased antigen presentation pathway by pyrimidine depletion mediated by CDK9/PTEFb was not validated by genetic KD or KO targeting by CDK9/PTEFb pathways.

We appreciate this comment, and we would like to explain why we did not pursue these approaches. According to DepMap, CRISPR/Cas9-mediated knockout of CDK9 in cancer cell lines is almost universally deleterious, scoring as “essential” in 99.8% (1093/1095) of all cell lines tested (see Author response image 1 below). This makes sense, as P-TEFb is required for productive RNA polymerase II elongation of most mammalian genes. As such, it was not feasible to generate cell lines with stable genetic knockout of CDK9 to test our hypothesis.

While knockdown of CDK9 by RNA interference could support our results, DepMap data seems to indicate that RNAi-mediated knockdown of CDK9 is generally ineffective in silencing its activity, as this perturbation scored as “essential” in only 6.2% (44/710) of tested cell lines. This suggests that incomplete depletion of CDK9 will likely not be sufficient to block APP induction downstream of nucleotide depletion. Furthermore, RNAi-mediated depletion of CDK9 may trigger transcriptional changes in the cell by virtue of its many documented protein-protein interactions, and it would be difficult to establish a consistent “time zero” at which point CDK9 protein depletion is substantial but secondary effects of this have not yet occurred to a significant degree. These factors constitute major limitations of experiments using RNAi-mediated knockdown of CDK9.

Author response image 1.

Essentiality score from CRISPR and RNAi perturbation of CDK9 in cancer cell lines https://depmap.org/portal/gene/CDK9?tab=overview&dependency=RNAi_merged

At any rate, we provide evidence that three different inhibitors of CDK9 (flavopiridol, dinaciclib, and AT7519) all inhibit our effect of interest (Fig 4B). The same results were observed using a previously validated CDK9-directed proteolysis targeting chimera (PROTAC2), and this was reversed by addition of excess pomalidomide (Fig 4C), which correlated with the presence/absence of CDK9 on western blot under the exact same conditions (Fig 4D).

It is formally possible that all CDK9 inhibitors we tested are blocking BQ-mediated APP induction by some shared off-target mechanism (or perhaps by two or more different off-target mechanisms) AND this CDK9-independent target also happens to be degraded by PROTAC2. However, this would be an extraordinarily non-parsimonious explanation for our results, and so we contend that we have provided compelling evidence for the requirement of CDK9 for BQ-mediated APP induction.

Finally, high concentrations of BQ have been reported to show off-target effects, sensitizing cancer cells to ferroptosis, and the authors should discuss whether the dose used in the in vivo study reached the ferroptotic sensitizing dose or not.

We are intrigued by the results shown to us by Reviewer #1 in the linked preprint (Mishima et al 2022, https://doi.org/10.21203/rs.3.rs-2190326/v1). We have also observed in our unpublished data that very high concentrations of BQ (>150µM) cause loss of cell viability that is not rescued by uridine supplementation and that occurs even in DHODH knockout cells. This effect of high-dose BQ must be DHODH-independent. We also agree that Mishima et al provide compelling evidence that the ferroptosis-sensitizing effect of high-dose BQ treatment is due (at least in large part) to inhibition of FSP1.

Although we showed that DHODH is strongly inhibited in tumor cells in vivo (Fig 5C), we did not directly measure the concentration of BQ in the tumor or plasma. Sykes et al (PMID: 27641501) found that the maximum plasma concentration (Cmax) for [BQ]free following a single IP administration in C57Bl6/J mice (15mg/kg) is approximately 3µM, while the Cmax for [BQ]total was around 215µM. Because polar drug molecules bound to serum proteins (predominantly albumin) are not available to bind other targets, [BQ]free is the relevant parameter.

Given a Cmax for [BQ]free of 3µM and half-life of 12.0 hours, we estimate that the steady-state [BQ]free with daily IP injections at this dose is around 4µM. Since we used an administration schedule of 10mg/kg every 24 hours, we estimate that the steady-state plasma [BQ]free in our system was 2.67µM (assuming initial Cmax of 2µM and half-life of 12.0 hours).

To derive an upper-bound estimate for the Cmax of [BQ]free over the 12-day treatment period (Fig 5A-D), we will use the observed data for 15mg/kg dose, and we will assume that 1) there is no clearance of BQ whatsoever and 2) that [BQ]free increases linearly with increasing [BQ]total. This yields a maximum free BQ concentration of 12 x 3 = 36µM.

Therefore, we consider it very unlikely that plasma concentrations of free BQ in our experiment exceeded the lower limit of the ferroptosis-sensitizing dose range reported by Mishima et al. However, without direct pharmacokinetic analysis, we cannot say for sure what the maximal [BQ]free was under our experimental conditions.

Reviewer #2 (Public Review):

In their manuscript entitled "DHODH inhibition enhances the efficacy of immune checkpoint blockade by increasing cancer cell antigen presentation", Mullen et al. describe an interesting mechanism of inducing antigen presentation. The manuscript includes a series of experiments that demonstrate that blockade of pyrimidine synthesis with DHODH inhibitors (i.e. brequinar (BQ)) stimulates the expression of genes involved in antigen presentation. The authors provide evidence that BQ mediated induction of MHC is independent of interferon signaling. A subsequent targeted chemical screen yielded evidence that CDK9 is the critical downstream mediator that induces RNA Pol II pause release on antigen presentation genes to increase expression. Finally, the authors demonstrate that BQ elicits strong anti-tumor activity in vivo in syngeneic models, and that combination of BQ with immune checkpoint blockade (ICB) results in significant lifespan extension in the B16-F10 melanoma model. Overall, the manuscript uncovers an interesting and unexpected mechanism that influences antigen presentation and provides an avenue for pharmacological manipulation of MHC genes, which is therapeutically relevant in many cancers. However, a few key experiments are needed to ensure that the proposed mechanism is indeed functional in vivo.

The combination of DHODH inhibition with ICB reflects more of an additive response instead of a synergistic combination. Moreover, the temporal separation of BQ and ICB raises the question of whether the induction of antigen presentation with BQ is persistent during the course of delayed ICB treatment. To confidently conclude that induction of antigen presentation is a fundamental component of the in vivo response to DHODH inhibition, the authors should examine whether depletion of immune cells can reduce the therapeutic efficacy of BQ in vivo.

We concur with this assessment.

Moreover, they should examine whether BQ treatment induces antigen presentation in non-malignant cells and APCs to determine the cancer specificity.

Although we showed that this occurs in HEK-293T cells, we appreciate that this cell line is not representative of human cells of any organ system in vivo. So, we agree it is important to determine if DHODH inhibition induces antigen presentation in human tissues and professional antigen presenting cells, and this is an excellent focus for future studies.

However, it should also be noted that increased antigen presentation in non-malignant host tissues would not be expected to generate an autoimmune response, because host tissues likely lack strong neoantigens, and whatever immunogenic peptides they may have would likely be presented via MHC-I at baseline (i.e. even in the absence of DHODH inhibitor treatment), since all nucleated cells express MHC-I.

This argument is strongly supported by clinical experience/data, as DHODH inhibitors (leflunomide and teriflunomide) are commonly used to treat rheumatoid arthritis and multiple sclerosis. While the pathophysiology of these autoimmune syndromes is complex, it is thought that both diseases are driven by aberrant T-cell attack on host tissues, mediated by incorrect recognition of host antigens presented via MHC-I (as well as MHC-II) as “foreign.”

If increased antigen presentation in host tissues (downstream of DHODH inhibition) could lead to a de novo autoimmune response, then administration of DHODH inhibitors would be expected to exacerbate T-cell driven autoimmune disease rather than ameliorate it. Randomized controlled trials have consistently found that treatment with DHODH inhibitors leads to improvement of rheumatoid arthritis and multiple sclerosis symptoms, which is the opposite of what one would expect if DHODH inhibitors are causing de novo autoimmune reactions in human patients.

Finally, although the authors show that DHODH inhibition induces expression of both MHC-I and MHC-II genes at the RNA level, only MHC-I is validated by flow cytometry given the importance of MHC-II expression on epithelial cancers, including melanoma, MHC-II should be validated as well.

We fully agree with this statement. We attempted to quantify cell surface MHC-II expression by FACS using the same method as for MHC-I (Figs 1G-H, 2D, and 3F). We did not detect cell surface MHC-II in any of our cancer cell lines, despite the use of high-dose interferon gamma and other stimulants (which robustly increase MHC-II mRNA in our system) in an attempt to induce expression. However, because we did not use cells known to express MHC-II as a positive control (e.g. B-cell leukemia cell lines or primary splenocytes), we do not know if our results are due to some technical failure (perhaps related to our protocol/reagents) or if they reflect a true absence of cell surface MHC-II in our cell lines.

If the latter is true, that implies that either 1) MHC-II mRNA is not translated or 2) that it is translated, but our cancer cell lines lack one or more elements of the machinery required for MHC-II antigen presentation.

In any case, it is important to determine if DHODH inhibition increases MHC-II at the cell surface of cancer cells using appropriate positive and negative controls, as this could have important implications for cancer immunotherapy.

[As a minor point, melanoma is not an epithelial cancer, as it is derived from neural crest lineage cells (melanocytes)]

Overall, the paper is clearly written and presented. With the additional experiments described above, especially in vivo, this manuscript would provide a strong contribution to the field of antigen presentation in cancer. The distinct mechanisms by which DHODH inhibition induces antigen presentation will also set the stage for future exploration into alternative methods of antigen induction.

Reviewer #3 (Public Review):

Mullen et al present an important study describing how DHODH inhibition enhances efficacy of immune checkpoint blockade by increasing cell surface expression of MHC I in cancer cells. DHODH inhibitors have been used in the clinic for many years to treat patients with rheumatoid arthritis and there has been a growing interest in repurposing these inhibitors as anti-cancer drugs. In this manuscript, the Singh group build on their previous work defining combinatorial strategies with DHODH inhibitors to improve efficacy. The authors identify an increase in expression of genes involved in the antigen presentation pathway and MHC I after BQ treatment and they narrow the mechanism to be strictly pyrimidine and CDK9/P-TEFb dependent. The authors rationalize that increased MHC I expression induced by DHODH inhibition might favor efficacy of dual immune checkpoint blockade. This combinatorial treatment prolonged survival in an immunocompetent B16F10 melanoma model.

[No comment from authors]

Previous studies have shown that DHODH inhibitors can increase expression of innate immunity-related genes but the role of DHODH and pyrimidine nucleotides in antigen presentation has not been previously reported. A strength of the manuscript is the use of multiple controls across a panel of cell lines to exclude off-target effects and to confirm that effects are exclusively dependent on pyrimidine depletion. Overall, the authors do a thorough characterization of the mechanism that mediates MHC I upregulation using multiple strategies. Furthermore, the in vivo studies provide solid evidence for combining DHODH inhibitors with immune checkpoint blockade.

No comment from authors

However, despite the use of multiple cell lines, most experiments are only performed in one cell line, and it is hard to understand why particular gene sets, cell lines or time points are selected for each experiment. It would be beneficial to standardize experimental conditions and confirm the most relevant findings in multiple cell lines.

We appreciate this comment, and we understand how the use of various cell lines may seem puzzling. We would like to explain how our cell line panel evolved over the course of the study. Our first indication that BQ caused APP upregulation came from transcriptomics experiments (Figs 1A-D, S1A) performed as part of a previous study investigating BQ resistance (Mullen et al, 2023 Cancer Letters). In that study, we used CFPAC-1 as a model for BQ sensitivity and S2-013 as a model for BQ resistance. We did RNA sequencing +/- BQ in these cell lines to look for gene expression patterns that might underlie resistance/sensitivity to BQ. When analyzing this data, we serendipitously discovered the APP/MHC phenomenon, which gave rise to the present study.

Our next step was to extend these findings to cancer cell lines of other histologies, and we prioritized cell lines derived from common cancer types for which immunotherapy (specifically ICB) are clinically approved. This is why A549 (lung adenocarcinoma), HCT116 (colorectal adenocarcinoma), A375 (cutaneous melanoma), and MDA-MB-231 (triple-negative breast cancer) cell lines were introduced.

Because PDAC is considered to have an especially “immune-cold” tumor microenvironment, we reasoned that even dramatically increasing cancer cell antigen presentation may be insufficient to elicit an effective anti-tumor immune response in vivo. So we shifted our focus towards melanoma, because a subset of melanoma patients is very responsive to ICB and loss of antigen presentation (by direct silencing or homozygous loss-of-function mutations in MHC-I components such as B2M, or by functional loss of IFN-JAK1/2-STAT signaling) has been shown to mediate ICB resistance in human melanoma patients. This is why we extended our findings to B16F10 murine melanoma cells, intending to use them for in vivo studies with syngeneic immunocompetent recipient mice.

The PDAC cell line MiaPaCa2 was introduced because a collaborator at our institution (Amar Natarajan) happened to have IKK2 knockout MiaPaCa2 cells, which allowed us to genetically validate our inhibitor results showing that IKK1 and IKK2 (crucial effectors for NF-kB signaling) are dispensable for our effect of interest.

Ultimately, realizing that our results spanned various human and murine cell lines, we chose to use HEK-293T cells to validate the general applicability of our findings to proliferating cells in 2D culture, since HEK-293T cells (compared to our cancer cell lines) have relatively few genetic idiosyncrasies and express MHC-I at baseline.

The differential in vivo survival depending on dosing schedule is interesting. However, this section could be strengthened with a more thorough evaluation of the tumors at endpoint.

Overall, this is an interesting manuscript proposing a mechanistic link between pyrimidine depletion and MHC I expression and a novel therapeutic strategy combining DHODH inhibitors with dual checkpoint blockade. These results might be relevant for the clinical development of DHODH inhibitors in the treatment of solid tumors, a setting where these inhibitors have not shown optimal efficacy yet.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The main issue is that it did not directly examine whether the increased antigen presentation by DHODH inhibition contributed to the potentiation of the efficacy of immune-check blockade (ICB). The additional effect of BQ in the xenograft tumor study was not examined to determine if it was due to increased antigen presentation toward the cancer cells or due to merely cell cycle arrest effect by pyrimidine depletion in the tumor cells. The different administration timing of ICB with BQ treatment (Fig 5E) would not be sufficient to answer this issue.

We agree with this assessment and, and we believe the experiment proposed by Reviewer #2 below (comparing the efficacy of BQ in Rag-null versus immunocompetent recipients) would address this question directly. We also think that using a more immunogenic cell line for this experiment (such as B16F10 transduced with ovalbumin or some other strong neoantigen) would be useful given the poor immunogenicity and lack of any defined strong neoantigen in B16F10 cells. An orthogonal approach would be to engraft cancer cells with or without B2M knockout into immunocompetent recipient mice (+/- BQ treatment) to further implicate MHC-I and antigen presentation. These questions will be addressed in future studies.

(2) Additionally, in the in vivo study, the increase in surface MHC1 in the protein level in by BQ treatment was not examined in the tumor samples, and it was not confirmed whether increased antigen presentation by BQ treatment actually promoted an anti-cancer immune response in immune cells. To support the story presented in the study, these data would be necessary.

We attempted to show this by immunohistochemistry, but unfortunately the anti-H2-Db antibody that we obtained for this purpose did not have satisfactory performance to assess this in our tissue samples harvested at necropsy.

(3) The mechanism of the increased antigen presentation pathway by pyrimidine depletion mediated by CDK9/PTEFb was not validated by genetic KD or KO targeting by CDK9/PTEFb pathways. In general, results only by the inhibitor assay have a limitation of off-target effects.

Please see our above reply to Reviewer #1 comment making this same point, where we spell out our rationale for not pursuing these experiments.

(4) High concentrations of BQ (> 50 uM) have been reported to show off-target effects, sensitizing cancer cells to ferroptosis, an iron-mediated lipid peroxidation-dependent cell death, independent of DHODH inhibition (https://www.researchsquare.com/article/rs-2190326/v1). It would be needed to discuss whether the dose used in the in vivo study reached the ferroptotic sensitizing dose or not.

Please see our above reply to Reviewer #1 comment making this same point, where we explain why we are very confident that the BQ dose administered in our animal experiments was far below the minimum reported BQ dose required to sensitize cancer cells to ferroptosis in vitro.

Reviewer #2 (Recommendations For The Authors):

Major Points

(1) According to the proposed model, BQ mediated induction of antigen presentation is a contributing factor to the efficacy of this therapeutic strategy. If this is true, then depletion of immune cells should reduce the therapeutic efficacy of BQ in vivo. The authors should perform the B16-F10 transplant experiments in either Rag null mice (if available) or with CD8/CD4 depletion. The expectation would be that T cell depletion (or MHC loss with genetic manipulation) should reduce the efficacy of BQ treatment. Absent this critical experiment, it is difficult to confidently conclude that induction of antigen presentation is a fundamental component of the in vivo response to DHODH inhibition.

We agree with this assessment and the proposed experiment comparing the response in Rag-null versus immunocompetent recipients. We also think that using a more immunogenic cell line for this experiment (such as B16F10 transduced with ovalbumin or some other strong neoantigen) would be useful given the poor immunogenicity and lack of any defined strong neoantigen in B16F10 cells. An orthogonal approach would be to engraft cancer cells with or without B2M knockout into immunocompetent recipient mice (+/- BQ treatment) to further implicate MHC-I and antigen presentation. These questions will be addressed in future studies.

(2) Does BQ treatment induce antigen presentation in non-malignant cells? APCs? If the induction of antigen presentation is not cancer specific and related to a pyrimidine depletion stress response, then there is a possibility that healthy tissues will also exhibit a similar phenotype, raising concerns about the specificity of a de novo immune response. The authors should examine antigen presentation genes in healthy tissues treated with BQ.

We agree it is important to examine if our findings regarding nucleotide depletion and antigen presentation are true of APCs and other non-transformed cells, but we are not so concerned about the possibility of raising an immune response against non-malignant host tissues, as explained above. We have reproduced the relevant section below:

“However, it should also be noted that increased antigen presentation in non-malignant host tissues would not be expected to generate an autoimmune response, because host tissues likely lack strong neoantigens, and whatever immunogenic peptides they may have would likely be presented via MHC-I at baseline, since all nucleated cells express MHC-I.

This argument is strongly supported by clinical experience/data, as DHODH inhibitors (leflunomide and teriflunomide) are commonly used to treat rheumatoid arthritis and multiple sclerosis. While the pathophysiology of these autoimmune syndromes is complex, it is thought that both diseases are driven by aberrant T-cell attack on host tissues, mediated by incorrect recognition of host antigens presented via MHC-I (as well as MHC-II) as “foreign.”

If increased antigen presentation in host tissues (downstream of DHODH inhibition) could lead to a de novo autoimmune response, then administration of DHODH inhibitors would be expected to exacerbate T-cell driven autoimmune disease rather than ameliorate it. Randomized controlled trials have consistently found that treatment with DHODH inhibitors leads to improvement of rheumatoid arthritis and multiple sclerosis symptoms, which is the opposite of what one would expect if DHODH inhibitors are causing de novo autoimmune reactions in human patients.”

(3) In the title, the authors claim that DHODH enhances the efficacy of ICB. However, the experiment shown in Figure 5D does not demonstrate this. The Kaplan Meier curves reflect more of an additive response versus a synergistic combination. Furthermore, the concurrent treatment of BQ and ICB seems to inhibit the efficacy of ICB due to BQ toxicity in immune cells. This result seems to contradict the title.

We do not agree with this assessment. Given that the effect of dual ICB alone was very marginal, while the effect of BQ monotherapy was quite marked, we cannot conclude from Fig 5 that BQ treatment inhibited ICB efficacy due to immune suppression.

(4) Related to Point 3, the temporal separation of BQ and ICB raises the question of whether the induction of antigen presentation with BQ is persistent during the course of delayed ICB treatment. One explanation for the results is that BQ treatment reduces tumor burden, and then a subsequent course of ICB also reduces tumor burden but not that the two therapies are functioning in synergy. To address this, the authors should measure the duration of BQ mediated induction of antigen presentation after stopping treatment.

We agree that the alternative explanation proposed by Reviewer #2 is possible and we appreciate the suggestion to test the stability of APP induction after stopping BQ treatment.

(5) In Figure 1, the authors show that DHODH inhibition induces expression of both MHC-I and MHC-II genes at the RNA level. However, they only validate MHC-I by flow cytometry. A simple experiment to evaluate the effect of BQ treatment on MHC-II surface expression would provide important additional mechanistic insight into the immunomodulatory effects of DHODH inhibition, especially given recent literature reinforcing the importance of MHC-II expression on epithelial cancers, including melanoma (Oliveira et al. Nature 2022).

We fully agree with this statement. We attempted to quantify cell surface MHC-II expression by FACS using the same method as for MHC-I (Figs 1G-H, 2D, and 3F). We did not detect cell surface MHC-II in any of our cancer cell lines, despite the use of high-dose interferon gamma and other stimulants (which robustly increase MHC-II mRNA in our system) in an attempt to induce expression. However, because we did not use cells known to express MHC-II as a positive control (e.g. B-cell leukemia cell lines or primary splenocytes), we do not know if our results are due to some technical failure (perhaps related to our protocol/reagents) or if they reflect a true absence of cell surface MHC-II in our cell lines.

If the latter is true, that implies that either 1) MHC-II mRNA is not translated or 2) that it is translated, but our cancer cell lines lack one or more elements of the machinery required for MHC-II antigen presentation.

In any case, it is important to determine if DHODH inhibition increases MHC-II at the cell surface of cancer cells using appropriate positive and negative controls, as this could have important implications for cancer immunotherapy.

[As a minor point, melanoma is not an epithelial cancer, as it is derived from neural crest lineage cells (melanocytes)]

Minor Points

(1) The authors show ChIP-seq tracks from Tan et al. for HLA-B. However, given the pervasive effect of Ter treatment across many HLA genes, the authors should either show tracks at additional loci, or provide a heatmap of read density across more loci. This would substantiate the mechanistic claim that RNA Pol II occupancy and activity across antigen presentation genes is the major driver of response to DHODH inhibition as opposed to mRNA stabilization/increased translation.

We appreciate this suggestion. We have changed Fig 4 by replacing the HLA-B track (old Fig 4E) with a representation of fold change (Ter/DMSO) in Pol II occupancy versus fold change (Ter/DMSO) in mRNA abundance for 23 relevant genes (new Fig 4G); both of these datasets were obtained from the Tan et al manuscript. This new figure panel (Fig 4G) also shows linear regression analysis demonstrating that Pol II occupancy and mRNA expression are significantly correlated for APP genes. While we recognize that this data in itself is not formal proof of our hypothesis, it does strongly support the notion that increased transcription is responsible for the increased mRNA abundance of APP genes that we have observed.

(2) A compelling way to demonstrate a change in antigen presentation is through mass spectrometry based immunopeptidomics. Performing immunopeptidomic analysis of BQ treated cell lines would provide substantial mechanistic insight into the outcome of BQ treatment. While this approach may be outside the scope of the current work, the authors should speculate on how this treatment may specifically alter the antigenic landscape where future directions would include empirical immunopeptidomics measurements.

We fully agree with this comment. While the abundance of cancer cell surface MHC-I is an important factor for anticancer immunity, another crucial factor is the identity of peptides that are presented. Treatments that cause presentation of more immunogenic peptides can enhance T-cell recognition even in the absence of a relative change in cell surface MHC-I abundance.

While we did not perform the immunopeptidomics experiments described, we can offer some speculation regarding this comment. As shown in Fig 1D-E, transcriptomics experiments suggest that immunoproteasome subunits (PSMB8, PSMB9, PSMB10) are upregulated upon DHODH inhibition. If this change in mRNA levels translates into greater immunoproteasome activity (which was not tested in our study), this would be expected to alter the repertoire of peptides available for presentation and could thereby change the immunopeptidome.

However, this hypothesis requires direct testing, and we hope future studies will delineate the effects of DHODH inhibition and other cancer therapies on the immunopeptidome, as this area of research will have important clinical implications.

(3) While the signaling through CDK9 seems convincing, it still does not provide a mechanistic link between depleted pyrimidines and CDK9 activity. The authors should speculate on the mechanism that signals to CDK9.

We agree with the assessment. A mechanistic link between depleted pyrimidines and CDK9 activity will be a subject of future studies.

(4) Related to minor point 2, the authors should consider a genetic approach to confirm the importance of CDK9. While the pharmacological approach, including multiple mechanistically distinct CDK9 inhibitors provides strong evidence, an additional experiment with genetic depletion of CDK9 (CRISPR KO, shRNA, etc) would provide compelling mechanistic confirmation.

Reviewer #1 raised this very same point, and we agree. Please see our reply to Reviewer #1, which details why we did not pursue this approach and argues that the evidence we present is compelling even in absence of genetic manipulation.

Additionally, please see the new Fig 4E and 4F, which is a repeat of Fig 4B using HCT116 cells. Figure 4E shows that, in this cell line, CDK9 inhibitors (flavopiridol, dinaciclib, and AT7519) block BQ-mediated APP induction, while PROTAC2 does not. Figure 4F shows that (for reasons we cannot fully explain) PROTAC2 does not lead to CDK9 degradation in HCT116 cells. This data strongly implicates CDK9, because it excludes a CDK9-degradation-independent effect of PROTAC2.

(5) Figure 2B needs a legend.

Thank you for pointing this out. We have added a legend to Fig 2B.

(6) The authors should comment in the discussion on how this strategy may be particularly useful in patients harboring genetic or epigenetic loss of interferon signaling, a known mechanism of ICB resistance. Perhaps DHODH inhibition could rescue MHC expression in cells that are deficient in interferon sensing.

Thank you for this suggestion! We have amended the Discussion section to mention this important point. Please see paragraph 2 of the revised Discussion section where we have added the following text:

“Because BQ-mediated APP induction does not require interferon signaling, this strategy may have particular relevance for clinical scenarios in which tumor antigen presentation is dampened by the loss or silencing of cancer cell interferon signaling, which has been demonstrated to confer both intrinsic and acquired ICB resistance in human melanoma patients.”

Reviewer #3 (Recommendations For The Authors):

The authors present convincing evidence of the mechanism by which pyrimidine nucleotides regulate MHC I levels and about the potential of combining DHODH inhibitors with dual immune checkpoint blockade (ICB). This is an interesting paper given the clinical relevance of DHODH inhibitors. The studies raise some questions, and some points might need clarifying as below:

• In Figure 2C, why do the authors focus on these two genes in the uridine rescue? These are important genes mediating antigen presentation, but it might be more interesting to see how H2-Db and H2-Kb expression correlate with the protein data shown in Fig 2D. Fig. 2C-2D is a relevant control, so it would be important to validate in a different cancer cell line (e.g. one of the PDAC cell lines used for the RNAseq).

We appreciate this comment. Although Fig 3C shows that BQ-induced expression of H2-Db, H2-Kb, and B2m is reversed by uridine (in B16F10 cells), we recognize that this was not the best placement for this data, as it can easily be overlooked here since uridine reversal is not the main point of Fig 3C. We have left Fig 3C as is, because we think that the uridine reversal demonstrated in that panel serves as a good internal positive control for reversal of BQ-mediated APP induction in that experiment.

We have repeated the experiments shown in the original Fig 2C and substituted the original Fig 2C with a new Fig 2C and Fig S2B, which show both Tap1 and Nlrc5 as well as H2-Db, H2-Kb, and B2m after treatment with either BQ (new Fig 2C) or teriflunomide (new Fig S2B). The original Fig S2B is now Fig S2C, and it shows that uridine has no effect on the expression of any of the genes assayed in the new Fig 2C or S2B.

The reversibility of cell surface MHC-I induction was also validated in HCT116 cells (Fig 3F). We included the uridine reversal in Fig 3F to avoid duplicating the control and BQ FACS data in multiple panels.

We have also added the qPCR data for HCT116 cells showing this same phenotype (at the mRNA level), which is the new Fig S2D.

We decided to prioritize HCT116 cells for our mechanistic studies (Figures S2D, S4A, and 4E-F) because previous reports indicate that it is diploid and therefore less genetically deranged compared to our other cancer cell lines.

• Figure 2F shows an elegant experiment to discard off-target effects related to cell death and to confirm that the increased MHC I expression is uniquely dependent on pyrimidines. DHODH has recently been involved in ferroptosis, a highly immunogenic type of cell death. What are the authors´ thoughts on BQ-induced ferroptosis as a possible contributor to the effects of ICB? Does BQ + ferroptosis inhibitor (ferrostatin) affect cell surface MHC I and/or expression of antigen processing genes?

The potential role of DHODH in ferroptosis protection (Mao et al 2021) has important implications, so we are glad that multiple reviewers raised questions concerning ferroptosis. We did not directly test the effect of ferroptosis inducing agents (with or without BQ) on MHC-I/APP expression, but that is certainly a worthwhile line of investigation.

The DHODH/ferroptosis issue is complicated by a study pointed out by Reviewer #1 that challenges the role of DHODH inhibition in BQ-mediated ferroptosis sensitization (Mishima et al, 2022). This study argues that high-dose BQ treatment causes FSP1 inhibition, and this underlies the effect of BQ on the cellular response to ferroptosis-inducing agents.

Regardless of whether BQ-induced ferroptosis-sensitization is dependent on DHODH, FSP1, or some other factor, the Mao and Mishima studies agree that a relatively high dose of BQ is required to observe these effects (100-200µM for most cell lines and >50µM even in the most ferroptosis-sensitive cell lines). As we explained above, we consider it very unlikely that the in vivo BQ exposure in our experiments (Fig 5) was high enough to cause significant ferroptosis, especially in the absence of any dedicated ferroptosis-inducing agent (which is typically required to cause ferroptosis even in the presence of high-dose BQ).

• The authors nail down the mechanism to CDK9 (Fig 4). However, all these experiments are performed in 293T cells. I would like to see a repeat of Fig. 4B in a cancer cell line (either PDAC or B16). Also, does BQ have any effect on CDK9 expression/protein levels?

We have added two figure panels that address this comment (new Fig 4E and 4F). Figure 4E (which is a repeat of Fig 4B with HCT116 cells) shows that CDK9 inhibitors (flavopiridol, AT7519, and dinaciclib) reverse BQ-mediated APP induction in HCT116 cells (this agrees with Fig S4A showing that flavopiridol reverses MHC induction by various nucleotide synthesis inhibitors in this cell line), but PROTAC2 does not. Figure 4F shows that PROTAC2 (for reasons we cannot explain) does not cause CDK9 degradation in HCT116 cells. This adds further support to our thesis that CDK9 is a critical mediator of BQ-mediated APP induction (because how else can this pattern of results be explained?). The text of the Results section has been amended to reflect this.

We chose to use HCT116 cells for this repeat experiment 1) to align with Fig S4A and 2) because, as previously mentioned, we consider HCT116 to be a good cell line for mechanistic studies because of its relative lack of idiosyncratic genetic features (compared to CFPAC-1, for example, which was derived from a patient with cystic fibrosis).

• What are the differences in tumor size for the experiment shown in Figure 5E? What about tumor cell death in the ICB vs. BQ+ICB groups?

Because this was a survival assay, direct comparisons of tumor volumes between groups was not possible at later time points, since mice that die or have to be euthanized are removed from their experimental group, which lowers the average group tumor burden at subsequent time points. Although tumor volume was the most common euthanasia criteria reached, a subset of mice were either found dead or had to be euthanized for other reasons attributed to their tumor burden (moribund state, inability to ambulate or stand, persistent bleeding from tumor ulceration, severe loss of body mass, etc.). This confounds any comparison of endpoint measurements (such as immunohistochemical quantification of tumor cell death markers, T-cell markers, etc.).

• The different response in the concurrent vs delayed treatment is very interesting. The authors suggest two possible mechanisms to explain this: "1) Concurrent BQ dampens the initial anticancer immune response generated by dual ICB, or b) cancer cell MHC-I and related genes are not maximally upregulated at the time of ICB administration with concurrent treatment". However, and despite the caveat of comparing the in vitro to the in vivo setting, Fig 2D shows upregulation of MHC I already at 24h of treatment in B16 cells. Have the authors checked T cell infiltration in the concurrent and delayed treatment setting?

For the same reasons described in response to the preceding comment, tumors harvested upon mouse death/euthanasia from our survival experiment were not suitable for cross-cohort comparison of tumor endpoint measurements. An additional experiment in which mice are necropsied at a prespecified time point (before any mice have died or reached euthanasia criteria, as in the experiment for Fig 5A-D) would be required to answer this question.

• Page 5, line 181 -do the authors mean "nucleotide salvage inhibitors" instead of "synthesis"?

We believe the reviewer is referring to the following sentence:

“The other drugs screened included nucleotide synthesis inhibitors (5-fluorouracil, methotrexate, gemcitabine, and hydroxyurea), DNA damage inducers (oxaliplatin, irinotecan, and cytarabine), a microtubule targeting drug (paclitaxel), a DNA methylation inhibitor (azacytidine), and other small molecule inhibitors (Fig 2F).”

In this context, we believe our use of “synthesis” instead of “salvage” is correct, because methotrexate and 5-FU inhibit thymidylate synthase (which mediates de novo dTTP synthesis), while gemcitabine and hydroxyurea inhibit ribonucleotide reductase (which mediates de novo synthesis of all dNTPs).

Author Response

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

eLife assessment

This study, utilizing CITE-Seq to explore CML, is considered a useful contribution to our understanding of treatment response. However, the reviewers express concern about the incomplete evidence due to the small sample size and recommend addressing these limitations. Strengthening the study with additional patient samples and validation measures would enhance its significance.

We thank the editors for the assessment of our manuscript. In view of the comments of the three reviewers, we have increased the number of CML patient samples analyzed to confirm all the major findings included in the manuscript. In total, more than 80 patient samples across different approaches have now been analyzed and incorporated in the revised manuscript.

To the best of our knowledge, this is the first single cell multiomics report in CML and differs substantially from the recent single cell omics-based reports where single modalities were measured one at a time (Krishnan et al., 2023; Patel et al., 2022). Thus, the sc-multiomic investigation of LSCs and HSCs from the same patient addresses a major gap in the field towards managing efficacy and toxicity of TKI treatment by enumerating CD26+CD35- LSCs and CD26-CD35+ HSCs burden and their ratio at diagnosis vs. 3 months of therapy. The findings suggest design of a simpler and cheaper FACS assay to simultaneously stratify CML patients for TKI efficacy as well as hematologic toxicity.

Reviewer 1:

Summary:

This manuscript by Warfvinge et al. reports the results of CITE-seq to generate singlecell multi-omics maps from BM CD34+ and CD34+CD38- cells from nine CML patients at diagnosis. Patients were retrospectively stratified by molecular response after 12 months of TKI therapy using European Leukemia Net (ELN) recommendations. They demonstrate heterogeneity of stem and progenitor cell composition at diagnosis, and show that compared to optimal responders, patients with treatment failure after 12 months of therapy demonstrate increased frequency of molecularly defined primitive cells at diagnosis. These results were validated by deconvolution of an independent previously published dataset of bulk transcriptomes from 59 CML patients. They further applied a BCR-ABL-associated gene signature to classify primitive Lin-CD34+CD38- stem cells as BCR:ABL+ and BCR:ABL-. They identified variability in the ratio of leukemic to non-leukemic primitive cells between patients, showed differences in the expression of cell surface markers, and determined that a combination of CD26 and CD35 cell surface markers could be used to prospectively isolate the two populations. The relative proportion of CD26-CD35+ (BCR:ABL-) primitive stem cells was higher in optimal responders compared to treatment failures, both at diagnosis and following 3 months of TKI therapy.

Strengths:

The studies are carefully conducted and the results are very clearly presented. The data generated will be a valuable resource for further studies. The strengths of this study are the application of single-cell multi-omics using CITE-Seq to study individual variations in stem and progenitor clusters at diagnosis that are associated with good versus poor outcomes in response to TKI treatment. These results were confirmed by deconvolution of a historical bulk RNAseq data set. Moreover, they are also consistent with a recent report from Krishnan et al. and are a useful confirmation of those results. The major new contribution of this study is the use of gene expression profiles to distinguish BCRABL+ and BCR-ABL- populations within CML primitive stem cell clusters and then applying antibody-derived tag (ADT) data to define molecularly identified BCR:ABL+ and BCR-ABL- primitive cells by expression of surface markers. This approach allowed them to show an association between the ratio of BCR-ABL+ vs BCR-ABL- primitive cells and TKI response and study dynamic changes in these populations following short-term TKI treatment.

Weaknesses:

One of the limitations of the study is the small number of samples employed, which is insufficient to make associations with outcomes with confidence. Although the authors discuss the potential heterogeneity of primitive stem, they do not directly address the heterogeneity of hematopoietic potential or response to TKI treatment in the results presented. Another limitation is that the BCR-ABL + versus BCR-ABL- status of cells was not confirmed by direct sequencing for BCR-ABL. The BCR-ABL status of cells sorted based on CD26 and CD35 was evaluated in only two samples. We also note that the surface markers identified were previously reported by the same authors using different single-cell approaches, which limits the novelty of the findings. It will be important to determine whether the GEP and surface markers identified here are able to distinguish BCR-ABL+ and BCR-ABL- primitive stem cells later in the course of TKI treatment. Finally, although the authors do describe differential gene expression between CML and normal, BCR:ABL+ and BCR:ABL-, primitive stem cells they have not as yet taken the opportunity to use these findings to address questions regarding biological mechanisms related to CML LSC that impact on TKI response and outcomes.

Reviewer #1 (Recommendations For The Authors):

Minor comment: Fig 4 legend -E and F should be C and D.

We thank the reviewer for positive assessment of our work. Here, we highlight the updates in the revised manuscript considering the feedback received.

Minor comment: Fig 4 legend -E and F should be C and D.

We have edited the revised manuscript accordingly

One of the limitations of the study is the small number of samples employed, which is insufficient to make associations with outcomes with confidence.

Although we performed CITE-seq for 9 CML patient samples at diagnosis, we extended our investigations to include additional samples (e.g., largescale deconvolution analysis of samples, Fig 3 C-E, qPCR for BCR::ABL1 status, Fig. 6A, and the ratio between CD35+ and CD26+ populations at diagnosis and during TKI therapy, Fig. 6C-D) as described in the manuscript.

In comparison to a scRNA-seq, multiomic CITE-seq involves preparation and sequencing of separate libraries corresponding to RNA and ADTs thereby being even more resource demanding limiting our capacity to process an extensive number of patient samples. To confirm our findings in a larger cohort we have therefore adopted a computational deconvolution approach, CIBERSORT to analyze a larger number of independent samples (n=59). This reflects a growing, sustainable trend to study larger number of patients in face of still prohibitively expensive but potentially insightful scomics approaches (For example, please see Zeng et al, A cellular hierarchy framework for understanding heterogeneity and predicting drug response in acute myeloid leukemia, Nature Medicine, 2022).

However, in view of the comment, we have now substantially increased the number of analyzed patients in the revised manuscript. These include increased number of patient samples to investigate the ratio between CD35 and CD26 marked populations at diagnosis, and 3 months of TKI therapy (from n=8 to n=12 with now 6 optimal responders and 5 treatment failure at diagnosis and after TKI therapy), qPCR for BCR::ABL1 expression status at diagnosis (from n=3 to n=9) , and followed up the BCR::ABL1 expression in three additional samples after TKI therapy. Moreover, we examined the CD26 and CD35 marked populations for expression of GAS2, one of our top candidate LSC signature genes in three additional samples at diagnosis and at 3m follow up. Thus, >80 patient samples across different approaches have been analyzed to strengthen all major conclusions of the study.

We emphasize that we were cautious in generalizing the observation obtained from any one approach and sought to confirm any major finding using at least one complementary method. As an example, although CITE-seq (n=9) showed altered frequency of all cell clusters between optimal and poor responders (Fig. 3B), we refrained from generalizing because our independent large-scale computational deconvolution analysis (n=59) only substantiated the altered proportion of primitive and myeloid cell clusters (Fig. 3E).

Although the authors discuss the potential heterogeneity of primitive stem, they do not directly address the heterogeneity of hematopoietic potential or response to TKI treatment in the results presented.

Thanks for noting the discussion on heterogeneity of the primitive stem cells. As described in the original manuscript, the figure 6 D-E showed a relationship between heterogeneity and TKI therapy response. The results showed that CD35+/CD26+ ratio within the HSC fraction associated with this therapy response. We have now increased the number of patient samples analyzed and present the updated results in the revised manuscript (now figure 6 C-D). These observations set the stage for assessing whether long term therapy outcome can also be influenced by heterogeneity at diagnosis.

We have shown the hematopoietic potential of HSCs marked by CD35 expression in an independent parallel study and therefore only mentioned it concisely in the current manuscript. A combination of scRNA-seq, scATAC-seq and cell surface proteomics showed CD35+ cells at the apex of healthy human hematopoiesis, containing an HSCspecific epigenetic signature and molecular program, as well as possessing self-renewal capacity and multilineage reconstitution in vivo and vitro. The preprint is available as Sommarin et al. ‘Single-cell multiomics reveals distinct cell states at the top of the human hematopoietic hierarchy’, Biorxiv; https://www.biorxiv.org/content/10.1101/2021.04.01.437998v2.full

We also note that the surface markers identified were previously reported by the same authors using different single-cell approaches, which limits the novelty of the findings.

Our current manuscript is indeed a continuation of and builds onto our previous paper (Warfvinge R et al. Blood, 2017). In contrast to our previous report which was limited to examination of only 96 genes per cell, CITE-seq allowed us to examine the molecular program of cells using unbiased global gene expression profiling. Finally, although CD26 appears, once again as a reliable marker of BCR::ABL1+ primitive cells, CD35 emerges as a novel and previously undescribed marker of BCR::ABL1- residual stem cells. A combination of CD35 and CD26 allowed us to efficiently distinguish between the two populations housed within the Lin-34+38/low stem cell immunophenotype.

Another limitation is that the BCR-ABL + versus BCR-ABL- status of cells was not confirmed by direct sequencing for BCR-ABL. The BCR-ABL status of cells sorted based on CD26 and CD35 was evaluated in only two samples

Single cell detection of fusion transcripts is challenging with low detection sensitivity in single cell RNA-seq as has been noted previously (Krishnan et al. Blood, 2023, Giustacchini et al. Nature Medicine, 2017, Rodriguez-Meira et al. Molecular Cell, 2019). However, this is likely to change with the inclusion of targetspecific probes in scRNA-seq library preparation protocols. Nonetheless, in view of the comment, we have included more patient samples (from the previous n=3 to current n=10 (including TKI treated samples) for direct assessment of BCR-ABL1 status by qPCR analysis; the updated results are included in the revised manuscript (Figure 6A).

It will be important to determine whether the GEP and surface markers identified here are able to distinguish BCR-ABL+ and BCR-ABL- primitive stem cells later in the course of TKI treatment.

We performed qPCR to check for BCR::ABL1 status, and the level of GAS2, one of the top genes expressed in CML cells within CD26+ and CD35+ cells at diagnosis and following 3 months of TKI therapy. The results showed that while CD26+ are BCR::ABL1+, the CD35+ cells are BCR::ABL1- at both time points. Moreover, the expression of LSC-specific gene, GAS2 was specific to BCR::ABL1+ CD26+ cells at both diagnosis as well as following 3 months of TKI therapy. The new results are presented in figure 6B in the revised manuscript.

Finally, although the authors do describe differential gene expression between CML and normal, BCR:ABL+ and BCR:ABL-, primitive stem cells they have not as yet taken the opportunity to use these findings to address questions regarding biological mechanisms related to CML LSC that impact on TKI response and outcomes.

We agree with the reviewer that our major focus here was to characterize the cellular heterogeneity coupled to treatment outcome and therefore we did not delve deep into the molecular mechanisms underlying TKI response. However, in response to this comment, as mentioned above, we noted that one of the top genes in BCR::ABL1 cells (Fig. 4 C; right; in red), GAS2 (Growth Specific Arrest 2) was expressed at both diagnosis and TKI therapy within CD26+ cells relative to CD35+ cells (updated figure 6B). Interestingly, GAS2 was also detected in CML LSCs in a recent scRNA-seq study (Krishnan et al. Blood, 2023) suggesting GAS2 upregulation could be a consistent molecular feature of CML cells. GAS2 has been previously noted as deregulated in CML (Janssen JJ et al. Leukemia, 2005, Radich J et al, PNAS, 2006), control of cell cycle, apoptosis, and response to Imatinib (Zhou et al. PLoS One, 2014). Future investigations are warranted to assess whether GAS2 could play a role in the outcome of long-term TKI therapy.

Reviewer 2:

Summary:

The authors use single-cell "multi-comics" to study clonal heterogeneity in chronic myeloid leukemia (CML) and its impact on treatment response and resistance. Their main results suggest 1) Cell compartments and gene expression signatures both shared in CML cells (versus normal), yet 2) some heterogeneity of multiomic mapping correlated with ELN treatment response; 3) further definition of s unique combination of CD26 and CD35 surface markers associated with gene expression defined BCR::ABL1+ LSCs and BCR::ABL1- HSCs. The manuscript is well-written, and the method and figures are clear and informative. The results fit the expanding view of cancer and its therapy as a complex Darwinian exercise of clonal heterogeneity and the selective pressures of treatments.

Strengths:

Cutting-edge technology by one of the expert groups of single-cell 'comics.

Weaknesses:

Very small sample sizes, without a validation set. The obvious main problem with the study is that an enormous amount of results and conjecture arise from a very small data set: only nine cases for the treatment response section (three in each of the ELN categories), only two normal marrows, and only two patient cases for the division kinetic studies. Thus, it is very difficult to know the "noise" in the system - the stability of clusters and gene expression and the normal variation one might expect, versus patterns that may be reproducibly study artifact, effects of gene expression from freezing-thawing, time on the bench, antibody labeling, etc. This is not so much a criticism as a statement of reality: these elegant experiments are difficult, timeconsuming, and very expensive. Thus in the Discussion, it would be helpful for the authors to just frankly lay out these limitations for the reader to consider. Also in the Discussion, it would be interesting for the authors to consider what's next: what type of validation would be needed to make these studies translatable to the clinic? Is there a clever way to use these data to design a faster/cheaper assay?

We thank the reviewer for appraisal of our manuscript. We take the opportunity to point out the updates in the revised manuscript in view of the comments.

Very small sample sizes, without a validation set. The obvious main problem with the study is that an enormous amount of results and conjecture arise from a very small data set: only nine cases for the treatment response section (three in each of the ELN categories), only two normal marrows, and only two patient cases for the division kinetic studies.

As the reviewer has noted the single cell omics experiments remain resource demanding thereby placing a limitation on the number of patients analyzed. As described above in response to the comments from reviewer 1, multiomic CITE-seq allows extraction of two modalities in comparison to a typical scRNA-seq, however, this also makes it even more limited in the number of samples processed in a sustainable way. This was one of the motivations to analyze a larger number of independent samples (n=59) while benefiting from the insights gained from CITE-seq (n=9). Furthermore, by analyzing CD34+ cells from bone marrow and peripheral blood of CML patients, including both responders and non-responders after one year of Imatinib therapy, we were able to significantly diversity the patient pool, which was lacking in our CITE-seq patient pool. As mentioned above, this reflects a growing trend to analyze larger number of patients while anchoring the analysis on prohibitively expensive but potentially insightful sc-omics approaches (For example, please see Zeng et al, A cellular hierarchy framework for understanding heterogeneity and predicting drug response in acute myeloid leukemia, Nature Medicine, 2022).

As emphasized above, we frequently sought to confirm the findings from one approach using a complementary method and independent samples. For example, although CITE-seq (n=9) showed altered frequency of all cell clusters between optimal and poor responders (Fig. 3B), we refrained from generalizing because an independent largescale computational deconvolution analysis (n=59) only substantiated the altered proportion of primitive and myeloid clusters.

In view of the comment, we have now increased the number of patients analyzed during the revision process. These include increased numbers to investigate the ratio between CD35+ and CD26+ populations at diagnosis, as well as 3 months of TKI therapy, qPCR for BCR::ABL1, and patients examined for GAS2, one of the top genes expressed in CML cells (see response to reviewer 1 for details). Altogether, >80 patient samples across different approaches were analyzed to strengthen the conclusions.

During the revision, we have analyzed cells from 8 CML patients for cell cycle using gene activity scores. This is in addition to the cell division kinetics data reported previously are now together described in the supplementary figures 9C-F.

It is very difficult to know the "noise" in the system - the stability of clusters and gene expression and the normal variation one might expect, versus patterns that may be reproducibly study artifact, effects of gene expression from freezing-thawing, time on the bench, antibody labeling, etc. This is not so much a criticism as a statement of reality: these elegant experiments are difficult, time-consuming, and very expensive. Thus in the Discussion, it would be helpful for the authors to just frankly lay out these limitations for the reader to consider.

We agree with the reviewer that sc-omics approaches can be noisy despite continuing efforts to denoise single cell datasets through both experimental and bioinformatic innovations. Therefore, we have updated the discussion as recommended by the reviewer (paragraph 5 in the discussion).

We also note that CITE-seq, in contrast to scRNA-seq alone provides dual features: surface marker/protein as well as RNA for annotating the same cluster. In our manuscript, for example, cell clusters in UMAP for normal BM; Fig 1B were described using both surface markers (Fig. 1C) and RNA (Fig. 1D) making the cluster identity robust. To further elaborate this approach, a new supplementary figure 1C shows annotations of clusters using both RNA and surface markers.

To potentially address the issue of stability of clusters and gene expression, we compared the marker genes for major clusters from nBM from this study (supplementary table 4, Warfvinge et al.) with those described recently in a scRNA-seq study by Krishnan et al. supplementary table 8, Blood, 2023 using Cell Radar, a tool that identifies and visualizes which hematopoietic cell types are enriched within a given gene set (description: https://github.com/KarlssonG/cellradar

Direct link: https://karlssong.github.io/cellradar/). To compare, we used our in-house gene list for the major clusters as well as mapped the same number of top marker genes based on log2FC from corresponding cluster from Krishnan et al. as inputs to Cell Radar. The Cell Radar plot outputs are shown below.

Author response image 1.

This approach showed broad similarities across clusters from this study with their counterparts from the other study suggesting the cluster identities reported here are likely to be robust. Please note these figures are for reviewer response only and not included in the final manuscript.

Also in the Discussion, it would be interesting for the authors to consider what's next: what type of validation would be needed to make these studies translatable to the clinic? Is there a clever way to use these data to design a faster/cheaper assay?

Our findings on CD26+ and CD35+ surface markers to enrich BCR::ABL1+ and BCR::ABL1- cells suggest a simpler, faster and cheaper FACS panel can possibly quantify leukemic and non-leukemic stem cells in CML patients. We anticipate that future investigations, clinical studies might examine whether CD26CD35+ cells could be plausible candidates for restoring normal hematopoiesis once the TKI therapy diminishes the leukemic load, and whether patients with low counts of CD35+ cells at diagnosis have a relatively higher chance of developing hematologic toxicity such as cytopenia during therapy.

We briefly mentioned this possibility in the discussion; however, we have now moved it to another paragraph to highlight the same. Please see paragraph 5 in the revised manuscript.

Reviewer 3:

Summary:

In this study, Warfvinge and colleagues use CITE-seq to interrogate how CML stem cells change between diagnosis and after one year of TKI therapy. This provides important insight into why some CML patients are "optimal responders" to TKI therapy while others experience treatment failure. CITE-seq in CML patients revealed several important findings. First, substantial cellular heterogeneity was observed at diagnosis, suggesting that this is a hallmark of CML. Further, patients who experienced treatment failure demonstrated increased numbers of primitive cells at diagnosis compared to optimal responders. This finding was validated in a bulk gene expression dataset from 59 CML patients, in which it was shown that the proportion of primitive cells versus lineage-primed cells correlates to treatment outcome. Even more importantly, because CITE-seq quantifies cell surface protein in addition to gene expression data, the authors were able to identify that BCR/ABL+ and BCR/ABL- CML stem cells express distinct cell surface markers (CD26+/CD35- and CD26-/CD35+, respectively). In optimal responders, BCR/ABL- CD26-/CD35+ CML stem cells were predominant, while the opposite was true in patients with treatment failure. Together, these findings represent a critical step forward for the CML field and may allow more informed development of CML therapies, as well as the ability to predict patient outcomes prior to treatment.

Strengths:

This is an important, beautifully written, well-referenced study that represents a fundamental advance in the CML field. The data are clean and compelling, demonstrating convincingly that optimal responders and patients with treatment failure display significant differences in the proportion of primitive cells at diagnosis, and the ratio of BCR-ABL+ versus negative LSCs. The finding that BCR/ABL+ versus negative LSCs display distinct surface markers is also key and will allow for a more detailed interrogation of these cell populations at a molecular level.

Weaknesses:

CITE-seq was performed in only 9 CML patient samples and 2 healthy donors. Additional samples would greatly strengthen the very interesting and notable findings.

Reviewer #3 (Recommendations For The Authors):

My only recommendation is to bolster findings with additional CML and healthy donor samples.

CITE-seq was performed in only 9 CML patient samples and 2 healthy donors. Additional samples would greatly strengthen the very interesting and notable findings.

We thank the reviewer for the positive assessment of our manuscript. As mentioned in response to comments from reviewer 1 and 2, CITE-seq remains an reource consuming single cell method potentially limiting the number of patients to be analyzed. However, during the revision process, we have increased the number of patient material analyzed for other assays; these include increased number to investigate the ratio between CD35+ and CD26+ populations at diagnosis, and 3 months of TKI therapy, qPCR for BCR::ABL1, and patients examined for GAS2, one of the top genes expressed in CML cells. Thus, >80 patient samples across different assays have been analyzed to strengthen the conclusions. (Please see comment to reviewer 1 for more details)

Author Response

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

We want to thank the reviewers for their thoughtful analysis and questions.

A brief overview of the changes to the manuscript is provided here, with individual responses to the reviewer comments following.

The methods section has been expanded to better explain the techniques used in our analyses. CTCF binding data section has likewise been expanded, to include more detail on the dataset and our analysis of its contents. All other requested clarifications have been added to areas of the results.

Beyond specific requests from the reviewers, we made the following changes.

We felt that a particular terminology choice on our part resulted in some confusion: the use of “SNPs” to refer to genetic variants within our Diversity Outbred samples. While we used SNPs that lay closest to the center of our haplotype predictions as our representative loci for each linkage disequilibrium block, this was done for computational purposes only. We did not focus most of our analyses on the haplotypes themselves, because of the uncertainty of which variants within an LD block actually participated in the genetic-epigenetic interactions we imputed.

Thus, we edited the text to remove mention of “SNPs” unless our analysis did directly and deliberately profile SNPs themselves. In all other cases, we now refer to “haplotypes”, “genetic variants”, or “variants”. This should help increase clarity in the manuscript as a whole.

A small error was discovered within the labelling and processing of regression model outputs in chromosome 14. A consistency check was run on all chromosomes, finding that only Chr 14 was affected. Chr 14 was rerun in its entirety to verify its results, with the previous results now archived within our databases uploaded on Synapse (see Methods for a link). All relevant calculations and figures were regenerated, resulting in an average shift of 1% or less across the manuscript. All analyses remain highly statistically significant.

Responses to comments from Reviewer #1

Methods

• Sequencing depth was retrieved from the original publication on the primary multiomics dataset. (Line 105-106)

• A line was added regarding initial mouse genome alignment for the original publication: we explain the GigaMUGA genotyping array, used for the DO mESC samples. For our ChIP-seq data, we reword to specify: we used liftovers from imputed strain-specific genomes to B6 mm10. (Lines 108-110; 116-120; 168-170)

• Aneuploidy removal is expanded upon in a similar fashion: the original QC identified chromosome-level gene expression differences to remove aneuploid samples. (Line 111)

• Mention of the pre-publication use of an alternative null model has been removed, given its lack of relevance to the rest of the text. While it was interesting to compare to the standard null model, it amounts to a side note that distracts from the focus of the paper. (Line 137-139).

• Descriptive subheadings have been added.

Results - Line 179 (now Line 191) now points to Methods.

• Line 189-200 (now Line 188-204): language altered to better explain our intent: We wished to perform an intrachromosomal scan across the whole genome for non-additive genetic-epigenetic interactions. However, there were computational limits to how many possible combinations of gene, haplotype, and ATAC-seq peak we could feasibly test. We thus generated a random subset of possible combinations. This was also performed to identify target regions for focused analyses.

• Line 195 (now line 206, expanded on in Line 210): Clarification added on the significance of our result: if non-additive genetic-epigenetic interactions were not a significant explanatory factor for gene expression, we would expect to see no enrichment of low p-value results. Instead, we see 0.07% of our models coming in at adj. p < 1x10-7.

• Line 199 (now Line 216): The requested calculations were run, and are now included in table S3. We found that within 4 Mb of a given gene, less than 10% of variants and ATAC peaks within clustered closer to each other than they did to the gene they affected.

Please note that this figure has a level of uncertainty due to linkage disequilibrium. Thus, rather than precisely answering the question “[are there haplotype-ATAC pairs] that are in the same locality but further away from the gene?”, we asked "is the ATAC peak closer than the gene to the point where we have the highest confidence of correctly calling the interacting genotype?". The relevant code has been deposited in our Synapse repository (see Methods for link).

• Line 205 (now restructured in Line 221-228): The text has been edited to specify our intent. We are referring to a set of TAD-focused regression models we generated (see Methods) that comprehensively included all possible interactions between genes, and all haplotypes and ATAC peaks within +/- 1 TAD of the gene.

• (Line 227): We specified that the previously-published TAD boundary dataset we used was retrieved from the Bing Ren lab’s Hi-C projects, which imputed locations of TAD boundaries in B6 mESCs.

• We have relabeled Figure 1 and tweaked the surrounding text to clear up some confusing aspects. The Euler plots in Figure 1D-E reflect the fact that each ATAC-seq peak and haplotype can be in multiple relationships with local genes and regulatory factors. Some of these relationships will be simple correlation between their presence and gene expression, while others may co-regulate alongside independent regulatory factors, or engage in non-additive regulatory interactions.

Because these non-additive regulatory interactions have not been comprehensively studied, we wished to determine whether there were any regulatory factors within our data that would not be detected as significant via more conventional methods, such as correlation analysis, mediation analysis, or regression analysis without an interaction term. Our Euler plots show that there are large subsets of both ATAC-seq peaks and haplotypes that are exclusively found in non-additive interactions. Thus, our justification for focusing on non-additive interactions for the rest of the paper.

• Line 256 (now Line 252-255): We further clarified the above in this section: correlation and mediation analyses were previously completed by the team which initially analyzed the DO mESC dataset (Skelly et al. 2020, Cell Stem Cell). They performed a correlation analysis between open chromatin and gene expression (Skelly et al. Fig. 2A), and identified expression quantitative trait loci (eQTL) (Skelly et al. Fig. 2E). We felt that more direct comparisons to the Skelly et al. data would distract readers from our focus on genetic-epigenetic interactions. Thus, we limited our discussion of non-interacting regulatory relationships to Figures 1-2, and a brief mention in Figure 5.

• Line 290 (now Line 337): We pulled promoter locations from the FANTOM5 database of mouse promoters, and included analysis in both the text and Figure S4A-B.

• (Line 475-476): we clarified “DO founder SNPs” to “SNPs from the non-reference DO founder strains”.

• Line 472 (restructured in Lines 531-564): We have expanded on this section, including answers to the reviewer’s questions regarding ChIP-seq peak counts, overlap with the TAD map we used for our other analyses, and expanded upon strain-specific CTCF binding we identified in our ChIP-seq analysis.

Responses to comments from Reviewer #2:

(1) Typo corrected.

(2) Lines 194-195 (now line 206, expanded on in Line 210): We have expanded upon the intent and expectations of our analysis. In summary: if non-additive genetic-epigenetic interactions were not a significant explanatory factor for gene expression, we would expect to see no enrichment of low p-value results. Thus, we would expect 0.0000001% of results to reach adj. p < 1x10-7. Instead, we see 0.07% of our models coming in at adj. p < 1x10-7, four orders of magnitude greater than expected.

(3) Lines 226-230 (Expanded on in Lines 252-276): We have relabeled Figure 1 and tweaked the surrounding text to clear up some confusing aspects. The percentages in the text are derived from the data summarized in the Euler plots in Figure 1D-E. These plots reflect the fact that each ATAC-seq peak and haplotype can be in multiple relationships with local genes and regulatory factors. Some of these relationships will be simple correlation between their presence and gene expression, while others may co-regulate alongside independent regulatory factors, or engage in non-additive regulatory interactions.

(4) Line 261-263 (now lines 299-300): A companion to Figure 2B has been added (Fig. S3), which provides interaction counts for each ATAC-seq peak that contributed to Figure 2B. A horizontal line is included to highlight the locations of the highly-interacting ATAC peaks.

(5) Analysis regarding Figure 3B had been removed from its original context. It has now been restored to the manuscript (Line 368-371).

Author Response

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

Reviewer #1 (Recommendation for the authors)

I only have one comment for improvement of this study and it has to do with the comparison of simulators that they conducted. There are many other simulators around now, including scDesign3, spaSim, SPIDER, SRTSIM, etc. Are any of those methods worth including in the comparison?

Indeed, many of the mentioned simulators did not exist when we initially developed synthspot, and upon closer examination, they are not directly comparable to our tool.

• scDesign3: The runtime of scDesign3 is quite long as a result of its generative model. The example provided in its tutorial only simulates 183 genes and takes over seven minutes when using four cores on a system with Intel Xeon E5-2640 CPUs running at 2.5GHz. In a small downsampling analysis, we simulated 10, 50, 100, and 150 genes with scDesign3 and observed runtimes of 30, 130, 245, and 360 seconds, respectively. This seems to indicate a linear relationship between the number of genes and the runtime, therefore rendering it unsuitable for simulating whole-transcriptome datasets for deconvolution.

• spaSim: spaSim focuses on modelling cell locations in different tissue structures but does not provide gene expression data. It is designed for testing cell colocalization capabilities rather than simulating gene expression.

• SPIDER: Although SPIDER appears to have some overlap with our work, it seems to be in the early stages of development. The GitHub repository contains only two scripts without any documentation, and the preprint does not provide instructions on how to use the tool.

• SRTSim: SRTSim explicitly states in its publication that it is not suitable for evaluating cell type deconvolution, as its focus is on simulating gene expression data without modelling cell type composition.

• scMultiSim: scMultiSim, like scDesign3, is limited in its capability to model the entire transcriptome.

Nonetheless, the inherent modularity of our Nextflow framework makes it possible for users to simply run the deconvolution methods on data that has been simulated by other simulators if need be.

Additionally, we have added the following rationale for why we developed synthspot in “Synthspot allows simulation of artificial tissue patterns”:

“On the other hand, general-purpose simulators are typically more focused on other inference tasks, such as spatial clustering and cell-cell communication, and are unsuitable for deconvolution. For instance, generative models and kinetic models like those of scDesign3 and scMultiSim are computationally intensive and unable to model entire transcriptomes. SRTSim focuses on modeling gene expression trends and does not explicitly model tissue composition, while spaSim only models tissue composition without gene expression.”

The other aspect of the simulation comparison that I'm missing is some kind of spatial metric. There are metrics about feature correlation, sample-sample correlation, library size, etc. But, what about spatial correlation (e.g., Moran's I or similar). Perhaps comparing the distribution of Moran's I across genes in a simulated and real dataset would be a good first start.

We would like to clarify that synthspot does not actually simulate the spatial location of spots, but synthetic regions where spots from the same region share similar compositions. Hence, incorporating a spatial metric in the comparison is not feasible. However, as RCTD is the only method that explicitly uses spot locations in its model (Supplementary Table 2, "Location information"), we believe that generating synthetic datasets with actual coordinates would not significantly impact the conclusions of the study.

Reviewer #2 (Public Review)

On the other hand, the authors state that in silver standard datasets one half of the scRNA-seq data was used for simulation and the other half was used as a reference for the algorithms, but the method of splitting the data, i.e., at random or proportionally by cell type, was not specified.

The data was split proportionally by cell type. To clarify this, we have included an additional sentence in the main text under the first paragraph of “Cell2location and RCTD perform well in synthetic data”, as well as in Figure S2.

Reviewer #2 (Recommendation for the authors)

Figure legends in Figures 3, 4 and across most Supplementary material are almost illegible. Please consider increasing font size for better readability.

Thank you for bringing this to our attention. The font size has been increased for all main and supplementary figures. Additionally, the supplementary figures have also been exported in higher resolution.

Supplementary Notes Figure 2c reads "... total count per sampled multiplied by..."

This has been adapted, as well as the captions of Supplementary Notes Figure 3c and 4c which had the same typo.

Review #3 (Public review)

The simulation setup has a significant weakness in the selection of reference single-cell RNAseq datasets used for generating synthetic spots. It is unclear why a mix of mouse and human scRNA-seq datasets were chosen, as this does not reflect a realistic biological scenario. This could call into question the findings of the "detecting rare cell types remains challenging even for top-performing methods" section of the paper, as the true "rare cell types" would not be as distinct as human skin cells in a mouse brain setting as simulated here.

We appreciate the reviewer’s concern and would like to clarify that within one simulated dataset, we never mix mouse and human scRNA-seq data together. The synthetic spots generated for the silver standards are always sampled from a single scRNA-seq or snRNA-seq dataset. Specifically, for each of the seven public scRNA-seq datasets, we generate synthetic datasets with one of nine abundance patterns, resulting in a total of 63 synthetic datasets. These abundance patterns only affect the sampling priors that are used—the spots are still created with combinations of cells sampled from the same dataset.

Furthermore, it is unclear why the authors developed Synthspot when other similar frameworks, such as SRTsim, exist. Have the authors explored other simulation frameworks?

While there are other simulation frameworks available now, synthspot was designed to specifically address the requirements of our study, offering unique capabilities that make it suitable for deconvolution evaluation. Moreover, many of the simulators did not exist when we initially developed our tool. We have added the following rationale for why we developed synthspot in “Synthspot allows simulation of artificial tissue patterns”:

“On the other hand, general-purpose simulators are typically more focused on other inference tasks, such as spatial clustering and cell-cell communication, and are unsuitable for deconvolution. For instance, generative models and kinetic models like those of scDesign3 and scMultiSim are computationally intensive and unable to model entire transcriptomes. SRTSim focuses on modeling gene expression trends and does not explicitly model tissue composition, while spaSim only models tissue composition without gene expression.”

In our response to Reviewer 1 copied below, we also outline specific reasons why other simulators were not suitable for our benchmark:

• scDesign3: The runtime of scDesign3 is quite long as a result of its generative model. The example provided in its tutorial only simulates 183 genes and takes over seven minutes when using four cores on a system with Intel Xeon E5-2640 CPUs running at 2.5GHz. In a small downsampling analysis, we simulated 10, 50, 100, and 150 genes with scDesign3 and observed runtimes of 30, 130, 245, and 360 seconds, respectively. This seems to indicate a linear relationship between the number of genes and the runtime, therefore rendering it unsuitable for simulating whole-transcriptome datasets for deconvolution.

• spaSim: spaSim focuses on modelling cell locations in different tissue structures but does not provide gene expression data. It is designed for testing cell colocalization capabilities rather than simulating gene expression.

• SPIDER: Although SPIDER appears to have some overlap with our work, it seems to be in the early stages of development. The GitHub repository contains only two scripts without any documentation, and the preprint does not provide instructions on how to use the tool.

• SRTSim: SRTSim explicitly states in its publication that it is not suitable for evaluating cell type deconvolution, as its focus is on simulating gene expression data without modelling cell type composition.

• scMultiSim: scMultiSim, like scDesign3, is limited in its capability to model the entire transcriptome.

Finally, we would have appreciated the inclusion of tissue samples with more complex structures, such as those from tumors, where there may be more intricate mixing between cell types and spot types.

We acknowledge the reviewer's suggestion and have incorporated a melanoma dataset from Karras et al. (2022) in response to this suggestion. This study profiled melanoma tumors by using both scRNA-seq and spatial technologies. The scRNA-seq consists of eight immune cell types and seven melanoma cell states. We have included this study as an additional silver standard and case study, the latter of which is presented in a separate section following the liver analysis (and a corresponding section in Methods).

We found that method performances on synthetic datasets generated from this melanoma dataset follow previous trends (Figure S3-S5). However, the inclusion of the case study led to the following changes in the overall rankings: cell2location and RCTD are now tied for first place (previously RCTD ranked first), and Seurat and SPOTlight have swapped places. Despite these changes, the core messages and conclusions of our paper remain unchanged. All relevant figures (Figures 1a, 2, 3a, 4a, 6b, 7a, S3-S6, S9) have been updated to incorporate these new analyses and results.

Review #3 (Recommendation for the authors)

To maintain consistency in the results, it is recommended to exclude the human scRNAseq set when generating synthetic spots. Furthermore, addressing the other significant weaknesses mentioned earlier would be beneficial.

Please refer to our response to the public review where we address the same remark.

It is essential to differentiate this work from previous benchmarking and simulation frameworks.

In addition to the rationale on why we developed our own framework (see response to the public review), we have included the following text in the discussion that highlights our versatile approach when using a real spatial dataset for evaluation:

“In the case studies, we demonstrated two approaches for evaluating deconvolution methods in datasets without an absolute ground truth. These approaches include using proportions derived from another sequencing or spatial technology as a proxy, and leveraging spot annotations, e.g., zonation or blood vessel annotations, that typically have already been generated for a separate analysis.”

Furthermore, we conducted an extra analysis in the liver case study, generating synthetic datasets with one experimental protocol and using the remaining two as separate references (Figure S13). This further illustrates the usefulness of our simulation framework, which we mentioned by appending this sentence in the discussion:

“As in our silver standards, users can select the abundance pattern most resembling the real tissue to generate the synthetic spatial dataset, as we have also demonstrated in the liver case study.”

Author Response

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

Reviewer #1

(1) Since you only included patients with early-onset preeclampsia in the study, I suggest revising the title to "Identification of novel syncytiotrophoblast membrane extracellular vesicle derived protein biomarkers in early-onset preeclampsia...."

We have changed our title to early-onset preeclampsia.

(2) Under methods, you state that placenta was obtained from women undergoing elective cesarean section. Was this because all the study patients were delivered before the onset of labor? Or were laboring patients specifically excluded from the study?

Indeed, labor influences the extracellular vesicles (EVs) generated. To ensure consistency in our samples and avoid this variable, we chose placentas obtained from elective cesarean sections (CS) for our study.

(3) In Table 1 on page 10, the 8th row (Birth weight grams) needs to be reformatted. The mean birthweights for normal pregnancy and preeclampsia should be the same.

We have reformatted the table and using ranges instead of brackets.

(4) In the legend for Table 1, the sentence beginning on page 10, line 227, and continuing onto page 11, line 228, does not make sense. Part of the sentence was omitted inadvertently.

We have modified this sentence to :

Detergent treatment, which could break down EVs, with NP-40 confirmed that the majority (99%) of our samples were largely vesicular since only 0.1 ± 0.12% of BODIPY FL N-(2-aminoethyl)-maleimide and PLAP double-positive events were detected (a reduction of 99%) (Figure 1E and 1H).'

(5) As you acknowledge, the sample size (12 patients) was small. This is understandable because early-onset preeclampsia occurs in <1% of parturients. You could collaborate with other centers in future studies to increase the sample size.

Thank you very much for your comment. We are willing to cooperate on future research and will try to expand our sample size in subsequent studies.

Reviewer #2 (Recommendations For The Authors):

(1) This is one of the many "catalogue" papers where placental exosome proteins in preeclampsia are profiled. Thus, the manuscript lacks novelty. The only novelty factor is the authors have isolated exosomes by a different method and even separated the small and large exosomes. However, there is no mention of how these exosomes differ from each other in terms of their functionality. Thus it is hard to judge the biological significance of this work.

We appreciate your insights regarding the novelty of our study. While numerous papers have profiled placental exosome proteins in preeclampsia, our methodology for enriching sSTB-EVs (exosomes) offers a distinct perspective. We believe that the separation of sSTB-EVs (exosomes) and medium/large STB-EVs (microvesicles) introduces a differentiation that extends beyond mere profiling, with implications for their functionality. There are previous studies showed that the different sizes of placenta EVs have distinct characteristics (Zabel RR, et al. Enrichment and characterization of extracellular vesicles from ex vivo one-sided human placenta perfusion. Am J Reprod Immunol. 2021 Aug;86(2)). Furthermore, the way cells internalize and respond to EVs may depend on the size of the EV (Zhuang X et al. Treatment of brain inflammatory diseases by delivering exosome encapsulated anti-inflammatory drugs from the nasal region to the brain. Mol Ther. 2011 Oct;19(10).) Therefore, it would be important for future studies to distinguish different sizes of EVs for the research.

(2) The authors must demonstrate that these two types of EVs are also produced in vivo by detecting them in the serum of women.

Thank you for the comment. Many previous studies have shown the two types of placental EVs in women's blood. Nakahara et al.'s (PMCID: PMC7755551) extensive review compiles studies that have specifically isolated various subtypes of placenta-derived EVs from maternal circulation. We have also readdressed it in the introduction.

(3) The authors must compare the proteomes of serum-derived placental exosomes and the proteome of the STBs isolated from the perfusion experiments to judge how overlapping the outcomes are from those produced naturally and those produced under ex vivo conditions.

We appreciate the reviewer's suggestion to compare the proteomes of serum-derived placental sSTB-EVs (exosomes) with those from STBs isolated through perfusion experiments. Indeed, such a comparison would provide valuable insights into the similarities and differences between naturally produced and ex vivo-generated sSTB-EVS (exosomes). However, isolating placental EVs from maternal circulation for comprehensive proteomic profiling presents challenges. It requires a significant amount of serum or plasma sample that will be sufficient to enable the isolation of placenta-specific EVs amongst numerous EVs in the circulation. In addition, it will require multiple intricate steps such as ultracentrifugation followed by immunoprecipitation. Each of these steps can potentially lead to the loss of EVs. Additionally, given the high concentration of lipoproteins in plasma relative to EVs, there's a significant risk of obtaining low-purity isolates from the outset. These challenges might compromise the comparability of results between placenta-specific EVs from maternal circulation and those from ex vivo perfusion. Nevertheless, we acknowledge the value of such an endeavor and will consider incorporating this aspect in future studies as the EV and proteomic methodology and technology improve and become more sensitive.

(4) I have a major issue with the chosen study subjects. While the study title and the manuscript mention preeclampsia, as per the inclusion criteria mentioned in lines 88-90, the patients will be HELLP syndrome. Please clarify what was used and modify the manuscript accordingly.

Thank you very much for finding this error. Our patients had none of the features that would qualify them for HELLP syndrome. We have edited to:

PE was defined as new (after 20 weeks) systolic blood pressure of 140 mmHg or diastolic pressure of 90 mmHg, proteinuria (protein/creatinine ratio of 30 mg/mmol or more). None of our patients had maternal acute kidney injury, liver dysfunction, neurological features, hemolysis, or thrombocytopenia.

(5) It is hard to reconcile how only 15 proteins were identified in the placental extract while 300+ in EVs. There is a methodological issue in the mass spec or extraction. With such widely different denominators in the total proteins identified, it is hard to compare the outcomes in terms of the three sample types.

We acknowledge the reviewer's concerns regarding the disparity in protein counts between the placental extract and the EVs. Ultimately, more is not necessarily better. Several factors might contribute to this discrepancy. Firstly, it is plausible that certain proteins exhibit selective affinity to varying sizes of EVs, leading to a more diverse range of proteins than the placental extract. We were also stringent in our analysis to enable us to select proteins whose biological differences are more likely to be reproducible with a different validatory method like a western blot. Additionally, although the placental extract might contain a higher total protein concentration, it doesn't necessarily translate to a richer diversity of disease-specific proteins. Considering these nuances when comparing protein outcomes across sample types is helpful.

(6) I am unable to understand the terms least differentially expressed and most differentially expressed. Do the authors mean upregulated and downregulated? Please clarify and use the terms appropriately by providing fold change values.

We appreciate the reviewer's request for clarification. We intended to provide a relative measure of expression for the terms 'least differentially expressed' and 'most differentially expressed'. The terms are roughly equitable to down- and upregulated. Regarding EVs, we avoid using the terms 'upregulated' and 'downregulated' as EVs act as transporters and do not possess regulatory functions per se. However, for the placenta, we recognize the relevance of these terms.

(7) The data presented is very superficial and lacks methodological details. The authors should provide the total number of targets achieved after mass spec. The cutoffs used the FDRs and other details.

We apologize for the omission. We have added these details to the method section.

(8) It is not clear how were these differentially abundant proteins identified. What was the cutoff used? Was it identified in all the replicates?

We apologize for the omission. We have added these details to the method section.

(9) How many samples were subjected to the discovery cohort, and how many were in the validation cohort? Were they the same or different? If the samples were different, how many PE samples had differentially abundant proteins by both methods?

The study utilized 12 samples for initial discovery and another 12 for western blot validation. The validation samples specifically targeted proteins of interest, rather than undergoing another comprehensive mass spectrometry analysis.

(10) It is striking that the authors report the expression of prostatic acid phosphatase in the placenta. In my understanding of placental biology, this gene or protein is not known to be expressed by the placenta. Please perform immunofluorescence to demonstrate that this protein is indeed produced in the STBs

Research has revealed that even though it's called prostate-specific antigen, it's created in tissues other than the prostate, such as the placenta. Here are a couple of references to support this claim: PMID: 10634405, PMID: 7533063, PMID: 8939403, and PMID: 8945610. Hence it is likely not beneficial to demonstrate what many researchers have already demonstrated.

(11) Please validate the differential abundance of these proteins in the exosomes isolated from the plasma of women with and without preeclampsia. A serial measurement will be of high value to determine how early as compared to hypertension, these biomarkers can predict preeclampsia.

We are validating each EV-carried marker individually in the circulation (plasma or serum), localizing them in the placenta, and performing downstream functional analysis. This article is already lengthy and would likely be too cumbersome to include the details of all individual proteins in this manuscript. However, we have already published papers on Siglec 6 (PMID: 32998819) and Neprilysin (PMID: 30929513), and others will be published soon. We agree that there will be a lot of value to serial measurement, not just in terms of how early as compared to hypertension, these biomarkers can predict preeclampsia but also as potentially a more sensitive or specific test. This would be the subject of subsequent papers.

(12) The authors are recommended to carry out immunofluorescence to localize the differentially abundant proteins in the placental sections and show that they are specific to STBs.

We have already provided a similar response earlier (see response to point 11). In addition, while it is preferable, the biomarkers don't necessarily need to be specific to STB. Not all biomarkers are mechanistic agents/targets, and not all mechanistic agents are biomarkers. However, mechanistic agents should preferably be placental-specific. For example, the total sFLT1, the most studied biomarker, is not exclusively synthesized in the placenta, even though the placental-specific isoform represents a small fraction of the total sFLT-1. For example, in the non-placental world, alkaline phosphatase (ALP) is not exclusively produced by the liver but is a ‘biomarker’ of cholestatic disease.

(13) Table 1 should give the range and SD could be given as + instead of the bracket.

Thank you for your suggestion. We have edited it accordingly.

(14) It is necessary to provide the gestational age of the onset of hypertension to get a judgment of how long these women were preeclamptic, culminating in HELLP.

We want to emphasize that none of our patients experienced HELLP syndrome. In the results section, we have included the gestational age at the time of diagnosis in the table for preeclampsia. It's crucial to understand that the gestational age at diagnosis is distinct from the gestational age when hypertension initially appeared. Detecting the exact gestational age of hypertension onset would be challenging, and it would likely require a prospective or randomized clinical trial with continuous monitoring, possibly on a daily basis. However, our study is retrospective. Thus we can only comment on the gestational age at diagnosis

(15) For newborns the term Sex is used and not gender

Thank you for your suggestion. We have edited it accordingly.

(16) Figure 2 is stretched and hard to read

Thank you for your suggestion. We have edited it accordingly by creating two separate images to promote readability.

(17) Line 278 change the sentence "there fifteen (15) proteins in the placenta" to "there were fifteen (15) proteins in the placenta"

Thank you for your suggestion. We have edited it accordingly.

(18) Line 288 you mean least and not lease

Thank you for your suggestion. We have edited it accordingly.

Author Response

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

eLife assessment

This important study advances our knowledge of how parasites evade the host complement immune system. The new cryo-EM structure of the trypanosome receptor ISG65 bound to complement component C3b is highly compelling and well-supported by biochemical experiments. This work will be of broad interest to parasitologists, immunologist, and structural biologists.

We thank the reviewers and editorial team for this assessment of our work.

Public Reviews:

Reviewer #1 (Public Review):

The authors set out to use structural biology (cryo-EM), surface plasmon resonance, and complement convertase assays to understand the mechanism(s) by which ISG65 dampens the cytoxicity/cellular clearance to/of trypanosomes opsonised with C3b by the innate immune system.

The cryo-EM structure adds significantly to the author's previous crystallographic data because the latter was limited to the C3d sub-domain of C3b. Further, the in vitro convertase assay adds an additional functional dimension to this study.

The authors have achieved their aims and the results support their conclusions.

The role of complement in immunity to T. brucei (or lack thereof) has been a significant question in molecular parasitology for over 30 years. The identification of ISG65 as the C3 receptor and now this study providing mechanistic insights represents a major advance in the field.

Reviewer #2 (Public Review):

This is an excellent paper that uses structural work to determine the precise role of one of the few invariant proteins on the surface of the African trypanosome. This protein, ISG65, was recently determined to be a complement receptor and specifically a receptor of C3, whose binding to ISG65 led to resistance to complement-mediated lysis. But the molecular mechanism that underlies resistance was unknown.

Here, through cryoEM studies, the authors reveal the interaction interface (two actually) between ISG65 and C3, and based on this, make inferences regarding downstream events in the complement cascade. Specifically, they suggest that ISG65 preferably binds the converted C3b (rather than the soluble C3). Moreover, while conversion to a C3bB complex is not blocked, the ability to bind complement receptors 1 and 3 is likely blocked.

Of course, all this is work on proteins in isolation and the remaining question is - can this in fact happen on the membrane? The VSG-coated membrane is supposed to be incredibly dense (packed at the limits of physical density) and so it is unclear whether the interactions that are implied by the structural work can actually happen on the membrane of a live trypanosome. This is not necessarily a dig but it should be addressed in the manuscript perhaps as a caveat.

We thank the reviewer for their positive response our work. We fully agree with the reviewer about the caveats which come from this work being done in a biochemical context. We have addressed this in lines 223-24 and 327-333.

Reviewer #3 (Public Review):

The authors investigate the mechanisms by which ISG65 and C3 recognize and interact with each other. The major strength is the identification of eco-site by determining the cryoEM structure of the complex, which suggests new intervention strategies. This is a solid body of work that has an important impact on parasitology, immunology, and structural biology.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

A paper by Sulzen et al was published online on 27th April in Nature Communications that has a similarity (the cryo-EM structure) to this paper. This does not detract from the value of this paper. The authors should, however, include a "compare and contrast" section in this paper to explain similarities and differences in the conclusions. For example, while this paper demonstrates that ISG65 does not prevent C3 convertase activity, the Sulzen paper suggests it does prevent C5 convertase activity. The compatibility of these conclusions should be discussed.

Two studies of ISG65 were published shortly after submission of this manuscript (Sulzen et al and Lorenzen et al) and we have added a brief comparison of the conclusions of these papers here. These mentions include lines 151, 155-6, 201-2, 274-278, 292-93 and 321-323. For a more in-depth comparison we have published an opinion piece in Trends in Parasitology, which discusses all three of these papers and which we also now reference here.

Could the authors comment as to whether they think the association of C3b with the unstructured region of ISG65 comes about via S-S shuffling? I.e., is C3B first thioester linked to VSG and then this rearranges to ISG65 through C3b-ISG65 proximity?

We thank the reviewer for the interesting suggestion. However, we are not aware of evidence showing that C3b, which has been conjugated to a target protein through its covalent ester bond, then becomes transferred to a second target protein. As ISG65 can bind to C3 as well as C3b, we think that the conjugate could form when ISG65-bound C3 converts to C3b, becomes reactive and, through proximity, is most likely to conjugate to ISG65. Whether this occurs to a substantial degree in trypanosomes, or whether it is more likely that ISG65 interacts with C3b which is already VSG-conjugated, requires further experiments. We have edited lines 217-222 to make this point more clearly.

Reviewer #3 (Recommendations For The Authors):

The authors previously reported that ISG65 C-terminus is so flexible and is not resolved in their 2022 ISG65-C3d (TED of C3b) crystal structure, which is the same case here in the cryo-EM structure of ISG65-C3b. Thus, I am wondering how C3b might find the flexible C-terminus and form a covalent bond.

We think that the answer to the reviewer’s question relates to local concentration. When two reactive compounds are not attached together, then they diffuse freely in three-dimensions and their likelihood of colliding and reacting is subject to the randomness of Brownian motion. However, if they bind together through an interaction distinct from the reactive residues, then this increases their relative local concentration and the likelihood of collision and reaction taking place. In the case of ISG65, this is coupled with the ability of ISG65 to bind to C3 before it converts to C3b and becomes reactive. The interaction of ISG65 with C3/C3b will therefore bring together the reactive residues and increases the probability that they will collide and form a conjugate. Our control with BSA, which does not bind to C3/C3b, and does not form these conjugates supports this conclusion. We have edited lines 217-222 to clarify.

I also find it puzzling that deleting L2 or L3 in ISG65, which they found forming additional contracts with CUB domain of C3b (12 times binding tighter), does not affect the ISG65-C3b conjugate formation in the in vitro C3 convertase formation assay.

When we consider the affinities that the L2 and L3 loop deletions variants have for ISG65, and the concentration of ISG65 in the C3 convertase assay, we would predict that the conjugates still form with the L2 and L3 variants. This binding would therefore increase the relative local concentration of the reactive residues and ensure preferential conjugate formation, as we observe.

(1) Page 2 bottom line, "In particular, loop 2 forms a direct contact with the CUB domain of ISG65, centered around an electrostatic", ISG65 should be C3b.

We thank the reviewer for spotting this. It has been corrected.

(2) Page 4, "We found that ISG65 does not complete with either factor B or Factor D and does not block the binding of factor Bb (Figure 3b). This suggests that the C3 convertase can form in the presence of ISG65", "complete" should be "compete".

It has been corrected.

(3) Page 4, "revealed that in the presence of ISG65 a high molecular weight band appeared, which we identified through mass spectrometry to be a conjugate of ISG65 with C3b". There is no mass spectrometry data in the manuscript to support this.

We agree with the reviewer that this data should be included in the paper and have now added it as Supplementary Table 3.

(4) Page 5, "By inhibiting binding of CR2 to C3d, ISG65 will reduce the likelihood that B-cell receptor binding to trypanosome antigens will result in B-cell activation and antibody production." - this sentence is a bit confusing.

We have clarified this point in lines 243-245.

(5) Related to Figure 2a. "This structure reveals the two distinct interfaces formed between ISG65 and C3b (Figure 2a)." It would be clearer to label where interface 1 and interface 2 are in Figure 2a.

We have now labelled interfaces 1 and 2 above the insets in Figure 2a.

(6) Related to Figure 2C. I suggest mutagenesis to validate ISG65 L2/L3 - C3b CUB domain interaction, i.e. mutate ISG65 (N188, R187, Y190) and perform SPR with C3b.

We agree with the reviewer that this experiment was a valuable validation of our structural data. To achieve this aim, we changed our SPR assay, coupling C3 variants to the chip surface in an orientation which would match their conjugation to a pathogen and allowing us to reliably compare the affinities of ISG65 variants. We then assessed the binding of ISG65, ISG65∆L2, and the ISG65L2N188A,H189A,Y190A proposed by the reviewer. As predicted from the structure, both loop 2 deletion and mutation reduced the affinity for C3b but did not affect the affinity for C3d, suggesting that the difference in affinity of ISG65 for C3b and C3d is due to the observed interface 2. This new data is described in lines 150-168 and is presented in Figure 2c.

(7) Related to Figure 3a. Is the C3b only structure in the presence of ISG65 the real C3b only? Discussion can be added.

Our cryoEM analysis of the ISG65-C3b mixture yielded three dimensional classes which contained clear density for ISG65 and those in which there was no density for ISG65. While the reviewer is technically correct, and we cannot be 100% sure that there is not an entirely disordered ISG65 attached to these ‘unbound’ C3b, we think that this is extremely unlikely. In either case, these ‘unbound’ C3b are indistinguishable from other structures of C3b and the argument in the paper stands. We have added a clause in lines 178-179 to make this point.

(8) Related to Figure 3e. There is no label for WT and deletion mutants. Also, L1 and L3 deletion does not seem to show on the gel.

We have added these labels.

Author Response

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

We thank the Reviewing Editor and two additional reviewers for the insightful input they gave us on the first version of our manuscript on allosteric activity regulation of the anaerobic ribonucleotide reductase from Prevotella copri. We have revised the manuscript in the light of the reviewers' comments. In particular, we have added additional experiments using hydrogen-deuterium exchange mass spectrometry (HDX-MS) to probe the accessibility and mobility of different parts of the protein structure in the apo-state and in the presence of dATP/CTP and ATP/CTP. The results strongly confirm the binding of nucleotides to the activity and specificity sites, as seen biochemically and structurally. In the question of mobility of the glycyl radical domain the HDX-MS experiments suggest an increased mobility in the presence of dATP, though the results are not as clear-cut as for the nucleotide binding. The HDX-MS analyses are complicated by the fact that they reflect all species in solution, which are evidently multiple for all states of PcNrdD. Finally, we have rephrased key parts of the results and discussion, and modified the title, to avoid any implication that we believe the glycyl radical domain becomes extensively disordered, rather that it becomes more mobile to the extent that it cannot be seen in the cryo-EM structures.

eLife assessment

This study advances our understanding of the allosteric regulation of anaerobic ribonucleotide reductases (RNRs) by nucleotides, providing valuable new structural insight into class III RNRs containing ATP cones. The cryo-EM structural characterization of the system is solid, but other aspects of the manuscript, which are incomplete, could be improved by including additional functional characterization and more evidence for the proposed mechanism of inhibition by dATP. The work will be of interest to biochemists and structural biologists working on ribonucleotide reductases and other allosterically regulated enzymes.

Public Reviews:

Reviewer #1 (Public Review):

The goal of this study is to understand the allosteric mechanism of overall activity regulation in an anaerobic ribonucleotide reductase (RNR) that contains an ATP-cone domain. Through cryo-EM structural analysis of various nucleotide-bound states of the RNR, the mechanism of dATP inhibition is found to involve order-disorder transitions in the active site. These effects appear to prevent substrate binding and a radical transfer needed to initiate the reaction.

Strengths of the manuscript include the comprehensive nature of the work - including numerous structures of different forms of the RNR and detailed characterization of enzyme activity to establish the parameters of dATP inhibition. The manuscript could be improved, however, by performing additional experiments to establish that the mechanism of inhibition can be observed in other contexts and it is not an artifact of the structural approach. Additionally, some of the presentations of biochemical data could be improved to comply with standard best practices.

The work is impactful because it reports initial observations about a potentially new mode of allosteric inhibition in this enzyme class. It also sets the stage for future work to understand the molecular basis for this phenomenon in more detail.

We thank the editor and reviewers for their positive evaluation of the potential impact of our work. We completely agree that hypotheses based on structural data require orthogonal experimental verification. However, the number and consistency of the cryo-EM structures speak in favour of the data being representative of conditions in solution. We feel that in particular cryo-EM data should be relatively free of artefacts, e.g. biased or incorrect relative domain orientations, compared to crystallography, where crystal packing effects can affect these parameters. As we write in response to Reviewer #2, it has been difficult to propose a direct structural mechanism for transmission of the allosteric signal from the a-site in the ATP-cone to the active site and GRD given that the ATP-cones and linker are disordered in the dATP-bound dimers and only partly ordered in the dATP-bound tetramers. Further verification experiments will be performed in future but are outside the scope of the present article.

We will improve the presentation of the biochemical data in a revised version.

General comments:

(1) It would be ideal to perform an additional experiment of some type to confirm the orderdisorder phenomena observed in the cryo-EM structures to rule out the possibility that it is an artifact of the structure determination approach. Circular dichroism might be a possibility?

Circular dichroism reports only on the approximate relative proportions of helix, sheet and loop structure in a protein, thus we believe that it would not be a sensitive enough tool to distinguish between ordered and disordered states. We are considering what alternative methods might be appropriate.

(2) Does the disordering phenomenon of one subunit in the ATP-bound structures have any significance - could it be related to half-of-sites activity? Does this RNR exhibit half-of-sites activity?

Half-of-sites activity has not been biochemically proven in any ribonucleotide reductase in spite of the fact that it was first suggested in 1987 (PMID: 3298261). However, strong structural indication was recently published in the form of the holo-complex of the class Ia ribonucleotide reductase from Escherichia coli, which is highly asymmetrical and in which productive contacts forming an intact proton-coupled electron transfer pathway are only formed between one of two pairs of monomers (PMID: 32217749). We have not been able to prove half-of-sites activity for PcNrdD due to low overall radical content, but the structural results are indeed consistent with such an activity.

(3) Does the disordering of the GRD with dATP bound have any long-term impact on the stability of the Gly radical? I realize that the authors tested the ability to form the Gly radical in the presence of dATP in Fig. 4 of the manuscript. But it looks like they only analyzed the samples after 20 min of incubation. Were longer time points analyzed?

Radical content was measured after 5 min and 20 min incubation; 5 min incubations (not included in the manuscript) consistently gave higher radical content compared to 20 min incubation. Longer time points were not analysed, as we assumed that the radical content would be even lower after 20 min.

(4) Did the authors establish whether the effect of dATP inhibition on substrate binding is reversible? If dATP is removed, can substrates rebind?

This is an interesting question. We measured KDs for dATP in the micromolar range and are hence confident that dATP binding is reversible. Our measurements do not, however, directly prove that inhibition of the enzyme is reversible. Nevertheless, it is worth noting that the protein as purified was precipitated and analysed by the UV-visible spectrum. The aspurified PcNrdD contained 30% nucleotide contamination. The as-purified sample was then analysed by HPLC and we identified a major peak, corresponding to dATP/dADP. Therefore, purification conditions had to be optimised to remove the nucleotides. This is evidence that PcNrdD that has “seen” dATP can subsequently bind substrates in the presence of ATP. We will describe the purification more clearly in a revision.

(5) In some figures (Fig. 6e, for example), the cryo-EM density map for the nucleotide component of the model is not continuous over the entire molecule. Can the authors comment on the significance of this phenomenon? Were the ligands validated in any way to ensure that the assignments were made correctly?

Indeed we sometimes saw discontinuous density for the nucleotides, both in the active site and in the specificity site. However, the break was almost always near the C5’ carbon atom, which is common to all nucleotides. While we cannot readily explain this phenomenon, the nucleotides refined well with full occupancy, giving B-factors similar to those of the surrounding protein atoms. The identity of the nucleotide could always be inferred from a) the size of the base (purine or pyrimidine); b) the known nucleotide combinations added to the protein before grid preparation; c) prior knowledge on the combinations of effector and substrate that have been found valid for all RNRs since the first studies of allosteric specificity regulation.

Reviewer #2 (Public Review):

This manuscript describes the functional and structural characterization of an anaerobic (Class III) ribonucleotide reductase (RNR) with an ATP cone domain from Prevotella copri (PcNrdD). Most significantly, the cryo-EM structural characterization revealed the presence of a flap domain that connects the ATP cone domain and the active site and provides structural insights about how nucleotides and deoxynucleotides bind to this enzyme. The authors also demonstrated the catalytic functions and the oligomeric states. However, many of the biochemical characterizations are incomplete, and it is difficult to make mechanistic conclusions from the reported structures. The reported nucleotide-binding constants may not be accurate because of the design of the assays, which complicates the interpretation of the effects of ATP and dATP on PcNrdD oligomeric states. Importantly, statistical information was missing in most of the biochemical data. Also, while the authors concluded that the dATP binding makes the GRD flexible based on the absence of cryo-EM density for GRD in the dATP-bound PcNrdD, no other supports were provided. There was also a concern about the relevance of the proposed GRD flexibility and the stability of Gly radical. Overall, the manuscript provides structural insights about Class III RNR with ATP cone domain and how it binds ATP and dATP allosteric effectors. However, ambiguity remains about the molecular mechanism by which the dATP binding to the ATP cone domain inhibits the Class III RNR activity.

Strengths:

(1) The manuscript reports the first near-atomic resolution of the structures of Class III RNR with ATP domain in complex with ATP and dATP. These structures revealed the NxN flap domain proposed to form an interaction network between the substrate, the linker to the ATP cone domain, the GRD, and loop 2 important for substrate specificity. The structures also provided insights into how ATP and dATP bind to the ATP cone domain of Class III RNR. Also, the structures suggested that the ATP cone domain is directly involved in the tetramer formation by forming an interaction with the core domain in the presence of dATP. These observations serve as an important basis for future study on the mechanism of Allosteric regulation of Class III RNR.

(2) The authors used a wide range of methodologies including activity assays, nucleotide binding assays, oligomeric state determination, and cryo-EM structural characterization, which were impressive and necessary to understand the complex allosteric regulation of RNR.

(3) The activity assays demonstrated the catalytic function of PcNrdD and its ability to be activated by ATP and low-concentration dATP and inhibited by high-concentration dATP.

(4) ITC and MST were used to show the ability of PcNrdD to bind NTP and dATP.

(5) GEMMA was used successfully to determine the oligomeric state of PcNrdD, which suggested that PcNrdD exists in dimeric and tetrameric forms, whose ratio is affected by ATP and/or dATP.

Weaknesses:

(1) Activity assays.

The activity assays were performed under conditions that may not represent the nucleotide reduction activity. The authors initiated the Gly radical formation and nucleotide reduction simultaneously. The authors also showed that the amount of Gly radical formation was different in the presence of ATP vs dATP. Therefore, it is possible that the observed Vmax is affected by the amount of Gly radical. In fact, some of the data fit poorly into the kinetic model. Also, the number of biological and technical replicates was not described, and no statistical information was provided for the curve fitting.

The highest turnover activity of PcNrdD measured in presence of ATP was 1.3 s-1 (470 nmol/min/mg), a kcat comparable to recently reported values for anaerobic and aerobic RNRs from Neisseria bacilliformis, Leeuwenhoekiella blandensis, Facklamia ignava, Thermus virus P74-23, and Aquifex aeolicus (PMID: 25157154, PMID: 29388911, PMID: 30166338, PMID: 34314684, PMID: 34941255). The general trend illustrated in Figure 1 is that ATP has an activating effect on enzyme activity, whereas high concentrations of dATP have an inactivating effect on activity, which cannot be explained by suboptimal assay conditions since our EPR results consistently show that more radical is formed in incubations with dATP compared to incubations with ATP. Curve fitting methods used are listed in Materials and Methods (as specified in the Figure 1 legend), and standard errors for all specified curve fitting results (from triplicate experiments) are shown in Figure 1.

(2) Binding assays.

The interpretation of the binding assays is complicated by the fact that dATP binds both a- and s-sites and ATP binds a- and active sites. dATP may also bind the active site as the product. It is unknown if ATP binds s-site in PcNrdD. Despite this complexity, the binding assays were performed under the condition that all the binding sites were available.

Therefore, it is not clear which event these assays are reporting.

Both ITC and MST experiments involving ATP and dATP binding to the a-site were performed in the presence of at least 1 mM GTP substrate (5 mM in MST) to fill the active site, and 1 mM dTTP effector to fill the s-site (specified in the legend to Figure 2). These conditions enable binding of ATP or dATP only to the a-site in the ATP-cone.

(3) Oligomeric states.

Due to the ambiguity in the kinetic parameters and the binding constants determined above, the effects of ATP and dATP on the oligomeric states are difficult to interpret. The concentrations of ATP used in these experiments (50 and 100 uM) were significantly lower than KL determined by the activity assays (780 uM), while it is close to the Kd values determined by ITC or MST (~25 uM). Since it is unclear what binding events ITC and MST are reporting, the data in Figure 3 does not provide support for the claimed effects of ATP binding. For the effects of dATP, the authors did not observe a significant difference in oligomeric states between 50 or 100 uM dATP alone vs 50 uM dATP and 100 uM CTP. The former condition has dATP ~ 2x higher than the Kd and KL (Figure 1b) and therefore could be considered as "inhibited". On the other hand, NrdD should be fully active under the latter condition. Therefore, these observations show no correlation between the oligomeric state and the catalytic activity.

The results in Figure 3 show that at in presence of 100 µM ATP plus 100 µM CTP the oligomeric equilibrium is 64% dimers plus 36% tetramers, and in presence of 50-100 µM dATP the oligomeric equilibrium is 32% dimers and 68% tetramers. We agree that there is no clear and strong correlation between oligomeric state and inhibition. We will also try to make it clearer in a revised version. Meanwhile, in order to add some clarity to our observations, SEC experiments at higher nucleotide concentrations will be done to strengthen our observations.

(4) Effects of dATP binding on GRD structure

One of the key conclusions of this manuscript is that dATP binding induces the dissociation of GRD from the active site. However, the structures did not provide an explanation for how the dATP binding affects the conformation of GRD or whether the dissociation of GRD is a direct consequence of dATP binding or it is due to the absence of nucleotide substrate. Also, Gly radical is unlikely to be stable when it is not protected from the bulk solvent. Therefore, it is unlikely that the GRD dissociates from the active site unless the inhibition by dATP is irreversible. Further evidence is needed to support the proposed mechanism of inhibition by dATP.

We admit that it has been difficult to propose a direct structural mechanism for transmission of the allosteric signal from the a-site in the ATP-cone to the active site and GRD given that the ATP-cones and linker are disordered in the dATP-bound dimers and that the linker can only be partly modelled in the dATP-bound tetramers. Most likely dATP binding causes a change in the dynamics of the linker region and NxN flap that directly affects substrate binding and simultaneously causes disorder of the GRD, given that all are part of a connected system (described as “nexus” in the manuscript). The structures determined in the presence of dATP and CTP show that CTP cannot bind in the absence of an ordered NxN flap.

In any case a major conclusion of the work is that dATP does not inhibit the anaerobic RNR by prevention of glycyl radical formation but by prevention of its subsequent transfer. We agree that further evidence is required to support the proposed mechanism, but given the extent of the data already presented in the manuscript, we feel that such studies should be the subject of a future publication.

(5) Functional support for the observed structures.

Evidence for connecting structural observations and mechanistic conclusions is largely missing. For example, the authors proposed that the interactions between the ATP cone domain and the core domain are responsible for tetramer formation. However, no biochemical evidence was provided to support this proposal. Similarly, the functional significance of the interaction through the NxN flap domain was not proved by mutagenesis experiments.

We did actually make mutants to verify the observed interactions, but several of them did not behave well in our hands, e.g. with regard to protein stability. Since we have no evidence that oligomerisation is coupled to inhibition, and since we did not observe any conservation between protein sequences in the interaction area, we chose not to pursue this point further. The main merit of the tetramer structures is that they allowed a high-resolution view of dATP binding to the ATP-cone and a comparison to previously-observed ATP-cones. Nevertheless, mutation experiments, also including the NxN flap, could be the subject of future work.

Reviewer #3 (Public Review):

The manuscript by Bimai et al describes a structural and functional characterization of an anaerobic ribonucleotide reductase (RNR) enzyme from the human microbe, P. copri. More specifically, the authors aimed to characterize the mechanism by how (d)ATP modulates nucleotide reduction in this anaerobic RNR, using a combination of enzyme kinetics, binding thermodynamics, and cryo-EM structural determination. One of the principal findings of this paper is the ordering of a NxN 'flap' in the presence of ATP that promotes RNR catalysis and the disordering of both this flap and the glycyl radical domain (GRD) when the inhibitory effector, dATP, binds. The latter is correlated with a loss of substrate binding, which is the likely mechanism for dATP inhibition. It is important to note that the GRD is remote (>30 Ang) from the binding site of the dATP molecule, suggesting long-range communication of the structural (dis)ordering. The authors also present evidence for a shift in oligomerization in the presence of dATP. The work does provide evidence for new insights/views into the subtle differences of nucleotide modulation (allostery) of RNR through long-range interactions.

The strengths of the work are the impressive, in-depth structural analysis of the various regulated forms of PcRNR by (d)ATP using cryo-EM. The authors present seven different models in total, with striking differences in oligomerization and (dis)ordering of select structural features, including the GRD that is integral to catalysis. The authors present several, complementary biochemical experiments (ITC, MST, EPR, kinetics) aimed at resolving the binding and regulatory mechanism of the enzyme by various nucleotides. The authors present a good breadth of the literature in which the focus of allosteric regulation of RNRs has been on the aerobic orthologues.

Given the resolution of some of the structures in the remote regions that appear to be of importance, the rigor of the work could have been improved by complementing this experimental studies with molecular dynamics (MD) simulations to reveal the dynamics of the GRD and loops/flaps at the active site.

We have discussed with expert colleagues the possibility of carrying out MD simulations on the different states in order to study the differential effects of ATP and dATP binding on the dynamics of the GRD. However, they felt that the chance of obtaining meaningful results was low, particularly since some structural elements are missing from the models for both forms, in particular the linker between the ATP-cone and the core.

The biochemical data supporting the loss of substrate binding with dATP association is compelling, but the binding studies of the (d)ATP regulatory molecules are not; the authors noted less-than-unity binding stoichiometries for the effectors.

Most of the methods used measure only binding strength, not the number of binding sites (N), whereas ITC also measures number of sites. N is dependent on the integrity of the protein, i.e. the number of protein molecules in a preparation that are involved in binding, and quite often gives lower values than the theoretical number of binding sites.

Also, the work would benefit from additional support for oligomerization changes using an additional biochemical/biophysical approach.

SEC (chromatography), GEMMA (mass spectrometry) and cryo-EM were used to study oligomerization. Since each method has restrictions on nucleotide concentrations as well as protein concentrations that can be used, the results are not directly comparable, but all three methods indicate nucleotide dependent oligomerization changes. The SEC results will be included in a revised version.

Overall, the authors have mostly achieved their overall aims of the manuscript. With focused modifications, including additional control experiments, the manuscript should be a welcomed addition to the RNR field

Recommendations for the authors: Reviewer #1 (Recommendations For The Authors):

(1) The last sentence of the abstract is not complete. The structures implicate a complex network of interactions in ... ? What do they implicate?

A couple of words seem to have been missed from the abstract. We have rewritten the end of the abstract to emphasise better that the dynamical transitions involve a linked network of interactions and not just the GRD.

(2) A reference is needed in the second sentence of the introduction.

We have added a reference as requested.

(3) Page 2, paragraph 2. The authors state "two beta subunits (NrdB) harboring a stable radical." This is not accurate. First of all, each beta subunit harbors its own cysteine oxidant.

And in several subclasses, that oxidant is not a stable radical but an oxidized metal cluster. Please revise to improve accuracy and also provide appropriate references.

We have revised the description and added a recent reference.

(4) Page 4, Fig. 1, panels C and D. The fit of the curve to the data is pretty poor. Is there an explanation? Could the data be improved in some way? In general, it is also best practice nowadays to show the individual data points in addition to the error bars in plots like the ones shown in Figure 1. Please modify the plots to include the individual data points in this figure - and probably also the subsequent figures showing binding data.

We have modified relevant panels in Figures 1, 2 and 5 as requested.

(5) Page 12, first paragraph. The authors state that one of the monomers in the ATP-CTP structure is well ordered and the other is less ordered. It would be ideal to show in a figure the basis for this conclusion using the cryo-EM maps. The "less ordered" monomer appears to be fully modeled.

Since the 2-fold axis of the dimer is vertical, the GRD of the left-hand monomer is hidden from view at the back of the molecule in Figure 6. For this monomer there was a small amount of density that allowed modelling of part of the glycyl radical loop (though not the tip containing the radical Gly itself) and the NxN flap, albeit with significantly higher mobility. We have illustrated this through an additional supplement for Figure 6 (figure supplement 2) in which the B-factors of the residues are shown both as a ribbon with radius proportional to the B-factor and through colouring. We hope that the four views in Figure 6 (figure supplement 2) together illustrate the relative mobility of different parts of the dimer.

It would also be ideal to show the basis for the conclusion that the entire GRD is disordered in the dATP-bound dimer structure.

Thank you for this suggestion. We have added a fifth supplement to Figure 8 in which we show the cryo-EM reconstruction for the dATP-bound dimer in two orientations, with the ATP-CTP-bound structure superimposed, which clearly shows that the entire GRD, the ATPcones, linker and NxN flap are all disordered in both monomers.

Reviewer #2 (Recommendations For The Authors):

(1) Units to describe enzyme activity.

• The unit for the specific activity in the main text (nmol/min•mg) is unusual. It is most likely a typo of nmol/min/mg or nmol/(min•mg).

We have changes to nmol/min/mg in the text.

• The unit for the Vmax is unusual and should not be confused with the specific activity. By definition, Vmax is the velocity of a reaction at a defined enzyme concentration/amount. For example, if an assay of 10 mg enzyme yielded 470 nmol of product in 1 min, Vmax is 470 nmol/min, whereas the specific activity is 47 nmol/min/mg.

The velocity as calculated above is ca 1.3 s-1. We have added kcat values to accompany the specific activities given.

(2) Steady-state kinetic analysis.

• The steady-state kinetic analysis in Figure 1 needs to be repeated. While the nonlinear curve fitting for Figure 1a is reasonable, those in Figures 1b, 1c, and 1d were outside the error range. Consequently, the reported kinetic parameters are unlikely accurate. The authors should repeat the assays with different enzyme preparation to account for all the errors. If the fit curve is still outside the error range, the kinetic model is likely incorrect, and the authors need to investigate different kinetic models.

The replotted Figure 1 now includes two different experiments for 1b (four replicates in total).

• The authors should report the number of replicates and the statistical data for the curve fitting.

The figure legend has been updated with statistical data for all curve fits, and the number of replicates has been added.

• The authors should report Vmax, Ki, and KL for Figure 1d.

Results in Figures 1c and 1d are less straightforward than those in Figures 1a and 1b where the s-site is filled with dTTP, favouring binding of GTP to the active site. The curve fit in Figure 1c is disturbed at high concentrations of ATP, which plausibly competes with the CTP substrate and results in inhibition by formed dATP. The curve fit in Figure 1d is less certain since reduction of substrate is low due to intrinsic CTP reduction in absence of effector and partially overlapping activation and inhibition effects of dATP.

• The authors should consider presenting the data in a log scale because of the complex nature of the activation/inhibition at the lower concentrations of dATP.

Log scale plots are included as insets in Figures 1b and 1d.

• The basal level of CPT reduction in the absence of an effector nucleotide should be reported with an error.

The error value has been added in the figure legend for the basal level of CTP reduction in the absence of effector.

(3) Equations for the kinetic analysis.

-The equations should be numbered and referred to in the Figure 1 legend.

All equations are specified and numbered in Materials and Methods. The equation used for each curve fit in the panels in Figure 1 is specified in the figure legend.

-KL must be defined in the main text. I suppose this is Kd for ATP or dATP. The equation for KL determination is missing brackets for dNTP.

KL (the concentration of an allosteric effector that gives half maximal enzyme activity) is defined in Materials and Methods where the equation is described. KL is not the same as KD (the dissociation constant for a ligand and its receptor). Brackets have been added to equation 1.

• I believe dNTP in the first equation is incorrect because ATP was the ligand for Figures 1A and 1C.

[dNTP] in the first equation has been changed to [NTP/dNTP] to indicate that both ribonucleotides and deoxyribonucleotides can bind.

• The second equation can be expressed as dATP as I believe this is the only ligand that inhibits the enzyme.

We prefer to keep the more general [dNTP] in the equation.

• The equation used for the fitting in Figure 1d must be defined more clearly than "a combination of the two equations".

The equation used for the curve fit in Figure 1d has been specified as equation 3 in Materials and Methods.

(4) Design of the activity assays

It is not clear if the activity assays report the rate of glycyl radical formation or nucleotide reduction. The authors mixed NrdD and NrdG and initiated the reaction by adding formate (essential for nucleotide reduction) and dithionite (Gly radical formation). The Gly radical formation is slow (in min time scale). The authors reported that ATP/dATP affected the rate of Gly radical formation and in the presence of ATP, Gly radical formation was incomplete even after 20 min. Therefore, it is possible that within the timescale of the activity assays (5 min), the reactions could be partially limited by the Gly radical formation, which may be the reason for the poor curve fitting.

Activity assays were performed with 5 min pre-incubation without dithionite and formate (no glycyl radical formation) and 10 min incubation after addition of dithionite and formate (glycyl radical formation plus substrate reduction). During earlier tests, NrdD and NrdG were first preincubated in the presence of dithionite (glycyl radical formation) and after addition of formate the substrate reduction was monitored during 20 min. These experiments resulted in lower enzyme activity, whereas higher activity was achieved only upon formate addition to the preincubation reaction. We suppose that the presence of dithionite, which is a strong reducing agent, affected NrdD stability and the reaction was stabilised by the presence of formate at an earlier stage of the reaction. For the EPR conditions used in the paper, 5 min incubation gave higher radical content compared to 20 min, and the reported activity assay gave highest activity after 10 min incubation; kcat of 1.3 s-1.

(5) Methods section for the activity assays.

• The concentration of dTTP, ATP, and dATP used in the assays must be described.

We thank the reviewer for pointing out this omission and we have now specified the concentrations used.

• Although the authors mentioned that they changed the concentration of dTTP, such data were not presented. Is this correct? Did the authors fix the dTTP concentration for the GTP reduction?

We apologise for the ambiguity and have specified that the dTTP concentration was fixed at 1 mM in the GTP experiments and that only the ATP or dATP concentrations were varied.

(6) Discrepancy between Ki/KL and Kd.

• There is a significant ambiguity remaining about the binding event that the ITC and MST results are reporting. Although dATP binds to both a- and s-sites and ATP binds to both active site and a-site, only a single binding event was observed in both cases. To distinguish the dATP binding to a- and s-sites and the active site, the authors should perform binding assays using mutant enzymes with only one of the binding sites available for dATP/ATP binding.

MST and ITC were performed in presence of substrate (1 mM GTP) and s-site effector (1 mM dTTP in ITC experiments, and 5 mM dTTP in MST experiments), thus dATP is blocked from binding to the s-site and ATP from binding to the active site.

• There are significant differences between Kd determined by MST or ITC and Ki/KL determined by the activity assays. Kd measurements were performed in the absence of the substrate nucleotides, while the assays required substrates. There may be complications from the presence of NrdG and the Gly radical formation. The authors must clearly describe all these complications and the discrepancy between Kd and Ki/KL.

MST, ITC and enzyme assays were all performed in the presence of substrate, and enzyme assays also contained NrdG, which was not present in the MST and ITC analyses. While KD is a thermodynamic constant representing the affinity of ligand to its binding site - in our case an effector nucleotide to the ATP-cone, KL is a kinetic constant (the allosteric effector concentration that gives half maximal activity) representing the relationship between the effector concentration and the reaction speed and is affected by the enzyme turnover number (kcat). The relationship between KD, KL and Ki is further complicated by conformational and possibly oligomeric state changes of NrdD upon binding of allosteric effectors, which occurs on a slower time scale than the rapid exchange of nucleotides in allosteric sites.

• The results of ATP/dATP copurification experiments shown in Figure 2 - figure supplement 1 show the preference of dATP binding over ATP. However, the results do not necessarily support the competition between ATP and dATP for binding to the ATP cone domain. It is still possible that dATP binding to the s-site diminishes the binding of ATP to the a-site.

Our aim was to exclude the possibility that ATP and dATP can bind to the ATP-cone at the same time and not to study competition between the two. Nevertheless, to eliminate the possibility that dATP binding to the s-site could affect nucleotide binding to the a-site, in two out of three conditions described in the supplementary figure, the experiments were performed in the presence of dTTP to prevent binding of dATP to the s-site.

(7) Oligomeric states.

• The authors must present the GEMMA results without ATP or dATP. Otherwise, the effects of ATP and dATP on the oligomeric state are not clear.

We cannot report GEMMA results without ATP or dATP because apo-PcNrdD was unstable in the GEMMA buffer and clogged the capillaries. Instead, SEC analysis was performed on apo-PcNrdD in a more suitable buffer and showed a homogeneous peak corresponding to a dimer (included as Figure 3 - figure supplement 1).

• Figure 3 does not support the induction of a2 upon ATP binding. The concentrations of ATP used in these experiments (50 and 100 uM) were significantly lower than KL determined by the activity assays (780 uM), while it is close to the Kd values determined by ITC or MST (~25 uM). Since it is unclear what binding events ITC and MST are reporting, the data in Figure 3 does not provide support for the claimed effects of ATP binding.

MST and ITC were performed in the presence of substrate (1 mM GTP) and s-site effector (1 mM dTTP in ITC experiments, and 5 mM dTTP in MST experiments), and they thus measure binding of ATP or dATP to the ATP cone. SEC analysis with 2 µM apo-PcNrdD and higher nucleotide concentrations (1 mM) was performed, confirming the presence of both dimers and tetramers in solution at different ratios depending on the addition of ATP or dATP. The SEC analysis, included as Figure 3 - figure supplement 1, confirms the existence of an equilibrium in solution.

• The effects of dATP must be presented more clearly. The authors did not observe a significant difference in oligomeric states between 50 or 100 uM dATP vs 50 uM dATP and 100 uM CTP. The former condition has dATP ~ 2x higher than the Kd and KL (Figure 1b) and therefore could be considered as "inhibited". On the other hand, NrdD should be fully active under the latter condition. The absence of difference in the oligomeric states between these two different conditions suggested to me that the oligomeric state does not regulate the NrdD activity. The authors seemed to indicate the same conclusion, but did not describe it clearly.

We agree that the oligomeric state most likely does not regulate the NrdD activity and hope to have explained this better in the revised version.

• Figure 3 legend mentioned a and b, but the figure was not labeled.

We have corrected this.

• The authors should triplicate the analysis and report the errors.

Five scans were added for each trace to increase the signal-to-noise level (included in figure legend).

(8) EPR characterization of Gly radical

• The amount of Gly radical must be quantified by EPR. The authors must report how much NrdD has Gly radical.

The concentration of NrdD (1 µM) in the activity assays is too low to be quantified by EPR. In the EPR experiment the glycyl radical content is given in the figure legend.

• The authors claim that the Gly radical environment was similar based on the doublet feature. However, the double feature comes from the hyperfine splitting with α proton whose orientation relative to the radical p-orbital would not be affected by the conformation or the environment. Thus, this conclusion is incorrect and must be removed.

We thank the reviewer for the clarifying comment and have removed our suggestion in the text.

(9) Gly711 should be shown in Fig. 6e to help readers understand the last paragraph on page 12.

The figure reference has been changed to Fig. 7, where this is shown more clearly. In Fig. 6e, inclusion of Gly711 would obscure other important information.

(10) GRD structure with dATP

The disorder of GRD in the presence of dATP does not agree with the formation of Gly radical under the same conditions. Gly radical is unlikely stable if it is extensively exposed to solvent. Most likely, the observed cryo-EM structures represent the conformation irrelevant to Gly radical formation.

We agree that the glycyl radical is unlikely to be stable if exposed to solvent. We believe that the GRD is not completely disordered but most likely made more mobile through rigid body movements of the domain to an extent that makes it invisible in the cryo-EM maps. It is most likely still in the vicinity of the active site, shielding the glycyl radical. Our new HDX-MS results show a small but tangible increase in mobility of the GRD in the presence of dATP compared to ATP. Of course the differences in dynamics remain to be confirmed. It is worth noting that the group of Catherine Drennan at MIT published a conference abstract more than a year ago that suggested a similar pattern of ordered/dynamic GRDs, based on crystal structures, though the details have not yet been published (https://doi.org/10.1096/fasebj.2022.36.S1.R3407).

We also agree that the cryo-EM structures do not show the GRD conformation relevant to Gly radical formation, as this has been shown spectroscopically for the GRE pyruvate formate lyase to require large conformational changes in the GRD and also the presence of the activase. However, revealing this conformation would be a completely different project. We postulate that inactivation proceeds by prevention of radical transfer to the substrate, not by prevention of its formation.

We have altered the wording in several places in the revised manuscript, including the title, to avoid using the term “disorder”, as this may imply (partial) unfolding, and we certainly do not wish to imply that.

(11) The difference between dATP and ATP binding

From the presented structures, it was not clear how the absence of 2'-OH affects the oligomeric state and the structure of the GRD. The low resolution of the ATP-bound structure precluded the comparison between the ATP and dATP-bound structures.

We agree that a detailed analysis of the differences between ATP- and dATP-bound structures requires higher resolution structures, particularly of the ATP-bound form. This will be the subject of future studies.

(12) Conclusion about the disordered GRD.

-The authors should describe the reason why the dATP binding affected the structure of GRD. The authors did not discuss why dATP binding affected the folding or mobility of GRD. Since this is the key conclusion of this manuscript and the authors are making this conclusion based on the absence of the ordered GRD structure (hence the negative results), the authors should carefully describe why the dATP binding does not allow the binding/folding of GRD in the position observed in the ATP-bound structure.

As mentioned in our response to point 4 in this reviewer’s Public Review, it is difficult to propose a direct structural mechanism for transmission of the allosteric signal from the a-site in the ATP-cone to the active site and GRD given that the ATP-cones and linker are disordered in the dATP-bound dimers and that the linker cannot be completely modelled even in the dATP-bound tetramers. Our first hypotheses were that the ATP-cone might work by a steric occlusion mechanism, but the reality appears more complex. Most likely dATP binding causes a change in the dynamics of the linker region and NxN flap that directly affects substrate binding and simultaneously causes higher mobility of the GRD, given that all are part of a connected system. The structures determined in the presence of dATP and CTP show that CTP cannot bind in the absence of an ordered NxN flap. We hope that future structural studies of NrdDs from other organisms may shed further light on this mechanism.

• The authors should test if the dATP inhibition is reversible for PcNrdD. If dATP binding induces dissociation of GRD from the active site and makes GRD flexible, Gly radical would most likely be quenched by formate or other components in the assay solution. If dATP inhibition is reversible, it is hard to believe that Gly radical dissociates completely from the active site.

As-purified PcNrdD contains dATP and can after removal of bound nucleotides bind substrate in presence of ATP. The as-purified PcNrdD protein contained 30% nucleotide contamination. After precipitation, HPLC analysis identified a major peak corresponding to dATP/dADP. Purification conditions were optimised to remove the nucleotides and we have added this information to the purification description.

(13) Functional support for the observed structures.

Similar to X-ray crystallography, cryo-EM is a highly selective method that requires the selection of particles that can be analyzed with sufficient resolution. This means that the analysis could be biased towards the protein conformations stable on the cryo-EM grid. Consequently, testing the structural observations by functional characterization of mutant enzymes is critical. However, the authors did not perform such functional characterizations and made conclusions purely based on the structural observations.

We acknowledge this limitation. We constructed several mutations located at the tetrameric interface between the ATP-cone and the core protein based on the cryo-EM structure of dATP loaded NrdD. Unfortunately, these mutant proteins were unstable and led to protein cleavage.

(14) Other minor points:

• In the introduction, the authors stated "The presence and function of the ATP-cone domain distinguish anaerobic RNRs from the other members of the large glycyl radical enzyme (GRE) family that are otherwise structurally and mechanistically related (Backman et al., 2017)." This statement is misleading because GREs are functionally diverse.

We have removed the words “and mechanistically” to reduce ambiguity.

• p. 12, e.g. should be removed.

We are not sure what is meant here. Does the reviewer mean p. 21 “The interactions are mostly hydrophobic but are reinforced by several H-bonds, e.g. between Gln3D-Gln458A, Ser53D–Gln458A, Arg11D-Asp468A, the main chain amide of Ile12D and Tyr557A.”?

Reviewer #3 (Recommendations For The Authors):

Overall, the work presents an impressive and in-depth structural view of the conformational changes stemming from the interactions of (d)ATP allosteric effector molecules that are interrelated to RNR function. The manuscript is written clearly and provides a solid overview of RNR chemistry. The cryo-EM data show striking differences between ATP and dATP bound forms, though in select regions, the resolution is not good enough for strong interpretations of the finer details.

(1) In cryo-EM structures, dATP appears to shift the oligomerization equilibrium from nearly all dimeric forms (absence of dATP) to a mixture of both dimeric and tetrameric species (presence of dATP). The examination of the oligomeric composition in solution using the GEMMA - a mass spectral technique - showed somewhat similar trends, though given the magnitude of the differences, it was less compelling. Have the authors considered a complementary solution technique, such as analytical SEC or dynamic light scattering that could provide further support for the change in oligomerization as observed in the cryo-EM?

SEC analysis with 2 µM apoPcNrdD and higher nucleotide concentrations (1 mM) was performed, confirming the presence of both dimer and tetramer in solution at different ratios depending on the addition of ATP or dATP. The SEC analysis, included as Figure 3 - figure supplement 1, confirms the existence of an equilibrium in solution.

(2) The protein as isolated from the final SEC shows a predominant peak corresponding to aggregate protein. It would be helpful if the authors ran an analytical SEC on the protein sample that is more refined to see how much soluble dimer/tetramer vs. aggregate protein there is. This could impact the kinetic and thermodynamic analysis of effector interactions. Further, the second major peak is labeled as 'monomer'. Is the protein isolated as a monomer and then forms dimer upon effector binding? It is unclear. The authors should consider presenting the SEC standards for the given column and buffer condition so that a reasonable estimate of the oligomerization status of the isolated protein can be assigned.

Can the reviewer possibly have believed that Figure 1 - supplementary figure 2a shows PcNrdD rather than PcNrdG? The figure supplement corresponds to the as-isolated SEC analysis of the activase (PcNrdG), which shows the presence of two main peaks of aggregates and monomer. The monomeric peak was reinjected and showed no presence of further aggregation states. Currently it is not known which oligomeric state the activase harbours upon binding to PcNrdD and glycyl radical formation. None of the other SEC figures in the MS has any predominant peak corresponding to aggregated protein.

(3) More details are needed for the ITC section. The ITC methods are not clear. What is the exact composition of the ligand solution being titrated into the protein solution? It is unclear how the less-than-unity binding stoichiometry was determined and what it means. Is the n value for the monomer, dimer, or tetramer forms? It is concerning that n < 1 is observed for dATP binding in the ITC whereas there are 3 dATP bound/subunit in the cryo-EM. For completeness, titration of a buffer into protein solution (no ligand) should be conducted and presented to demonstrate that the heats produced in Figure 2 correspond to the ligand only (and not a buffer mismatch).

ITC experiments were performed in the presence of 1 mM GTP (c-site) and 1 mM dTTP (ssite). Unlike other parameters in ITC analyses, the N value is usually the least accurate of all fitted parameters and strongly depends on the concentration of the active protein in the sample. N values described in the current study are in the same range as values reported for ATP-cones in other RNRs and NrdR (Rozman Grinberg & al 2018a, 2018b, 2022 McKethan and Spiro 2013). The results most likely reflect two high-affinity binding sites for dATP and one high affinity binding site for ATP. Different nucleotide concentrations were used in the cryoEM and ITC experiments.

(4) It is intriguing that the binding of dATP doesn't quell the glycyl radical. In fact, it appears that, as the authors suggest, the amount of glycyl radical might be increased in these samples. However, the cryo-EM data indicates that the GRD is disordered. It is unclear how these would be correlated, as one would not expect a disordered structural element to maintain such a potent oxidant.

As already written above, we do not wish to imply that the GRD is completely or even highly disordered, just that its dynamics increase in the presence of dATP. Otherwise we completely agree that a very exposed Gly radical is incompatible with its stability. It could be that the amount of disorder is exaggerated somewhat by the vitrification process in cryo-EM. We have tried to reword some of the text to emphasise higher mobility rather than disorder.

It has been difficult to propose a direct structural mechanism for transmission of the allosteric signal from the a-site in the ATP-cone to the active site and GRD given that the ATP-cones and linker are disordered in the dATP-bound dimers and that the linker can not be completely modelled even in the dATP-bound tetramers. We initially thought that a steric occlusion mechanism might be at play, but the reality appears more complex. Most likely dATP binding causes a change in the dynamics of the linker region and NxN flap that directly affects substrate binding and simultaneously causes higher mobility of the GRD, given that all are part of a connected system. The structures determined in the presence of dATP and CTP show that CTP cannot bind in the absence of an ordered NxN flap. We hope that future structural studies of NrdDs from other organisms may shed further light on this mechanism.

(5) It is a bit difficult to keep track of the myriad of structural information and differences amongst the various nucleotide-dependent conditions. It would be useful for the authors to add a summary figure that depicts the various oligomers, orientations, and (dis)ordered structural elements with cartoon representations.

Thank you for this suggestion. It has been added as Figure 11.

(6) The mechanism by which (d)ATP binding changes the (dis)ordering of select loops based on the current cryo-EM data is unclear (even the authors agree). The addition of molecular dynamics (MD) simulations on two different structures to reveal the network or structural communication would be a great addition to the work and validate the structural data.

We have discussed this with a colleague who is an expert in MD. Their advice was that such simulations would be very difficult given that some amino-acids are missing in both of the relevant starting structures (ATP-CTP and dATP-CTP dimer) and could give very variable results. Thus we chose to do complementary experiments with hydrogen-deuterium exchange mass spectrometry (HDX-MS) instead. The results are included in the revised manuscript.

Minor points

(1) There are some conflicting reports as to whether P. copri is considered a human 'pathogen'. According to Yeoh, et al Scientific Reports 2022, P. copri is one of the predominant microbes in the human gut and is linked to a positive impact on metabolism. Perhaps the addition of a citation that provides support for it as a pathogen would clarify the statement on p. 3.

We have added a recent reference (Nii T, Maeda Y, Motooka D, et al. (2023) Genomic repertoires linked with pathogenic potency of arthritogenic Prevotella copri isolated from the gut of patients with rheumatoid arthritis. Ann Rheum Dis 82: 621-629. doi: 10.1136/annrheumdis-2022-222881).

(2) In Figure 3, the number of dimers/tetramers for dATP (100 uM) does not add up to 100.

What is the other 2%?

Thank you for pointing this out - it has been corrected.

(3) The data in Figures 5C and D do show slight changes that could be fit and interpreted as a 'weak' interaction. Thus, the statement on p 9 "where dATP-loaded PcNrdD could bind neither GTP nor CTP" should be changed to indicate that the interactions are weak (or that the nucleotides weakly associate).

The text and the figure have been changed according to the reviewer’s suggestion.

Author Response

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

Reviewer #1 (Recommendations For The Authors):

Firstly, the authors place a great deal of emphasis on the impact of the Hif1-a inhibitor PX-478. The literature surrounding this inhibitor and its mode of action indicates that it is not a direct inhibitor of activity but that its greatest impact is on the production of Hif1-a. The authors do include another inhibitor as a control, Echinomycin, but it does not appear to be as biologically active and the panel of experiments conducted with this is extremely limited. I would be more comfortable with a full Seahorse experimental panel for Echinomycin, similar to SFig 2.G as performed with PX-478.

We thank the reviewer for their comment highlighting the different mechanisms of action of the HIF-1α inhibitors used in this article. While echinomycin inhibits the binding of HIF-1α to the hypoxia response element (HRE) thereby blocking HIF-1a DNA binding capability, PX-478 inhibits HIF-1α deubiquitination, decreases HIF-1α mRNA expression, and reduces HIF-1α translation. We have included a paragraph explaining this phenomenon in the new version of the manuscript (page 9). In addition, we extended the panel of experiments performed with echinomycin, which confirmed a marked inhibition of the glycolytic pathway when DCs were stimulated with irradiated Mtb in the presence of echinomycin as assessed by SCENITH (new Figure S3H).

Similarly, it would be of value to have Seahorse profiling that directly excludes FAO from the metabolic profile through the use of Etomoxir as an inhibitor of fatty acid oxidation, which one would assume would have no impact on the metabolic response.

In order to estimate the contribution of FAO towards fueling protein synthesis in DCs stimulated with iMtb, the FAO inhibitor etomoxir was incorporated to the SCENITH method as previously described (Adamik et al., 2022). Overall, FAO dependence was found to be less than 10% in DCs, regardless of their activation state. While mitochondrial dependence is reduced after iMtb stimulation, there is no difference in FAO dependence, suggesting that OXPHOS is primarily driven by glucose in iMtb-stimulated cells. This is consistent with HIF1α-induced increase of glucose metabolism-related genes. We have adjusted the results section to include this new result (new Figure S1).

Aside from these minor points, I believe this to be a rigorous study.

Reviewer #2 (Recommendations For The Authors):

In Fig. 1 and Fig. 2, the authors conclude that Mtb rewires the metabolism of Mo-DCs and induces both glycolysis and OXPHOS. The data shows that infection with iMtb or Mtb increases glucose uptake and lactate release, suggesting an increase in glycolysis. However, an increase in lactate is not a measure of glycolysis. Lactate is a byproduct of glycolysis; the end product of glycolysis is pyruvate.

We are grateful for the reviewer's comment, as it gives us the opportunity to explain the conceptual framework on which we based our study. Traditionally, pyruvate has been considered to be the end product of glycolysis when oxygen is present and lactate the end product under hypoxic conditions. Numerous studies have shown that lactate is produced even under aerobic conditions (Brooks, 2018). Therefore, we frame this work in accordance with this view that states that glycolysis begins with glucose as its substrate and terminates with the production of lactate as its main end product (Rogatzki, Ferguson, Goodwin, & Gladden, 2015; Schurr, 2023; Schurr & Schurr, 2017).

Secondly, since the authors have access to the Agilent Extracellular Flux Analyzer, they should have performed detailed ECAR/OCR measurements to conclusively demonstrate that both glycolysis and OXPHOS are increased in Mo. This is especially important for OXPHOS because the only readout shown for OXPHOS is an increase in mitochondrial mass (figure 1 G, H), which is not acceptable. Overall, the data does not indicate that Mtb triggers OXPHOS in the dendritic cells. It only indicates dead iMtb increases the mass of mitochondria in DCs.

The reviewer’s advice is well appreciated. However, we would like to clarify what may be a misunderstanding; that is, the assays alluded to by the reviewer were not performed on monocytes but on DCs. As advised by the reviewer, we now include the OCR measurements by Seahorse and describe the figures according to their order of appearance in the new version of the manuscript.

What happens to the mitochondrial mass when infected with live Mtb?

In response to the reviewer’s question, we determined the mitochondrial mass in infected DCs with live Mtb. In contrast to DCs treated with irradiated Mtb, those infected with live bacteria showed a clear reduction of their mitochondrial mass (modified Figure 1G). This result indicates that, although both Mtb-infected and irradiated Mtb-exposed DCs show a clear increase in their glycolytic activity, divergent responses are observed in terms of mitochondrial mass.

It will be best if the authors indicate in the figure headings that dead Mtb was used.

We agree with the reviewer. For figures 1-3, we applied the term “Mtb” in the figure headings since both irradiated and viable bacteria were used for the corresponding experiments. In figures 4-5, the term “iMtb” (alluding to irradiated Mtb) was used in the figure headings as suggested by the reviewer. For the remaining figures, the term “iMtb” was indicated in their legends when dead bacteria weres used to stimulate DCs.

E.g., Figure 1F; what does live Mtb do to GLUT1 levels etc etc?

In response to the reviewer’s question, we included new data about Glut1 expression in DCs infected with live Mtb in the latest version of the manuscript. In line with the increase in glucose uptake shown in figure 1B, we observed an increase in the percentage of Glut1 positive DCs upon Mtb infection (new Figure 1F, lower panels). The increase in Glut1 expression strengthens the notion that DCs activates their glycolytic activity in response to the infection, as demonstrated by the elevated release of lactate, glucose consumption, HIF-1α expression, LDHA expression (Figure 1) and glycolytic activity (Figure 2, SCENITH results with viable Mtb). Therefore, these data strongly support the induction of glycolysis by Mtb (either viable or irradiated) in DCs.

Also, we found that they were still able to activate CD4+ T cells from PPD+ donors in response to iMtb. This activation of CD4 T cells with iMtb in the presence of a HIF-1alpha inhibitor is expected, as iMtb is dead and not virulent. What happens when the cells are infected with live virulent Mtb?

We would like to clarify the main purpose of the DC-T cells co-culture assays in the presence of the HIF-1α inhibitors. To characterize the impact of HIF-1α on DC functionality, we assessed the capacity of DCs to activate autologous CD4+ T cells when stimulated with iMtb in the presence of HIF-1α inhibitors. To this end, we used iMtb merely as a source of antigens to load DCs and evaluate the effect of HIF-1α inhibition on the activation of antigen-specific T cell. The use of viable Mtb may introduce confounding factors, such as pathogen-triggered inhibitory mechanisms (e.g., EsxH secretion by Mtb, (Portal-Celhay et al., 2016)), which would prevent us from reaching conclusions about the role of HIF-1α. Thus, we consider that the use of live bacteria for this experiment is out of the scope of this manuscript.

The authors demonstrated that CD16+ monocytes from TB patients have higher glycolytic capacity than healthy controls Fig 7. The authors should differentiate TB patient monocytes into DCs and measure their bioenergetics to test if infection alters their glycolysis and OXPHOS.

In agreement with the reviewer, the determination of metabolic pathways in DCs differentiated from monocytes of TB patients is a key aspect of this work. Accordingly, the bioenergetic determinations of DCs generated from monocytes from TB patients versus healthy subjects are now illustrated in Figures 6F (lactate release) and 6G (SCENITH profile).

In the discussion, the authors state that "pathologically active glycolysis in monocytes from TB patients leads to poor glycolytic induction and migratory capacities of monocyte-derived DCs." However, the data from Fig. 1 and 2 show that treatment with iMtb or Mtb induces glycolysis in MoDCs. How do the authors explain these contrasting results?

We thank the reviewer for pointing out this issue. Figures 1 and 2 show DCs differentiated from monocytes of healthy donors (HS). In this case, DCs from HS respond to Mtb by inducing a glycolytic and migratory profile. Yet, in the case of monocytes isolated from TB patients, these cells exhibit an early glycolytic profile from the beginning of differentiation, ultimately yielding DCs with low glycolytic capacity and low migratory activity in response to Mtb. We included this explanation in the discussion (page 18) to better clarify this issue.

Also, the term "pathological" active glycolysis (Introduction and Discussion) is an inappropriate term.

As requested by the reviewer, we excluded the term “pathological” to describe the phenomenon reported in this study.

Lastly, it should be shown whether the DCs generated from CD16+ monocyte from TB patients generate tolerogenic and/or aberrant DCs, which have lower glycolytic and migration capacity compared to the CD16- monocyte population. In Figure 7B, the authors should discuss why the CD16+ monocyte population has lower glycolytic capacity compared to CD16- monocytes in healthy donors. Furthermore, in contrast to the TB patients, do DCs generated from CD16+ monocyte in healthy donors have increased glycolytic and migration capacity compared to CD16- monocyte (because these monocytes showed lower glycolytic capacity)? Furthermore, if there is no difference in glycolytic capacity among the three monocyte populations in TB patients, on what basis was it concluded that DCs generated only from the CD16+ monocyte population may be the cause of lower migration capacity? The authors state in Figure 7F that the DMOG pretreatment matches the situation where the Mo-DCs from TB patients showed reduced migration. Did the authors check the Hif-1alpha levels in monocytes obtained from TB patients?

We appreciate this in-depth analysis by the reviewer because it allows us to clarify some interpretations of the SCENITH results in Figure 7B. It is important to keep in mind that with the SCENITH technique we can only infer about the relative contributions between the metabolic pathways, without alluding to the absolute magnitudes of such contributions. In this regard, it is key to note that the amount of lactate released during the first hours of the TB monocyte culture is much higher than that released by monocytes from healthy subjects (HS, Figure 7A), even when most of monocytes, which are CD14+ CD16-, have comparable glycolytic capacities between HS and TB. Another example to illustrate how to interpret SCENITH results can be found in Figure 2, where a lower mitochondrial dependence is observed in iMtb-stimulated DCs (Figure 2A), while the absolute ATP production associated to OXPHOS is indeed higher as measured by Seahorse (Figure 2D). Therefore, the glycolytic capacity is not a direct readout of the magnitude of glycolysis, but of its contribution to total metabolism. The low levels of lactate released from HS monocytes likely reflects their low activation state and low metabolic activity compared to TB monocytes. In this regard, we have previously demonstrated that monocytes from pulmonary TB patients display an activated phenotype (Balboa et al., 2011). The fact that there is no difference between the glycolytic capacities of TB and HS CD16- monocytes indicates that their proportional contributions to protein synthesis are comparable (again, without inferring about their absolute values, which may be very different).

Beyond the previous clarification, the reviewer's proposal to isolate subsets of monocytes is a very interesting idea. However, the experimental approach is very difficult based on the amount of blood we can obtain from patients. The cohort of patients included in this work comprises very severe patients and we are given up to 15-20 ml of peripheral blood from each. This volume of blood yields up to 10 million PBMC with approximately 1 million monocytes. If we separate the monocyte subsets, the recovered cells per condition will be insufficient to perform the intended assays.

Nevertheless, we incorporate new evidence that TB disease is associated with an increased activation and glycolytic profile of circulating CD16+ monocytes.

i) First, we show that the baseline glycolytic capacity of CD16+ monocytes correlates with time since the onset of TB-related symptoms (new Figure 7C).

ii) Second, we performed high-throughput GeneSet Enrichment Analysis (GSEA) on transcriptomic data (GEO accession number: GSE185372) of CD14+CD16-, CD14+CD16+ and CD14dimCD16+ monocytes isolated from individuals with active TB, latent TB (IGRA+), as well as from TB negative healthy controls (IGRA-). We found enrichments that, unlike oxidative phosphorylation, glycolysis tends to increase in active TB in both CD14+CD16+ and CD14dimCD16+ monocytes (new Figure 7D).

iii) We measured the expression of HIF-1α in monocyte subsets by FACS and found that this transcription factor is expressed at higher levels in CD16+ monocyte subsets from TB patients compared to their counterparts from healthy donors (new Figure 8 A). We consider this result justifies the assays shown in Figure 8B-C, in which we prematurely activated HIF-1α in healthy donor monocytes during early differentiation to DCs and measured its impact on the migration of the generated DCs.

In the Discussion, the authors mention that circulating monocytes from TB patients differentiate from DCs with low immunogenic potential. However, the authors have not shown any immunological defect in any of their data with monocytes from TB patients. In the proxy model mentioned in Figure 7, they have in fact shown that these preconditioned DCs have higher CD86 expression. Can the authors explain/show data to justify the statement in the first paragraph of the Discussion?

We agree with the reviewer on this observation. Our findings are limited to the generation of DCs with low migratory potential (low chemotactic activity towards CCL21 of DC differentiated from TB patient monocytes shown in figure 6H and of DC generated from pre-conditioned monocytes shown in figure 8C). We have modified that part of the discussion to better clarify this point, replacing migratory with immunogenic.

The authors should note that oxamate is a competitive inhibitor of the enzyme lactate dehydrogenase and not glycolysis. Also, LDHA catalyzes the conversion from pyruvate to lactate and not the other way around (Results, page 6).

This comment relates to the first one by the reviewer, in which the dogma of glycolysis was discussed. According to the new conception of glycolysis, it begins with glucose as its substrate and terminates with the production of lactate as its main end product.

The following statements by the authors on page 6 are incorrect: "Because irradiated and viable Mtb induced comparable activation of glycolysis, we subsequently performed all our assays with irradiated Mtb only in the rest of the study due to biosafety reasons." and: "To our knowledge, this is the first study addressing the metabolic status and migratory activity of Mo-DCs from TB patients."

We deleted the first sentence and reworded the second sentence as "To our knowledge, this is the first study to address how the metabolic status of monocytes from TB patients influences the migratory activity of further differentiated DCs".

The Discussion reads as if live Mtb was used in the experiments, which is not the case. This should be corrected.

We changed Mtb for iMtb when it was the case in the discussion. In some cases, Mtb stimulation was used instead of Mtb infection.

Minor Comments:

(1) In Figure 1F legend "Quantification of Glut1+ cells plotted to the right". The underlined part should be "plotted below".

It was corrected.

(2) In Figure 1H. Please describe the quantitation method and describe how many cells or the number/size of fields were used to quantitate mitochondria.

For mitochondrial morphometric analysis, TEM images were quantified with the ImageJ “analyze particles” plugin in thresholded images, with size (μm2) settings from 0.001 to infinite. For quantification, 8–10 cells of random fields (1000x magnification) per condition were analyzed. We included this information in the methods section of the new version of the manuscript.

(3) Please mention the number of independent experimental repeats for each experimental data set and figure.

In each figure, the number of independent experiments is indicated by individual dots.

(4) In Figure 2A legend, "PER; left panel" should be PER; lower panel and "OCR; right panel" should be OCR; upper panel.

It was corrected.

References for reviewers

Adamik, J., Munson, P. V., Hartmann, F. J., Combes, A. J., Pierre, P., Krummel, M. F., … Butterfield, L. H. (2022). Distinct metabolic states guide maturation of inflammatory and tolerogenic dendritic cells. Nature Communications 2022 13:1, 13(1), 1–19. https://doi.org/10.1038/s41467-022-32849-1

Balboa, L., Romero, M. M., Basile, J. I., Sabio y Garcia, C. A., Schierloh, P., Yokobori, N., … Aleman, M. (2011). Paradoxical role of CD16+CCR2+CCR5+ monocytes in tuberculosis: efficient APC in pleural effusion but also mark disease severity in blood. Journal of Leukocyte Biology. https://doi.org/10.1189/jlb.1010577

Brooks, G. A. (2018). Cell Metabolism The Science and Translation of Lactate Shuttle Theory. Cell Metab. https://doi.org/10.1016/j.cmet.2018.03.008

Portal-Celhay, C., Tufariello, J. M., Srivastava, S., Zahra, A., Klevorn, T., Grace, P. S., … Philips, J. A. (2016). Mycobacterium tuberculosis EsxH inhibits ESCRT-dependent CD4+ T-cell activation. Nature Microbiology, 2, 16232. https://doi.org/10.1038/NMICROBIOL.2016.232

Rogatzki, M. J., Ferguson, B. S., Goodwin, M. L., & Gladden, L. B. (2015). Lactate is always the end product of glycolysis. Frontiers in Neuroscience, 9(FEB), 125097. https://doi.org/10.3389/FNINS.2015.00022/BIBTEX

Schurr, A. (2023). From rags to riches: Lactate ascension as a pivotal metabolite in neuroenergetics. Frontiers in Neuroscience, 17, 1145358. https://doi.org/10.3389/FNINS.2023.1145358/BIBTEX

Schurr, A., & Schurr, A. (2017). Lactate, Not Pyruvate, Is the End Product of Glucose Metabolism via Glycolysis. Carbohydrate. https://doi.org/10.5772/66699

Author Response

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

Thank you for your continued review and for providing insightful suggestions. Below, I share some unpublished new findings related to the MYRF ChIP, comment on the potential interplay between myrf-1 and myrf-2, and describe the modifications we've implemented to address the reviewers' comments.

(1) MYRF-1 ChIP

Our collaboration with the modERN (Model Organism Encyclopedia of Regulatory Networks) project has recently yielded MYRF ChIP data. The results demonstrate clear and consistent MYRF binding across samples, notably on the lin-4 promoter. Given the significant detail and extensive description required to adequately present these findings, we have decided it is impractical to include them in the current paper. These results will be more suitably published in a separate ongoing study focused on MYRF's regulatory targets during larval development.

(2) Inter-regulation between myrf-1 and myrf-2

We acknowledge the interpretation that myrf-2 may act as a genetic antagonist to myrf-1, as suggested by the delayed arrest in myrf-1; myrf-2 double mutants and a trend towards increased lin-4 expression in myrf-2 mutants. Additionally, our unpublished data suggest an elevated myrf-2 expression peak in myrf-1 null mutants during the L1-L2 transition, indicating a potential mutual repressive interaction between myrf1 and myrf-2.

On the other hand, myrf-1 and myrf-2 exhibit functional redundancy in DD synaptic rewiring and lin-4 expression. A gain of function in myrf-2 promotes early DD synaptic rewiring. Furthermore, three independent co-immunoprecipitation analyses targeting myrf-1::gfp, myrf-2::gfp, and pan-1::gfp confirm a tight association between myrf-1 and myrf-2 in vivo. These findings challenge the notion of myrf-2 primarily antagonizing myrf-1, or vice versa.

We propose a model where myrf-1 and myrf-2 collaborate and are functionally redundant, with compensatory elevated expression when one paralog is absent. For instance, the loss of myrf-1 triggers upregulation of myrf-2, which, though insufficient on its own, accelerates the transcriptional program and exacerbates system deterioration, leading to accelerated death. How exactly this takes place is currently unclear. We notice the MYRF binding on both myrf-1 and myrf-2 genes in MYRF-ChIP.

Given the complexity of these interactions, we have chosen not to delve deeply into this discussion in the paper without more direct evidence, which would require detailed analysis.

(3) Revisions Addressing Reviewer Suggestions

(a) We have revised our interpretation of the mScarlet signal changes in myrf-1(ybq6) and myrf-2(ybq42) mutants to reflect a more nuanced understanding of their potential genetic relationship, as highlighted in the main text.

“The mScarlet signals exhibit a marked reduction in the putative null mutant myrf-1(ybq6) (Figure 1D, E). Intriguingly, in the putative null myrf-2(ybq42) mutants, there is a noticeable trend towards increased mScarlet signals, although this increase does not reach statistical significance (Figure 2C, D).”

(b) In response to feedback on Figure 2 and the characterization of lin-4(umn84) mutants, we've included a new series of images showing lin-4(umn84)/+ and lin-4(umn84) signals through larval stages, presented as Figure 2 Figure Supplement 2. This addition clarifies the functional status of lin-4 nulls in our study.

“Our observations revealed that mScarlet signals were not detected early L1 larvae (Figure 2C-F; Figure 2 Figure Supplement 2).”

(c) To improve the clarity of Fig 6, we've added indicator arrows in the red, green, and merge channels, enhancing the visualization of the signals.

We appreciate the opportunity to clarify these points and hope that our revisions and additional data address the concerns raised.

Author Response

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

We greatly appreciate the reviewers' and editors' comments and suggestions on our manuscript "Transposable elements regulate thymus development and function." We performed additional analyses to validate our results and rephrased some manuscript sections according to the comments. We believe these changes significantly increase the solidity of our conclusions. Our point-by-point answer to the reviewers' and editors' comments is detailed below. New data and analyses are shown in Figure 1d, Figure 2g and h, Figure 5e and f, Figure 1 – figure supplement 1, Figure 2 – figure supplement 2, Figure 3 – figure supplement 1 and 2, Figure 4 – figure supplement 2, Figure 5 – figure supplement 1, as well as the corresponding text sections.

Reviewer #1:

(1) The authors sometimes made overstatements largely due to the lack or shortage of experimental evidence.

For example in figure 4, the authors concluded that thymic pDCs produced higher copies of TE-derived RNAs to support the constitutive expression of type-I interferons in thymic pDCs, unlike peripheral pDCs. However, the data was showing only the correlation between the distinct TE expression pattern in pDCs and the abundance of dsRNAs. We are compelled to say that the evidence is totally too weak to mention the function of TEs in the production of interferon. Even if pDCs express a distinct type and amount of TE-derived transcripts, it may be a negligible amount compared to the total cellular RNAs. How many TE-derived RNAs potentially form the dsRNAs? Are they over-expressed in pDCs?

The data interpretation requires more caution to connect the distinct results of transcriptome data to the biological significance.

We contend that our manuscript combines the attributes of a research article (novel concepts) and a resource article (datasets of TEs implicated in various aspects of thymus function). The critical strength of our work is that it opens entirely novel research perspectives. We are unaware of previous studies on the role of TEs in the human thymus. The drawback is that, as with all novel multi-omic systems biology studies, our work provides a roadmap for a multitude of future mechanistic studies that could not be realized at this stage. Indeed, we performed wet lab experiments to validate some but not all conclusions: i) presentation of TE-derived MAPs by TECs and ii) formation of dsRNAs in thymic pDCs. In response to Reviewer #1, we performed supplementary analyses to increase the robustness of our conclusions. Also, we indicated when conclusions relied strictly on correlative evidence and clarified the hypotheses drawn from our observations.

Regarding the Reviewer's questions about TE-derived dsRNAs, LINE, LTR, and SINE elements all have the potential to generate dsRNAs, given their highly repetitive nature and bi-directional transcription (1). As ~32% of TE subfamilies are overexpressed in pDCs, we hypothesized that these TE sequences might form dsRNA structures in these cells. To address the Reviewer's concerns regarding the amount of TE-derived RNAs among total cellular RNAs, we also computed the percentage of reads assigned to TEs in the different subsets of thymic APCs (see Reviewer 1 comment #4).

(2) Lack of generality of specific examples. This manuscript discusses the whole genomic picture of TE expression. In addition, one good way is to focus on the specific example to clearly discuss the biological significance of the acquisition of TEs for the thymic APC functions and the thymic selection.

In figure 2, the authors focused on ETS-1 and its potential target genes ZNF26 and MTMR3, however, the significance of these genes in NK cell function or development is unclear. The authors should examine and discuss whether the distinct features of TEs can be found among the genomic loci that link to the fundamental function of the thymus, e.g., antigen processing/presentation.

We thank the Reviewer for this highly relevant comment. We investigated the genomic loci associated with NK cell biology to determine if ETS1 peaks would overlap with TE sequences in protein-coding genes' promoter region. Figure 2h illustrates two examples of ETS1 significant peaks overlapping TE sequences upstream of PRF1 and KLRD1. PRF1 is a protein implicated in NK cell cytotoxicity, whereas KLRD1 (CD94) dimerizes with NKG2 and regulates NK cell activation via interaction with the nonclassical MHC-I molecule HLA-E (2, 3). Thus, we modified the section of the manuscript addressing these results to include these new analyses:

"Finally, we analyzed publicly available ChIP-seq data of ETS1, an important TF for NK cell development (4), to confirm its ability to bind TE sequences. Indeed, 19% of ETS1 peaks overlap with TE sequences (Figure 2g). Notably, ETS1 peaks overlapped with TE sequences (Figure 2h, in red) in the promoter regions of PRF1 and KLRD1, two genes important for NK cells' effector functions (2, 3)."

(3) Since the deep analysis of the dataset yielded many intriguing suggestions, why not add a discussion of the biological reasons and significance? For example, in Figure 1, why is TE expression negatively correlated with proliferation? cTEC-TE is mostly postnatal, while mTEC-TE is more embryonic. What does this mean?

We thank the Reviewer for this comment. To our knowledge, the relationship between cell division and transcriptional activity of TEs has not been extensively studied in the literature. However, a recent study has shown that L1 expression is induced in senescent cells. We therefore added the following sentences to our Discussion:

"The negative correlation between TE expression and cell cycle scores in the thymus is coherent with recent data showing that transcriptional activity of L1s is increased in senescent cells (5). A potential rationale for this could be to prevent deleterious transposition events during DNA replication and cell division."

We also added several discussion points regarding the regulation of TEs by KZFPs to answer concerns raised by Reviewer 2 (see Reviewer 2 comment #1).

(4) To consolidate the experimental evidence about pDCs and TE-derived dsRNAs, one option is to show the amount of TE-derived RNA copies among total RNAs. The immunohistochemistry analysis in figure 4 requires additional data to demonstrate that overlapped staining was not caused by technical biases (e.g. uneven fixation may cause the non-specifically stained regions/cells). To show this, authors should have confirmed not only the positive stainings but also the negative staining (e.g. CD3, etc.). Another possible staining control was showing that non-pDC (CD303- cell fractions in this case) cells were less stained by the ds-RNA probe.

We thank the Reviewer for this suggestion. We computed the proportion of reads in each cell assigned to two groups of sequences known to generate dsRNAs: TEs and mitochondrial genes (1). These analyses showed that the proportion of reads assigned to TEs is higher in pDCs than other thymic APCs by several orders of magnitude (~20% of all reads). In contrast, reads derived from mitochondrial genes had a lower abundance in pDCs. We included these results in Figure 4 – figure supplement 2 and included the following text in the Results section entitled "TE expression in human pDCs is associated with dsRNA structures":

"To evaluate if these dsRNAs arise from TE sequences, we analyzed in thymic APC subsets the proportion of the transcriptome assigned to two groups of genomic sequences known as important sources of dsRNAs, TEs and mitochondrial genes (1). Strikingly, whereas the percentage of reads from mitochondrial genes was typically lower in pDCs than in other thymic APCs, the proportion of the transcriptome originating from TEs was higher in pDCs (~22%) by several orders of magnitude (Figure 4 – figure supplement 2)."

As a negative control for the immunofluorescence experiments, we used CD123- cells. Indeed, flow cytometry analysis of the magnetically enriched CD303+ fraction was around 90% pure, as revealed by double staining with CD123 and CD304 (two additional markers of pDCs): CD123- cells were also CD304-/lo, showing that these cells are non-pDCs. Thus, we decided to compare the dsRNA signal between CD123+ cells (pDCs) and CD123- cells (non-pDCs). The difference between CD123+ and CD123- cells was striking (Figure 4d).

Author response image 1.

Reviewer #1 (Recommendations For The Authors):

It was sometimes difficult for me to recognize the dot plots representing low expression against the white background. e.g., figure 1 supplement 1.

We thank the Reviewer for their comment, and we modified Figure 1 – figure supplement 1 as well as Figure 3 – figure 3 supplement 2 to improve the contrast between dots and background.

Reviewer #2:

Reviewer #2 (Recommendations For The Authors):

(1) In the abstract, results and discussion, the following conclusions are drawn that are not supported by the data: a) TEs interact with multiple transcription factors in thymic cells, b) TE expression leads to dsRNA formation, activation of RIG-I/MDA5 and secretion of IFN-alpha, c) TEs are regulated by cell proliferation and expression of KZFPs in the thymus. All these statements derive from correlations. Only one TF has ChIP-seq data associated with it, dsRNA formation and/or IFN-alpha secretion could be independent of TE expression, and whilst KZFPs most likely regulate TEs in the thymus, the data do not demonstrate it. The authors also seem to suggests that AIRE, FEZF2 and CHD4 regulate TEs directly, but binding is not shown. The manuscript needs a thorough revision to be absolutely clear about the correlative nature of the described associations.

We agree with Reviewer #2 that some of the conclusions in our initial manuscript were not fully supported by experimental data. In the revised manuscript, we clearly indicated when conclusions relied strictly on correlative evidence and clarified the hypotheses drawn from our observations. Regarding the regulation of TE expression by AIRE, FEZF2, and CHD4, we reanalyzed publicly available ChIP-seq data of AIRE and FEZF2 in murine mTECs. For AIRE, we confirmed that ~30% of AIRE's statistically significant peaks overlap with TE sequences (see Reviewer 2, comment #6 for more details on read alignment and peak calling), confirming its ability to bind to TE sequences directly. We added these results to the main figures (Figure 5f) and modified the "AIRE, CHD4, and FEZF2 regulate distinct sets of TE sequences in murine mTECs" as follows:

“[…]. As a proof of concept, we validated that 31.42% of AIRE peaks overlap with TE sequences by reanalyzing ChIP-seq data, confirming AIRE's potential to bind TE sequences (Figure 5f)."

A reanalysis of FEZF2's ChIP-seq data yielded no significant peaks while using stringent criteria. For this reason, we decided to exclude these data and only use AIRE as a proof of concept.

Regarding KZFPs, we agree with Reviewer #2 that their impact on TE expression is probably significantly underestimated in our data. A potential reason for this is that KZFP expression is typically low; thus, transcriptomic signals from KZFPs could have been missed by the low depth of scRNA-seq. We mentioned this point in the Discussion:

"On the other hand, the contribution of KZFPs to TE regulation in the thymus is likely underestimated due to their typically low expression (6) and scRNA-seq's limit of detection."

(2) On the technical side, there are many dangers about analyzing RNA-seq data at the subfamily level and without stringent quality control checks. Outputs may be greatly confounded by pervasive transcription (see PMID 31425522), DNA contamination, and overlap of TEs with highly expressed genes. Whether TE transcripts are independent units or part of a gene also has important implications for the conclusions drawn. I would say that for most purposes of this work, an analysis restricted to independent TE transcripts, with appropriate controls for DNA contamination, would provide great reassurances that the results from subfamily-level analyses are sound. Showing examples from the genome browser throughout would also help.

We agree with the Reviewer that contamination could have interfered with TE quantification. We used FastQ Screen (7) to evaluate the contamination of our human scRNA-seq data. As illustrated in the Figure below, most reads aligned with the human genome, and there were no reads uniquely assigned to another species analyzed, confirming the high purity of our dataset.

Author response image 2.

As stated by the Reviewer, pervasive expression is another factor that can lead to overestimation of TE expression. To evaluate if pervasive expression impacted the results of our differential expression analysis of TEs between APC subsets, we visualized read alignment to TE sequences using a genome browser. We selected two samples containing the highest numbers of mTEC(II) and pDCs (T07_TH_EPCAM and FCAImmP7277556, respectively) and used STAR to align reads to the human genome (GRCh38). We then visualized read alignment to randomly selected loci of two subfamilies identified as overexpressed by mTEC(II) or pDCs (HERVE-int and Harlequin-int, respectively). The examples below show that the signal detected is specific to the TE sequences located in introns. Even though this visualization cannot guarantee that pervasive expression did not affect TE quantification in any way, it increases the confidence that the signal detected by our analyses genuinely originates from TE expression.

Author response image 3.

Author response image 4.

Author response image 5.

Author response image 6.

Author response image 7.

(3) Related to the above, it would be useful to describe in the main text (and methods) how multi-mapping reads are being handled. It wasn't clear to me how kallisto handles this, and it has implications for the results. In the analysis suggested above, only uniquely mapped reads would have to be used, despite its limitations.

We agree with the Reviewer that this information regarding assignment of multimapping reads is important. Kallisto uses an expectation-maximization (EM) algorithm to deal with multimapping reads, a strategy used by several algorithms developed to study TE expression (8). Briefly, the EM algorithm reassigns multimapping reads based on the number of uniquely mapped reads assigned to each sequence. Thus, we added the following details to the methods section:

"Preprocessing of the scRNA-seq data was performed with the kallisto (9), which uses an expectation-maximization algorithm to reassign multimapping reads based on the frequency of unique mappers at each sequence, and bustools workflow."

(4) Whilst I liked the basic idea, I am not convinced that correlating TE and TF expression is a good strategy for identifying TE-TF associations at enhancers. Enhancers express very low levels of short transcripts, which I doubt would be detected in low-depth scRNA-seq data. The transcripts the authors are using to make such associations may therefore have nothing to do with the enhancer roles of TEs. I would limit these analyses to cell types for which there is histone modification data and correlate TF expression with that instead.

We agree with the Reviewer that it would have been interesting to correlate the expression of TFs with signals of histone marks at TE sequences. However, we could not perform this analysis because we did not have matched data of histone marks throughout thymic development. Therefore, we adopted an alternative, well-suited strategy.

Our strategy to identify TE enhancer candidates is depicted in Figure 2a: i) correlation between the expression of the TF and the TE subfamily, ii) presence of the TF binding motif in the sequence of the TE enhancer candidate, and iii) colocalization of the TE enhancer candidate with significant peaks of H3K27ac and H3K4me3 in the same cell type from the ENCODE Consortium ChIP-seq data. We limited our analyses to the eight cell types present both in our dataset and the ENCODE Consortium: B cells, CD4 Single Positive T cells (CD4 SP), CD8 Single Positive T cells (CD8 SP), dendritic cells (DC), monocytes and macrophages (Mono/Macro), NK cells, Th17, and Treg.

(5) Figure 2G: binding of ETS1 is unconvincing. Were there statistically meaningful peaks called in these regions? It would be good to also show a metaplot/heatmap of ETS1 profile over all elements of relevant subfamilies. Showing histone marks on the genome browser snapshots would also be useful. Is there any transcriptional evidence that the specific Alus shown act as alternative promoters?

We agree with the Reviewer that the examples provided were not particularly convincing. Thus, we reanalyzed the data to determine if statistically significant ETS1 peaks (see the answer to Reviewer 2's comment #6 for details on the methods) located near gene transcription start sites overlapped with TEs. We thereby provided examples of significant ETS1 peaks overlapping TE sequences in the promoter region of two prototypical NK cell protein-coding genes (Figure 2h).

(6) Why was -k 10 used with bowtie2? This will map the same read to multiple locations in the genome, increasing read density at more repetitive (younger) TEs. The authors should use either default settings, being clear about the outcome (random assignment of multimapping reads to one location), or use only uniquely aligned reads.

We thank the Reviewer for their comment and agree that using the -k 10 parameter with bowtie2 was not optimal for TE analysis. To improve the strength of our analyses, we reanalyzed all ChIP-seq data of our manuscript (Figure 2g and h, Figure 5e and f) using the following strategy: alignment with bowtie2 using default parameters except –very-sensitive, multimapping read removal with samtools view -q 10, removal of duplicate reads with samtools markdup -r, peaks calling was performed with macs2 with the -m 5 50 parameter, and peaks overlapping ENCODE's blacklist regions were removed with bedtools intersect.

These new analyses strengthen our evidence that TEs interact with multiple genes that regulate thymic development and function. We updated the results sections concerning ChIP-seq data analyses and the Methods section to include this information:

"ChIP-seq reads were aligned to the reference Homo sapiens genome (GRCh38) using bowtie2 (version 2.3.5) (10) with the --very-sensitive parameter. Multimapping reads were removed using the samtools view function with the -q 10 parameter, and duplicate reads were removed using the samtools markdup function with the -r parameter (11). Peak calling was performed with macs2 with the -m 5 50 parameter (12). Peaks overlapping with the ENCODE blacklist regions (13) were removed with bedtools intersect (14) with default parameters. Overlap of ETS1 peaks with TE sequences was determined using bedtools intersect with default parameters. BigWig files were generated using the bamCoverage function of deeptools2 (15), and genomic tracks were visualized in the USCS Genome Browser (16)."

(7) Figure 1d needs a y axis scale. Could the authors also provide details of how the random distribution of TE expression was generated?

We agree that the Reviewer that Figure 1d was incomplete and made the appropriate modifications. Regarding the random distribution, we reproduced our dataset containing the expression of 809 TE subfamilies in 18 cell populations. For each combination of TE subfamily and cell type, we randomly assigned an "expression pattern" as identified by the hierarchical clustering of Figure 1b. Then, we computed the maximal occurrence of an expression pattern across cell types for each TE subfamily to generate the distribution curve in Figure 1d. We added the following details to the Methods section to clarify how the random distribution was generated:

"As a control, a random distribution of the expression of 809 TE subfamilies in 18 cell populations was generated. A cluster (cluster 1, 2, or 3) was randomly attributed for each combination of TE subfamily and cell type, and the maximal occurrence of a given cluster across cell types was then computed for each TE subfamily. Finally, the distributions of LINE, LTR, and SINE elements were compared to the random distribution with Kolmogorov-Smirnov tests."

(8) The motif analysis requires a minimum of 1 locus from each TE subfamily containing it in order to be reported, but this seems like a really low threshold that will output a lot of noise. What is the rationale here?

We agree with the Reviewer that this threshold might appear low. Nonetheless, these analyses ultimately aimed to identify TE promoter and enhancer candidates. Hence, we did not want to put an arbitrary threshold at a higher value (e.g., a certain number or percentage of all loci of a given TE subfamily), as this might create a bias based on the total number of loci of a given TE subfamily. Moreover, our rationale was that a TE locus might act as a promoter/enhancer even if it is the only locus of its subfamily containing a TF binding site.

Even though this strategy might have created some noise in the analyses of interactions between TFs and TEs of Figure 2 (panels a-e), we are confident that our bootstrap strategy efficiently removed low-quality identifications based on low correlations values or expression of TF and TE in low percentages of cells. Additionally, the subsequent analyses on TE promoter and enhancer candidates were performed exclusively for the TE loci containing TF binding sites to avoid adding noise to these analyses.

(9) Figure 4e: is this a log2 enrichment? If not, the enrichments for some of the gene sets are not so high.

The enrichment values represented in Figure 4e are not log-transformed. It is essential to highlight that gene set enrichment values were computed for each possible pair of thymic APCs (e.g., pDC vs. cDC1, pDC vs. mTEC(II), etc.), and the values represented in Figure 4e are an average of each comparison pictured at the bottom of the UpSet plot.

However, we agree with Reviewer 2 that the average enrichment value is not extremely high. We thus made the following modifications to the Results section ("TE expression in human pDCs is associated with dsRNA structures") to better represent it:

"Notably, thymic pDCs harbored moderate yet significant enrichment of gene signatures of RIG-I and MDA5-mediated IFN ɑ/β signaling compared to all other thymic APCs (Figure 4e and Supplementary file 1 – Table 8)."

(10) Please be clear on results subtitles when these refer to mouse.

We apologize for the confusion and modified the subtitles to clarify if the results refer to mouse or human data.

(11) Figure 1 - figure supplement 2: "assignation" should be 'assignment'.

We thank the Reviewer for their keen eye and changed the title of Figure 1 – figure supplement 2.

(1) Sadeq S, Al-Hashimi S, Cusack CM, Werner A. Endogenous Double-Stranded RNA. Noncoding RNA. 2021;7(1).

(2) Kim N, Kim M, Yun S, Doh J, Greenberg PD, Kim TD, et al. MicroRNA-150 regulates the cytotoxicity of natural killers by targeting perforin-1. J Allergy Clin Immunol. 2014;134(1):195-203.

(3) Gunturi A, Berg RE, Forman J. The role of CD94/NKG2 in innate and adaptive immunity. Immunol Res. 2004;30(1):29-34.

(4) Taveirne S, Wahlen S, Van Loocke W, Kiekens L, Persyn E, Van Ammel E, et al. The transcription factor ETS1 is an important regulator of human NK cell development and terminal differentiation. Blood. 2020;136(3):288-98.

(5) De Cecco M, Ito T, Petrashen AP, Elias AE, Skvir NJ, Criscione SW, et al. L1 drives IFN in senescent cells and promotes age-associated inflammation. Nature. 2019;566(7742):73-8.

(6) Huntley S, Baggott DM, Hamilton AT, Tran-Gyamfi M, Yang S, Kim J, et al. A comprehensive catalog of human KRAB-associated zinc finger genes: insights into the evolutionary history of a large family of transcriptional repressors. Genome Res. 2006;16(5):669-77.

(7) Wingett SW, Andrews S. FastQ Screen: A tool for multi-genome mapping and quality control. F1000Res. 2018;7:1338.

(8) Lanciano S, Cristofari G. Measuring and interpreting transposable element expression. Nat Rev Genet. 2020;21(12):721-36.

(9) Bray NL, Pimentel H, Melsted P, Pachter L. Near-optimal probabilistic RNA-seq quantification. Nat Biotechnol. 2016;34(5):525-7.

(10) Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods. 2012;9(4):357-9.

(11) Danecek P, Bonfield JK, Liddle J, Marshall J, Ohan V, Pollard MO, et al. Twelve years of SAMtools and BCFtools. Gigascience. 2021;10(2).

(12) Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, et al. Model-based analysis of ChIP-Seq (MACS). Genome Biol. 2008;9(9):R137.

(13) Amemiya HM, Kundaje A, Boyle AP. The ENCODE Blacklist: Identification of Problematic Regions of the Genome. Sci Rep. 2019;9(1):9354.

(14) Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics. 2010;26(6):841-2.

(15) Ramirez F, Ryan DP, Gruning B, Bhardwaj V, Kilpert F, Richter AS, et al. deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res. 2016;44(W1):W160-5.

(16) Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM, et al. The human genome browser at UCSC. Genome Res. 2002;12(6):996-1006.

Author Response

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

Reviewer #1:

The very detailed insights gained by the authors into allosteric regulation require very specialized techniques in this study. This poses a challenge to communicate the methods, the results, and the meaning of the results to a broader audience. In some places, the authors overcome this challenge better than in others.

Following this reviewer’s suggestions, we have extensively revised the text, making the text more understandable to a broader audience.

The manuscript does not show up on BioRxiv.

The manuscript is now deposited in Biorxv (doi: 10.1101/2023.09.12.557419)

Fig3: GS-ES2 transition: the changes appear minimal in the illustration.

As suggested by this reviewer, we have re-examined the GS-ES2 transition and clearly defined the structural characteristics of the conformationally excited state 2 (ES2) state. As shown in the revised Fig.3 of the main text, the ground state (GS) features a π-π packing between the aromatic rings of F100 and Y156, as well as a cation-π stacking between R308 and F102. In the ES2 state, these above interactions are disrupted, while a new π-π packing interaction is formed between F100 and F102. We added new comments in the main text clarifying these structural interactions that characterize each state.

GS-ES1 transition: how is the K72-E91 salt bridge disrupted? How do you define the formation/disruption of a salt bridge? The current figure does not make this very clear and the K72-E91 salt bridge appears to be intact in ES1. Maybe the authors could replace the dotted K72-E91 line with a dotted line and distance?

As stated above, we revised Fig. 3 highlighting the differences between the two states. The K72 and E91 salt bridge is formed when the distance between Nε of K72 and Oε of E91 is shorter than 4.0 Å (the typical cutoff for a salt bridge). In the ES1 state, the outward movement of the αC helix increases the distance over 4.5 Å, disrupting the salt bridge.

L251: Could the authors remind the reader why they are only comparing V104 and I150? Could they give a little context as to why they consider the agreement to be good? It appears that they would be statistically different, so a little context for what comprises a good agreement in the literature may be helpful.

Our mutagenesis studies show that V104 and I150 are key residues for allosteric communication, and if mutated, result in well-folded but inactive kinases (Sci Adv. doi: 10.1126/sciadv.1600663). Importantly, V104 and I150 show two distinct populations in the CEST experiments that can be directly related to the GS and ES states. Regarding the fitting of these residues, we obtained a good agreement with the direction of the chemical shifts, which supports the hypothesized GS -> ES structural transition. The lack of a quantitative agreement between the chemical shifts of the experimental and simulated excited state is not surprising for two reasons a) all state-of-the art simulations fall short in sampling slow conformational interconversions, and b) the uncertainty of the SHIFTX algorithm for the prediction of 13C chemical shifts of methyl groups is quite large. Finally, we would like to point out that most NMR relaxation-dispersion experiments (CEST and CPMG) are performed for the backbone 15N, 13Calpha and 1H resonances, which have been used to calculate the structures of the intermediate states (Neudecker, P. et. al Science, 2012, 336,doi: 10.1126/science.1214203) and yield reasonable agreement with the prediction for metastable states derived from Markov Models (Olsson, S. J. Am. Chem. Soc., 2017,139,doi:10.1021/jacs.6b09460). To the best of our knowledge, there is no literature reporting on calculations of the 13C CEST profiles for methyl groups from MD simulations, and remarkably, we found a reasonably good agreement between experimental and predicted chemical shifts (see Fig.5C).

Just to clarify: the calculated CS values are informed by experimental CS values that were used in the calculation?

We used the backbone chemical shifts as the restraints only in the metadynamics simulations. We used the chemical shifts of the methyl groups and their corresponding excited states to verify the ES2 state.

Figure 8: in its current form this potentially exciting result is lost on the average reader.

we modified Fig. 8 of the main text, making the intra- and inter-residue correlations visible to the reader.

Reviewer #2:

While the alphaC-beta4 loop is a conserved feature of protein kinases, the residues within this loop vary across various kinase families and groups, enabling group and family-specific control of activity through cis and trans acting elements. F102 in PKA interacts with co-conserved residues in the C-tail, which has been proposed to function as a cis regulatory element. The authors should elaborate on the conformational changes in the C-tail, particularly in the arginine that packs against F102, in the results and discussion. This would further extend the impact and scope of the manuscript, which is currently confined to PKA.

As suggested by this reviewer, we re-analyzed the time-dependent interactions between F102 and R308 at the C-tail. As this reviewer suspected, these interactions differentiate the ES2 from the GS state. In the GS state, there is a stable cation-π interaction between F102 and R308, which becomes transient in the ES2 state (Fig. 3). For the F100A mutant, the interactions between F102 and R308 have lower occurrence relative to the WT enzyme, i.e., a weaker interaction between the αC-β4 loop and the C-tail (see new Figure 6 - figure supplement 1). The latter supports our conclusion that the structural coupling between the C-tail and the two lobes of the enzyme decreases for the F100A mutant. We added more comments in the main text.

FAIR standards of making the data accessible and reproducible are not directly addressed.

We have deposited all our NMR data on the Data Repository Site at the University of Minnesota, DRUM (https://hdl.handle.net/11299/261043).

The MD data and conformational states would be a valuable resource for the community and should be shared via some open-source repositories.

Due to the large size of the simulations (>500 GB), we could not deposit them in the Data Repository Site at the University of Minnesota (DRUM). We are actively working with the personnel at DRUM to upload all the trajectories in an alternate site. However, these data will be available to the public immediately upon request.

The authors state that ES1 and ES2 states are novel and not observed in previous crystal structures. The authors should quantify this through comparisons with PKA inactive states and with other AGC kinases.

We apologize for the confusion. We now clarify that the ES1 is a well-known inactivation pathway. As suggested by this reviewer, we now report a few examples of active and inactive conformations of PKA-C and other kinases (see new Figure 3 – figure supplement 2.). Briefly, ES1 corresponds to the typical αC-out conformation found for PKA-C bound to inhibitors or in R194A mutant. A similar conformation is present for Src, Abl, and CDK2. The C-out conformation features a disrupted β3K-αCE salt bridge, which is key for active kinases. In contrast, the transition GS-ES2 is not present in the inactive conformations deposited in the PDB.

Based on the results, can the authors speculate on the impact of oncogenic mutations in the alphaCbeta4 loop mutations in PKA?

We now include additional comments and another citation that further supports our findings. In short, the activation of a kinase is generated by mutation insertions that stabilize the αC-β4 loop as pointed out by Kannan and Zhang (see references 28, 30, and 68). In contrast, mutations that destabilize this allosteric site (e.g., F100A) are inactivating, disrupting the structural couplings of the two lobes (our work).

Reviewer #3:

The manuscript is somewhat difficult to read even for kinase experts, and even harder for the layman. The difficulty partially arises from mixing technical description of the simulations with structural interpretation of the results, which is more intuitive, and partially arises from the assumption that readers are familiar with kinase architecture and its key elements (the aC helix, the APE motif, etc).

We revised the text and modified Fig. 1 in the main text to make the paper more accessible to the general audience.

The authors haven't done a good job describing the ES2 state intuitively. From my examination of the figures, it appears that in the ES2 state, the kinase domain is more elongated and the N and the C lobes are relatively less engaged than in the ground state. This may or may not be exactly, but a more intuitive description of the ES2 state is needed.

As suggested by this reviewer, we include a better description of the ES2 state of the kinase and the structural details of the inactivation pathway. Also, we checked the radius of gyration of the two lobes for GS and ES2. ES2 is slightly more elongated with an Rg of 20.3 ± 0.1 Å as compared to the GS state (20.0 ± 0.2 Å). This marginal difference is consistent with our characterization of the local packing around the C-4 loop, in which the lack of stable interaction with E and C-tail in the ES2 state makes the overall structure less compact.

The authors need to introduce and give a brief description of technical terms such as CV (collective variable), PC (principal component) etc.

We now specify both collective variables and principal components and include those definitions in the Method section. Briefly, to characterize the complex conformational transitions of PKA-C, we utilize collective variables (Figure 2 – figure supplement 1). We chose these variables based on structural motifs described in the literature to define local and global structural transitions (Camilloni C., Vendruscolo, M, Biochemistry, 2015,54,7470; Kukic, P. et al. Structure, 2015,23, 745). On the other hand, we utilized the principal component analysis to compare the conformational changes of the kinase in the same two-dimensional space, revealing the two lowest frequencies that define the global motions of the enzyme (Figures 7C, D, and E).

The following paper should be discussed as it discussed similar ATP/substrate binding of Src kinase based on an extensive network that largely overlaps with the discussed PKA network. Foda, et al. "A dynamically coupled allosteric network underlies binding cooperativity in Src kinase." Nature communications 6.1 (2015): 5939.

We apologize for missing this citation. Indeed, it makes our finding more general as allosteric cooperativity is key in other kinases such as Src and ERK2. We included this in the Discussion section.

The CHESCA analysis appears to be an add-on that doesn't add much value. It is difficult to direct. I'd suggest considering removing it to the SI.

We understand this concern. We rewrote part of the paper to make the NMR analysis of the correlated chemical shifts described by the CHESCA matrices linked to the MD calculations.

Author Response

Reviewer #1 (Public Review):

Theoretical principles of viscous fluid mechanics are used here to assess likely mechanisms of transport in the ER. A set of candidate mechanisms is evaluated, making good use of imaging to represent ER network geometries. Evidence is provided that the contraction of peripheral sheets provides a much more credible mechanism than the contraction of individual tubules, junctions, or perinuclear sheets.

The work has been conducted carefully and comprehensively, making good use of underlying physical principles. There is a good discussion of the role of slip; sensible approximations (low volume fraction, small particle size, slender geometries, pragmatic treatment of boundary conditions) allow tractable and transparent calculations; clear physical arguments provide useful bounds; stochastic and deterministic features of the problem are well integrated.

We thank the reviewer for their positive assessment of our work.

There are just a couple of areas where more discussion might be warranted, in my view.

(1) The energetic cost of tubule contraction is estimated, but I did not see an equivalent estimate for the contraction of peripheral sheets. It might be helpful to estimate the energetic cost of viscous dissipation in generated flows at higher frequencies.

This is a good point. We will also include an energetic cost estimate for the contractions of peripheral sheets in the revised manuscript.

The mechanism of peripheral sheet contraction is unclear: do ATP-driven mechanisms somehow interact with thermal fluctuations of membranes?

The new energetic estimates in the revision might help constrain possible hypotheses for the mechanism(s) driving peripheral sheet contraction, and suggest if a dedicated ATP-driven mechanism is required.

(2) Mutations are mentioned in the abstract but not (as far as I could see) later in the manuscript. It would be helpful if any consequences for pathologies could be developed in the text.

We are grateful for this suggestion. The need to rationalise pathology associated with the subtle effects of ER-morphogens’ mutations is indeed pointed out as one factor motivating the study of the interplay between ER structure and performance. In the revised manuscript, we plan to include a brief discussion potentially linking ER morphogenes’ malfunction to luminal transport, integrating additional freshly published data.

Reviewer #2 (Public Review):

Summary:

This study explores theoretically the consequences of structural fluctuations of the endoplasmic reticulum (ER) morphology called contractions on molecular transport. Most of the manuscript consists of the construction of an interesting theoretical flow field (physical model) under various hypothetical assumptions. The computational modeling is followed by some simulations

Strengths:

The authors are focusing their attention on testing the hypothesis that a local flow in the tubule could be driven by tubular pinching. We recall that trafficking in the ER is considered to be mostly driven by diffusion at least at a spatial scale that is large enough to account for averaging of any random flow occurring from multiple directions [note that this is not the case for plants].

We thank the reviewer. We have indeed explored here the possibilities of active transport, focusing especially on transport over the length scale of single tubules, as a result of structural fluctuations, and found tubular pinching to be ineffective compared to e.g. peripheral sheets fluctuations. In the revised version we plan to add text mentioning what is known about the ER in plants.

Weaknesses:

The manuscript extensively details the construction of the theoretical model, occupying a significant portion of the manuscript. While this section contains interesting computations, its relevance and utility could be better emphasized, perhaps warranting a reorganization of the manuscript to foreground this critical aspect.

Overall, the manuscript appears highly technical with limited conclusive insights, particularly lacking predictions confirmed by experimental validation. There is an absence of substantial conclusions regarding molecular trafficking within the ER.

We sought to balance the theoretical/computational details of our model with the biophysical conclusions drawn from its predictions. Given the model's complexity and novelty, it was essential to elucidate the theoretical underpinnings comprehensively, in order to allow others to implement it in the future with additional, or different, parameters. To maintain clarity and focus in the main text, we have judiciously relegated extensive technical details to the methods section or supplementary materials, and divided the text into stand-alone section headings allowing the reader to skip through to conclusions.

The primary focus of our manuscript is to introduce and explore, via our theoretical model, the interplay between ER structure dynamics and molecular transport. Our approach, while in silico, generates concrete predictions about the physical processes underpinning luminal motion within the ER. For instance, our findings challenge the previously postulated role of small tubular contractions in driving luminal flow, instead highlighting the potential significance of local flat ER areas—empirically documented entities—for facilitating such motion.

Furthermore, by deducing what type of transport may or may not occur within the range of possible ER structural fluctuations, our model offers detailed predictions designed to bridge the gap between theoretical insight and experimental verification. These predictions detail the spatial and temporal parameters essential for effective transport, delineating plausible values for these parameters. We hope that the model’s predictions will invite experimentalists to devise innovative methodologies to test them. We plan to introduce text edits to the revised version to clarify these.

Author Response

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

Recommendations

Recommendation #1: Address potential confounds in the experimental design:

(1a) Confounding factors between baseline to early learning. While the visual display of the curved line remains constant, there are at least three changes between these two phases: 1) the presence of reward feedback (the focus of the paper); 2) a perturbation introduced to draw a hidden, mirror-symmetric curved line; 3) instructions provided to use reward feedback to trace the line on the screen (intentionally deceitful). As such, it remains unclear which of these factors are driving the changes in both behavior and bold signals between the two phases. The absence of a veridical feedback phase in which participants received reward feedback associated with the shown trajectory seems like a major limitation.

(1b) Confounding Factors Between Early and Late Learning. While the authors have focused on interpreting changes from early to late due to the explore-exploit trade-off, there are three additional factors possibly at play: 1) increasing fatigue, 2) withdrawal of attention, specifically related to individuals who have either successfully learned the perturbation within the first few trials or those who have simply given up, or 3) increasing awareness of the perturbation (not clear if subjective reports about perturbation awareness were measured.). I understand that fMRI research is resource-intensive; however, it is not clear how to rule out these alternatives with their existing data without additional control groups. [Another reviewer added the following: Why did the authors not acquire data during a control condition? How can we be confident that the neural dynamics observed are not due to the simple passage of time? Or if these effects are due to the task, what drives them? The reward component, the movement execution, increased automaticity?]

We have opted to address both of these points above within a single reply, as together they suggest potential confounding factors across the three phases of the task. We would agree that, if the results of our pairwise comparisons (e.g., Early > Baseline or Late > Early) were considered in isolation from one another, then these critiques of the study would be problematic. However, when considering the pattern of effects across the three task phases, we believe most of these critiques can be dismissed. Below, we first describe our results in this context, and then discuss how they address the reviewers’ various critiques.

Recall that from Baseline to Early learning, we observe an expansion of several cortical areas (e.g., core regions in the DMN) along the manifold (red areas in Fig. 4A, see manifold shifts in Fig. 4C) that subsequently exhibit contraction during Early to Late learning (blue areas in Fig. 4B, see manifold shifts in Fig. 4D). We show this overlap in brain areas in Author response image 1 below, panel A. Notably, several of these brain areas appear to contract back to their original, Baseline locations along the manifold during Late learning (compare Fig. 4C and D). This is evidenced by the fact that many of these same regions (e.g., DMN regions, in Author response image 1 panel A below) fail to show a significant difference between the Baseline and Late learning epochs (see Author response image 1 panel B below, which is taken from supplementary Fig 6). That is, the regions that show significant expansion and subsequent contraction (in Author response image 1 panel A below) tend not to overlap with the regions that significantly changed over the time course of the task (in Author response image 1 panel B below).

Author response image 1.

Note that this basic observation above is not only true of our regional manifold eccentricity data, but also in the underlying functional connectivity data associated with individual brain regions. To make this second point clearer, we have modified and annotated our Fig. 5 and included it below. Note the reversal in seed-based functional connectivity from Baseline to Early learning (leftmost brain plots) compared to Early to Late learning (rightmost brain plots). That is, it is generally the case that for each seed-region (A-C) the areas that increase in seed-connectivity with the seed region (in red; leftmost plot) are also the areas that decrease in seed-connectivity with the seed region (in blue; rightmost plot), and vice versa. [Also note that these connectivity reversals are conveyed through the eccentricity data — the horizontal red line in the rightmost plots denote the mean eccentricity of these brain regions during the Baseline phase, helping to highlight the fact that the eccentricity of the Late learning phase reverses back towards this Baseline level].

Author response image 2.

Critically, these reversals in brain connectivity noted above directly counter several of the critiques noted by the reviewers. For instance, this reversal pattern of effects argues against the idea that our results during Early Learning can be simply explained due to the (i) presence of reward feedback, (ii) presence of the perturbation or (iii) instructions to use reward feedback to trace the path on the screen. Indeed, all of these factors are also present during Late learning, and yet many of the patterns of brain activity during this time period revert back to the Baseline patterns of connectivity, where these factors are absent. Similarly, this reversal pattern strongly refutes the idea that the effects are simply due to the passage of time, increasing fatigue, or general awareness of the perturbation. Indeed, if any of these factors alone could explain the data, then we would have expected a gradual increase (or decrease) in eccentricity and connectivity from Baseline to Early to Late learning, which we do not observe. We believe these are all important points when interpreting the data, but which we failed to mention in our original manuscript when discussing our findings.

We have now rectified this in the revised paper, where we now write in our Discussion:

“Finally, it is important to note that the reversal pattern of effects noted above suggests that our findings during learning cannot be simply attributed to the introduction of reward feedback and/or the perturbation during Early learning, as both of these task-related features are also present during Late learning. In addition, these results cannot be simply explained due to the passage of time or increasing subject fatigue, as this would predict a consistent directional change in eccentricity across the Baseline, Early and Late learning epochs.”

However, having said the above, we acknowledge that one potential factor that our findings cannot exclude is that they are (at least partially) attributable to changes in subjects’ state of attention throughout the task. Indeed, one can certainly argue that Baseline trials in our study don’t require a great deal of attention (after all, subjects are simply tracing a curved path presented on the screen). Likewise, for subjects that have learned the hidden shape, the Late learning trials are also likely to require limited attentional resources (indeed, many subjects at this point are simply producing the same shape trial after trial). Consequently, the large shift in brain connectivity that we observe from Baseline to Early Learning, and the subsequent reversion back to Baseline-levels of connectivity during Late learning, could actually reflect a heightened allocation of attention as subjects are attempting to learn the (hidden) rewarded shape. However, we do not believe that this would reflect a ‘confound’ of our study per se — indeed, any subject who has participated in a motor learning study would agree that the early learning phase of a task is far more cognitively demanding than Baseline trials and Late learning trials. As such, it is difficult to disentangle this ‘attention’ factor from the learning process itself (and in fact, it is likely central to it).

Of course, one could have designed a ‘control’ task in which subjects must direct their attention to something other than the learning task itself (e.g., divided attention paradigm, e.g., Taylor & Thoroughman, 2007, 2008, and/or perform a secondary task concurrently (Codol et al., 2018; Holland et al., 2018), but we know that this type of manipulation impairs the learning process itself. Thus, in such a case, it wouldn’t be obvious to the experimenter what they are actually measuring in brain activity during such a task. And, to extend this argument even further, it is true that any sort of brain-based modulation can be argued to reflect some ‘attentional’ process, rather than modulations related to the specific task-based process under consideration (in our case, motor learning). In this regard, we are sympathetic to the views of Richard Andersen and colleagues who have eloquently stated that “The study of how attention interacts with other neural processing systems is a most important endeavor. However, we think that over-generalizing attention to encompass a large variety of different neural processes weakens the concept and undercuts the ability to develop a robust understanding of other cognitive functions.” (Andersen & Cui, 2007, Neuron). In short, it appears that different fields/researchers have alternate views on the usefulness of attention as an explanatory construct (see also articles from Hommel et al., 2019, “No one knows what attention is”, and Wu, 2023, “We know what attention is!”), and we personally don’t have a dog in this fight. We only highlight these issues to draw attention (no pun intended) that it is not trivial to separate these different neural processes during a motor learning study.

Nevertheless, we do believe these are important points worth flagging for the reader in our paper, as they might have similar questions. To this end, we have now included in our Discussion section the following text:

“It is also possible that some of these task-related shifts in connectivity relate to shifts in task-general processes, such as changes in the allocation of attentional resources (Bédard and Song, 2013; Rosenberg et al., 2016) or overall cognitive engagement (Aben et al., 2020), which themselves play critical roles in shaping learning (Codol et al., 2018; Holland et al., 2018; Song, 2019; Taylor and Thoroughman, 2008, 2007; for a review of these topics, see Tsay et al., 2023). Such processes are particularly important during the earlier phases of learning when sensorimotor contingencies need to be established. While these remain questions for future work, our data nevertheless suggest that this shift in connectivity may be enabled through the PMC.”

Finally, we should note that, at the end of testing, we did not assess participants' awareness of the manipulation (i.e., that they were, in fact, being rewarded based on a mirror image path). In hindsight, this would have been a good idea and provided some value to the current project. Nevertheless, it seems clear that, based on several of the learning profiles observed (e.g., subjects who exhibited very rapid learning during the Early Learning phase, more on this below), that many individuals became aware of a shape approximating the rewarded path. Note that we have included new figures (see our responses below) that give a better example of what fast versus slower learning looks like. In addition, we now note in our Methods that we did not probe participants about their subjective awareness re: the perturbation:

“Note that, at the end of testing, we did not assess participants’ awareness of the manipulation (i.e., that they were, in fact, being rewarded based on a mirror image path of the visible path).”

Recommendation #2: Provide more behavioral quantification.

(2a) The authors chose to only plot the average learning score in Figure 1D, without an indication of movement variability. I think this is quite important, to give the reader an impression of how variable the movements were at baseline, during early learning, and over the course of learning. There is evidence that baseline variability influences the 'detectability' of imposed rotations (in the case of adaptation learning), which could be relevant here. Shading the plots by movement variability would also be important to see if there was some refinement of the moment after participants performed at the ceiling (which seems to be the case ~ after trial 150). This is especially worrying given that in Fig 6A there is a clear indication that there is a large difference between subjects' solutions on the task. One subject exhibits almost a one-shot learning curve (reaching a score of 75 after one or two trials), whereas others don't seem to really learn until the near end. What does this between-subject variability mean for the authors' hypothesized neural processes?

In line with these recommendations, we have now provided much better behavioral quantification of subject-level performance in both the main manuscript and supplementary material. For instance, in a new supplemental Figure 1 (shown below), we now include mean subject (+/- SE) reaction times (RTs), movement times (MTs) and movement path variability (our computing of these measures are now defined in our Methods section).

As can be seen in the figure, all three of these variables tended to decrease over the course of the study, though we note there was a noticeable uptick in both RTs and MTs from the Baseline to Early learning phase, once subjects started receiving trial-by-trial reward feedback based on their movements. With respect to path variability, it is not obvious that there was a significant refinement of the paths created during late learning (panel D below), though there was certainly a general trend for path variability to decrease over learning.

Author response image 3.

Behavioral measures of learning across the task. (A-D) shows average participant reward scores (A), reaction times (B), movement times (C) and path variability (D) over the course of the task. In each plot, the black line denotes the mean across participants and the gray banding denotes +/- 1 SEM. The three equal-length task epochs for subsequent neural analyses are indicated by the gray shaded boxes.

In addition to these above results, we have also created a new Figure 6 in the main manuscript, which now solely focuses on individual differences in subject learning (see below). Hopefully, this figure clarifies key features of the task and its reward structure, and also depicts (in movement trajectory space) what fast versus slow learning looks like in the task. Specifically, we believe that this figure now clearly delineates for the reader the mapping between movement trajectory and the reward score feedback presented to participants, which appeared to be a source of confusion based on the reviewers’ comments below. As can be clearly observed in this figure, trajectories that approximated the ‘visible path’ (black line) resulted in fairly mediocre scores (see score color legend at right), whereas trajectories that approximated the ‘reward path’ (dashed black line, see trials 191-200 of the fast learner) resulted in fairly high scores. This figure also more clearly delineates how fPCA loadings derived from our functional data analysis were used to derive subject-level learning scores (panel C).

Author response image 4.

Individual differences in subject learning performance. (A) Examples of a good learner (bordered in green) and poor learner (bordered in red). (B) Individual subject learning curves for the task. Solid black line denotes the mean across all subjects whereas light gray lines denote individual participants. The green and red traces denote the learning curves for the example good and poor learners denoted in A. (C) Derivation of subject learning scores. We performed functional principal component analysis (fPCA) on subjects’ learning curves in order to identify the dominant patterns of variability during learning. The top component, which encodes overall learning, explained the majority of the observed variance (~75%). The green and red bands denote the effect of positive and negative component scores, respectively, relative to mean performance. Thus, subjects who learned more quickly than average have a higher loading (in green) on this ‘Learning score’ component than subjects who learned more slowly (in red) than average. The plot at right denotes the loading for each participant (open circles) onto this Learning score component.

The reviewers note that there are large individual differences in learning performance across the task. This was clearly our hope when designing the reward structure of this task, as it would allow us to further investigate the neural correlates of these individual differences (indeed, during pilot testing, we sought out a reward structure to the task that would allow for these intersubject differences). The subjects who learn early during the task end up having higher fPCA scores than the subjects who learn more gradually (or learn the task late). From our perspective, these differences are a feature, and not a bug, and they do not negate any of our original interpretations. That is, subjects who learn earlier on average tend to contract their DAN-A network during the early learning phase whereas subjects who learn more slowly on average (or learn late) instead tend to contract their DAN-A network during late learning (Fig. 7).

(2b) In the methods, the authors stated that they scaled the score such that even a perfectly traced visible path would always result in an imperfect score of 40 patients. What happens if a subject scores perfectly on the first try (which seemed to have happened for the green highlighted subject in Fig 6A), but is then permanently confronted with a score of 40 or below? Wouldn't this result in an error-clamp-like (error-based motor adaptation) design for this subject and all other high performers, which would vastly differ from the task demands for the other subjects? How did the authors factor in the wide between-subject variability?

We think the reviewers may have misinterpreted the reward structure of the task, and we apologize for not being clearer in our descriptions. The reward score that subjects received after each trial was based on how well they traced the mirror-image of the visible path. However, all the participant can see on the screen is the visible path. We hope that our inclusion of the new Figure 6 (shown above) makes the reward structure of the task, and its relationship to movement trajectories, much clearer. We should also note that, even for the highest performing subject (denoted in Fig. 6), it still required approximately 20 trials for them to reach asymptote performance.

(2c) The study would benefit from a more detailed description of participants' behavioral performance during the task. Specifically, it is crucial to understand how participants' motor skills evolve over time. Information on changes in movement speed, accuracy, and other relevant behavioral metrics would enhance the understanding of the relationship between behavior and brain activity during the learning process. Additionally, please clarify whether the display on the screen was presented continuously throughout the entire trial or only during active movement periods. Differences in display duration could potentially impact the observed differences in brain activity during learning.

We hope that with our inclusion of the new Supplementary Figure 1 (shown above) this addresses the reviewers’ recommendation. Generally, we find that RTs, MTs and path variability all decrease over the course of the task. We think this relates to the early learning phase being more attentionally demanding and requiring more conscious effort, than the later learning phases.

Also, yes, the visible path was displayed on the screen continuously throughout the trial, and only disappeared at the 4.5 second mark of each trial (when the screen was blanked and the data was saved off for 1.5 seconds prior to commencement of the next trial; 6 seconds total per trial). Thus, there were no differences in display duration across trials and phases of the task. We have now clarified this in the Methods section, where we now write the following:

“When the cursor reached the target distance, the target changed color from red to green to indicate that the trial was completed. Importantly, other than this color change in the distance marker, the visible curved path remained constant and participants never received any feedback about the position of their cursor.”

(2d) It is unclear from plots 6A, 6B, and 1D how the scale of the behavioral data matches with the scaling of the scores. Are these the 'real' scores, meaning 100 on the y-axis would be equivalent to 40 in the task? Why then do all subjects reach an asymptote at 75? Or is 75 equivalent to 40 and the axis labels are wrong?

As indicated above, we clearly did a poor job of describing the reward structure of our task in our original paper, and we now hope that our inclusion of Figure 6 makes things clear. A ‘40’ score on the y-axis would indicate that a subject has perfectly traced the visible path whereas a perfect ‘100’ score would indicate that a subject has perfectly traced the (hidden) mirror image path.

The fact that several of the subjects reach asymptote around 75 is likely a byproduct of two factors. Firstly, the subjects performed their movements in the absence of any visual error feedback (they could not see the position of a cursor that represented their hand position), which had the effect of increasing motor variability in their actions from trial to trial. Secondly, there appears to be an underestimation among subjects regarding the curvature of the concealed, mirror-image path (i.e., that the rewarded path actually had an equal but opposite curvature to that of the visible path). This is particularly evident in the case of the top-performing subject (illustrated in Figure 6A) who, even during late learning, failed to produce a completely arched movement.

(2e) Labeling of Contrasts: There is a consistent issue with the labeling of contrasts in the presented figures, causing confusion. While the text refers to the difference as "baseline to early learning," the label used in figures, such as Figure 4, reads "baseline > early." It is essential to clarify whether the presented contrast is indeed "baseline > early" or "early > baseline" to avoid any misinterpretation.

We thank the reviewers for catching this error. Indeed, the intended label was Early > Baseline, and this has now been corrected throughout.

Recommendation #3. Clarify which motor learning mechanism(s) are at play.

(3a) Participants were performing at a relatively low level, achieving around 50-60 points by the end of learning. This outcome may not be that surprising, given that reward-based learning might have a substantial explicit component and may also heavily depend on reasoning processes, beyond reinforcement learning or contextual recall (Holland et al., 2018; Tsay et al., 2023). Even within our own data, where explicit processes are isolated, average performance is low and many individuals fail to learn (Brudner et al., 2016; Tsay et al., 2022). Given this, many participants in the current study may have simply given up. A potential indicator of giving up could be a subset of participants moving straight ahead in a rote manner (a heuristic to gain moderate points). Consequently, alterations in brain networks may not reflect exploration and exploitation strategies but instead indicate levels of engagement and disengagement. Could the authors plot the average trajectory and the average curvature changes throughout learning? Are individuals indeed defaulting to moving straight ahead in learning, corresponding to an average of 50-60 points? If so, the interpretation of brain activity may need to be tempered.

We can do one better, and actually give you a sense of the learning trajectories for every subject over time. In the figure below, which we now include as Supplementary Figure 2 in our revision, we have plotted, for each subject, a subset of their movement trajectories across learning trials (every 10 trials). As can be seen in the diversity of these trajectories, the average trajectory and average curvature would do a fairly poor job of describing the pattern of learning-related changes across subjects. Moreover, it is not obvious from looking at these plots the extent to which poor learning subjects (i.e., subjects who never converge on the reward path) actually ‘give up’ in the task — rather, many of these subjects still show some modulation (albeit minor) of their movement trajectories in the later trials (see the purple and pink traces). As an aside, we are also not entirely convinced that straight ahead movements, which we don’t find many of in our dataset, can be taken as direct evidence that the subject has given up.

Author response image 5

Variability in learning across subjects. Plots show representative trajectory data from each subject (n=36) over the course of the 200 learning trials. Coloured traces show individual trials over time (each trace is separated by ten trials, e.g., trial 1, 10, 20, 30, etc.) to give a sense of the trajectory changes throughout the task (20 trials in total are shown for each subject).

We should also note that we are not entirely opposed to the idea of describing aspects of our findings in terms of subject engagement versus disengagement over time, as such processes are related at some level to exploration (i.e., cognitive engagement in finding the best solution) and exploitation (i.e., cognitively disengaging and automating one’s behavior). As noted in our reply to Recommendation #1 above, we now give some consideration of these explanations in our Discussion section, where we now write:

“It is also possible that these task-related shifts in connectivity relates to shifts in task-general processes, such as changes in the allocation of attentional resources (Bédard and Song, 2013; Rosenberg et al., 2016) or overall cognitive engagement (Aben et al., 2020), which themselves play critical roles in shaping learning (Codol et al., 2018; Holland et al., 2018; Song, 2019; Taylor and Thoroughman, 2008, 2007; for a review of these topics, see Tsay et al., 2023). Such processes are particularly important during the earlier phases of learning when sensorimotor contingencies need to be established. While these remain questions for future work, our data nevertheless suggest that this shift in connectivity may be enabled through the PMC.”

(3b) The authors are mixing two commonly used paradigms, reward-based learning, and motor adaptation, but provide no discussion of the different learning processes at play here. Which processes were they attempting to probe? Making this explicit would help the reader understand which brain regions should be implicated based on previous literature. As it stands, the task is hard to interpret. Relatedly, there is a wealth of literature on explicit vs implicit learning mechanisms in adaptation tasks now. Given that the authors are specifically looking at brain structures in the cerebral cortex that are commonly associated with explicit and strategic learning rather than implicit adaptation, how do the authors relate their findings to this literature? Are the learning processes probed in the task more explicit, more implicit, or is there a change in strategy usage over time? Did the authors acquire data on strategies used by the participants to solve the task? How does the baseline variability come into play here?

As noted in our paper, our task was directly inspired by the reward-based motor learning tasks developed by Dam et al., 2013 (Plos One) and Wu et al., 2014 (Nature Neuroscience). What drew us to these tasks is that they allowed us to study the neural bases of reward-based learning mechanisms in the absence of subjects also being able to exploit error-based mechanisms to achieve learning. Indeed, when first describing the task in the Results section of our paper we wrote the following:

“Importantly, because subjects received no visual feedback about their actual finger trajectory and could not see their own hand, they could only use the score feedback — and thus only reward-based learning mechanisms — to modify their movements from one trial to the next (Dam et al., 2013; Wu et al., 2014).”

If the reviewers are referring to ‘motor adaptation’ in the context in which that terminology is commonly used — i.e., the use of sensory prediction errors to support error-based learning — then we would argue that motor adaptation is not a feature of the current study. It is true that in our study subjects learn to ‘adapt’ their movements across trials, but this shaping of the movement trajectories must be supported through reinforcement learning mechanisms (and, of course, supplemented by the use of cognitive strategies as discussed in the nice review by Tsay et al., 2023). We apologize for not being clearer in our paper about this key distinction and we have now included new text in the introduction to our Results to directly address this:

“Importantly, because subjects received no visual feedback about their actual finger trajectory and could not see their own hand, they could only use the score feedback — and thus only reward-based learning mechanisms — to modify their movements from one trial to the next (Dam et al., 2013; Wu et al., 2014). That is, subjects could not use error-based learning mechanisms to achieve learning in our study, as this form of learning requires sensory errors that convey both the change in direction and magnitude needed to correct the movement.”

With this issue aside, we are well aware of the established framework for thinking about sensorimotor adaptation as being composed of a combination of explicit and implicit components (indeed, this has been a central feature of several of our other recent neuroimaging studies that have explored visuomotor rotation learning, e.g., Gale et al., 2022 PNAS, Areshenkoff et al., 2022 elife, Standage et al., 2023 Cerebral Cortex). However, there has been comparably little work done on these parallel components within the domain of reinforcement learning tasks (though see Codol et al., 2018; Holland et al., 2018, van Mastrigt et al., 2023; see also the Tsay et al., 2023 review), and as far as we can tell, nothing has been done to date in the reward-based motor learning area using fMRI. By design, we avoided using descriptors of ‘explicit’ or ‘implicit’ in our study because our experimental paradigm did not allow a separate measurement of those two components to learning during the task. Nevertheless, it seems clear to us from examining the subjects’ learning curves (see supplementary figure 2 above), that individuals who learn very quickly are using strategic processes (such as action exploration to identify the best path) to enhance their learning. As we noted in an above response, we did not query subjects after the fact about their strategy use, which admittedly was a missed opportunity on our part.

Author response image 6.

With respect to the comment on baseline variability and its relationship to performance, this is an interesting idea and one that was explored in the Wu et al., 2014 Nature Neuroscience paper. Prompted by the reviewers, we have now explored this idea in the current data set by testing for a relationship between movement path variability during baseline trials (all 70 baseline trials, see Supplementary Figure 1D above for reference) and subjects’ fPCA score on our learning task. However, when we performed this analysis, we did not observe a significant positive relationship between baseline variability and subject performance. Rather, we actually found a trend towards a negative relationship (though this was non-significant; r=-0.2916, p=0.0844). Admittedly, we are not sure what conclusions can be drawn from this analysis, and in any case, we believe it to be tangential to our main results. We provide the results (at right) for the reviewers if they are interested. This may be an interesting avenue for exploration in future work.

Recommendation #4: Provide stronger justification for brain imaging methods.

(4a) Observing how brain activity varies across these different networks is remarkable, especially how sensorimotor regions separate and then contract with other, more cognitive areas. However, does the signal-to-noise ratio in each area/network influence manifold eccentricity and limit the possible changes in eccentricity during learning? Specifically, if a region has a low signal-to-noise ratio, it might exhibit minimal changes during learning (a phenomenon perhaps relevant to null manifold changes in the striatum due to low signal-to-noise); conversely, regions with higher signal-to-noise (e.g., motor cortex in this sensorimotor task) might exhibit changes more easily detected. As such, it is unclear how to interpret manifold changes without considering an area/network's signal-to-noise ratio.

We appreciate where these concerns are coming from. First, we should note that the timeseries data used in our analysis were z-transformed (mean zero, 1 std) to allow normalization of the signal both over time and across regions (and thus mitigate the possibility that the changes observed could simply reflect mean overall signal changes across different regions). Nevertheless, differences in signal intensity across brain regions — particularly between cortex and striatum — are well-known, though it is not obvious how these differences may manifest in terms of a task-based modulation of MR signals.

To examine this issue in the current data set, we extracted, for each subject and time epoch (Baseline, Early and Late learning) the raw scanner data (in MR arbitrary units, a.u.) for the cortical and striatal regions and computed the (1) mean signal intensity, (2) standard deviation of the signal (Std) and (3) temporal signal to noise ratio (tSNR; calculated by mean/Std). Note that in the fMRI connectivity literature tSNR is often the preferred SNR measure as it normalizes the mean signal based on the signal’s variability over time, thus providing a general measure of overall ‘signal quality’. The results of this analysis, averaged across subjects and regions, is shown below.

Author response image 7.

Note that, as expected, the overall signal intensity (left plot) of cortex is higher than in the striatum, reflecting the closer proximity of cortex to the receiver coils in the MR head coil. In fact, the signal intensity in cortex is approximately 38% higher than that in the striatum (~625 - 450)/450). However, the signal variation in cortex is also greater than striatum (middle plot), but in this case approximately 100% greater (i.e., (~5 - 2.5)/2.5)). The result of this is that the tSNR (mean/std) for our data set and the ROI parcellations we used is actually greater in the striatum than in cortex (right plot). Thus, all else being equal, there seems to have been sufficient tSNR in the striatum for us to have detected motor-learning related effects. As such, we suspect the null effects for the striatum in our study actually stem from two sources.

The first likely source is the relatively lower number of striatal regions (12) as compared to cortical regions (998) used in our analysis, coupled with our use of PCA on these data (which, by design, identifies the largest sources of variation in connectivity). In future studies, this unbalance could be rectified by using finer parcellations of the striatum (even down to the voxel level) while keeping the same parcellation of cortex (i.e., equate the number of ‘regions’ in each of striatum and cortex). The second likely source is our use of a striatal atlas (the Harvard-Oxford atlas) that divides brain regions based on their neuroanatomy rather than their function. In future work, we plan on addressing this latter concern by using finer, more functionally relevant parcellations of striatum (such as in Tian et al., 2020, Nature Neuroscience). Note that we sought to capture these interrelated possible explanations in our Discussion section, where we wrote the following:

“While we identified several changes in the cortical manifold that are associated with reward-based motor learning, it is noteworthy that we did not observe any significant changes in manifold eccentricity within the striatum. While clearly the evidence indicates that this region plays a key role in reward-guided behavior (Averbeck and O’Doherty, 2022; O’Doherty et al., 2017), there are several possible reasons why our manifold approach did not identify this collection of brain areas. First, the relatively small size of the striatum may mean that our analysis approach was too coarse to identify changes in the connectivity of this region. Though we used a 3T scanner and employed a widely-used parcellation scheme that divided the striatum into its constituent anatomical regions (e.g., hippocampus, caudate, etc.), both of these approaches may have obscured important differences in connectivity that exist within each of these regions. For example, areas such the hippocampus and caudate are not homogenous areas but themselves exhibit gradients of connectivity (e.g., head versus tail) that can only be revealed at the voxel level (Tian et al., 2020; Vos de Wael et al., 2021). Second, while our dimension reduction approach, by design, aims to identify gradients of functional connectivity that account for the largest amounts of variance, the limited number of striatal regions (as compared to cortex) necessitates that their contribution to the total whole-brain variance is relatively small. Consistent with this perspective, we found that the low-dimensional manifold architecture in cortex did not strongly depend on whether or not striatal regions were included in the analysis (see Supplementary Fig. 6). As such, selective changes in the patterns of functional connectivity at the level of the striatum may be obscured using our cortex x striatum dimension reduction approach. Future work can help address some of these limitations by using both finer parcellations of striatal cortex (perhaps even down to the voxel level)(Tian et al., 2020) and by focusing specifically on changes in the interactions between the striatum and cortex during learning. The latter can be accomplished by selectively performing dimension reduction on the slice of the functional connectivity matrix that corresponds to functional coupling between striatum and cortex.”

(4b) Could the authors clarify how activity in the dorsal attention network (DAN) changes throughout learning, and how these changes also relate to individual differences in learning performance? Specifically, on average, the DAN seems to expand early and contract late, relative to the baseline. This is interpreted to signify that the DAN exhibits lesser connectivity followed by greater connectivity with other brain regions. However, in terms of how these changes relate to behavior, participants who go against the average trend (DAN exhibits more contraction early in learning, and expansion from early to late) seem to exhibit better learning performance. This finding is quite puzzling. Does this mean that the average trend of expansion and contraction is not facilitative, but rather detrimental, to learning? [Another reviewer added: The authors do not state any explicit hypotheses, but only establish that DMN coordinates activity among several regions. What predictions can we derive from this? What are the authors looking for in the data? The work seems more descriptive than hypothesis-driven. This is fine but should be clarified in the introduction.]

These are good questions, and we are glad the reviewers appreciated the subtlety here. The reviewers are indeed correct that the relationship of the DAN-A network to behavioral performance appears to go against the grain of the group-level results that we found for the entire DAN network (which we note is composed of both the DAN-A and DAN-B networks). That is, subjects who exhibited greater contraction from Baseline to Early learning and likewise, greater expansion from Early to Late learning, tended to perform better in the task (according to our fPCA scores). However, on this point it is worth noting that it was mainly the DAN-B network which exhibited group-level expansion from Baseline to Early Learning whereas the DAN-A network exhibited negligible expansion. This can be seen in Author response image 8 below, which shows the pattern of expansion and contraction (as in Fig. 4), but instead broken down into the 17-network parcellation. The red asterisk denotes the expansion from Baseline to Early learning for the DAN-B network, which is much greater than that observed for the DAN-A network (which is basically around the zero difference line).

Author response image 8.

Thus, it appears that the DAN-A and DAN-B networks are modulated to a different extent during the task, which likely contributes to the perceived discrepancy between the group-level effects (reported using the 7-network parcellation) and the individual differences effects (reported using the finer 17-network parcellation). Based on the reviewers’ comments, this seems like an important distinction to clarify in the manuscript, and we have now described this nuance in our Results section where we now write:

“...Using this permutation testing approach, we found that it was only the change in eccentricity of the DAN-A network that correlated with Learning score (see Fig. 7C), such that the more the DAN-A network decreased in eccentricity from Baseline to Early learning (i.e., contracted along the manifold), the better subjects performed at the task (see Fig. 7C, scatterplot at right). Consistent with the notion that changes in the eccentricity of the DAN-A network are linked to learning performance, we also found the inverse pattern of effects during Late learning, whereby the more that this same network increased in eccentricity from Early to Late learning (i.e., expanded along the manifold), the better subjects performed at the task (Fig. 7D). We should note that this pattern of performance effects for the DAN-A — i.e., greater contraction during Early learning and greater expansion during Late learning being associated with better learning — appears at odds with the group-level effects described in Fig. 4A and B, where we generally find the opposite pattern for the entire DAN network (composed of the DAN-A and DAN-B subnetworks). However, this potential discrepancy can be explained when examining the changes in eccentricity using the 17-network parcellation (see Supplementary Figure 8). At this higher resolution level we find that these group-level effects for the entire DAN network are being largely driven by eccentricity changes in the DAN-B network (areas in anterior superior parietal cortex and premotor cortex), and not by mean changes in the DAN-A network. By contrast, our present results suggest that it is the contraction and expansion of areas of the DAN-A network (and not DAN-B network) that are selectively associated with differences in subject learning performance.”

Finally, re: the reviewers’ comments that we do not state any explicit hypotheses etc., we acknowledge that, beyond our general hypothesis stated at the outset about the DMN being involved in reward-based motor learning, our study is quite descriptive and exploratory in nature. Such little work has been done in this research area (i.e., using manifold learning approaches to study motor learning with fMRI) that it would be disingenuous to have any stronger hypotheses than those stated in our Introduction. Thus, to make the exploratory nature of our study clear to the reader, we have added the following text (in red) to our Introduction:

“Here we applied this manifold approach to explore how brain activity across widely distributed cortical and striatal systems is coordinated during reward-based motor learning. We were particularly interested in characterizing how connectivity between regions within the DMN and the rest of the brain changes as participants shift from learning the relationship between motor commands and reward feedback, during early learning, to subsequently using this information, during late learning. We were also interested in exploring whether learning-dependent changes in manifold structure relate to variation in subject motor performance.”

We hope these changes now make it obvious the intention of our study.

(4c) The paper examines a type of motor adaptation task with a reward-based learning component. This, to me, strongly implicates the cerebellum, given that it has a long-established crucial role in adaptation and has recently been implicated in reward-based learning (see work by Wagner & Galea). Why is there no mention of the cerebellum and why it was left out of this study? Especially given that the authors state in the abstract they examine cortical and subcortical structures. It's evident from the methods that the authors did not acquire data from the cerebellum or had too small a FOV to fully cover it (34 slices at 4 mm thickness 136 mm which is likely a bit short to fully cover the cerebellum in many participants). What was the rationale behind this methodological choice? It would be good to clarify this for the reader. Related to this, the authors need to rephrase their statements on 'whole-brain' connectivity matrices or analyses - it is not whole-brain when it excludes the cerebellum.

As we noted above, we do not believe this task to be a motor adaptation task, in the sense that subjects are not able to use sensory prediction errors (and thus error-based learning mechanisms) to improve their performance. Rather, by denying subjects this sensory error feedback they are only able to use reinforcement learning processes, along with cognitive strategies (nicely covered in Tsay et al., 2023), to improve performance. Nevertheless, we recognize that the cerebellum has been increasingly implicated in facets of reward-based learning, particularly within the rodent domain (e.g., Wagner et al., 2017; Heffley et al., 2018; Kostadinov et al., 2019, etc.). In our study, we did indeed collect data from the cerebellum but did not include it in our original analyses, as we wanted (1) the current paper to build on prior work in the human and macaque reward-learning domain (which focuses solely on striatum and cortex, and which rarely discusses cerebellum, see Averbeck & O’Doherty, 2022 & Klein-Flugge et al., 2022 for recent reviews), and, (2) allow this to be a more targeted focus of future work (specifically we plan on focusing on striatal-cerebellar interactions during learning, which are hypothesized based on the neuroanatomical tract tracing work of Bostan and Strick, etc.). We hope the reviewers respect our decisions in this regard.

Nevertheless, we acknowledge that based on our statements about ‘whole-brain’ connectivity and vagueness about what we mean by ‘subcortex,’ that this may be confusing for the reader. We have now removed and/or corrected such references throughout the paper (however, note that in some cases it is difficult to avoid reference to “whole-brain” — e.g., “whole-brain correlation map” or “whole-brain false discovery rate correction”, which is standard terminology in the field).

In addition, we are now explicit in our Methods section that the cerebellum was not included in our analyses.

“Each volume comprised 34 contiguous (no gap) oblique slices acquired at a ~30° caudal tilt with respect to the plane of the anterior and posterior commissure (AC-PC), providing whole-brain coverage of the cerebrum and cerebellum. Note that for the current study, we did not examine changes in cerebellar activity during learning.”

(4d) The authors centered the matrices before further analyses to remove variance associated with the subject. Why not run a PCA on the connectivity matrices and remove the PC that is associated with subject variance? What is the advantage of first centering the connectivity matrices? Is this standard practice in the field?

Centering in some form has become reasonably common in the functional connectivity literature, as there is considerable evidence that task-related (or cognitive) changes in whole-brain connectivity are dwarfed by static, subject-level differences (e.g., Gratton, et al, 2018, Neuron). If covariance matrices were ordinary scalar values, then isolating task-related changes could be accomplished simply by subtracting a baseline scan or mean score; but because the space of covariance matrices is non-Euclidean, the actual computations involved in this subtraction are more complex (see our Methods). However, fundamentally (and conceptually) our procedure is simply ordinary mean-centering, but adapted to this non-Euclidean space. Despite the added complexity, there is considerable evidence that such computations — adapted directly to the geometry of the space of covariance matrices — outperform simpler methods, which treat covariance matrices as arrays of real numbers (e.g. naive substraction, see Dodero et al. & Ng et al., references below). Moreover, our previous work has found that this procedure works quite well to isolate changes associated with different task conditions (Areshenkoff et al., 2021, Neuroimage; Areshenkoff et al., 2022, elife).

Although PCA can be adapted to work well with covariance matrix valued data, it would at best be a less direct solution than simply subtracting subjects' mean connectivity. This is because the top components from applying PCA would be dominated by both subject-specific effects (not of interest here), and by the large-scale connectivity structure typically observed in component based analyses of whole-brain connectivity (i.e. the principal gradient), whereas changes associated with task-condition (the thing of interest here) would be buried among the less reliable components. By contrast, our procedure directly isolates these task changes.

References cited above:

Dodero, L., Minh, H. Q., San Biagio, M., Murino, V., & Sona, D. (2015, April). Kernel-based classification for brain connectivity graphs on the Riemannian manifold of positive definite matrices. In 2015 IEEE 12th international symposium on biomedical imaging (ISBI) (pp. 42-45). IEEE.

Ng, B., Dressler, M., Varoquaux, G., Poline, J. B., Greicius, M., & Thirion, B. (2014). Transport on Riemannian manifold for functional connectivity-based classification. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2014: 17th International Conference, Boston, MA, USA, September 14-18, 2014, Proceedings, Part II 17 (pp. 405-412). Springer International Publishing.

(4e) Seems like a missed opportunity that the authors just use a single, PCA-derived measure to quantify learning, where multiple measures could have been of interest, especially given that the introduction established some interesting learning-related concepts related to exploration and exploitation, which could be conceptualized as movement variability and movement accuracy. It is unclear why the authors designed a task that was this novel and interesting, drawing on several psychological concepts, but then chose to ignore these concepts in the analysis.

We were disappointed to hear that the reviewers did not appreciate our functional PCA-derived measure to quantify subject learning. This is a novel data-driven analysis approach that we have previously used with success in recent work (e.g., Areshenkoff et al., 2022, elife) and, from our perspective, we thought it was quite elegant that we were able to describe the entire trajectory of learning across all participants along a single axis that explained the majority (~75%) of the variance in the patterns of behavioral learning data. Moreover, the creation of a single behavioral measure per participant (what we call a ‘Learning score’, see Fig. 6C) helped simplify our brain-behavior correlation analyses considerably, as it provided a single measure that accounts for the natural auto-correlation in subjects’ learning curves (i.e., that subjects who learn quickly also tend to be better overall learners by the end of the learning phase). It also avoids the difficulty (and sometimes arbitrariness) of having to select specific trial bins for behavioral analysis (e.g., choosing the first 5, 10, 20 or 25 trials as a measure of ‘early learning’, and so on). Of course, one of the major alternatives to our approach would have involved fitting an exponential to each subject’s learning curves and taking measures like learning rate etc., but in our experience we have found that these types of models don’t always fit well, or derive robust/reliable parameters at the individual subject level. To strengthen the motivation for our approach, we have now included the following text in our Results:

“To quantify this variation in subject performance in a manner that accounted the auto-correlation in learning performance over time (i.e., subjects who learned more quickly tend to exhibit better performance by the end of learning), we opted for a pure data-driven approach and performed functional principal component analysis (fPCA; (Shang, 2014)) on subjects’ learning curves. This approach allowed us to isolate the dominant patterns of variability in subject’s learning curves over time (see Methods for further details; see also Areshenkoff et al., 2022).”

In any case, the reviewers may be pleased to hear that in current work in the lab we are using more model-based approaches to attempt to derive sets of parameters (per participant) that relate to some of the variables of interest described by the reviewers, but that we relate to much more dynamical (shorter-term) changes in brain activity.

(4f) Overall Changes in Activity: The manuscript should delve into the potential influence of overall changes in brain activity on the results. The choice of using Euclidean distance as a metric for quantifying changes in connectivity is sensitive to scaling in overall activity. Therefore, it is crucial to discuss whether activity in task-relevant areas increases from baseline to early learning and decreases from early to late learning, or if other patterns emerge. A comprehensive analysis of overall activity changes will provide a more complete understanding of the findings.

These are good questions and we are happy to explore this in the data. However, as mentioned in our response to query 4a above, it is important to note that the timeseries data for each brain region was z-scored prior to analysis, with the aim of removing any mean changes in activity levels (note that this is a standard preprocessing step when performing functional connectivity analysis, given that mean signal changes are not the focus of interest in functional connectivity analyses).

To further emphasize these points, we have taken our z-scored timeseries data and calculated the mean signal for each region within each task epoch (Baseline, Early and Late learning, see panel A in figure below). The point of showing this data (where each z-score map looks near identical across the top, middle and bottom plots) is to demonstrate just how miniscule the mean signal changes are in the z-scored timeseries data. This point can also be observed when plotting the mean z-score signal across regions for each epoch (see panel B in figure below). Here we find that Baseline and Early learning have a near identical mean activation level across regions (albeit with slightly different variability across subjects), whereas there is a slight increase during late learning — though it should be noted that our y-axis, which measures in the thousandths, really magnifies this effect.

To more directly address the reviewers’ comments, using the z-score signal per region we have also performed the same statistical pairwise comparisons (Early > Baseline and Late>Early) as we performed in the main manuscript Fig. 4 (see panel C in Author response image 9 below). In this plot, areas in red denote an increase in activity from Baseline to Early learning (top plot) and from Early to Late learning (bottom plot), whereas areas in blue denote a decrease for those same comparisons. The important thing to emphasize here is that the spatial maps resulting from this analysis are generally quite different from the maps of eccentricity that we report in Fig. 4 in our paper. For instance, in the figure below, we see significant changes in the activity of visual cortex between epochs but this is not found in our eccentricity results (compare with Fig. 4). Likewise, in our eccentricity results (Fig. 4), we find significant changes in the manifold positioning of areas in medial prefrontal cortex (MPFC), but this is not observed in the activation levels of these regions (panel C below). Again, we are hesitant to make too much of these results, as the activation differences denoted as significant in the figure below are likely to be an effect on the order of thousandths of a z-score (e.g., 0.002 > 0.001), but this hopefully assuages reviewers’ concerns that our manifold results are solely attributable to changes in overall activity levels.

We are hesitant to include the results below in our paper as we feel that they don’t add much to the interpretation (as the purpose of z-scoring was to remove large activation differences). However, if the reviewers strongly believe otherwise, we would consider including them in the supplement.

Author response image 9.

Examination of overall changes in activity across regions. (A) Mean z-score maps across subjects for the Baseline (top), Early Learning (middle) and Late learning (bottom) epochs. (B) Mean z-score across brain regions for each epoch. Error bars represent +/- 1 SEM. (C) Pairwise contrasts of the z-score signal between task epochs. Positive (red) and negative (blue) values show significant increases and decreases in z-score signal, respectively, following FDR correction for region-wise paired t-tests (at q<0.05).

Author Response

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

Reviewer #1 (Public Review):

Summary:

The authors aim to address a critical challenge in the field of bioinformatics: the accurate and efficient identification of protein binding sites from sequences. Their work seeks to overcome the limitations of current methods, which largely depend on multiple sequence alignments or experimental protein structures, by introducing GPSite, a multi-task network designed to predict binding residues of various molecules on proteins using ESMFold.

Strengths:

• Benchmarking. The authors provide a comprehensive benchmark against multiple methods, showcasing the performances of a large number of methods in various scenarios.

• Accessibility and Ease of Use. GPSite is highlighted as a freely accessible tool with user-friendly features on its website, enhancing its potential for widespread adoption in the research community.

RE: We thank the reviewer for acknowledging the contributions and strengths of our work!

Weaknesses:

• Lack of Novelty. The method primarily combines existing approaches and lacks significant technical innovation. This raises concerns about the original contribution of the work in terms of methodological development. Moreover, the paper reproduces results and analyses already presented in previous literature, without providing novel analysis or interpretation. This further diminishes the contribution of this paper to advancing knowledge in the field.

RE: The novelty of this work is primarily manifested in four key aspects. Firstly, although we have employed several existing tools such as ProtTrans and ESMFold to extract sequence features and predict protein conformations, these techniques were hardly explored in the field of binding site prediction. We have successfully demonstrated the feasibility of substituting multiple sequence alignments with language model embeddings and training with predicted structures, providing a new solution to overcome the limitations of current methods for genome-wide applications. Secondly, though a few methods tend to capture geometric information based on protein surfaces or atom graphs, surface calculation and property mapping are usually time-consuming, while massage passing on full atom graphs is memory-consuming and thus challenging to process long sequences. Besides, these methods are sensitive towards details and errors in the predicted structures. To facilitate large-scale annotations, we have innovatively applied geometric deep learning to protein residue graphs for comprehensively capturing backbone and sidechain geometric contexts in an efficient and effective manner (Figure 1). Thirdly, we have not only exploited multi-task learning to integrate diverse ligands and enhance performance, but also shown its capability to easily extend to the binding site prediction of other unseen ligands (Figure 4 D-E). Last but not least, as a “Tools and Resources” article, we have provided a fast, accurate and user-friendly webserver, as well as constructed a large annotation database for the sequences in Swiss-Prot. Leveraging this database, we have conducted extensive analyses on the associations between binding sites and molecular functions, biological processes, and disease-causing mutations (Figure 5), indicating the potential of our tool to unveil unexplored biology underlying genomic data.

We have now revised the descriptions in the “The geometry-aware protein binding site predictor (GPSite)” section to highlight the novelty of our work in a clearer manner:

“In conclusion, GPSite is distinguished from the previous approaches in four key aspects. First, profiting from the effectiveness and low computational cost of ProtTrans and ESMFold, GPSite is liberated from the reliance on MSA and native structures, thus enabling genome-wide binding site prediction. Second, unlike methods that only explore the Cα models of proteins 25,40, GPSite exploits a comprehensive geometric featurizer to fully refine knowledge in the backbone and sidechain atoms. Third, the employed message propagation on residue graphs is global structure-aware and time-efficient compared to the methods based on surface point clouds 21,22, and memory-efficient unlike methods based on full atom graphs 23,24. Residue-based message passing is also less sensitive towards errors in the predicted structures. Last but not least, instead of predicting binding sites for a single molecule type or learning binding patterns separately for different molecules, GPSite applies multi-task learning to better model the latent relationships among different binding partners.”

• Benchmark Discrepancies. The variation in benchmark results, especially between initial comparisons and those with PeSTo. GPSite achieves a PR AUC of 0.484 on the global benchmark but a PR AUC of 0.61 on the benchmark against PeSTo. For consistency, PeSTo should be included in the benchmark against all other methods. It suggests potential issues with the benchmark set or the stability of the method. This inconsistency needs to be addressed to validate the reliability of the results.

RE: We thank the reviewer for the constructive comments. Since our performance comparison experiments involved numerous competitive methods whose training sets are disparate, it was difficult to compare or rank all these methods fairly using a single test set. Given the substantial overlap between our protein-binding site test set and the training set of PeSTo, we meticulously re-split our entire protein-protein binding site dataset to generate a new test set that avoids any overlap with the training sets of both GPSite and PeSTo and performed a separate evaluation, where GPSite achieves a higher AUPR than PeSTo (0.610 against 0.433). This is quite common in this field. For instance, in the study of PeSTo (Nat Commun 2023), the comparisons of PeSTo with MaSIF-site, SPPIDER, and PSIVER were conducted using one test set, while the comparison with ScanNet was performed on a separate test set.

Based on the reviewer’s suggestion, we have now replaced this experiment with a direct comparison with PeSTo using the datasets from PeSTo, in order to enhance the completeness and convincingness of our results. The corresponding descriptions are now added in Appendix 1-note 2, and the results are added in Appendix 2-table 4. For convenience, we also attach the note and table here:

“Since 340 out of 375 proteins in our protein-protein binding site test set share > 30% identity with the training sequences of PeSTo, we performed a separate comparison between GPSite and PeSTo using the training and test datasets from PeSTo. By re-training with simply the same hyperparameters, GPSite achieves better performance than PeSTo (AUPR of 0.824 against 0.797) as shown in Appendix 2-table 4. Furthermore, when using ESMFold-predicted structures as input, the performance of PeSTo decreases substantially (AUPR of 0.691), and the superiority of our method will be further reflected. As in 24, the performance of ScanNet is also included (AUPR of 0.720), which is also largely outperformed by GPSite.”

Author response table 1.

Performance comparison of GPSite with ScanNet and PeSTo on the protein-protein binding site test set from PeSTo 24

Note: The performance of ScanNet and PeSTo are directly obtained from 24. PeSTo* denotes evaluation using the ESMFold-predicted structures as input. The metrics provided are the median AUPR, median AUC and median MCC. The best/second-best results are indicated by bold/underlined fonts.

• Interface Definition Ambiguity. There is a lack of clarity in defining the interface for the binding site predictions. Different methods are trained using varying criteria (surfaces in MaSIF-site, distance thresholds in ScanNet). The authors do not adequately address how GPSite's definition aligns with or differs from these standards and how this issue was addressed. It could indicate that the comparison of those methods is unreliable and unfair.

RE: We thank the reviewer for the comments. The precise definition of ligand-binding sites is elucidated in the “Benchmark datasets” section. Specifically, the datasets of DNA, RNA, peptide, ATP, HEM and metal ions used to train GPSite were collected from the widely acknowledged BioLiP database [PMID: 23087378]. In BioLiP, a binding residue is defined if the smallest atomic distance between the target residue and the ligand is <0.5 Å plus the sum of the Van der Waal’s radius of the two nearest atoms. Meanwhile, most comparative methods regarding these ligands were also trained on data from BioLiP, thereby ensuring fair comparisons.

However, since BioLiP does not include data on protein-protein binding sites, studies for protein-protein binding site prediction may adopt slightly distinct label definitions, as the reviewer suggested. Here, we employed the protein-protein binding site data from our previous study [PMID: 34498061], where a protein-binding residue was defined as a surface residue (relative solvent accessibility > 5%) that lost more than 1 Å2 absolute solvent accessibility after protein-protein complex formation. This definition was initially introduced in PSIVER [PMID: 20529890] and widely applied in various studies (e.g., PMID: 31593229, PMID: 32840562). SPPIDER [PMID: 17152079] and MaSIF-site [PMID: 31819266] have also adopted similar surface-based definitions as PSIVER. On the other hand, ScanNet [PMID: 35637310] employed an atom distance threshold of 4 Å to define contacts while PeSTo [PMID: 37072397] used a threshold of 5 Å. However, it is noteworthy that current methods in this field including ScanNet (Nat Methods 2022) and PeSTo (Nat Commun 2023) directly compared methods using different label definitions without any alignment in their benchmark studies, likely due to the subtle distinctions among these definitions. For instance, the study of PeSTo directly performed comparisons with ScanNet, MaSIF-site, SPPIDER, and PSIVER. Therefore, we followed these previous works, directly comparing GPSite with other protein-protein binding site predictors.

In the revised “Benchmark datasets” section, we have now provided more details for the binding site definitions in different datasets to avoid any potential ambiguity:

“The benchmark datasets for evaluating binding site predictions of DNA, RNA, peptide, ATP, and HEM are constructed from BioLiP”; “A binding residue is defined if the smallest atomic distance between the target residue and the ligand is < 0.5 Å plus the sum of the Van der Waal’s radius of the two nearest atoms”; “Besides, the benchmark dataset of protein-protein binding sites is directly from 26, which contains non-redundant transient heterodimeric protein complexes dated up to May 2021. Surface regions that become solvent inaccessible on complex formation are defined as the ground truth protein-binding sites. The benchmark datasets of metal ion (Zn2+, Ca2+, Mg2+ and Mn2+) binding sites are directly from 18, which contain non-redundant proteins dated up to December 2021 from BioLiP.”

While GPSite demonstrates the potential to surpass state-of-the-art methods in protein binding site prediction, the evidence supporting these claims seems incomplete. The lack of methodological novelty and the unresolved questions in benchmark consistency and interface definition somewhat undermine the confidence in the results. Therefore, it's not entirely clear if the authors have fully achieved their aims as outlined.

The work is useful for the field, especially in disease mechanism elucidation and novel drug design. The availability of genome-scale binding residue annotations GPSite offers is a significant advancement. However, the utility of this tool could be hampered by the aforementioned weaknesses unless they are adequately addressed.

RE: We thank the reviewer for acknowledging the advancement and value of our work, as well as pointing out areas where improvements can be made. As discussed above, we have now carried out the corresponding revisions in the revised manuscript to enhance the completeness and clearness of our work.

Reviewer #2 (Public Review):

Summary:

This work provides a new framework, "GPsite" to predict DNA, RNA, peptide, protein, ATP, HEM, and metal ions binding sites on proteins. This framework comes with a webserver and a database of annotations. The core of the model is a Geometric featurizer neural network that predicts the binding sites of a protein. One major contribution of the authors is the fact that they feed this neural network with predicted structure from ESMFold for training and prediction (instead of native structure in similar works) and a high-quality protein Language Model representation. The other major contribution is that it provides the public with a new light framework to predict protein-ligand interactions for a broad range of ligands.

The authors have demonstrated the interest of their framework with mostly two techniques: ablation and benchmark.

Strengths:

• The performance of this framework as well as the provided dataset and web server make it useful to conduct studies.

• The ablations of some core elements of the method, such as the protein Language Model part, or the input structure are very insightful and can help convince the reader that every part of the framework is necessary. This could also guide further developments in the field. As such, the presentation of this part of the work can hold a more critical place in this work.

RE: We thank the reviewer for recognizing the contributions of our work and for noting that our experiments are thorough.

Weaknesses:

• Overall, we can acknowledge the important effort of the authors to compare their work to other similar frameworks. Yet, the lack of homogeneity of training methods and data from one work to the other makes the comparison slightly unconvincing, as the authors pointed out. Overall, the paper puts significant effort into convincing the reader that the method is beating the state of the art. Maybe, there are other aspects that could be more interesting to insist on (usability, interest in protein engineering, and theoretical works).

RE: We sincerely appreciate the reviewer for the constructive and insightful comments. As to the concern of training data heterogeneity raised by the reviewer, it is noteworthy that current studies in this field, such as ScanNet (Nat Methods 2022) and PeSTo (Nat Commun 2023), directly compare methods trained on different datasets in their benchmark experiments. Therefore, we have adhered to the paradigm in these previous works. According to the detailed recommendations by the reviewer, we have now improved our manuscript by incorporating additional ablation studies regarding the effects of training procedure and language model representations, as well as case studies regarding the predicted structure’s quality and GPSite-based function annotations. We have also refined the Discussion section to focus more on the achievements of this work. A comprehensive point-by-point response to the reviewer’s recommendations is provided below.

Reviewer #2 (Recommendations For The Authors):

Major comments:

Overall I think the work is slightly deserved by its presentation. Some improvements could be made to the paper to better highlight the significance of your contribution.

RE: We thank the reviewer for recognizing the significance of our work!

• Line 188: "As expected, the performance of these methods mostly decreases substantially utilizing predicted structures for testing because they were trained with high-quality native structures.

This is a major ablation that was not performed in this case. You used the predicted structure to train, while the other did not. One better way to assess the interest of this approach would be to compare the performance of a network trained with only native structure to compare the leap in performance with and without this predicted structure as you did after to assess the interest of some other aspect of your method such as single to multitask.

RE: We thank the reviewer for the valuable recommendation. We have now assessed the benefit of training with predicted instead of native structures, which brings an average AUPR increase of 4.2% as detailed in Appendix 1-note 5 and Appendix 2-table 9. For convenience, we also attach the note and table here:

“We examined the performance under different training and evaluation settings as shown in Appendix 2-table 9. As expected, the model yields exceptional performance (average AUPR of 0.656) when trained and evaluated using native structures. However, if this model is fed with predicted structures of the test proteins, the performance substantially declines to an average AUPR of 0.573. This trend aligns with the observations for other structure-based methods as illustrated in Figure 2. More importantly, in the practical scenario where only predicted structures are available for the target proteins, training the model with predicted structures (i.e., GPSite) results in superior performance than training the model with native structures (average AUPR of 0.594 against 0.573), probably owing to the consistency between the training and testing data. For completeness, the results in Appendix 3-figure 2 are also included where GPSite is tested with native structures (average AUPR of 0.637).”

Author response table 2.

Performance comparison on the ten binding site test sets under different training and evaluation settings

Note: The numbers in this table are AUPR values. “Pep” and “Pro” denote peptide and protein, respectively. “Avg” means the average AUPR values among the ten test sets. “native” and “predicted” denote applying native and predicted structures as input, respectively.

• Line 263: "ProtTrans consistently obtains competitive or superior performance compared to the MSA profiles, particularly for the target proteins with few homologous sequences (Neff < 2)."

This seems a bit far-fetched. If we see clearly in the figure that the performances are far superior for Neff < 2. The performances seem rather similar for higher Neff. Could the author evaluate numerically the significance of the improvement? MSA profiles outperform GPSite on 4 intervals and I don't know the distribution of the data.

RE: We thank the reviewer for the valuable suggestion. We have now revised this sentence to avoid any potential ambiguity:

“As evidenced in Figure 4B and Appendix 2-table 8, ProtTrans consistently obtains competitive or superior performance compared to the MSA profile. Notably, for the target proteins with few homologous sequences (Neff < 2), ProtTrans surpasses MSA profile significantly with an improvement of 3.9% on AUC (P-value = 4.3×10-8).”

The detailed significance tests and data distribution are now added in Appendix 2-table 8 and attached below as Author response-table 3 for convenience:

Author response table 3.

Performance comparison between GPSite and the baseline model using MSA profile for proteins with different Neff values in the combined test set of the ten ligands

Note: Significance tests are performed following the procedure in 12,25. If P-value < 0.05, the difference between the performance is considered statistically significant.

• Line 285: "We first visualized the distributions of residues in this dataset using t-SNE, where the residues are encoded by raw feature vectors encompassing ProtTrans embeddings and DSSP structural properties, or latent embedding vectors from the shared network of GPSite. "

Wouldn't embedding from single-task be more relevant to show the interest of multi-task training here? Is the difference that big when comparing embeddings from single-task training to embeddings from multi-task training? Otherwise, I think the evidence from Figure 4e is sufficient, the interest of multitasking could be well-shown by single-task vs. multi-task AUPR and a few examples or predictions that are improved.

RE: We thank the reviewer for the comment. In the second paragraph of the “The effects of protein features and model designs” section, we have compared the performance of multi-task and single-task learning. However, the visualization results in Figure 4D are related to the third paragraph, where we conducted a downstream exploration of the possibility to extend GPSite to other unseen ligands. This is based on the hypothesis that the shared network in GPSite may have captured certain common ligand-binding mechanisms during the preceding multi-task training process. We visualized the distributions of residues in an unseen carbohydrate-binding site dataset using t-SNE, where the residues are encoded by raw feature vectors (ProtTrans and DSSP), or latent embedding vectors from the shared network trained before. Although the shared network has not been specifically trained on the carbohydrate dataset, the latent representations from GPSite effectively improve the discriminability between the binding and non-binding residues as shown in Figure 4D. This finding indicates that the shared network trained on the initial set of ten molecule types has captured common binding mechanisms and may be applied to other unseen ligands.

We have now added more descriptions in this paragraph to avoid potential ambiguity:

“Residues that are conserved during evolution, exposed to solvent, or inside a pocket-shaped domain are inclined to participate in ligand binding. During the preceding multi-task training process, the shared network in GPSite should have learned to capture such common binding mechanisms. Here we show how GPSite can be easily extended to the binding site prediction for other unseen ligands by adopting the pre-trained shared network as a feature extractor. We considered a carbohydrate-binding site dataset from 54 which contains 100 proteins for training and 49 for testing. We first visualized the distributions of residues in this dataset using t-SNE 55, where the residues are encoded by raw feature vectors encompassing ProtTrans embeddings and DSSP structural properties, or latent embedding vectors from the shared network of GPSite trained on the ten molecule types previously.”

• Line291: "Employing these informative hidden embeddings as input features to train a simple MLP exhibits remarkable performance with an AUC of 0.881 (Figure 4E), higher than that of training a single-task version of GPSite from scratch (AUC of 0.853) or other state-of-the-art methods such as MTDsite and SPRINT-CBH."

Is it necessary to introduce other methods here? The single-task vs multi-task seems enough for what you want to show?

RE: We thank the reviewer for the comment. As discussed above, here we aim to show the potential of GPSite for the binding site prediction of unseen ligand (i.e., carbohydrate) by adopting the pre-trained shared network as a feature extractor. Thus, we think it’s reasonable to also include the performance of other state-of-the-art methods in this carbohydrate benchmark dataset as baselines.

• Line 321: "Specifically, a protein-level binding score can be generated for each ligand by averaging the top k predicted scores among all residues. Empirically, we set k to 5 for metal ions and 10 for other ligands, considering that the binding interfaces of metal ions are usually smaller."

Since binding sites are usually not localized on one single amino-acid, we can expect that most of the top k residues are localized around the same area of the protein both spatially and along the sequence. Is it something you observe and could consider in your method?

RE: We thank the reviewer for the comment. We employed a straightforward method (top-k average) to convert GPSite’s residue-level annotations into protein-level annotations, where k was set empirically based on the distributions of the numbers of binding residues per sequence observed in the training set. We have not put much effort in optimizing this strategy since it mainly serves as a proof-of-concept experiment (Figure 5 A-C) to show the potential of GPSite in discriminating ligand-binding proteins. We have now revised this sentence to better explain how we selected k:

“Specifically, a protein-level binding score indicating the overall binding propensity to a specific ligand can be generated by averaging the top k predicted scores among all residues. Empirically, we set k to 5 for metal ions and 10 for other ligands, considering the distributions of the numbers of binding residues per sequence observed in the training set.”

As for the question raised by the reviewer, we can indeed expect that most of the top k predicted binding residues tend to cluster into several but not necessarily one area. For instance, certain macromolecules like DNA may interact with several protein surface patches due to their elongated structures (e.g., Author esponse-figure 1A). Another case may be a protein binding to multiple molecules of the same ligand type (e.g., Author response-figure 1B).

Author response image 1.

The structures of 4XQK (A) and 4KYW (B) in PDB.

• Line 327: The accuracy of the GPSite protein-level binding scores is further validated by the ROC curves in Figure 5B, where GPSite achieves satisfactory AUC values for all ligands except protein (AUC of 0.608).

Here may be a good place to compare yourself with others, do other frameworks experience the same problem? If so, AUC and AUPR are not relevant here, can you expose some recall scores for example?

RE: We thank the reviewer for the valuable recommendation. We have conducted comprehensive method comparisons in the preceding “GPSite outperforms state-of-the-art methods” section, where GPSite surpasses all existing frameworks across various ligands. Here, the genome-wide analyses of Swiss-Prot in Figure 5 serve as a downstream demonstration of GPSite’s capacity for large-scale annotations. We didn’t compare with other methods since most of them are time-consuming or memory-consuming, thus unavailable to process sequences of substantial quantity or length. For example, it takes about 8 min for the MSA-based method GraphBind to annotate a protein with 500 residues, while it just takes about 20 s for GPSite (see Appendix 3-figure 1 for detailed runtime comparison). It is also challenging for the atom-graph-based method PeSTo to process structures more than 100 kDa (~1000 residues) on a 32 GB GPU as the authors suggested, while GPSite can easily process structures containing up to 2500 residues on a 16 GB GPU.

Regarding the recall score mentioned by the reviewer, GPSite achieves a recall of 0.95 (threshold = 0.5) for identifying protein-binding proteins. This indicates that GPSite can accurately identify positive samples, but it also tends to misclassify negative samples as positive. In our original manuscript, we claimed that “This may be ascribed to the fact that protein-protein interactions are ubiquitous in living organisms while the Swiss-Prot function annotations are incomplete”. To better support this claim, we have now added two examples in Appendix 1-note 7, where GPSite confidently predicted the presences of the “protein binding” function (GO:0005515). Notably, this function was absent in these two proteins in the Swiss-Prot database at the time of manuscript preparation (release: 2023-05-03), but has been included in the latest release of Swiss-Prot (release: 2023-11-08). For convenience, we also attach the note here:

“As depicted in Figure 5A, GPSite assigns relatively high prediction scores to the proteins without “protein binding” function in the Swiss-Prot annotations, leading to a modest AUC value of 0.608 (Figure 5B). This may be ascribed to the fact that protein-protein interactions are ubiquitous in living organisms while the Swiss-Prot function annotations are incomplete. To support this hypothesis, we present two proteins as case studies, both sharing < 20% sequence identity with the protein-binding training set of GPSite. The first case is Aminodeoxychorismate synthase component 2 from Escherichia coli (UniProt ID: P00903). GPSite confidently predicted this protein as a protein-binding protein with a high prediction score of 0.936. Notably, this protein was not annotated with the “protein binding” function (GO:0005515) or any of its GO child terms in the Swiss-Prot database at the time of manuscript preparation (https://rest.uniprot.org/unisave/P00903?format=txt&versions=171, release: 2023-05-03). However, in the latest release of Swiss-Prot (https://rest.uniprot.org/unisave/P00903?format=txt&versions=174, release: 2023-11-08) during manuscript revision, this protein is annotated with the “protein heterodimerization activity” function (GO:0046982), which is a child term of “protein binding”. In fact, the heterodimerization activity of this protein has been validated through experiments in the year of 1996 (PMID: 8679677), indicating the potential incompleteness of the Swiss-Prot annotations. The other case is Hydrogenase-2 operon protein HybE from Escherichia coli (UniProt ID: P0AAN1), which was also predicted as a protein-binding protein by GPSite (score = 0.909). Similarly, this protein was not annotated with the “protein binding” function in the Swiss-Prot database at the time of manuscript preparation (https://rest.uniprot.org/unisave/P0AAN1?format=txt&versions=108). However, in the latest release of Swiss-Prot (https://rest.uniprot.org/unisave/P0AAN1?format=txt&versions=111), this protein is annotated with the “preprotein binding” function (GO:0070678), which is a child term of “protein binding”. In fact, the preprotein binding function of this protein has been validated through experiments in the year of 2003 (PMID: 12914940). These cases demonstrate the effectiveness of GPSite for completing the missing function annotations in Swiss-Prot.”

• Line 381: 'Despite the noteworthy advancements achieved by GPSite, there remains scope for further improvements. Given that the ESM Metagenomic Atlas 34 provides 772 million predicted protein structures along with pre-computed language model embeddings, self-supervised learning can be employed to train a GPSite model for predicting masked sequence and structure attributes, or maximizing the similarity between the learned representations of substructures from identical proteins while minimizing the similarity between those from different proteins using a contrastive loss function training from scratch. Additional opportunities for upgrade exist within the network architecture. For example, a variational Expectation-Maximization (EM) framework 58 can be adopted to handle the hierarchical graph structure inherent in proteins, which contains the top view of the residue graph and the bottom view of the atom graph inside a residue. Such an EM procedure enables training two separate graph neural networks for the two views while simultaneously allowing interaction and mutual enhancement between the two modules. Meta-learning could also be explored in this multi-task scenario, which allows fast adaptation to unseen tasks with limited labels.'

I think this does not belong here. It feels like half of your discussion is not talking about the achievements of this paper but future very specific directions. Focus on the take-home arguments (performances of the model, ability to predict a large range of tasks, interest in key components of your model, easy use) of the paper and possible future direction but without being so specific.

RE: We thank the reviewer for the valuable suggestion. We have now simplified the discussions on the future directions notably:

“Despite the noteworthy advancements achieved by GPSite, there remains scope for further improvements. GPSite may be improved by pre-training on the abundant predicted structures in ESM Metagenomic Atlas, and then fine-tuning on binding site datasets. Besides, the hidden embeddings from ESMFold may also serve as informative protein representations. Additional opportunities for upgrade exist within the network architecture. For example, a variational Expectation-Maximization framework can be adopted to handle the hierarchical atom-to-residue graph structure inherent in proteins. Meta-learning could also be explored in this multi-task scenario, which allows fast adaptation to unseen tasks with limited labels.”

• Overall there is also a lack of displayed structure. You should try to select a few examples of binding sites that were identified correctly by your method and not by others, if possible get some insights on why. Also, some negative examples could be interesting so as to have a better idea of the interest.

RE: We thank the reviewer for the valuable recommendation. We have performed a case study for the structure of the glucocorticoid receptor in Figure 3 D-H to illustrate a potential reason for the robustness of GPSite. Moreover, we have now added a case study in Appendix 1-note 3 and Appendix 3-figure 5 to explain why GPSite sometimes is not as accurate as the state-of-the-art structure-based method. For convenience, we also attach the note and figure here:

“Here we present an example of an RNA-binding protein, i.e., the ribosome biogenesis protein ERB1 (PDB: 7R6Q, chain m), to illustrate the impact of predicted structure’s quality. As shown in Appendix 3-figure 5, ERB1 is an integral component of a large multimer structure comprising protein and RNA chains (i.e., the state E2 nucleolar 60S ribosome biogenesis intermediate). Likely due to the neglect of interactions from other protein chains, ESMFold fails to predict the correct conformation of the ERB1 chain (TM-score = 0.24). Using this incorrect predicted structure, GPSite achieves an AUPR of 0.580, lower than GraphBind input with the native structure (AUPR = 0.636). However, the performance of GraphBind substantially declines to an AUPR of 0.468 when employing the predicted structure as input. Moreover, if GPSite adopts the native structure for prediction, a notable performance boost can be obtained (AUPR = 0.681).”

Author response image 2.

The prediction results of GPSite and GraphBind for the ribosome biogenesis protein ERB1. (A) The state E2 nucleolar 60S ribosome biogenesis intermediate (PDB: 7R6Q). The ribosome biogenesis protein ERB1 (chain m) is highlighted in blue, while other protein chains are colored in gray. The RNA chains are shown in orange. (B) The RNA-binding sites on ERB1 (colored in red). (C) The ESMFold-predicted structure of ERB1 (TM-score = 0.24). The RNA-binding sites are also mapped onto this predicted structure (colored in red). (D-G) The prediction results of GPSite and GraphBind for the predicted and native ERB1 structures. The confidence of the predictions is represented with a gradient of color from blue for non-binding to red for binding.

Minor comments:

• Line 169: "Note that since our test sets may partly overlap with the training sets of these methods, the results reported here should be the upper limits for the existing methods."

Yes, but they were potentially not trained on the most recent structures in that case. These methods could also see improved performance with an updated training set.

RE: We thank the reviewer for the comment. We have now deleted this sentence.

• Line176: "Since 358 of the 375 proteins in our protein-binding site test set share > 30% identity with the training sequences of PeSTo, we re-split our protein-binding dataset to generate a test set of 65 proteins sharing < 30% identity with the training set of PeSTo for a fair evaluation."

Too specific to be here in my opinion.

RE: We thank the reviewer for the comment. We have now moved these details to Appendix 1-note 2. The description in the main text here is now more concise:

“Given the substantial overlap between our protein-binding site test set and the training set of PeSTo, we conducted separate training and comparison using the datasets of PeSTo, where GPSite still demonstrates a remarkable improvement over PeSTo (Appendix 1-note 2).”

• Figure 2. The authors should try to either increase Fig A's size or increase the font size. This could probably be done by compressing the size of Figure C into a single figure.

RE: We thank the reviewer for the suggestion. We have now increased the font size in Figure A. Besides, the figures in the final version of the manuscript should be clearer where we could upload SVG files.

• Have you tried using embeddings from more structure-aware pLM such as ESM Fold embeddings (fine-tuned) or ProstTrans (that may be more recent than this study)?

RE: We thank the reviewer for the insightful comment. We have not yet explored the embeddings from structure-aware pLM, but we acknowledge its potential as a promising avenue for future investigation. We have now added this point in our Discussion section:

“Besides, the hidden embeddings from ESMFold may also serve as informative protein representations.”

Reviewer #3 (Public Review):

Summary

The authors of this work aim to address the challenge of accurately and efficiently identifying protein binding sites from sequences. They recognize that the limitations of current methods, including reliance on multiple sequence alignments or experimental protein structure, and the under-explored geometry of the structure, which limit the performance and genome-scale applications. The authors have developed a multi-task network called GPSite that predicts binding residues for a range of biologically relevant molecules, including DNA, RNA, peptides, proteins, ATP, HEM, and metal ions, using a combination of sequence embeddings from protein language models and ESMFold-predicted structures. Their approach attempts to extract residual and relational geometric contexts in an end-to-end manner, surpassing current sequence-based and structure-based methods.

Strengths

• The GPSite model's ability to predict binding sites for a wide variety of molecules, including DNA, RNA, peptides, and various metal ions.

• Based on the presented results, GPSite outperforms state-of-the-art methods in several benchmark datasets.

• GPSite adopts predicted structures instead of native structures as input, enabling the model to be applied to a wider range of scenarios where native structures are rare.

• The authors emphasize the low computational cost of GPSite, which enables rapid genome-scale binding residue annotations, indicating the model's potential for large-scale applications.

RE: We thank the reviewer for recognizing the significance and value of our work!

Weaknesses

• One major advantage of GPSite, as claimed by the authors, is its efficiency. Although the manuscript mentioned that the inference takes about 5 hours for all datasets, it remains unclear how much improvement GPSite can offer compared with existing methods. A more detailed benchmark comparison of running time against other methods is recommended (including the running time of different components, since some methods like GPSite use predicted structures while some use native structures).

RE: We thank the reviewer for the valuable suggestion. Empirically, it takes about 5-20 min for existing MSA-based methods to make predictions for a protein with 500 residues, while it only takes about 1 min for GPSite (including structure prediction). However, it is worth noting that some predictors in our benchmark study are solely available as webservers, and it is challenging to compare the runtime between a standalone program and a webserver due to the disparity in hardware configurations. Therefore, we have now included comprehensive runtime comparisons between the GPSite webserver and other top-performing servers in Appendix 3-figure 1 to illustrate the practicality and efficiency of our method. For convenience, we also attach the figure here as Author response-figure 3. The corresponding description is now added in the “GPSite outperforms state-of-the-art methods” section:

“Moreover, GPSite is computationally efficient, achieving comparable or faster prediction speed compared to other top-performing methods (Appendix 3-figure 1).”

Author response image 3.

Runtime comparison of the GPSite webserver with other top-performing servers. Five protein chains (i.e., 8HN4_B, 8USJ_A, 8C1U_A, 8K3V_A and 8EXO_A) comprising 100, 300, 500, 700, and 900 residues, respectively, were selected for testing, and the average runtime is reported for each method. Note that a significant portion of GPSite’s runtime (75 s, indicated in orange) is allocated to structure prediction using ESMFold.

• Since the model uses predicted protein structure, the authors have conducted some studies on the effect of the predicted structure's quality. However, only the 0.7 threshold was used. A more comprehensive analysis with several different thresholds is recommended.

RE: We thank the reviewer for the comment. We assessed the effect of the predicted structure's quality by evaluating GPSite’s performance on high-quality (TM-score > 0.7) and low-quality (TM-score ≤ 0.7) predicted structures. We did not employ multiple thresholds (e.g., 0.3, 0.5, and 0.7), as the majority of proteins in the test sets were accurately predicted by ESMFold. Specifically, as shown in Figure 3B, Appendix 3-figure 3 and Appendix 2-table 5, the numbers of proteins with TM-score ≤ 0.7 are small in most datasets (e.g., 42 for DNA and 17 for ATP). Consequently, there is insufficient data available for analysis with lower thresholds, except for the RNA test set. Notably, Figure 3C presents a detailed inspection of the 104 proteins with TM-score < 0.5 in the RNA test set. Within this subset, GPSite consistently outperforms the state-of-the-art structure-based method GraphBind with predicted structures as input, regardless of the prediction quality of ESMFold. Only in cases where structures are predicted with extremely low quality (TM-score < 0.3) does GPSite fall behind GraphBind input with native structures. This result further demonstrates the robustness of GPSite. We have now added clearer explanations in the “GPSite is robust for low-quality predicted structures” section:

“Figure 3B and Appendix 3-figure 3 show the distributions of TM-scores between native and predicted structures calculated by US-align in the ten benchmark datasets, where most proteins are accurately predicted with TM-score > 0.7 (see also Appendix 2-table 5)”; “Given the infrequency of low-quality predicted structures except for the RNA test set, we took a closer inspection of the 104 proteins with predicted structures of TM-score < 0.5 in the RNA test set.”

• To demonstrate the robustness of GPSite, the authors performed a case study on human GR containing two zinc fingers, where the predicted structure is not perfect. The analysis could benefit from more a detailed explanation of why the model can still infer the binding site correctly even though the input structural information is slightly off.

RE: We thank the reviewer for the comment. We have actually explained the potential reason for the robustness of GPSite in the second paragraph of the “GPSite is robust for low-quality predicted structures” section. In summary, although the whole structure of this protein is not perfectly predicted, the local structures of the binding domains of peptide, DNA and Zn2+ are actually predicted accurately as evidenced by the superpositions of the native and predicted structures in Figure 3D and 3E. Therefore, GPSite can still make reliable predictions. We have now revised this paragraph to explain these more clearly:

“Figure 3D shows the structure of the human glucocorticoid receptor (GR), a transcription factor that binds DNA and assembles a coactivator peptide to regulate gene transcription (PDB: 7PRW, chain A). The DNA-binding domain of GR also consists of two C4-type zinc fingers to bind Zn2+ ions. Although the structure of this protein is not perfectly predicted (TM-score = 0.72), the local structures of the binding domains of peptide and DNA are actually predicted accurately as viewed by the superpositions of the native and predicted structures in Figure 3D and 3E. Therefore, GPSite can correctly predict all Zn2+ binding sites and precisely identify the binding sites of DNA and peptide with AUPR values of 0.949 and 0.924, respectively (Figure 3F, G and H).”

• To analyze the relatively low AUC value for protein-protein interactions, the authors claimed that it is "due to the fact that protein-protein interactions are ubiquitous in living organisms while the Swiss-Prot function annotations are incomplete", which is unjustified. It is highly recommended to support this claim by showing at least one example where GPSite's prediction is a valid binding site that is not present in the current Swiss-Prot database or via other approaches.

RE: We thank the reviewer for the valuable recommendation. To support this claim, we have now added two examples in Appendix 1-note 7, where GPSite confidently predicted the presences of the “protein binding” function (GO:0005515). Notably, this function was absent in these two proteins in the Swiss-Prot database at the time of manuscript preparation (release: 2023-05-03), but has been included in the latest release of Swiss-Prot (release: 2023-11-08). For convenience, we also attach the note below:

“As depicted in Figure 5A, GPSite assigns relatively high prediction scores to the proteins without “protein binding” function in the Swiss-Prot annotations, leading to a modest AUC value of 0.608 (Figure 5B). This may be ascribed to the fact that protein-protein interactions are ubiquitous in living organisms while the Swiss-Prot function annotations are incomplete. To support this hypothesis, we present two proteins as case studies, both sharing < 20% sequence identity with the protein-binding training set of GPSite. The first case is Aminodeoxychorismate synthase component 2 from Escherichia coli (UniProt ID: P00903). GPSite confidently predicted this protein as a protein-binding protein with a high prediction score of 0.936. Notably, this protein was not annotated with the “protein binding” function (GO:0005515) or any of its GO child terms in the Swiss-Prot database at the time of manuscript preparation (https://rest.uniprot.org/unisave/P00903?format=txt&versions=171, release: 2023-05-03). However, in the latest release of Swiss-Prot (https://rest.uniprot.org/unisave/P00903?format=txt&versions=174, release: 2023-11-08) during manuscript revision, this protein is annotated with the “protein heterodimerization activity” function (GO:0046982), which is a child term of “protein binding”. In fact, the heterodimerization activity of this protein has been validated through experiments in the year of 1996 (PMID: 8679677), indicating the potential incompleteness of the Swiss-Prot annotations. The other case is Hydrogenase-2 operon protein HybE from Escherichia coli (UniProt ID: P0AAN1), which was also predicted as a protein-binding protein by GPSite (score = 0.909). Similarly, this protein was not annotated with the “protein binding” function in the Swiss-Prot database at the time of manuscript preparation (https://rest.uniprot.org/unisave/P0AAN1?format=txt&versions=108). However, in the latest release of Swiss-Prot (https://rest.uniprot.org/unisave/P0AAN1?format=txt&versions=111), this protein is annotated with the “preprotein binding” function (GO:0070678), which is a child term of “protein binding”. In fact, the preprotein binding function of this protein has been validated through experiments in the year of 2003 (PMID: 12914940). These cases demonstrate the effectiveness of GPSite for completing the missing function annotations in Swiss-Prot.”

• The authors reported that many GPSite-predicted binding sites are associated with known biological functions. Notably, for RNA-binding sites, there is a significantly higher proportion of translation-related binding sites. The analysis could benefit from a further investigation into this observation, such as the analyzing the percentage of such interactions in the training site. In addition, if there is sufficient data, it would also be interesting to see the cross-interaction-type performance of the proposed model, e.g., train the model on a dataset excluding specific binding sites and test its performance on that class of interactions.

RE: We thank the reviewer for the suggestion. We would like to clarify that the analysis in Figure 5C was conducted at “protein-level” instead of “residue-level”. As described in the second paragraph of the “Large-scale binding site annotation for Swiss-Prot” section, a protein-level ligand-binding score was assigned to a protein by averaging the top k residue-level predicted binding scores. This protein-level score indicates the overall binding propensity of the protein to a specific ligand. We gathered the top 20,000 proteins with the highest protein-level binding scores for each ligand and found that their biological process annotations from Swiss-Prot were consistent with existing knowledge. We have now revised the corresponding sentence to explain these more clearly:

“Exploiting the residue-level binding site annotations, we could readily extend GPSite to discriminate between binding and non-binding proteins of various ligands. Specifically, a protein-level binding score indicating the overall binding propensity to a specific ligand can be generated by averaging the top k predicted scores among all residues.”

As for the cross-interaction-type performance raised by the reviewer, we have now conducted cross-type evaluations to investigate the specificity of the ligand-specific MLPs and the inherent similarities among different ligands in Appendix 1-note 6 and Appendix 2-table 10. For convenience, we also attach the note and table here:

“We conducted cross-type evaluations by applying different ligand-specific MLPs in GPSite for the test sets of different ligands. As shown in Appendix 2-table 10, for each ligand-binding site test set, the corresponding ligand-specific network consistently achieves the best performance. This indicates that the ligand-specific MLPs have specifically learned the binding patterns of particular molecules. We also noticed that the cross-type performance is reasonable for the ligands sharing similar properties. For instance, the DNA-specific MLP exhibits a reasonable AUPR when predicting RNA-binding sites, and vice versa. Similar trends are also observed between peptide and protein, as well as among metal ions as expected. Interestingly, the cross-type performance between ATP and HEM is also acceptable, potentially attributed to their comparable molecular weights (507.2 and 616.5, respectively).”

Author response table 4.

Cross-type performance by applying different ligand-specific MLPs in GPSite for the test sets of different ligands

Note: “Pep” and “Pro” denote peptide and protein, respectively. The numbers in this table are AUPR values. The best/second-best result in each test set is indicated by bold/underlined font.

Author Response

eLife assessment

The authors report that optogenetic inhibition of hippocampal axon terminals in retrosplenial cortex impairs the performance of a delayed non-match to place task. The significance of findings elucidating the role of hippocampal projections to the retrosplenial cortex in memory and decision-making behaviors is important. However, the strength of evidence for the paper's claims is currently incomplete.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This is a study on the role of the retrosplenial cortex (RSC) and the hippocampus in working memory. Working memory is a critical cognitive function that allows temporary retention of information for task execution. The RSC, which is functionally and anatomically connected to both primary sensory (especially visual) and higher cognitive areas, plays a key role in integrating spatial-temporal context and in goal-directed behaviors. However, the specific contributions of the RSC and the hippocampus in working memory-guided behaviors are not fully understood due to a lack of studies that experimentally disrupt the connection between these two regions during such behaviors.

In this study, researchers employed eArch3.0 to silence hippocampal axon terminals in the RSC, aiming to explore the roles of these brain regions in working memory. Experiments were conducted where animals with silenced hippocampal axon terminals in the RSC performed a delayed non-match to place (DNMP) task. The results indicated that this manipulation impaired memory retrieval, leading to decreased performance and quicker decision-making in the animals. Notably, the authors observed that the effects of this impairment persisted beyond the light-activation period of the opsin, affecting up to three subsequent trials. They suggest that disrupting the hippocampal-RSC connection has a significant and lasting impact on working memory performance.

Strengths:

They conducted a study exploring the impact of direct hippocampal inputs into the RSC, a region involved in encoding spatial-temporal context and transferring contextual information, on spatial working memory tasks. Utilizing eArch3.0 expressed in hippocampal neurons via the viral vector AAV5-hSyn1-eArch3.0, they aimed to bilaterally silence hippocampal terminals located at the RSC in rats pre-trained in a DNMP task. They discovered that silencing hippocampal terminals in the RSC significantly decreased working memory performance in eArch+ animals, especially during task interleaving sessions (TI) that alternated between trials with and without light delivery. This effect persisted even in non-illuminated trials, indicating a lasting impact beyond the periods of direct manipulation. Additionally, they observed a decreased likelihood of correct responses following TI trials and an increased error rate in eArch+ animals, even after incorrect responses, suggesting an impairment in error-corrective behavior. This contrasted with baseline sessions where no light was delivered, and both eArch+ and control animals showed low error rates.

Weaknesses:

While I agree with the authors that the role of hippocampal inputs to the RSC in spatial working memory is understudied and merits further investigation, I find that the optogenetic experiment, a core part of this manuscript that includes viral injections, could be improved. The effects were rather subtle, rendering some of the results barely significant and possibly too weak to support major conclusions.

We thank Reviewer#1 for carefully and critically reading our manuscript, and for the valuable comments provided. The judged “subtlety” of the effects stems from a perspective according to which a quantitatively lower effect bears less biological significance for cognition. We disagree with this perspective and find it rather reductive for several reasons.

Once seen in the context of the animal’s ecology, subtle impairments can be life-threatening precisely because of their subtlety, leading the animal to confidently rely on a defective capacity, for such events as remembering the habitual location of a predator, or food source.

Also, studies in animal cognition often undertake complete, rather than graded, suppression of a given mechanism (in the same sense as that of “knocking out” a gene that is relevant for behaviour), leading to a gravelly, rather that gradually, impaired model system, to the point of not allowing a hypothetical causal link to be mechanistically revealed beyond its mere presence. This often hinders a thorough interpretation of the perturbed factor’s role. If a caricatural analogy is allowed, it would be as if we were to study the role of an animal’s legs by chopping them both off and observing the resulting behaviour.

In our study we conclude that silencing HIPP inputs in RSC perturbs cognition enough to impair behaviour while not disabling the animal entirely, as such allowing for behaviour to proceed, and for our observation of graded, decreased (not absent), proficiency under optogenetic silencing. So rather than weak, we would say the results are statistically significant, and biologically realistic.

Additionally, no mechanistic investigation was conducted beyond referencing previous reports to interpret the core behavioral phenotypes.

We fully agree with this being a weakness, as we wish we could have done more mechanistic studies to find out exactly what is Arch activation doing to HIPP-RSC transmission, which neurons are being affected, and perhaps in the future dissect its circuit determinants. We have all these goals very present and hope we can address them soon.

Reviewer #2 (Public Review):

The authors examine the impact of optogenetic inhibition of hippocampal axon terminals in the retrosplenial cortex (RSP) during the performance of a working memory T-maze task. Performance on a delayed non-match-to-place task was impaired by such inhibition. The authors also report that inhibition is associated with faster decision-making and that the effects of inhibition can be observed over several subsequent trials. The work seems reasonably well done and the role of hippocampal projections to retrosplenial cortex in memory and decision-making is very relevant to multiple fields. However, the work should be expanded in several ways before one can make firm conclusions on the role of this projection in memory and behavior.

We thank Reviewer#2 for carefully and critically reading our manuscript, and for the valuable comments provided.

(1) The work is very singular in its message and the experimentation. Further, the impact of the inhibition on behaviour is very moderate. In this sense, the results do not support the conclusion that the hippocampal projection to retrosplenial cortex is key to working memory in a navigational setting.

As we have mentioned in response to Reviewer#1, the judged “very moderate” effect stems from a perspective according to which a quantitatively lower effect bears less biological significance for cognition, precluding its consideration as “key” for behaviour. We disagree with this perspective and find it rather reductive for several reasons. Once seen in the context of the animal’s ecology, quantitatively lower impairments in working memory are no less key for this cognitive capacity, and can be life-threatening precisely because of their subtlety, leading the animal to confidently rely on a defective capacity, for such events as remembering the habitual location of a predator, or food source. Furthermore, studies in animal cognition often undertake complete, rather than graded, suppression of a given mechanism (in the same sense as “knocking out” a gene that is relevant for behaviour), leading to a gravelly, rather that gradually, impaired model system, to the point of not allowing a hypothetical causal link to be mechanistically revealed beyond its mere presence. This often hinders a thorough interpretation of its role.

In our study we conclude that silencing HIPP inputs in RSC perturbs behaviour enough to impair behaviour while not disabling the animal entirely, as such allowing for behaviour to proceed, and our observation of graded, decreased (not absent), proficiency under optogenetic silencing. So rather than weak, we would say the results are statistically significant, and biologically realistic.

(2) There are no experiments examining other types of behavior or working memory. Given that the animals used in the studies could be put through a large number of different tasks, this is surprising. There is no control navigational task. There is no working memory test that is non-spatial. Such results should be presented in order to put the main finding in context.

It is hard to gainsay this point. The more thorough and complete a behavioural characterization is, the more informative is the study, from every angle you look at it. While we agree that other forms of WM would be quite interesting in this context, we also cannot ignore the fact that DNMP is widely tested as a WM task, one that is biologically plausible, sensitive to perturbations of neural circuitry know to be at play therein, and fully accepted in the field. Faced with the impossibility of running further studies, for lack of additional funding and human resources, we chose to run this task.

A control navigational task would, in our understanding, be used to assess whether silencing HIPP projections to RSC would affect (spatial?) navigation, rather than WM, thus explaining the observed impairment. To this we have the following to say: Spatial Navigation is a very basic cognitive function, one that relies on body orientation relative to spatial context, on keeping an updated representation of such spatial context, (“alas”, as memory), and on guiding behaviour according to acquired knowledge about spatial context. Some of these functions are integral to spatial working memory, as such, they might indeed be affected.

Dissecting the determinants of spatial WM is indeed an ongoing effort, one that was not the intention of the current study, but also one that we have very present, in hope we can address in the future.

A non-spatial WM task would indeed vastly solidify our claims beyond spatial WM, onto WM. We have, for this reason, changed the title of the manuscript which now reads “spatial working memory”.

(3) The actual impact of the inhibition on activity in RSP is not provided. While this may not be strictly necessary, it is relevant that the hippocampal projection to RSP includes, and is perhaps dominated by inhibitory inputs. I wonder why the authors chose to manipulate hippocampal inputs to RSP when the subiculum stands as a much stronger source of afferents to RSP and has been shown to exhibit spatial and directional tuning of activity. The points here are that we cannot be sure what the manipulation is really accomplishing in terms of inhibiting RSP activity (perhaps this explains the moderate impact on behavior) and that the effect of inhibiting hippocampal inputs is not an effective means by which to study how RSP is responsive to inputs that reflect environmental locations.

We fully agree that neural recordings addressing the effect of silencing on RSC neural activity is relevant. We do wish we could have provided more mechanistic studies, to find out exactly what is Arch activation doing to HIPP-RSC transmission, which neurons are being affected, and thus dissecting its circuit determinants. We have all these goals very present and hope we can address them soon. Subiculum, which we mention in the Introduction, is indeed a key player in this complex circuitry, one whose hypothetical influence is the subject of experimental studies which will certainly reveal many other key elements.

(4) The impact of inhibition on trials subsequent to the trial during which optical stimulation was actually supplied seems trivial. The authors themselves point to evidence that activation of the hyperpolarizing proton pump is rather long-lasting in its action. Further, each sample-test trial pairing is independent of the prior or subsequent trials. This finding is presented as a major finding of the work, but would normally be relegated to supplemental data as an expected outcome given the dynamics of the pump when activated.

We disagree that this finding is “trivial”, and object to the considerations of “normalcy”, which we are left wondering about.

In lack of neurophysiological experiments (for the reasons stated above) to address this interesting finding, we chose to interpret it in light of (the few) published observations, such being the logical course of action in scientific reporting, given the present circumstances.

Evidence for such a prolonged effect in the context of behaviour is scarce (to our knowledge only the one we cite in the manuscript). As such, it is highly relevant to report it, and give it the relevance we do in our manuscript, rather than “relegating it to supplementary data”, as the reviewer considers being “normal”.

In the DNMP task the consecutive sample-test pairs are explicitly not independent, as they are part of the same behavioural session. This is illustrated by the simple phenomenon of learning, namely the intra-session learning curves, and the well-known behavioral trial-history effects. The brain does not simply erase such information during the ITI.

(5) In the middle of the first paragraph of the discussion, the authors make reference to work showing RSP responses to "contextual information in egocentric and allocentric reference frames". The citations here are clearly deficient. How is the Nitzan 2020 paper at all relevant here?

Nitzan 2020 reports the propagation of information from HIPP to CTX via SUB and RSC, thus providing a conduit for mnemonic information between the two structures, alternative to the one we target, thus providing thorough information concerning the HIPP-RSC circuitry at play during behaviour.

Alexander and Nitz 2015 precisely cite the encoding, and conjunction, of two types of contextual information, internal (ego-) and external (allocentric).

The subsequent reference is indeed superfluous here.

We thank the Reviewer#2 for calling our attention to the fact that references for this information are inadequate and lacking. We have now cited (Gill et al., 2011; Miller et al., 2019; Vedder et al., 2017) and refer readers to the review (Alexander et al., 2023) for the purpose of illustrating the encoding of information in the two reference frames. In addition, we have substantially edited the Introduction and Discussion sections, and suppressed unnecessary passages.

(6) The manuscript is deficient in referencing and discussing data from the Smith laboratory that is similar. The discussion reads mainly like a repeat of the results section.

Please see above. We thank Reviewer#2 for this comment, we have now re-written the Discussion such that it is less of a summary of the Results and more focused on their implications and future directions.

Author Response

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

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Hats off to the authors for taking time to decipher the seemingly subtle but important differences between the Gnai2/3 double mutant and Ptx mutant phenotypes. These results further illustrate the dynamic requirement of Gnai/0 in hair bundle establishment. I have some minor suggestions for the authors to consider and it is up to the authors to decide whether to incorporate them:

We decided to make the current (revised) version the version of record, and we explain why below. Please include these comments in the review+rebuttal material.

(1) The abstract could be modified to reflect the revised interpretations of the results.

Response: the abstract is high-level and the changes in interpretation in the revised manuscript do not modify the message there. Briefly, the abstract only states that Gnai2; Gnai3 double mutants recapitulate two defects previously only observed with pertussis toxin. There is no claim about the timing or dose of GNAI proteins involved.

(2) The three rows of OHCs are like a different beast from each other. Mireille Montcouquiol's lab has demonstrated that there is a differential requirement for Gnai3 in hair bundle orientation among the three rows of OHCs. The results described in this manuscript support this notion as well.

To clarify, Gnai3 inactivation does not affect OHC orientation. Only pertussis toxin, and in this work Gnai2; Gnai3 double mutants, do. The Montcouquiol lab showed different degree of OHC1, OHC2 and OHC3 misorientation upon use of pertussis toxin in vitro using cochlear explants (Ezan et al 2013). We showed the same thing in vivo using transgenic models (Tarchini et al 2013; Tarchini et al 2016). The different OHC responses by row and corresponding citations are mentioned in several locations in the manuscript, including first on line 112 in the Introduction and in Fig. 1C in a graphical summary.

(3) I wonder if "compensate" or "redundancy" may be a better term to use than "rescue" in the Discussion and figure.

Use of “rescue” in the Discussion is line 603 and 604. We think that “rescue” is appropriate to refer to the ability of GNAI2 to compensate for the loss of GNAI1 and GNAI3 in mutant context. We would argue that these different wordings are largely interchangeable and do not change the message.

---

Author Response

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

We really appreciate the time the reviewers spent reading and commenting on the original manuscript. Although they were positive already, we decided to spend some time to address the main comments with new experiments as thoroughly as possible in a new manuscript version. We also heavily edited some sections accordingly.: 1) we delayed pertussis toxin activation in hair cells with Atoh1-Cre to show that the resulting misorientation phenotype is delayed compared to FoxG1-Cre results, as also seen in Gnai2; Gnai3 double mutants. It follows that Gnai2; Gnai3 and pertussis mutants do share a similar misorientation profile, and that GNAI proteins are required to normally reverse OHC1-2 (from medial to lateral), but also to maintain the lateral orientation, at least transiently. 2) We experimentally verified that one of our GNAI antibodies can indeed detect GNAI1, and consequently that absence of signal in Gnai2; Gnai3 double mutants is evidence that GNAI1 is not involved in apical hair cell polarization. We believe these changes strengthen the manuscript and its conclusions.

Reviewer #1 (Public Review):

A subclass of inhibitory heterotrimeric guanine nucleotide-binding protein subunits, GNAI, has been implicated in sensory hair cell formation, namely the establishment of hair bundle (stereocilia) orientation and staircase formation. However, the former role of hair bundle orientation has only been demonstrated in mutants expressing pertussis toxin, which blocks all GNAI subunits, but not in mutants with a single knockout of any of the Gnai genes, suggesting that there is a redundancy among various GNAI proteins in this role. Using various conditional mutants, the authors concluded that GNAI3 is the primary GNAI proteins required for hair bundle morphogenesis, whereas hair bundle orientation requires both GNAI2 and GNAI3.

Strength

Various compound mutants were generated to decipher the contribution of individual GNAI1, GNAI2, GNAI3 and GNAIO in the establishment of hair bundle orientation and morphogenesis. The study is thorough with detailed quantification of hair bundle orientation and morphogenesis, as well as auditory functions.

Weakness

While the hair bundle orientation phenotype in the Foxg1-cre; Gnai2-/-; Gnai3 lox/lox (double mutants) appear more severe than those observed in Ptx cKO mutants, it may be an oversimplification to attribute the differences to more GNAI function in the Ptx cko mutants. The phenotypes between the double mutants and Ptx cko mutants appear qualitatively different. For example, assuming the milder phenotypes in the Ptx cKO is due to incomplete loss of GNAI function, one would expect the Ptx phenotype would be reproducible by some combination of compound mutants among various Gnai genes. Such information was not provided. Furthermore, of all the double mutant specimens analyzed for hair bundle orientation (Fig. 8), the hair bundle/kinocilium position started out normally in the lateral quadrant at E17.5 but failed to be maintained by P0. This does not appear to be the case for Ptx cKO, in which all affected hair cells showed inverted orientation by E17.5. It is not clear whether this is the end-stage of bundle orientation in Ptx cKO, and the kinocilium position started out normal, similar to the double mutants before the age of analysis at E17.5. Understanding these differences may reveal specific requirements of individual GNAI subunits or other factors are being affected in the Ptx mutants.

This criticism was very useful and prompted new experiments as well as a change in data presentation and a fundamental rewrite regarding hair cell orientation. These changes are detailed below. Of note, however, please let us clarify that the original manuscript did show that the ptxA orientation phenotype is reproduced to some extent in Gnai2; Gnai3 double mutants (previously Fig. 8 and corresponding text line 505). We showed that OHC1-2 are also inverted in the double mutant, although at a later differentiation stage. We recognize that similarities in hair cell misorientation between ptxA and Gnai2; Gnai3 DKO were not explained and discussed well enough. This part of the manuscript has been re-worked extensively, and we hope that along with new results, comparisons between mutant models are easier to follow and understand. We notably fully adopted the idea that there are qualitative differences between ptxA and Gnai2; Gnai3 mutants, and not only a difference in the remaining “dose” of GNAI activity.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Comments related to clarification of the weakness:

(1) In general, hair bundle orientation in the double mutants is established in the lateral quadrant of the cochlea before being inverted (Fig. 8). These results are intriguing because the lateral orientation is the correct position for these hair bundles normally and Gnai proteins are thought to be required to get the kinocilium to the lateral position. This process appears to proceed normally in the double mutants but the kinocilium reverted to the medial default position over time, which suggests that Gnai2 and Gnai3 are only required for the maintenance and not the establishment of the kinocilium in the lateral position. Is this phenotype qualitatively similar in the Ptx cKO?

We addressed these issues with two types of modifications to the data:

(1) We modified the eccentricity threshold used at E17.5 in Fig. 8 (orientation) to be more stringent, using 0.4 (instead of 0.25 previously) in both controls and mutants. This means that we now only graph the orientation of cells where eccentricity is more marked. The rationale is that at early stages, it is challenging to distinguish immature vs defective near-symmetrical cells. We kept a threshold of 0.25 at P0 when the hair cell apical surface is larger and better differentiated (Fig. 8C-D). Importantly, the dataset remains rigorously identical. This change usefully highlights that a large proportion of OHC1 is in fact inverted (oriented medially) at E17.5 in Gnai2; Gnai3 double mutants at the cochlear mid, as also seen in the ptxA model at the same stage and position (see new Fig. 8A). At the E17.5 base (Fig. 8B), a slightly more mature position, the outcome is unchanged (the majority of OHC1 are inverted using either a 0.25 or 0.4 threshold in double mutants and in ptxA).

Interestingly however, the orientation trend is unchanged for OHC2: OHC2 remain oriented largely laterally (i.e. normally) at the E17.5 mid and base in Gnai2; Gnai3 double mutants even with a raised eccentricity thresholds, whereas by contrast OHC2 in ptxA are inverted at these stage and positions. In the double mutant, OHC2 only become inverted at the P0 base (Fig. 8D). This suggests that there are similarities (OHC1) but also differences (OHC2s) between the two mouse models, and that double mutants show a delay in adopting an inverted orientation compared to ptxA. Of note, OHC2 have been shown to differentiate later than OHC1 (for example, Anniko 1983 PMID:6869851).

(2) To directly test the idea that the misorientation phenotype (inverted OHC1-2) is comparable between the two models but delayed in Gnai2; Gnai3 mutants, we performed a new experiment and added new results in the manuscript. We delayed ptxA action by using Atoh1-Cre (postmitotic hair cells) instead of FoxG1-Cre (otic progenitors). Remarkably, this produced a pattern of OHC1-2 misorientation more similar to Gnai2; Gnai3 mutants: at the E17.5 base and P0 apex, OHC2 were still largely oriented laterally (normally) in Atoh1-Cre; ptxA as in Gnai2; Gnai3 mutants whereas at the P0 base a large proportion of OHC2 were inverted (Fig. 8 Supp 1B). OHC1 were inverted at all stages and positions in the Atoh1-Cre as in the FoxG1-Cre; ptxA model. For Atoh1-Cre; ptxA, we only illustrated OHC1 and OHC2 and did not add E17.5 mid or P0 mid results because other cell types and stage/positions did not provide additional insight. In addition, we are well aware that the full FoxG1-Cre; ptxA and Gnai2; Gnai3 results for 4 cells types (IHC, OHC1-3) and 5 stages/positions is already a lot of data for cell orientation.

These results suggest that:

(a) The normal reversal of OHC1-2 to adopt a lateral orientation needs to be maintained, at least transiently, and that maintenance also relies on GNAI/O (Results starting line 529. Disussion line 621).

(b) ptxA is more severe than Gnai2; Gnai3 when it comes to OHC1-2 orientation (Figure 9, role b). Oppositely, Gnai2; Gnai3 is obviously more severe when it comes to symmetry-breaking (Fig. 9, role a) and hair bundle morphogenesis (Fig. 9, c). It follows that the two early GNAI/O activities are qualitatively different and not just based on dose. This is essentially what this Reviewer correctly pointed out, and we have fully edited both Results and Discussion accordingly. We now speculate that the difference may lie in the identity of the necessary GNAI/O protein for each role. Any GNAI/O proteins acting as a switch downstream of the GPR156 receptor may relay orientation information (Fig. 9, role b), making ptxA a particularly effective disruption strategy since it downregulates all GNAI/O proteins. In contrast, symmetry-breaking may rely more specifically on GNAI2 and GNAI3, and ptxA is not expected to achieve a loss-of-function of GNAI2 and GNAI3 as extensive as a double targeted genetic inactivation of the corresponding genes. Please see new Results starting line 526 and Discussion starting line 603. We consequently abandoned the notion that increased doses of GNAI/O is required for each role, and we also clarify that symmetry-breaking (a) and orientation (b) occur at the same time (Fig. 9).

(2) P0 may not be late enough a stage to access phenotype maturity in the double mutants. For example, it is not clear from the basal PO results whether the IHC will acquire an inverted phenotype or just misorientation in the lateral side.

For context, the OHC1-2 misorientation pattern in the ptxA model at P0 does represent the end stage, as the same pattern is observed in adults (illustrated in Fig. 2A). In addition, OHC1-2 that express ptxA are inverted as soon as they break planar symmetry, and this was established at E16.5 in a previous publication where ptxA and Gpr156 misorientation patterns were compared and shown to be identical (Kindt et al., 2021 Supp. fig. 5C-D). However, we clearly failed to mention these important results in the original manuscript. We now cite Figure 2 for adult defects (line 522), and provide a citation for OHC1-2 inversion being observed from earliest stage of hair cell differentiation (Kindt et al., 2021) (line 519).

The vast majority of Gnai2; Gnai3 double mutants die before weaning but the single specimen we managed to collect at P21 also showed inverted OHC1-2 (representative example in Fig. 2A). Again, we previously failed to point out this important result. We now do so line 214 and 555. This is another evidence that OHC1-2 misorientation is in fact similar in the ptxA and Gnai2; Gnai3 models (but milder and delayed in the latter).

When it comes to IHCs and OHC3s however, the situation is less clear. These cell types are mildly misoriented in ptxA and Gpr156 mutants, but IHCs in particular appear severely misoriented in Gnai2; Gnai3 mutants based on the position of the basal body (Fig. 8). However, very dysmorphic hair bundles can pull on the basal body via the kinocilium and affect its position, which obscures hair cell orientation inferred from the basal body and subsequent interpretations. We do not delve on IHC and OHC3 and their orientation in Gnai2; Gnai3 mutants in the revision since we do not observe similar orientation defects in a different mouse model and lack sufficient adult data.

Suggestions to improve upon the manuscript for readers:

(1) Line 294, indicate on the figure the staining in bare zone and tips of stereocilia on row 1.

Pertains to Figure 4. In A, we now point out the bare zone and stereocilia tips with arrow and arrowheads, respectively (as in other figures).

(2) Fig.8 schematic diagram, the labels of the line and 90o side by side is misleading.

We added black ticks for 0, 90, 180, 270 degree references. In contrast, the hair cell angle represented was switched to magenta.

(3) Fig. 7 legend, redundancy towards the end of the paragraph.

Thank you for catching this issue. A large portion of the legend was indeed accidentally repeated and is now deleted.

(4) Line 490-493, Another plausible explanation is that other factors besides Gnai2 and Gnai3 are involved in breaking symmetry during bundle establishment.

We now acknowledge that other proteins besides GNAI/O may be involved (Discussion line 614). That said, the notion that we do not achieve sufficient and/or early enough GNAI loss is supported for example by the Beer-Hammer 2018 study where no defects in symmetry-breaking or orientation were reported in their Gnai2 flox/flox; Gnai3 flox/flox model (Discussion new Line 637).

(5) Line 518, the base were largely inverted (Figure 8B). Should Fig 8A be cited instead of 8B?

Fig. 8B has graphs for the E17.5 cochlear base where OHC1-2 are inverted in both ptxA and Gnai2;3 DKO models. Fig. 8A has graphs of the E17.5 cochlear mid (less differentiated hair cells) where an inversion was not obvious previously, but is now clear although only partial in Gnai2; Gnai3 DKO (see above; raised eccentricity threshold). In the context of the previous text, this citation was thus correct. However, this section has been heavily modified to better compare Gnai2; Gnai3 DKO and ptxA and is hopefully less confusing in the revised version.

Reviewer #2 (Public Review):

Jarysta and colleagues set out to define how similar GNAI/O family members contribute to the shape and orientation of stereocilia bundles on auditory hair cells. Previous work demonstrated that loss of particular GNAI proteins, or inhibition of GNAIs by pertussis toxin, caused several defects in hair bundle morphogenesis, but open questions remained which the authors sought to address. Some of these questions include whether all phenotypes resulting from expression of pertussis toxin stemmed from GNAI inhibition; which GNAI family members are most critical for directing bundle development; whether GNAI proteins are needed for basal body movements that contribute to bundle patterning. These questions are important for understanding how tissue is patterned in response to planar cell polarity cues.

To address questions related to the GNAI family in auditory hair cell development, the authors assembled an impressive and nearly comprehensive collection of mouse models. This approach allowed for each Gnai and Gnao gene to be knocked out individually or in combination with each other. Notably, a new floxed allele was generated for Gnai3 because loss of this gene in combination with Gnai2 deletion was known to be embryonic lethal. Besides these lines, a new knockin mouse was made to conditionally express untagged pertussis toxin following cre induction from a strong promoter. The breadth and complexity involved in generating and collecting these strains makes this study unique, and likely the authoritative last word on which GNAI proteins are needed for which aspect of auditory hair bundle development.

Appropriate methods were employed by the authors to characterize auditory hair bundle morphology in each mouse line. Conclusions were carefully drawn from the data and largely based on excellent quantitative analysis. The main conclusions are that GNAI3 has the largest effect on hair bundle development. GNAI2 can compensate for GNAI3 loss in early development but incompletely in late development. The Gnai2 Gnai3 double mutant recapitulates nearly all the phenotypic effects associated with pertussis toxin expression and also reveals a role for GNAIs in early movement of the basal body. Although these results are not entirely unexpected based on earlier reports, the current results both uncover new functions and put putative functions on more solid ground.

Based on this study, loss of GNAI1 and GNAO show a slight shortening of the tallest row of stereocilia but no other significant changes to bundle shape. Antibody staining shows no change in GNAI localization in the Gnai1 knockout, suggesting that little to no protein is found in hair cells. One caveat to this interpretation is that the antibody, while proposed to cross-react with GNAI1, is not clearly shown to immunolabel GNAI1. More than anything, this reservation mostly serves to illustrate how challenging it is to nail down every last detail. In turn, the comprehensive nature of the current study seems all the more impressive.

(1) The original manuscript quantified stereocilia properties in Gnai1 and Gnai2 single mutants, and in Gnai1; Gnai2 double mutants using non-parametric t-tests (Mann-Whitney) for comparisons. This approach indeed suggested subtle reduction in row 1 height in IHCs in all 3 mutants. We did not quantify stereocilia features in Gnao1 mutants but could not observe defects (new Fig. 2 Supp. 1E-F). In fact, we could not observe defects in Gnai1 and Gnai2 single mutants, and in Gnai1; Gnai2 double mutants either. For this reason we have been ambivalent about reporting defects for Gnai1 and Gnai2 single and Gnai1; Gnai2 double mutants.

In the revision, we applied a nested (hierarchical) t-test to avoid pseudo-replication (Eisner 2021; PMID: 33464305; https://pubmed.ncbi.nlm.nih.gov/33464305/). In our data, the nested t-tests structure measurements by animal instead of having all stereocilia or other cell measurements treated as independent values. This more stringent approach no longer finds row 1 height reduction significant in single Gnai1 or Gnai2 mutants, or in Gnai1; Gnai2 double mutants. We modified the text accordingly in Results and Discussion. Nested t-tests were applied uniformly across the manuscript and, besides IHC measurements in Fig. 2, now also apply to bare zone surface area in Fig. 6 and eccentricity in Fig. 7. For these experiments in contrast, previous conclusions are not changed. We think that this more careful statistical treatment is a closer representation of the data in term of the conclusions we can safely make.

(2) The reviewer's criticism about antibody specificity is accurate and fair, and is fully addressed in the revised manuscript. First, we provide a phylogeny cartoon as Figure 1A to compare the GNAI/O proteins and highlight how closely related they are in sequence. To validate the assumption that our approach would detect GNAI1 if it were present in hair cells, we took a new dual experimental approach in the revision. First, we electroporated Gnai1, Gnai2 and Gnai3 expression constructs in the E13.5 inner ear and tested whether the two GNAI antibodies used in the study can detect ectopic GNAI1 in Kolliker organ. This revealed that “ptGNAI2” detects GNAI1 very well (in addition to GNAI2), but that “scbtGNAI3” does not detect GNAI1 efficiently (although it does detect GNAI3 very well). To verify in vivo that “ptGNAI2” can detect endogenous GNAI1, we immunolabeled the gallbladder epithelium in Gnai1 mutants and littermate controls using the “ptGNAI2” antibody. Based on IMPC consortium data* about the Gnai1 LacZ mouse strain, Gnai1 is specifically expressed in the adult gallbladder. We could verify that signals detected in the Gnai1 mutants were visually reduced in comparison to littermate controls. We now added this validation step in Results line 309 and the data in Fig. 4 Supp. 1A-B).

*https://www.mousephenotype.org/data/genes/MGI:95771

Reviewer #2 (Recommendations For The Authors):

Minor comments that may marginally improve clarity.

Abstract line 24: delete "nor polarized" because polarization cannot be assessed since the protein is undetectable.

This is a fair point, now deleted.

Consider revising: Lines 80-82; 188-202 (the order in which the mutants were presented was hard to follow for me); 239-240.

Lines 80-82: Used to read as "Ptx recapitulates severe stereocilia stunting and immature-looking hair bundles observed when GPSM2 or both GNAI2 and GNAI3 are inactivated."

Line 88: Was now changed to "Ptx provokes immature-looking hair bundles with severely stunted stereocilia, mimicking defects in Gpsm2 mutants and Gnai2; Gnai3 double mutants".

Lines 188-202: This was the first paragraph describing adult stereocilia defects in the different Gnai/o mouse strains. We completely rewrote the entire section to reflect the order in which the strains appear in Figure 2, hopefully making the text easier to follow because it better matches panels in Fig. 2 . We also made several other modifications to streamline comparisons and better introduce the orientation defects that are later detailed at neonate stages.

Lines 239-240: Used to read "GNAI2 makes a clear contribution since stereocilia defects increase in severity when GNAI loss extends from GNAI3 to both GNAI2 and GNAI3".

Line 247: Was now changed for "GNAI2 makes a clear contribution since Gnai3neo stereocilia defects dramatically increase in severity when GNAI2 is absent as well in Gnai2; Gnai3 double mutants."

Line 164: hardwired is unclear. Conserved?

We modified this sentence as follows: Line 171: "We reasoned that apical HC development is probably highly constrained and less likely to be influenced by genetic heterogeneity compared to susceptibility to disease, for example."

Line 299: It is not clear why GNAI1 is a better target than GNAI3. This phrase is repeated in line 303, I suspect inadvertently. Is there evidence that this antibody detects GNAI1, perhaps in another tissue? Line 308: GNAI1 may also not be detected by this antibody.

Please see point 2 above. We removed these hypothetical statements entirely and we instead now experimentally show that one of the two commercial antibodies used can readily detect GNAI1 (yet does not detect signal in hair cells when GNAI2 and GNAI3 are absent in Fig. 4F).

Author Response

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

Reviewer #2 (Public Review):

Major Weaknesses:

The assertion that MOCAT can be rapidly applied in hospital pathology departments seems overstated due to the limited availability of light-sheet microscopes outside research labs. In the first rebuttal letter, authors explain the limitations of other microscopes more readily available in hospitals. This explanation relies on your own investigations and practical experience on the matter, so including them in some part of the manuscript would be beneficial.

We appreciate the reviewer's comments and have added a discussion on the limitations of microscopes that are more readily available in hospitals in our text:

Revised manuscript, line 305-316:

“3.3 Microscopy options for imaging centimeter-sized specimens

Optical sectioning techniques are crucial for obtaining high-quality volumetric images. Techniques such as confocal microscopes, multi-photon microscopy, and light-sheet microscopy filter out-of-focus signals, resulting in sharp images of individual planes. In our study, we used light-sheet microscopy and multi-point confocal (i.e., spinning disc) for imaging centimeter-sized specimens because of their scanning speeds. While two-photon and confocal microscopy offer high-resolution imaging of smaller volumes, they are not ideal for scanning entire tissues because of their prolonged scanning times.”

Non-optical sectioning wide-field fluorescence microscopes, like the Olympus BX series or ZEISS Axio imager series, can also be used to scan samples up to about 3.5mm thick with long working distance objective lenses. In these cases, deconvolution algorithms are required to eliminate out-of-focus signals. However, it should be noted that the epifluorescence system might reduce fluorescent intensity in deeper regions within the samples.”

Refractive index matching is a critical point in the protocol, the one providing final transparency. Authors utilized the commercial solutions NFC1 and NFC2 (Nebulem, Taiwan) with a known refractive index, but for which its composition is non-disclosable. My knowledge on the organic chemistry around refractive index matching is limited, but if users don't really know what is going on in this final step, the whole protocol would rely on a single world-wide provider and troubleshooting would be fishing. I suggest that you try to validate the approach with solutions of known composition, or at least provide the solutions sold by other providers.

We appreciate the reviewer's suggestions. Based on our experience, the CUBIC-R solution developed by Ueda's team also serves as an effective RI-matching solution in the MOCAT pipeline. Its only drawback is the potential reddening of the specimen, likely due to the light-responsive component, antipyrine. We have now added this information to the Methods section:

Revised manuscript, line 492-496:

“Refractive index (RI) matching. Before imaging, the specimens were RI-matched by being immersed in NFC1 (RI = 1.47) and NFC2 (RI = 1.52) solutions (Nebulum, Taipei, Taiwan). Each immersion lasted for one day at room temperature. Alternatively, RI-matching can also be accomplished by immersing specimens in a 1:1 dilution of CUBIC-R[28] for one day, followed by pure CUBIC-R for an additional day.“

Reviewer #2 (Recommendations For The Authors):

A comment on the name of the protocol, MOCAT. I am sorry to bring this now, and not before. But, I strongly recommend another name for the procedure. My concern is that the present name "MOCAT" refers to the problem, and NOT to the actual solution provided by you. See, the problem to solve is: to perform Multiplex labeling Of Centimeter-sized Archived Tissue (MOCAT), but it says nothing about HOW you did it: heat-induced antigen retrieval and Tween20-delipidation for centimeter-scale FFPE specimens. In summary, I strongly recommend that the acronym of the procedure refers more to the "solution" than to the "problem", and for me this is important because otherwise the acronym is not fair with present and future techniques pretending to provide a novel solution to the same problem. Another way to put it is that researchers can own their proposed solutions, but they do not own the problem to be solved.

We appreciate the reviewer's suggestions. In response to their concerns, we have renamed the procedure presented in this study as Heat-Induced FFPE-based Tissue Clearing, with the acronym HIF-Clear. This change reflects the critical step in our procedure. Corresponding updates have also been made in the manuscript.

Author Response

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

Reviewer #1 (Recommendations For The Authors):

This manuscript aims to understand the biological mechanisms underlying neuropsychiatric symptoms in Parkinson's disease by characterizing subtypes of neurons in the dorsal raphe nucleus and defining their susceptibility to the degeneration of dopaminergic and adrenergic systems in the brain. This study was well-designed, the results were presented beautifully, and the manuscript was well-written. Here are some comments that may help to improve the overall quality of this work.

We thank the reviewer for the kind comments.

Major concerns:

The current study utilized an intrastriatal 6-OHDA injection, which raises the possibility that the observed electrophysiological and morphological changes of DRN5-HT and DRNDA neurons (Figs 3-6) may be due to the direct effects of 6-OHDA to DRN5-HT and DRNDA neurons projecting to the dorsal striatum (at least for DRN5-HT neurons). This possibility requires further clarification and discussion.

6-OHDA is a catecholamine neurotoxin with low selectivity for serotonin neurons. However, changes in the levels of serotonin have been observed with high doses of 6OHDA. In our study, we used lower concentrations of 6-OHDA, which did not affect the levels of serotonin (Suppl. Fig 4D), or the number of DRN5-HT neurons (Suppl. Fig. 5B). Concerning the possible effect of 6-OHDA on DRNDA neurons, we did not observe any modification in the number of these cells in response to the administration of 6-OHDA (Suppl. Fig. 5C), (lines 170-175).

How does the loss of nigrostriatal dopamine neurons affect the electrophysiology and morphology of DRNDA neurons (Figs. 5-6)? What are the potential circuit mechanisms?

The dopaminergic system in the midbrain and the DRN constitute two highly interconnected nuclei and hence there are multiple possible circuit mechanisms that could explain how loss of nigrostriatal dopaminergic neurons affects DRNDA neurons: First, DRNDA neurons are directly innervated by dopaminergic neurons in the SNc and VTA and hence loss of SNc inputs might evoke acute as well as homeostatic changes in DRNDA (Lin et al., 2020; Pinto et al., 2019). Second, midbrain dopaminergic neurons are in turn innervated by the DRN (Watabe-Uchida et al., 2012) and loss of postsynaptic dopaminergic neurons might affect all neuron types in the DRN that target the midbrain. Finally, GABAergic populations in the midbrain have been shown to target DRN5-HT neurons and might potentially also target other local cell types such as DRNDA (Li et al., 2019). Another possible pathway is the bidirectional connection between the striatum and the DRN (Pollak-Dorocic et al, 2014). DA depletion in the striatum may affect the GABAergic projection to the DRN and in turn modify the properties of postsynaptic DRN neurons.

The potential circuit mechanisms are now included in the introduction (lines 58-59).

Whether these intrastriatal 6-OHDA mice exhibited nonmotor deficits (e.g., anxiety) that may be related to the observed changes in the DRN? Such behavioral data would enhance the overall conclusions of this work.

The PD model utilized in this study displays non-motor deficits, including depression- and anxiety-like behavior (Masini et al. 2021, Ztaou et al., 2018). This is now highlighted in the manuscript (lines 167-169).

Minor issues:

The panels of Fig. 2 should be re-labelled to match the descriptions in the main text (L. 142-158).

Fig.2 now matches the descriptions in the main text.

Fig 4D was missing from the figure, which does not match the descriptions in the main text (L. 193-204:)

Fig. 4D includes the parameters describing the dendritic branching and starts with the last graph on the right in the second row of the panel.

Line 409: Extra "as" after "average"

Corrected in revised manuscript.

Fig 3G: Missed asterisks.

Corrected in revised manuscript (Fig. 3G)

Details of how action parameters were quantified should be stated and specified in the methods.

We have now added a section called ‘Quantification of electrophysiological parameters’ in the methods where we explain how the electrophysiological properties are defined and quantified (lines 407-439).

"Parkinson's disease" in the title should be revised to "parkinsonism"

Corrected in revised manuscript.

Reviewer #2 (Recommendations For The Authors):

(1) Throughout the paper, there are numerous inaccuracies and inconsistencies in the figures, which impede the clear understanding of this paper. For example, there are discrepancies between the labeling of the main figures (sub-panels) and the corresponding manuscript (Figure 2, Figure 4).

Corrected in the revised manuscript.

The statistical presentations are inaccurate in several figures (Figure 3E, 3G), making it difficult to distinguish which data is statistically meaningful. Furthermore, the number of cells presented in each figure is ambiguous in the figure legend. It would be better to avoid expressions such as 'n = 28 - 43 cells per group', as in line 456 (Figure 1I). Please provide the exact number of cells for each graph.

We agree with the reviewer, and we have now added the precise n numbers for each panel in the corresponding legends in Fig 1, Fig 3, and Fig 5. Please note that some analysis was restricted to recordings where neurons fired close to their average spontaneous firing frequency (e.g. 1Hz for DRN5-HT) to allow for a fair comparison of the data across groups and that therefore the n numbers vary in different panels.

In some figures, the value of n in the graph seems different from the value of n in the figure legends (Figure 2G-I, Figure 4, Figure 6). Collectively, these inaccurate figures and the manuscript weaken the general credibility of the data presented.

We apologize for the misunderstanding, but in the type of chosen graph, equal values are overlapped. The numbers described in the figure legend are correct.

(2) Some of the authors' claims in this paper are not supported by quantitative analysis, but only by sample recording traces or simple descriptions. For example, in line 97, the authors mentioned, "no differences when comparing TH-positive to TH-negative neurons".

But there are no data actually analyzing these two groups in Supplementary Figure 2A.

In addition, in line 103, there is a claim that "DRN DA neurons showed that they share several properties characteristics of other DA populations located in the SNc and the ventral tegmental area". However, this claim is backed up only by a few sample traces in Figure 1E.

The statement (lines 110-111), "a relative constant action potential (AP) amplitude", is also not supported by appropriate quantitative analysis but only by sample recording traces.

In our study we found a small subset of DAT-tdTomato positive neurons which did not stain positive for TH after the slice recordings. In 5 of 6 of these neurons (recorded in sham), the electrophysiological properties did not differ from other TH-positive neurons. This is visualized in Suppl. Fig 2A. The absence of any statistical difference was also confirmed by a Mann Whiteny U test comparing the TH negative to the TH positive DRNDA neurons (no significant differences in all 6 of 6 properties shown in Suppl. Fig 2A). Additionally, all these cells were DAT-positive, further supporting their classification as dopaminergic neurons. Therefore, we suspect that the lack of TH staining is likely caused by the tissue processing itself. Please note that all our immunohistochemistry was run on slices after several hours of patch-clamping procedures. Finally, including or excluding this small subset of neurons in the present study does not change any of the results presented and data was therefore pooled. We have now clarified this in more detail in the results section and in Suppl. Fig 2A (lines 100-103).

We have moved the comparison of hallmark properties found in DRNDA neurons as well as in dopaminergic neurons in the midbrain from the results section to the discussion (lines 281-283).

The claim that DRN5HT neurons have a comparatively constant action potential amplitude compared to DRNDA neurons is supported by quantitative analysis shown in Fig 1I (left panel, “AP drop rate”), while the representative example traces are shown in Fig 1G.

(3) In the legend of Figure 2, the mouse used in this experiment is mentioned with two different names (wild-type mice in line 463 and sham-lesion mice in line 465). Is this a mistake? Or did the authors intentionally use the brain samples from sham-lesion mice for Figure 2?

Figure 2 shows data in control conditions (Sham-lesion in our case), both from wild-type and Dat-Tomato. The text has been changed to avoid misunderstandings.

(4) While the primary claim of this paper is the differential alterations of DRN 5-HT and DA neurons in a mouse PD model, the observed changes in the DRN neurons of the 'DA only lesion model' are comparatively minor to the 'DA and NA lesions model'. Therefore, it looks like NA depletion has a more critical role in the DRN neurons of 6OHDA-lesion mice than DA depletion. To understand the results of this paper better, it would be great if the authors can provide additional data from the 'NA only lesion model'.

We agree with the reviewer, and we have now added a new set of experiments in which we selectively lesioned noradrenergic cells by injecting 6-OHDA unilaterally into the LC. The new data are presented in supplementary figure 6 in the revised manuscript. We find that selective lesioning of the NA system affects DRNDA and DRN5-HT neurons mildly, suggesting that the concomitant lesion of the DA and NA systems is particularly impactful (possibly because of interactions between these two systems).

(5) In Figure 3B and Figure 5B, only the 6-OHDA+DMI group shows significant differences from the sham group. This finding might be attributed to the effect of DMI itself, not to the nigrostriatal DA degeneration without NA degeneration. Thus, adding the 'DMI-only group' in all experiments will strengthen the conclusion of this paper.

The effect of one acute administration of desipramine was temporally limited to the stereotactic intervention (line 373-375), which was performed several weeks before the electrophysiological and morphological analyses. Given that the half-life of desipramine is approximately 24 hrs (Nagy and Johansson, 1975), we believe that its impact was limited to the neuroprotection of NA-neurons from 6-OHDA toxicity.

(6) DRN 5-HT neurons are known to exhibit cellular heterogeneity, and in particular their electrophysiological properties are quite heterogeneous (Bernat Kocsis. 2006; J.V. Schweimer. et al. 2011). Furthermore, 5-HT neurons in the distinct subregions of the DRN display different membrane properties (LaTasha K. Crawford, 2010). Therefore, not all DRN 5-HT neurons can be regarded as electrophysiologically identical. Given that the molecular identity of all recorded cells was confirmed with neurobiotin in this paper, it would be better to show that recorded cells are not biased toward certain subregions of DRN.

In addition, providing more comprehensive descriptions of the electrophysiological features used in PCA analysis would be beneficial in understanding the electrophysiological profiling of DRN neurons explained in this paper.

Although several studies have revealed electrophysiological and molecular heterogeneity within the DRN5-HT population, we did not observe any significant differences within the DRN5-HT neurons recorded in this study. We compared the properties of DRN5HT neurons recorded more anterior to those recorded in the posterior

DRN as well as neurons found in more ventral locations to those in more dorsal locations (data not shown). We would like to point out that the largest differences within serotonergic neuron populations described by previous studies were often found when comparing those located in the medial raphe nucleus (MRN) to those found in the DRN. Calizo et al., (2011) showed for example significant differences in the input resistance and AHP amplitude between MRN5HT and DRN5HT neurons. These two properties as well as the AP amplitude, AP threshold, AP duration, and tau did however not differ between DRN subregions in their study - and neither in ours. We extended our Suppl. Fig 1 and mapped the location of DRN5HT and DRNDA neurons recorded in sham (Suppl. Fig 1D).

Overall, we’ve sampled neurons along the anterior-posterior and dorsal-ventral axes of the DRN, while on the medial-lateral axis, recorded DRN neurons were located medially.

We agree with the reviewer that a comprehensive description of the electrophysiological features was missing in the manuscript, and we have therefore added a new section in the materials and methods where we explain in detail how each parameter was measured and analyzed (‘Quantification of electrophysiological parameters’, lines 407-439). This section also provides detailed information about the five properties underlying the PCA shown in figure 1 (i.e. delay to the first action potential, action potential drop rate, action potential rise time, duration of the afterhyperpolarization, and capacitance).

(7) Some sample images presented in this paper contain information that can conflict with the previous research. In Figures 4B and 6B, TH expression was significantly increased in the DMI pretreatment group compared to the control group. However, several studies have shown that the administration of DMI decreases TH expression levels (Komori et al.1992; Nestler et al.1990). Therefore, it would be great if the authors further explained how the pretreatment of DMI with 6-OHDA affects TH level within the DRN.

Figure 4B and 6B do not show any quantification of TH expression. The difference observed in the representative pictures is casual and due to the variable expression of TH across the slice. Moreover, as mentioned in the response to point 5, mice were subjected to a single injection of DMI immediately preceding the stereotactic intervention (line 373375). In contrast, the increase in TH expression reported by Komori et al. 1992 and Nestler et al. 1990 was observed in response to chronic (two weeks) administration of DMI.

(8) This paper lacks direct evidence to demonstrate whether DMI pretreatment could effectively protect against NA depletion. Therefore, in addition to TH expression levels, it is important to provide data to confirm the intact NA levels (or NA axons) after DMI treatment.

NA levels in the striatum were measured by Enzyme-linked immunosorbent assay and reported in Suppl.Fig.4 in the revised manuscript.

(9) It would be great if the authors specifically explained why 6-OHDA was injected into the striatum (neither MFB nor SNc) to make a mouse model of PD.

Mice were injected in the dorsal striatum to produce a partial bilateral lesion of the dopamine and noradrenaline systems. This model reproduces the initial stages of PD and also recapitulates several non-motor symptoms of PD, including affective disorders, which may be related to changes in serotonergic and dopaminergic transmission in the dorsal raphe. In contrast, injections in the MFB and SNc quickly produce a severe motor phenotype closer to a late stage of the disease and cannot be done bilaterally. <br /> The striatal model has been successfully used in other publications (Kravitz et al., 2010, Masini et al., 2021, Ztaou et al., 2018, Chen et al., 2014, Branchi et al., 2008, Marques et al. 2019, Tadaiesky et al., 2008, Matheus et al., 2016, Silva et al., 2016).

(10) Supplementary Figures 2 and 3 were erroneously cut on the right side. These figure images should be replaced with the correct ones.

We thank the reviewer for noticing and we have now replaced the figures with the correct ones.

(11) There should be more explanations about tdTomato-positive but non-TH neurons in Supplementary Figure 2. It is strange to regard TH-negative neurons as DA neurons although these neurons have DA neuron-like electrophysiological properties. If these tdTomato-positive but non-TH neurons cannot release DA, can we say these are DA neurons?

In our study we found a small subset of DAT-tdTomato positive neurons which did not stain positive for TH afterwards. In 5 of 6 of these neurons (recorded in sham), the electrophysiological properties did not differ from other TH-positive neurons. This is visualized in Suppl. Fig 2A. The absence of any statistical difference was also confirmed by a Mann Whiteny U test comparing the TH-negative to the TH-positive DRNDA neurons (no significant differences in all 6 of 6 properties shown in SF2A). Additionally, all these cells were DAT-positive, further supporting their classification as dopaminergic neurons. Therefore, we suspect that the lack of TH staining is likely caused by the tissue processing itself. Please note that all our immunohistochemistry was run on slices after several hours of patch-clamping procedures. Finally, including or excluding this small subset of neurons in the present study does not change any of the results presented and data was therefore pooled. We have now clarified this in more detail in the results section and in Suppl. Fig 2A (lines 100-103).

Reviewer #3 (Recommendations For The Authors):

The authors report using a parametric statistical test, the t-test. The t-test makes the assumption that the data are normally distributed. Most biological data is not distributed normally, and with smaller datasets, it is difficult to say whether the underlying distribution would be normally distributed. I would recommend using the non-parametric versions of the same test (eg Mann-Whitney U test), which is likely to give a similar result while being more conservative given the potential for non-normal distribution.

All electrophysiological data were first tested for normality before running the corresponding statistical test (either t-test for normal distributed data or Mann-Whitney U test for non-normally distributed data). The morphological data are now analyzed by the Mann-Whitney U test (lines 484-494).

The authors state that mice were treated with 6-OHDA at 3 months, then brain slices were prepared 3 weeks later, making them about 4 months old. I could not find the age of sham/control mice and 6-OHDA/desipramine mice in the methods section. Were sham/controls and 6-OHDA slices prepared in an interleaved fashion?

Sham and 6-OHDA+DMI mice underwent surgery at 3 months and the brain slices were prepared 3 weeks later, as the 6-OHDA mice. We have now clarified this in the methods (line 381).

While desipramine is relatively selective as a norepinephrine reuptake inhibitor, it also can prevent serotonin reuptake. Could this mechanism also protect DRN neurons from the effects of 6-OHDA?

Even if desipramine has some affinity for the serotonin reuptake, this affinity is 100-fold less than the one described for the noradrenaline reuptake (Richelson and Pfenning, 1984, Gillman, 2007). Moreover, in our study the 6-OHDA injection in the dorsal striatum did not cause any direct damage to the DRN5-HT, as shown by the 5-HT measurement and DRN5-HT counting (Suppl. Fig. 4D, Suppl. Fig. 5A,B), so we can exclude that the effects observed in the DMI+6-OHDA group are related to a protection of the serotonergic system exerted by a single injection of desipramine.

On line 168, the authors use the abbreviation NA for noradrenergic. Was this abbreviation previously defined in the manuscript?

Yes, the abbreviation is defined in the introduction (line 73).

On line 45, the authors cite that the DRN-5HT subpopulation accounts for 30-50% of the DRN neurons. It would be helpful to know approximately what percentage of the DRN neurons belong to the DRNDA subpopulation as well.

To the best of our knowledge, there is unfortunately no detailed analysis of the prevalence of DRNDA neurons in mice available. Previous studies in rats have estimated that this population comprises around 1000 neurons (Descarries et al., 1986). According to Calizo et al. (2011), the number of any non-serotonergic neuron population (releasing dopamine or other neurotransmitters) in the DRN is one third to one tenth less than the number of DRN5-HT neurons. But please note that this study was also performed in rats (line 55).

While I appreciate that the authors did not over-interpret their findings, it would be useful to comment (in the Discussion) on how their findings could/should be used in interpreting other studies using 6-OHDA, as well as the relationship of their findings to loss of 5-HT and/or DRN neurons in Parkinson's Disease itself.

In the manuscript, we refer to the utility of the 6-OHDA model for the study of a wide range of non-motor symptoms. We have now described, in this model, how the loss of midbrain dopaminergic and noradrenergic neurons affects the electrophysiological and morphological properties of DRN5-HT and DRNDA neurons. This information will allow for a more precise assessment of the mechanisms involved in the affective and cognitive aspects of PD symptomatology (lines 354-356).

Author Response

We are writing this response letter with regards to the insightful feedback you provided on our manuscript titled: "A metabolic modeling-based framework for predicting trophic dependencies in native rhizobiomes of crop plants" submitted for consideration in eLife.

We sincerely appreciate the thorough and constructive reviews, seeing and fitting the intentions behind our work. We intend to fully address all points raised by the reviewers in our revised manuscript. Specifically, we plan to incorporate targeted revisions to address concerns raised during the review process, with focus on process benchmarking and validation of our framework to enhance its reliability and accuracy.

We believe that the current revision would improve the consistency and quality of the framework, making it a suitable tool for the characterization of microbial trophic interactions in diverse biological landscapes.

Thank you once again for both your time and dedication in reviewing our manuscript, as well as the constructive review.

Author Response

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

Recommendations for the authors:

(1) Substantial revision of the claims and interpretation of the results is needed, especially in the setting of additional data showing enhanced erythrophagocytosis with decreased RBC lifespan.

Thank you for your valuable feedback and suggestion for a substantial revision of the claims and interpretation of our results. We acknowledge the importance of considering additional data that shows enhanced erythrophagocytosis with decreased RBC lifespan. In response, we have revised our manuscript and incorporated additional experimental data to support and clarify our findings.

(1) In our original manuscript, we reported a decrease in the number of splenic red pulp macrophages (RPMs) and phagocytic erythrocytes after hypobaric hypoxia (HH) exposure. This conclusion was primarily based on our observations of reduced phagocytosis in the spleen.

(2) Additional experimental data on RBC labeling and erythrophagocytosis:

• Experiment 1 (RBC labeling and HH exposure)

We conducted an experiment where RBCs from mice were labeled with PKH67 and injected back into the mice. These mice were then exposed to normal normoxia (NN) or HH for 7 or 14 days. The subsequent assessment of RPMs in the spleen using flow cytometry and immunofluorescence detection revealed a significant decrease in both the population of splenic RPMs (F4/80hiCD11blo, new Figure 5A and C) and PKH67-positive macrophages after HH exposure (as depicted in new Figure 5A and C-E). This finding supports our original claim of reduced phagocytosis under HH conditions.

Author response image 1.

-Experiment 2 (erythrophagocytosis enhancement)

To examine the effects of enhanced erythrophagocytosis, we injected Tuftsin after administering PKH67-labelled RBCs. Our observations showed a significant decrease in PKH67 fluorescence in the spleen, particularly after Tuftsin injection compared to the NN group. This result suggests a reduction in RBC lifespan when erythrophagocytosis is enhanced (illustrated in new Figure 7, A-B).

Author response image 2.

(3) Revised conclusions:

• The additional data from these experiments support our original findings by providing a more comprehensive view of the impact of HH exposure on splenic erythrophagocytosis.

• The decrease in phagocytic RPMs and phagocytic erythrocytes after HH exposure, along with the observed decrease in RBC lifespan following enhanced erythrophagocytosis, collectively suggest a more complex interplay between hypoxia, erythrophagocytosis, and RBC lifespan than initially interpreted.

We think that these revisions and additional experimental data provide a more robust and detailed understanding of the effects of HH on splenic erythrophagocytosis and RBCs lifespan. We hope that these changes adequately address the concerns raised and strengthen the conclusions drawn in our manuscript.

(2) F4/80 high; CD11b low are true RPMs which the cells which the authors are presenting, i.e. splenic monocytes / pre-RPMs. To discuss RPM function requires the presentation of these cells specifically rather than general cells in the proper area of the spleen.

Thank you for your feedback requesting a substantial revision of our claims and interpretation, particularly considering additional data showing enhanced erythrophagocytosis with decreased RBC lifespan. In response, we have thoroughly revised our manuscript and included new experimental data that further elucidate the effects of HH on RPMs and erythrophagocytosis.

(1) Re-evaluation of RPMs population after HH exposure:

• Flow cytometry analysis (new Figure 3G, Figure 5A and B): We revisited the analysis of RPMs (F4/80hiCD11blo) in the spleen after 7 and 14 days of HH exposure. Our revised flow cytometry data consistently showed a significant decrease in the RPMs population post-HH exposure, reinforcing our initial findings.

Author response image 3.

Author response image 4.

• In situ expression of RPMs (Figure S1, A-D):

We further confirmed the decreased population of RPMs through in situ co-staining with F4/80 and CD11b, and F4/80 and CD68, in spleen tissues. These results clearly demonstrated a significant reduction in F4/80hiCD11blo (Figure S1, A and B) and F4/80hiCD68hi (Figure S1, C and D) cells following HH exposure.

Author response image 5.

(2) Single-cell sequencing analysis of splenic RPMs:

• We conducted a single-cell sequencing analysis of spleen samples post 7 days of HH exposure (Figure S2, A-C). This analysis revealed a notable shift in the distribution of RPMs, predominantly associated with Cluster 0 under NN conditions, to a reduced presence in this cluster after HH exposure.

• Pseudo-time series analysis indicated a transition pattern change in spleen RPMs, with a shift from Cluster 2 and Cluster 1 towards Cluster 0 under NN conditions, and a reverse transition following HH exposure (Figure S2, B and D). This finding implies a decrease in resident RPMs in the spleen under HH conditions.

(3) Consolidated findings and revised interpretation:

• The comprehensive analysis of flow cytometry, in situ staining, and single-cell sequencing data consistently indicates a significant reduction in the number of RPMs following HH exposure.

• These findings, taken together, strongly support the revised conclusion that HH exposure leads to a decrease in RPMs in the spleen, which in turn may affect erythrophagocytosis and RBC lifespan.

Author response image 6.

In conclusion, our revised manuscript now includes additional experimental data and analyses, strengthening our claims and providing a more nuanced interpretation of the impact of HH on spleen RPMs and related erythrophagocytosis processes. We believe these revisions and additional data address your concerns and enhance the scientific validity of our study.

(3) RBC retention in the spleen should be measured anyway quantitatively, eg, with proper flow cytometry, to determine whether it is increased or decreased.

Thank you for your query regarding the quantitative measurement of RBC retention in the spleen, particularly in relation to HH exposure. We have utilized a combination of techniques, including flow cytometry and histological staining, to investigate this aspect comprehensively. Below is a summary of our findings and methodology.

(1) Flow cytometry analysis of labeled RBCs:

• Our study employed both NHS-biotin (new Figure 4, A-D) and PKH67 labeling (new Figure 4, E-H) to track RBCs in mice exposed to HH. Flow cytometry results from these experiments (new Figure 4, A-H) showed a decrease in the proportion of labeled RBCs over time, both in the blood and spleen. Notably, there was a significantly greater reduction in the amplitude of fluorescently labeled RBCs after NN exposure compared to the reduced amplitude of fluorescently labeled RBCs observed in blood and spleen under HH exposure. The observed decrease in labeled RBCs was initially counterintuitive, as we expected an increase in RBC retention due to reduced erythrophagocytosis. However, this decrease can be attributed to the significantly increased production of RBCs following HH exposure, diluting the proportion of labeled cells.

• Specifically, for blood, the biotin-labeled RBCs decreased by 12.06% under NN exposure and by 7.82% under HH exposure, while the PKH67-labeled RBCs decreased by 9.70% under NN exposure and by 4.09% under HH exposure. For spleen, the biotin-labeled RBCs decreased by 3.13% under NN exposure and by 0.46% under HH exposure, while the PKH67-labeled RBCs decreased by 1.16% under NN exposure and by 0.92% under HH exposure. These findings suggest that HH exposure leads to a decrease in the clearance rate of RBCs.

Author response image 7.

(2) Detection of erythrophagocytosis in spleen:

To assess erythrophagocytosis directly, we labeled RBCs with PKH67 and analyzed their uptake by splenic macrophages (F4/80hi) after HH exposure. Our findings (new Figure 5, D-E) indicated a decrease in PKH67-positive macrophages in the spleen, suggesting reduced erythrophagocytosis.

Author response image 8.

(3) Flow cytometry analysis of RBC retention:

Our flow cytometry analysis revealed a decrease in PKH67-positive RBCs in both blood and spleen (Figure S4). We postulated that this was due to increased RBC production after HH exposure. However, this method might not accurately reflect RBC retention, as it measures the proportion of PKH67-labeled RBCs relative to the total number of RBCs, which increased after HH exposure.

Author response image 9.

(4) Histological and immunostaining analysis:

Histological examination using HE staining and band3 immunostaining in situ (new Figure 6, A-D, and G-H) revealed a significant increase in RBC numbers in the spleen after HH exposure. This was further confirmed by detecting retained RBCs in splenic single cells using Wright-Giemsa composite stain (new Figure 6, E and F) and retained PKH67-labelled RBCs in spleen (new Figure 6, I and J).

Author response image 10.

(5) Interpreting the data:

The comprehensive analysis suggests a complex interplay between increased RBC production and decreased erythrophagocytosis in the spleen following HH exposure. While flow cytometry indicated a decrease in the proportion of labeled RBCs, histological and immunostaining analyses demonstrated an actual increase in RBCs retention in the spleen. These findings collectively suggest that while the overall RBCs production is upregulated following HH exposure, the spleen's capacity for erythrophagocytosis is concurrently diminished, leading to increased RBCs retention.

(6) Conclusion:

Taken together, our results indicate a significant increase in RBCs retention in the spleen post-HH exposure, likely due to reduced residual RPMs and erythrophagocytosis. This conclusion is supported by a combination of flow cytometry, histological staining, and immunostaining techniques, providing a comprehensive view of RBC dynamics under HH conditions. We think these findings offer a clear quantitative measure of RBC retention in the spleen, addressing the concerns raised in your question.

(4) Numerous other methodological problems as listed below.

We appreciate your question, which highlights the importance of using multiple analytical approaches to understand complex physiological processes. Please find below our point-by-point response to the methodological comments.

Reviewer #1 (Recommendations For The Authors):

(1) Decreased BM and spleen monocytes d/t increased liver monocyte migration is unclear. there is no evidence that this happens or why it would be a reasonable hypothesis, even in splenectomized mice.

Thank you for highlighting the need for further clarification and justification of our hypothesized decrease in BM and spleen monocytes due to increased monocyte migration to the liver, particularly in the context of splenectomized mice. Indeed, our study has not explicitly verified an augmentation in mononuclear cell migration to the liver in splenectomized mice.

Nonetheless, our investigations have revealed a notable increase in monocyte migration to the liver after HH exposure. Noteworthy is our discovery of a significant upregulation in colony stimulating factor-1 (CSF-1) expression in the liver, observed after both 7 and 14 days of HH exposure (data not included). This observation was substantiated through flow cytometry analysis (as depicted in Figure S4), which affirmed an enhanced migration of monocytes to the liver. Specifically, we noted a considerable increase in the population of transient macrophages, monocytes, and Kupffer cells in the liver following HH exposure.

Author response image 11.

Considering these findings, we hypothesize that hypoxic conditions may activate a compensatory mechanism that directs monocytes towards the liver, potentially linked to the liver’s integral role in the systemic immune response. In accordance with these insights, we intend to revise our manuscript to reflect the speculative nature of this hypothesis more accurately, and to delineate the strategies we propose for its further empirical investigation. This amendment ensures that our hypothesis is presented with full consideration of its speculative basis, supported by a coherent framework for future validation.

(2) While F4/80+CD11b+ population is decreased, this is mainly driven by CD11b and F4/80+ alone population is significantly increased. This is counter to the hypothesis.

Thank you for addressing the apparent discrepancy in our findings concerning the F4/80+CD11b+ population and the increase in the F4/80+ alone population, which seems to contradict our initial hypothesis. Your observation is indeed crucial for the integrity of our study, and we appreciate the opportunity to clarify this matter.

(1) Clarification of flow cytometry results:

• In response to the concerns raised, we revisited our flow cytometry experiments with a focus on more clearly distinguishing the cell populations. Our initial graph had some ambiguities in cell grouping, which might have led to misinterpretations.

• The revised flow cytometry analysis, specifically aimed at identifying red pulp macrophages (RPMs) characterized as F4/80hiCD11blo in the spleen, demonstrated a significant decrease in the F4/80 population. This finding is now in alignment with our immunofluorescence results.

Author response image 12.

Author response image 13.

(2) Revised data and interpretation:

• The results presented in new Figure 3G and Figure 5 (A and B) consistently indicate a notable reduction in the RPMs population following HH exposure. This supports our revised understanding that HH exposure leads to a decrease in the specific macrophage subset (F4/80hiCD11blo) in the spleen.

We’ve updated our manuscript to reflect these new findings and interpretations. The revised manuscript details the revised flow cytometry analysis and discusses the potential mechanisms behind the observed changes in macrophage populations.

(3) HO-1 expression cannot be used as a surrogate to quantify number of macrophages as the expression per cell can decrease and give the same results. In addition, the localization of effect to the red pulp is not equivalent to an assertion that the conclusion applies to macrophages given the heterogeneity of this part of the organ and the spleen in general.

Thank you for your insightful comments regarding the use of HO-1 expression as a surrogate marker for quantifying macrophage numbers, and for pointing out the complexity of attributing changes in HO-1 expression specifically to macrophages in the splenic red pulp. Your observations are indeed valid and warrant a detailed response.

(1) Role of HO-1 in macrophage activity:

• In our study, HO-1 expression was not utilized as a direct marker for quantifying macrophages. Instead, it was considered an indicator of macrophage activity, particularly in relation to erythrophagocytosis. HO-1, being upregulated in response to erythrophagocytosis, serves as an indirect marker of this process within splenic macrophages.

• The rationale behind this approach was that increased HO-1 expression, induced by erythrophagocytosis in the spleen’s red pulp, could suggest an augmentation in the activity of splenic macrophages involved in this process.

(2) Limitations of using HO-1 as an indicator:

• We acknowledge your point that HO-1 expression per cell might decrease, potentially leading to misleading interpretations if used as a direct quantifier of macrophage numbers. The variability in HO-1 expression per cell indeed presents a limitation in using it as a sole indicator of macrophage quantity.

• Furthermore, your observation about the heterogeneity of the spleen, particularly the red pulp, is crucial. The red pulp is a complex environment with various cell types, and asserting that changes in HO-1 expression are exclusive to macrophages could oversimplify this complexity.

(3) Addressing the concerns:

• To address these concerns, we propose to supplement our HO-1 expression data with additional specific markers for macrophages. This would help in correlating HO-1 expression more accurately with macrophage numbers and activity.

• We also plan to conduct further studies to delineate the specific cell types in the red pulp contributing to HO-1 expression. This could involve techniques such as immunofluorescence or immunohistochemistry, which would allow us to localize HO-1 expression to specific cell populations within the splenic red pulp.

We’ve revised our manuscript to clarify the role of HO-1 expression as an indirect marker of erythrophagocytosis and to acknowledge its limitations as a surrogate for quantifying macrophage numbers.

(4) line 63-65 is inaccurate as red cell homeostasis reaches a new steady state in chronic hypoxia.

Thank you for pointing out the inaccuracy in lines 63-65 of our manuscript regarding red cell homeostasis in chronic hypoxia. Your feedback is invaluable in ensuring the accuracy and scientific integrity of our work. We’ve revised lines 63-65 to accurately reflect the understanding.

(5) Eryptosis is not defined in the manuscript.

Thank you for highlighting the omission of a definition for eryptosis in our manuscript. We acknowledge the significance of precisely defining such key terminologies, particularly when they play a crucial role in the context of our research findings. Eryptosis, a term referenced in our study, is a specialized form of programmed cell death unique to erythrocytes. Similar with apoptosis in other cell types, eryptosis is characterized by distinct physiological changes including cell shrinkage, membrane blebbing, and the externalization of phosphatidylserine on the erythrocyte surface. These features are indicative of the RBCs lifecycle and its regulated destruction process.

However, it is pertinent to note that our current study does not extensively delve into the mechanisms or implications of eryptosis. Our primary focus has been to elucidate the effects of HH exposure on the processes of splenic erythrophagocytosis and the resultant impact on the lifespan of RBCs. Given this focus, and to maintain the coherence and relevance of our manuscript, we have decided to exclude specific discussions of eryptosis from our revised manuscript. This decision aligns with our aim to provide a clear and concentrated exploration of the influence of HH exposure on RBCs dynamics and splenic function.

We appreciate your input, which has significantly contributed to enhancing the clarity and accuracy of our manuscript. The revision ensures that our research is presented with a focused scope, aligning closely with our experimental investigations and findings.

(6) Physiologically, there is no evidence that there is any "free iron" in cells, making line 89 point inaccurate.

Thank you for highlighting the concern regarding the reference to "free iron" in cells in line 89 of our manuscript. The term "free iron" in our manuscript was intended to refer to divalent iron (Fe2+), rather than unbound iron ions freely circulating within cells. We acknowledge that the term "free iron" might lead to misconceptions, as it implies the presence of unchelated iron, which is not physiologically common due to the potential for oxidative damage. To rectify this and provide clarity, we’ve revised line 89 of our manuscript to reflect our meaning more accurately. Instead of "free iron," we use "divalent iron (Fe2+)" to avoid any misunderstanding regarding the state of iron in cells. We also ensure that any implications drawn from the presence of Fe2+ in cells are consistent with current scientific literature and understanding.

(7) Fig 1f no stats

We appreciate your critical review and suggestions, which help in improving the accuracy and clarity of our research. We’ve revised statistic diagram of new Figure 1F.

(8) Splenectomy experiments demonstrate that erythrophagocytosis is almost completely replaced by functional macrophages in other tissues (likely Kupffer cells in the liver). there is only a minor defect and no data on whether it is in fact the liver or other organs that provide this replacement function and makes the assertions in lines 345-349 significantly overstated.

Thank you for your critical assessment of our interpretation of the splenectomy experiments, especially concerning the role of erythrophagocytosis by macrophages in other tissues, such as Kupffer cells in the liver. We appreciate your observation that our assertions may be overstated and acknowledge the need for more specific data to identify which organs compensate for the loss of splenic erythrophagocytosis.

(1) Splenectomy experiment findings:

• Our findings in Figure 2D do indicate that in the splenectomized group under NN conditions, erythrophagocytosis is substantially compensated for by functional macrophages in other tissues. This is an important observation that highlights the body's ability to adapt to the loss of splenic function.

• However, under HH conditions, our data suggest that the spleen plays an important role in managing erythrocyte turnover, as indicated by the significant impact of splenectomy on erythrophagocytosis and subsequent erythrocyte dynamics.

(2) Addressing the lack of specific organ identification:

• We acknowledge that our study does not definitively identify which organs, such as the liver or others, take over the erythrophagocytosis function post-splenectomy. This is an important aspect that needs further investigation.

• To address this, we also plan to perform additional experiments that could more accurately point out the specific tissues compensating for the loss of splenic erythrophagocytosis. This could involve tracking labeled erythrocytes or using specific markers to identify macrophages actively engaged in erythrophagocytosis in various organs.

(3) Revising manuscript statements:

Considering your feedback, we’ve revised the statements in lines 345-349 (lines 378-383 in revised manuscript) to enhance the scientific rigor and clarity of our research presentation.

(9) M1 vs M2 macrophage experiments are irrelevant to the main thrust of the manuscript, there are no references to support the use of only CD16 and CD86 for these purposes, and no stats are provided. It is also unclear why bone marrow monocyte data is presented and how it is relevant to the rest of the manuscript.

Thank you for your critical evaluation of the relevance and presentation of the M1 vs. M2 macrophage experiments in our manuscript. We appreciate your insights, especially regarding the use of specific markers and the lack of statistical analysis, as well as the relevance of bone marrow monocyte data to our study's main focus.

(1) Removal of M1 and M2 macrophage data:

Based on your feedback and our reassessment, we agree that the results pertaining to M1 and M2 macrophages did not align well with the main objectives of our manuscript. Consequently, we have decided to remove the related content on M1 and M2 macrophages from the revised manuscript. This decision was made to ensure that our manuscript remains focused and coherent, highlighting our primary findings without the distraction of unrelated or insufficiently supported data.

The use of only CD16 and CD86 markers for M1 and M2 macrophage characterization, without appropriate statistical analysis, was indeed a methodological limitation. We recognize that a more comprehensive set of markers and rigorous statistical analysis would be necessary for a meaningful interpretation of M1/M2 macrophage polarization. Furthermore, the relevance of these experiments to the central theme of our manuscript was not adequately established. Our study primarily focuses on erythrophagocytosis and red pulp macrophage dynamics under hypobaric hypoxia, and the M1/M2 polarization aspect did not contribute significantly to this narrative.

(2) Clarification on bone marrow monocyte data:

Regarding the inclusion of bone marrow monocyte data, we acknowledge that its relevance to the main thrust of the manuscript was not clearly articulated. In the revised manuscript, we provide a clearer rationale for its inclusion and how it relates to our primary objectives.

(3) Commitment to clarity and relevance:

We are committed to ensuring that every component of our manuscript contributes meaningfully to our overall objectives and research questions. Your feedback has been instrumental in guiding us to streamline our focus and present our findings more effectively.

We appreciate your valuable feedback, which has led to a more focused and relevant presentation of our research. These changes enhance the clarity and impact of our manuscript, ensuring that it accurately reflects our key research findings.

(10) Biotinolated RBC clearance is enhanced, demonstrating that RBC erythrophagocytosis is in fact ENHANCED, not diminished, calling into question the founding hypothesis that the manuscript proposes.

Thank you for your critical evaluation of our data on biotinylated RBC clearance, which suggests enhanced erythrophagocytosis under HH conditions. This observation indeed challenges our founding hypothesis that erythrophagocytosis is diminished in this setting. Below is a summary of our findings and methodology.

(1) Interpretation of RBC labeling results:

Both the previous results of NHS-biotin labeled RBCs (new Figure 4, A-D) and the current results of PKH67-labeled RBCs (new Figure 4, E-H) demonstrated a decrease in the number of labeled RBCs with an increase in injection time. The production of RBCs, including bone marrow and spleen production, was significantly increased following HH exposure, resulting in a consistent decrease in the proportion of labeled RBCs via flow cytometry detection both in the blood and spleen of mice compared to the NN group. However, compared to the reduced amplitude of fluorescently labeled RBCs observed in blood and spleen under NN exposure, there was a significantly weaker reduction in the amplitude of fluorescently labeled RBCs after HH exposure. Specifically, for blood, the biotin-labeled RBCs decreased by 12.06% under NN exposure and by 7.82% under HH exposure, while the PKH67-labeled RBCs decreased by 9.70% under NN exposure and by 4.09% under HH exposure. For spleen, the biotin-labeled RBCs decreased by 3.13% under NN exposure and by 0.46% under HH exposure, while the PKH67-labeled RBCs decreased by 1.16% under NN exposure and by 0.92% under HH exposure.

Author response image 14.

(2) Increased RBCs production under HH conditions:

It's important to note that RBCs production, including from bone marrow and spleen, was significantly increased following HH exposure. This increase in RBCs production could contribute to the decreased proportion of labeled RBCs observed in flow cytometry analyses, as there are more unlabeled RBCs diluting the proportion of labeled cells in the blood and spleen.

(3) Analysis of erythrophagocytosis in RPMs:

Our analysis of PKH67-labeled RBCs content within RPMs following HH exposure showed a significant reduction in the number of PKH67-positive RPMs in the spleen (new Figure 5). This finding suggests a decrease in erythrophagocytosis by RPMs under HH conditions.

Author response image 15.

(4) Reconciling the findings:

The apparent contradiction between enhanced RBC clearance (suggested by the reduced proportion of labeled RBCs) and reduced erythrophagocytosis in RPMs (indicated by fewer PKH67-positive RPMs) may be explained by the increased overall production of RBCs under HH. This increased production could mask the actual erythrophagocytosis activity in terms of the proportion of labeled cells. Therefore, while the proportion of labeled RBCs decreases more significantly under HH conditions, this does not necessarily indicate an enhanced erythrophagocytosis rate, but rather an increased dilution effect due to higher RBCs turnover.

(5) Revised interpretation and manuscript changes:

Given these factors, we update our manuscript to reflect this detailed interpretation and clarify the implications of the increased RBCs production under HH conditions on our observations of labeled RBCs clearance and erythrophagocytosis. We appreciate your insightful feedback, which has prompted a careful re-examination of our data and interpretations. We hope that these revisions provide a more accurate and comprehensive understanding of the effects of HH on erythrophagocytosis and RBCs dynamics.

(11) Legend in Fig 4c-4d looks incorrect and Fig 4e-4f is very non-specific since Wright stain does not provide evidence of what type of cells these are and making for a significant overstatement in the contribution of this data to "confirming" increased erythrophagocytosis in the spleen under HH exposure (line 395-396).

Thank you for your insightful observations regarding the data presentation and figure legends in our manuscript, particularly in relation to Figure 4 (renamed as Figure 6 in the revised manuscript) and the use of Wright-Giemsa composite staining. We appreciate your constructive feedback and acknowledge the importance of presenting our data with utmost clarity and precision.

(1) Amendments to Figure legends:

We recognize the necessity of rectifying inaccuracies in the legends of the previously labeled Figure 4C and D. Corrections have been meticulously implemented to ensure the legends accurately contain the data presented. Additionally, we acknowledge the error concerning the description of Wright staining. The method employed in our study is Wright-Giemsa composite staining, which, unlike Wright staining that solely stains cytoplasm (RBC), is capable of staining both nuclei and cytoplasm.

(2) Addressing the specificity of Wright-Giemsa Composite staining:

Our approach involved quantifying RBC retention using Wright-Giemsa composite staining on single splenic cells post-perfusion at 7 and 14 days post HH exposure. We understand and appreciate your concerns regarding the nonspecific nature of Wright staining. Although Wright stain is a general hematologic stain and not explicitly specific for certain cell types, its application in our study aimed to provide preliminary insights. The spleen cells, devoid of nuclei and thus likely to be RBCs, were stained and observed post-perfusion, indicating RBC retention within the spleen.

(3) Incorporating additional methods for RBC identification:

To enhance the specificity of our findings, we integrated supplementary methods for RBC identification in the revised manuscript. We employed band3 immunostaining (in the new Figure 6, C-D and G-H) and PKH67 labeling (Figure 6, I-J) for a more targeted identification of RBCs. Band3, serving as a reliable marker for RBCs, augments the specificity of our immunostaining approach. Likewise, PKH67 labeling affords a direct and definitive means to assess RBC retention in the spleen following HH exposure.

Author response image 16. same as 10

(4) Revised interpretation and manuscript modifications:

Based on these enhanced methodologies, we have refined our interpretation of the data and accordingly updated the manuscript. The revised narrative underscores that our conclusions regarding reduced erythrophagocytosis and RBC retention under HH conditions are corroborated by not only Wright-Giemsa composite staining but also by band3 immunostaining and PKH67 labeling, each contributing distinctively to our comprehensive understanding.

We are committed to ensuring that our manuscript precisely reflects the contribution of each method to our findings and conclusions. Your thorough review has been invaluable in identifying and rectifying areas for improvement in our research report and interpretation.

(12) Ferroptosis data in Fig 5 is not specific to macrophages and Fer-1 data confirms the expected effect of Fer-1 but there is no data that supports that Fer-1 reverses the destruction of these cells or restores their function in hypoxia. Finally, these experiments were performed in peritoneal macrophages which are functionally distinct from splenic RPM.

Thank you for your critique of our presentation and interpretation of the ferroptosis data in Figure 5 (renamed as Figure 9 in the revised manuscript), as well as your observations regarding the specificity of the experiments to macrophages and the effects of Fer-1. We value your input and acknowledge the need to clarify these aspects in our manuscript.

(1) Clarification on cell type used in experiments:

• We appreciate your attention to the details of our experimental setup. The experiments presented in Figure 9 were indeed conducted on splenic macrophages, not peritoneal macrophages, as incorrectly mentioned in the original figure legend. This was an error in our manuscript, and we have revised the figure legend accordingly to accurately reflect the cell type used.

(2) Specificity of ferroptosis data:

• We recognize that the data presented in Figure 9 need to be more explicitly linked to the specific macrophage population being studied. In the revised manuscript, we ensure that the discussion around ferroptosis data is clearly situated within the framework of splenic macrophages.

• We also provide additional methodological details in the 'Methods' section to reinforce the specificity of our experiments to splenic macrophages.

(3) Effects of Fer-1 on macrophage function and survival:

• Regarding the effect of Fer-1, we agree that while our data confirms the expected effect of Fer-1 in inhibiting ferroptosis, we have not provided direct evidence that Fer-1 reverses the destruction of macrophages or restores their function in hypoxia.

• To address this, we propose additional experiments to specifically investigate the impact of Fer-1 on the survival and functional restoration of splenic macrophages under hypoxic conditions. This would involve assessing not only the inhibition of ferroptosis but also the recovery of macrophage functionality post-treatment.

(4) Revised interpretation and manuscript changes:

• We’ve revised the relevant sections of our manuscript to reflect these clarifications and proposed additional studies. This includes modifying the discussion of the ferroptosis data to more accurately represent the cell types involved and the limitations of our current findings regarding the effects of Fer-1.

• The revised manuscript presents a more detailed interpretation of the ferroptosis data, clearly describing what our current experiments demonstrate and what remains to be investigated.

We are grateful for your insightful feedback, which has highlighted important areas for improvement in our research presentation. We think that these revisions will enhance the clarity and scientific accuracy of our manuscript, ensuring that our findings and conclusions are well-supported and precisely communicated.

Reviewer #2 (Recommendations For The Authors):

The following questions and remarks should be considered by the authors:

(1) The methods should clearly state whether the HH was discontinued during the 7 or 14 day exposure for cleaning, fresh water etc. Moreover, how was CO2 controlled? The procedure for splenectomy needs to be described in the methods.

Thank you for your inquiry regarding the specifics of our experimental methods, particularly the management of HH exposure and the procedure for splenectomy. We appreciate your attention to detail and the importance of these aspects for the reproducibility and clarity of our research.

(1) HH exposure conditions:

In our experiments, mice were continuously exposed to HH for the entire duration of 7 or 14 days, without interruption for activities such as cleaning or providing fresh water. This uninterrupted exposure was crucial for maintaining consistent hypobaric conditions throughout the experiment. The hypobaric chamber was configured to ensure a ventilation rate of 25 air exchanges per minute. This high ventilation rate was effective in regulating the concentration of CO2 inside the chamber, thereby maintaining a stable environment for the mice.

(2) The splenectomy was performed as follows:

After anesthesia, the mice were placed in a supine position, and their limbs were fixed. The abdominal operation area was skinned, disinfected, and covered with a sterile towel. A median incision was made in the upper abdomen, followed by laparotomy to locate the spleen. The spleen was then carefully pulled out through the incision. The arterial and venous directions in the splenic pedicle were examined, and two vascular forceps were used to clamp all the tissue in the main cadre of blood vessels below the splenic portal. The splenic pedicle was cut between the forceps to remove the spleen. The end of the proximal hepatic artery was clamped with a vascular clamp, and double or through ligation was performed to secure the site. The abdominal cavity was then cleaned to ensure there was no bleeding at the ligation site, and the incision was closed. Post-operatively, the animals were housed individually. Generally, they were able to feed themselves after recovering from anesthesia and did not require special care.

We hope this detailed description addresses your queries and provides a clear understanding of the experimental conditions and procedures used in our study. These methodological details are crucial for ensuring the accuracy and reproducibility of our research findings.

(2) The lack of changes in MCH needs explanation? During stress erythropoiesis some limit in iron availability should cause MCH decrease particularly if the authors claim that macrophages for rapid iron recycling are decreased. Fig 1A is dispensable. Fig 1G NN control 14 days does not make sense since it is higher than 7 days of HH.

Thank you for your inquiry regarding the lack of changes in Mean Corpuscular Hemoglobin (MCH) in our study, particularly in the context of stress erythropoiesis and decreased macrophage-mediated iron recycling. We appreciate the opportunity to provide further clarification on this aspect.

(1) Explanation for stable MCH levels:

• Our research identified a decrease in erythrophagocytosis and iron recycling in the spleen following HH exposure. Despite this, the MCH levels remained stable. This observation can be explained by considering the compensatory roles of other organs, particularly the liver and duodenum, in maintaining iron homeostasis.

• Specifically, our investigations revealed an enhanced capacity of the liver to engulf RBCs and process iron under HH conditions. This increased hepatic erythrophagocytosis likely compensates for the reduced splenic activity, thereby stabilizing MCH levels.

(2) Role of hepcidin and DMT1 expression:

Additionally, hypoxia is known to influence iron metabolism through the downregulation of Hepcidin and upregulation of Divalent Metal Transporter 1 (DMT1) expression. These alterations lead to enhanced intestinal iron absorption and increased blood iron levels, further contributing to the maintenance of MCH levels despite reduced splenic iron recycling.

(3) Revised Figure 1 and data presentation

To address the confusion regarding the data presented in Figure 1G, we have made revisions in our manuscript. The original Figure 1G, which did not align with the expected trends, has been removed. In its place, we have included a statistical chart of Figure 1F in the new version of Figure 1G. This revision will provide a clearer and more accurate representation of our findings.

(4) Manuscript updates and future research:

• We update our manuscript to incorporate these explanations, ensuring that the rationale behind the stable MCH levels is clearly articulated. This includes a discussion on the role of the liver and duodenum in iron metabolism under hypoxic conditions.

• Future research could explore in greater detail the mechanisms by which different organs contribute to iron homeostasis under stress conditions like HH, particularly focusing on the dynamic interplay between hepatic and splenic functions.

We thank you for your insightful question, which has prompted a thorough re-examination of our findings and interpretations. We believe that these clarifications will enhance the overall understanding of our study and its implications in the context of iron metabolism and erythropoiesis under hypoxic conditions.

(3) Fig 2 the difference between sham and splenectomy is really marginal and not convincing. Is there also a difference at 7 days? Why does the spleen size decrease between 7 and 14 days?

Thank you for your observations regarding the marginal differences observed between sham and splenectomy groups in Figure 2, as well as your inquiries about spleen size dynamics over time. We appreciate this opportunity to clarify these aspects of our study.

(1) Splenectomy vs. Sham group differences:

• In our experiments, the difference between the sham and splenectomy groups under HH conditions, though subtle, was consistent with our hypothesis regarding the spleen's role in erythrophagocytosis and stress erythropoiesis. Under NN conditions, no significant difference was observed between these groups, which aligns with the expectation that the spleen's contribution is more pronounced under hypoxic stress.

(2) Spleen size dynamics and peak stress erythropoiesis:

• The observed splenic enlargement prior to 7 days can be attributed to a combination of factors, including the retention of RBCs and extramedullary hematopoiesis, which is known to be a response to hypoxic stress.

• Prior research has elucidated that splenic stress-induced erythropoiesis, triggered by hypoxic conditions, typically attains its zenith within a timeframe of 3 to 7 days. This observation aligns with our Toluidine Blue (TO) staining results, which indicated that the apex of this response occurs at the 7-day mark (as depicted in Figure 1, F-G). Here, the culmination of this peak is characteristically succeeded by a diminution in extramedullary hematopoiesis, a phenomenon that could elucidate the observed contraction in spleen size, particularly in the interval between 7 and 14 days.

• This pattern of splenic response under prolonged hypoxic stress is corroborated by studies such as those conducted by Wang et al. (2021), Harada et al. (2015), and Cenariu et al. (2021). These references collectively underscore that the spleen undergoes significant dynamism in reaction to sustained hypoxia. This dynamism is initially manifested as an enlargement of the spleen, attributable to escalated erythropoiesis and erythrophagocytosis. Subsequently, as these processes approach normalization, a regression in spleen size ensues.

We’ve revised our manuscript to include a more detailed explanation of these splenic dynamics under HH conditions, referencing the relevant literature to provide a comprehensive context for our findings. We will also consider performing additional analysis or providing further data on spleen size changes at 7 days to support our observations and ensure a thorough understanding of the splenic response to hypoxic stress over time.

(4) Fig 3 B the clusters should be explained in detail. If the decrease in macrophages in Fig 3K/L is responsible for the effect, why does splenectomy not have a much stronger effect? How do the authors know which cells died in the calcein stained population in Fig 3D?

Thank you for your insightful questions regarding the details of our data presentation in Figure 3, particularly about the identification of cell clusters and the implications of macrophage reduction. We appreciate the opportunity to address these aspects and clarify our findings.

(1) Explanation of cell clusters in Figure 3B:

• In the revised manuscript, we have included detailed notes for each cell population represented in Figure 3B (Figure 3D in revised manuscript). These notes provide a clearer understanding of the cell types present in each cluster, enhancing the interpretability of our single-cell sequencing data.

• This detailed annotation will help readers to better understand the composition of the splenic cell populations under study and how they are affected by hypoxic conditions.

(2) Impact of splenectomy vs. macrophage reduction:

• The interplay between the reduction in macrophage populations, as evidenced by our single-cell sequencing data, and the ramifications of splenectomy presents a multifaceted scenario. Notably, the observed decline in macrophage numbers following HH exposure does not straightforwardly equate to a comparable alteration in overall splenic function, as might be anticipated with splenectomy.

• In the context of splenectomy under HH conditions, a significant escalation in the RBCs count was observed, surpassing that in non-splenectomized mice exposed to HH. This finding underscores the spleen's critical role in modulating RBCs dynamics under HH. It also indirectly suggests that the diminished phagocytic capacity of the spleen following HH exposure contributes to an augmented RBCs count, albeit to a lesser extent than in the splenectomy group. This difference is attributed to the fact that, while the number of RPMs in the spleen post-HH is reduced, they are still present, unlike in the case of splenectomy, where they are entirely absent.

• Splenectomy entails the complete removal of the spleen, thus eliminating a broad spectrum of functions beyond erythrophagocytosis and iron recycling mediated by macrophages. The nuanced changes observed in our study may be reflective of the spleen's diverse functionalities and the organism's adaptive compensatory mechanisms in response to the loss of this organ.

(3) Calcein stained population in Figure 3D:

• Regarding the identification of cell death in the calcein-stained population in Figure 3D (Figure 3A in revised manuscript), we acknowledge that the specific cell types undergoing death could not be distinctly determined from this analysis alone.

• The calcein staining method allows for the visualization of live (calcein-positive) and dead (calcein-negative) cells, but it does not provide specific information about the cell types. The decrease in macrophage population was inferred from the single-cell sequencing data, which offered a more precise identification of cell types.

(4) Revised manuscript and data presentation:

• Considering your feedback, we have revised our manuscript to provide a more comprehensive explanation of the data presented in Figure 3, including the nature of the cell clusters and the interpretation of the calcein staining results.

• We have also updated the manuscript to reflect the removal of Figure 3K/L results and to provide a more focused discussion on the relevant findings.

We are grateful for your detailed review, which has helped us to refine our data presentation and interpretation. These clarifications and revisions will enhance the clarity and scientific rigor of our manuscript, ensuring that our conclusions are well-supported and accurately conveyed.

(5) Is the reduced phagocytic capacity in Fig 4B significant? Erythrophagocytosis is compromised due to the considerable spontaneous loss of labelled erythrocytes; could other assays help? (potentially by a modified Chromium release assay?). Is it necessary to stimulated phagocytosis to see a significant effect?

Thank you for your inquiry regarding the significance of the reduced phagocytic capacity observed in Figure 4B, and the potential for employing alternative assays to elucidate erythrophagocytosis dynamics under HH conditions.

(1) Significance of reduced phagocytic capacity:

The observed reduction in the amplitude of fluorescently labeled RBCs in both the blood and spleen under HH conditions suggests a decrease in erythrophagocytosis. This is indicative of a diminished phagocytic capacity, particularly when contrasted with NN conditions.

(2) Investigation of erythrophagocytosis dynamics:

To delve deeper into erythrophagocytosis under HH, we employed Tuftsin to enhance this process. Following the injection of PKH67-labeled RBCs and subsequent HH exposure, we noted a significant decrease in PKH67 fluorescence in the spleen, particularly marked after the administration of Tuftsin. This finding implies that stimulated erythrophagocytosis can influence RBCs lifespan.

(3) Erythrophagocytosis under normal and hypoxic conditions:

Under normal conditions, the reduction in phagocytic activity is less apparent without stimulation. However, under HH conditions, our findings demonstrate a clear weakening of the phagocytic effect. While we established that promoting phagocytosis under NN conditions affects RBC lifespan, the impact of enhanced phagocytosis under HH on RBCs numbers was not explicitly investigated.

(4) Potential for alternative assays:

Considering the considerable spontaneous loss of labeled erythrocytes, alternative assays such as a modified Chromium release assay could provide further insights. Such assays might offer a more nuanced understanding of erythrophagocytosis efficiency and the stability of labeled RBCs under different conditions.

(5) Future research directions:

The implications of these results suggest that future studies should focus on comparing the effects of stimulated phagocytosis under both NN and HH conditions. This would offer a clearer picture of the impact of hypoxia on the phagocytic capacity of macrophages and the subsequent effects on RBC turnover.

In summary, our findings indicate a diminished erythrophagocytic capacity, with enhanced phagocytosis affecting RBCs lifespan. Further investigation, potentially using alternative assays, would be beneficial to comprehensively understand the dynamics of erythrophagocytosis in different physiological states.

(6) Can the observed ferroptosis be influenced by bi- and not trivalent iron chelators?

Thank you for your question regarding the potential influence of bi- and trivalent iron chelators on ferroptosis under hypoxic conditions. We appreciate the opportunity to discuss the implications of our findings in this context.

(1) Analysis of iron chelators on ferroptosis:

In our study, we did not specifically analyze the effects of bi- and trivalent iron chelators on ferroptosis under hypoxia. However, our observations with Deferoxamine (DFO), a well-known iron chelator, provide some insights into how iron chelation may influence ferroptosis in splenic macrophages under hypoxic conditions.

(2) Effect of DFO on oxidative stress markers:

Our findings showed that under 1% O2, there was an increase in Malondialdehyde (MDA) content, a marker of lipid peroxidation, and a decrease in Glutathione (GSH) content, indicative of oxidative stress. These changes are consistent with the induction of ferroptosis, which is characterized by increased lipid peroxidation and depletion of antioxidants. Treatment with Ferrostatin-1 (Fer-1) and DFO effectively reversed these alterations. This suggests that DFO, like Fer-1, can mitigate ferroptosis in splenic macrophages under hypoxia, primarily by impacting MDA and GSH levels.

Author response image 17.

(3) Potential role of iron chelators in ferroptosis:

The effectiveness of DFO in reducing markers of ferroptosis indicates that iron availability plays a crucial role in the ferroptotic process under hypoxic conditions. It is plausible that both bi- and trivalent iron chelators could influence ferroptosis, given their ability to modulate iron availability within cells. Since ferroptosis is an iron-dependent form of cell death, chelating iron, irrespective of its valence state, could potentially disrupt the process by limiting the iron necessary for the generation of reactive oxygen species and lipid peroxidation.

(4) Additional research and manuscript updates:

Our study highlights the need for further research to explore the differential effects of various iron chelators on ferroptosis, particularly under hypoxic conditions. Such studies could provide a more comprehensive understanding of the role of iron in ferroptosis and the potential therapeutic applications of iron chelators. We update our manuscript to include these findings and discuss the potential implications of iron chelation in the context of ferroptosis under hypoxic conditions. This will provide a broader perspective on our research and its significance in understanding the mechanisms of ferroptosis.

Author Response

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

eLife assessment

This valuable study provides insights into the IDA peptide with dual functions in development and immunity. The approach used is solid and helps to define the role of IDA in a two-step process, cell separation followed by activation of innate defenses. The main limitation of the study is the lack of direct evidence linking signaling by IDA and its HAE receptors to immunity. As such the work remains descriptive but it will nevertheless be of interest to a wide range of plant cell biologists.

We thank the reviewers for thoroughly reading our manuscript. We have used their comments and suggestions- to improve the manuscript. Below is a response to the reviewer's comments.

Public Reviews:

Reviewer #1 (Public Review):

The paper titled 'A dual function of the IDA peptide in regulating cell separation and modulating plant immunity at the molecular level' by Olsson Lalun et al., 2023 aims to understand how IDAHAE/HSL2 signalling modulates immunity, a pathway that has previously been implicated in development. This is a timely question to address as conflicting reports exist within the field. IDL6/7 have previously been shown to negatively regulate immune signalling, disease resistance and stress responses in leaf tissue, however IDA has been shown to positively regulate immunity through the shedding of infected tissues. Moreover, recently the related receptor NUT/HSL3 has been shown to positively regulate immune signalling and disease resistance. This work has the potential to bring clarity to this field, however the manuscript requires some additional work to address these questions. This is especially the case as it contracts some previous work with IDL peptides which are perceived by the same receptor complexes.

Can IDA induce pathogen resistance? Does the infiltration of IDA into leaf tissue enhance or reduce pathogen growth? Previously it has been shown that IDL6 makes plants more susceptible. Is this also true for IDA? Currently cytoplasmic calcium influx and apoplastic ROS as overinterpreted as immune responses - these can also be induced by many developmental cue e.g. CLE40 induced calcium transients. Whilst gene expression is more specific is also true that treatment with synthetic peptides, which are recognised by LRR-RKs, can induce immune gene expression, especially in the short term, even when that is not there in vivo function e.g. doi.org/10.15252/embj.2019103894.

We thank the reviewer for the concerns raised and agree that further experiments including pathogen assays would strengthen the link between IDA signaling and immunity and we plan for such experiments in future work. We have however, modified the discussion to include the possible role of IDA induced Ca2+ and ROS during development. We have recently published a preprint (accepted for publication in JXB) ( (Galindo-Trigo et al., 2023, https://doi.org/10.1101/2023.09.12.557497)) strengthening the link between IDA and defense by identifying WRKY transcription factors that regulate IDA expression through a Y1H assay.

This paper shows that receptors other than hae/hsl2 are genetically required to induce defense gene expression, it would have been interesting to see what phenotype would be associated with higher order mutants of closely related haesa/haesa-like receptors. Indeed recently HSL1 has been shown to function as a receptor for IDA/IDL peptides. Could the triple mutant suppress all response? Could the different receptors have distinct outputs? For example for FRK1 gene expression the hae hsl2 mutant has an enhanced response. Could defence gene expression be primarily mediated by HSL1 with subfunctionalisation within this clade?

We agree that it would be interesting to also include HSL1 in our studies. However, the focus of this study has been on HAE and HSL2 and we wanted to explore their role in IDA induced defense responses. Including HSL1 in these studies will require generation of multiple transgenic lines and repeating most of the experiments and are experiments we will consider in a follow up study together with pathogen assays (that would also address the main concern raised in the comment above). We have however, modified the text to include the known function of HSL1 and discuss the possibility of subfunctionalisation of this receptor clade.

One striking finding of the study is the strong additive interaction between IDA and flg22 treatment on gene expression. Do the authors also see this for co-treatment of different peptides with flg22, or is this unique function of IDA? Is this receptor dependent (HAE/HSL1/HSL2)?

This is a good question. Since our study focuses on the IDA signaling pathway we preferentially tested if the additive effect observed between flg22 and mIDA was also observed when mIDA was combined with another peptide involved in defense. The endogenous peptide PIP1, has previously been shown to amplify flg22 signaling (Hou et al 2014, doi:10.1371/journal.ppat.1004331 ). In this study it is shown that co-treatment with flg22 and PIP1 gives increased resistance to Pseudomonas PstDC3000 compared to when plants are treated with each peptide separately. In the same study, the authors also show reduced flg22 induce transcriptional activity of two defense related genes WRKY33 and PR in the receptor like kinase7 (rlk7) mutant (the receptor perceiving PIP1) (). To investigate whether PIP1 would give the same additive effect with mIDA as that observed between flg22 and mIDA, we co-treated seedlings with PIP1 and mIDA. We observed no enhanced transcriptional activity of FRK1, MYB51 and PEP3 in tissue from plants treated with both PIP1 and mIDA peptides compared to single exposure. These results are presented in supplementary figure 11. In conclusion we do not think mIDA acts as a general amplifier of all immune elicitors in plants.

It is interesting how tissue specific calcium responses are in response to IDA and flg22, suggesting the cellular distribution of their cognate receptors. However, one striking observation made by the authors as well, is that the expression of promoter seems to be broader than the calcium response. Indicating that additional factors are required for the observed calcium response. Could diffusion of the peptide be a contributing factor, or are only some cells competent to induce a calcium response?

It is interesting that the authors look for floral abscission phenotypes in cngc and rbohd/f mutants to conclude for genetic requirement of these in floral abscission. Do the authors have a hypothesis for why they failed to see a phenotype for the rbohd/f mutant as was published previously? Do you think there might be additional players redundantly mediating these processes?

It is a possibility that diffusion of the peptide plays a role in the observed response. In a biological context we would assume that the local production of the peptides plays an important role in the cellular responses. In our experimental setup, we add the peptide externally and we can therefore assume that the overlaying cells get in contact with the peptide before cells in the inner tissues and this could be affecting the response recorded However, our results show that there is a differences between flg22 and mIDA induced responses even when the application of the peptides is performed in the same manner, indicating that the difference in the response is not primarily due to the diffusion rate of the peptides but is likely due to different factors being present in different cells. To acquire a better picture of the distribution of receptor expression in the root tissue and to investigate in which cells the receptors have an overlapping expression pattern, we have included results in figure 6 showing plant lines co-expressing transcriptional reporters of FLS2 and HAE or HSL2.

Can you observe callose deposition in the cotyledons of the 35S::HAE line? Are the receptors expressed in native cotyledons? This is the only phenotype tested in the cotyledons.

We thank the reviewer for this valuable comment. We have now conducted callose deposition assay on the 35S:HAE line. And Indeed, we observe callose depositions when cotyledons from a 35S:HAE line is treated with mIDA. We have included these results in figure 4 and have adjusted the text regarding the callose assay accordingly. In addition, we have analyzed the promoter activity of pHAE in cotelydons and we observe weak promoter activity. These results are included as supplementary figure 1d.

Are flg22-induced calcium responses affected in hae hsl2?

The experiment suggested by the reviewer is an important control to ensure that the hae hsl2-Aeq line can respond to a Ca2+ inducing peptide signaling through a different receptor than HAE or HSL2. One would expect to see a Ca2+ response in this line to the flg22 peptide. We performed this experiment and surprisingly we could not detect a flgg22 induced Ca2+ signal in the hae hsl2 mutnt. As it is unlikely that the Ca2+ response triggered by flg22 is dependent on HAE and HSL2 we have to assume that the lack of response is due to a malfunction of the Aeq sensor in this line. As a control to measure the amount of Aeq present in the cells we treat the Aeq seedlings with 2 M CaCl2 and measure the luminescence constantly for 180 seconds (Ranf et al., 2012, DOI10.1093/mp/ssr064). The CaCl2 treatment disrupts the cells and releases the Aeq sensor into the solution where it will react with Ca2+ and release the total possible response in the sample (Lmax) in form of a luminescent peak. When treating the hae hsl2-Aeq line with CaCl2we observe a luminescent peak, indicating the presence of the sensor, however, the response is reduced compared to WT seedlings expressing Aeq. Given the sensitivity of FLS2 to flg22 one would still expect to see a Ca2+ peak in the hae hsl2-Aeq line even if the amount of sensor is reduced. Given that this is not the case, we have to assume that localization or conformation of the sensor is somehow affected in this line or that there is another biological explanation that we cannot explain at the moment.

We have therefore opted on omitting the results using the hae hsl2 Aeq lines from the manuscript and are in the process of mutating HAE and HSL2 by CRISPR-Cas9 in the Aeq background to verify that the mIDA triggered Ca2+ response is dependent on HAE and HSL2.

Reviewer #2 (Public Review):

Lalun and co-authors investigate the signalling outputs triggered by the perception of IDA, a plant peptide regulating organs abscission. The authors observed that IDA perception leads to a transient influx of Ca2+, to the production of reactive oxygen species in the apoplast, and to an increase accumulation of transcripts which are also responsive to an immunogenic epitope of bacterial flagellin, flg22. The authors show that IDA is transcriptionally upregulated in response to several biotic and abiotic stimuli. Finally, based on the similarities in the molecular responses triggered by IDA and elicitors (such as flg22) the authors proposed that IDA has a dual function in modulating abscission and immunity. The manuscript is rather descriptive and provide little information regarding IDA signalling per se. A potential functional link between IDA signalling and immune signalling remains speculative.

We thank the reviewer for the concerns raised and agree that further experiments including pathogen assays would strengthen the link between IDA signaling and immunity and plan for such experiments in future work.

Reviewer #3 (Public Review):

Previously, it has been shown the essential role of IDA peptide and HAESA receptor families in driving various cell separation processes such as abscission of flowers as a natural developmental process, of leaves as a defense mechanism when plants are under pathogenic attack or at the lateral root emergence and root tip cell sloughing. In this work, Olsson et al. show for the first time the possible role of IDA peptide in triggering plant innate immunity after the cell separation process occurred. Such an event has been previously proposed to take place in order to seal open remaining tissue after cell separation to avoid creating an entry point for opportunistic pathogens.

The elegant experiments in this work demonstrate that IDA peptide is triggering the defenseassociated marker genes together with immune specific responses including release of ROS and intracellular CA2+. Thus, the work highlights an intriguing direct link between endogenous cell wall remodeling and plant immunity. Moreover, the upregulation of IDA in response to abiotic and especially biotic stimuli are providing a valuable indication for potential involvement of HAE/IDA signalling in other processes than plant development.

We are pleased that the reviewer finds our findings linking IDA to defense interesting and would like to thank the reviewer for this positive feedback.

Strengths:

The various methods and different approaches chosen by the authors consolidates the additional new role for a hormone-peptide such as IDA. The involvement of IDA in triggering of the immunity complex process represents a further step in understanding what happens after cell separation occurs. The Ca2+ and ROS imaging and measurements together with using the haehsl2 and haehsl2 p35S::HAE-YFP genotypes provide a robust quantification of defense responses activation. While Ca2+ and ROS can be detected after applying the IDA treatment after the occurrence of cell separation it is adequately shown that the enzymes responsible for ROS production, RBOHD and RBOHF, are not implicated in the floral abscission.

Furthermore, IDA production is triggered by biotic and abiotic factors such as flg22, a bacterial elicitor, fungi, mannitol or salt, while the mature IDA is activating the production of FRK1, MYB51 and PEP3, genes known for being part of plant defense process.

Thank you.

Weaknesses:

Even though there is shown a clear involvement of IDA in activating the after-cell separation immune system, the use of p35S:HAE-YFP line represent a weak point in the scientific demonstration. The mentioned line is driving the HAE receptor by a constitutive promoter, capable of loading the plant with HAE protein without discriminating on a specific tissue. Since it is known that IDA family consist of more members distributed in various tissues, it is very difficult to fully differentiate the effects of HAE present ubiquitously.

We agree on this statement. Nevertheless, it is important to note that the responses we have observed are not detectable in WT plants that do not (over)express the HAE receptors. Suggesting that the ROS and callose deposition are induced by the addition of mIDA peptide and not the potential presence of the endogenous IDL peptides.

The co-localization of HAE/HSL2 and FLS2 receptors is a valuable point to address since in the present work, the marker lines presented do not get activated in the same cell types of the root tissues which renders the idea of nanodomains co-localization (as hypothetically written in the discussion) rather unlikely.

Thank you for raising an important aspect of our study. It is true that not all cells in the root which have promoter activity for FLS2 also exhibit promoter activity for either HAE or HSL2. However, we have observed that certain cells in the roots show promoter activity for both receptors. In the revised version of the manuscript, we have included plants expression a transcriptional promoter for both FLS2 and HAE or HSL2 using different fluorescent proteins. We have investigated overlapping promoter activity both at sites of lateral roots, in the tip of the primary root and in the abscission zone. Our results show overlapping expression of the transcriptional reporters in certain cells, indicating that FLS2 and HAE or HSL2 are likely to be found in some of the same cells during plant development. We also observe cells where only one or none of the promoters are active.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Supplementary Figure 3: re-labelling of y axis; 200 than 200,00 for clarity.

This has been addressed.

Supplementary Figure 2: It would be good to include the age of the seedlings used to study calcium influx in the legend.

This has been addressed.

Supplementary Figure 1: rephrase 'IDA induces ROS production in Arabidopsis'.

This has been addressed.

The use of chelating agents to establish the need of calcium from extracellular space is a clear experiment supporting the calcium response phenotype specific to IDA treatment in seedlings. Removing the last asparagine (N) and using it as a peptide that fails to elicit calcium response could simply be because of the peptide is smaller in length or different chemical properties. Therefore, a scrambled sequence would have been a better control.

We thank the reviewer for the suggestion of using a scrambled peptide as a negative control, however we find it unlikely that mIDA∆N69 could induce any activity based on previous work. Results from crystal structure of mIDA bound to the HAE receptor and ligand-receptor interaction studies (10.7554/eLife.15075 ) show that the last asparagine in the mIDA peptide is essential for detectable binding to the HAE receptor and that a peptide lacking this amino acid does not have any activity. We will however, in future experiments also include a scrambled version of the peptide as an additional control.

Reviewer #2 (Recommendations For The Authors):

Please find below specific comments:

(1) Most of the molecular outputs triggered by IDA can be considered as common molecular marks of plant peptides signalling, they do not represent strong evidences of a potential function of IDA in modulating immunity. For instance, perception of CIF peptides, which control the establishment of the Casparian strips, regulate the production of reactive oxygen species, and the transcription of genes associated with immune responses (Fujita et al., The EMBO Journal 2020). It should also be considered that FRK1, whose function remains unknown, may be involved in both immunity and abscission and that the upregulation of FRK1 upon IDA treatment is not indicative of active modulation of immune signalling by IDA.

This is a fair point raised by the reviewer and we now address in the manuscript that ROS and Ca2+ are hallmarks of both plant development and defense. The function of FRK1 is not known however, it is unlikely that the upregulation of FRK1 in response to mIDA plays a role in the developmental progression of abscission as it is not temporally regulated during the abscission process, thus making it an unlikely candidate in the regulation of cell separation (Cai & Lashbrook, 2008, https://doi.org/10.1104/pp.107.110908). We do however agree that further experiments including pathogen assays would strengthen the link between IDA signaling and immunity and plan for such experiments in future work.

(2) It remains unknown whether IDA modulate immunity. For instance, does IDA perception promote resistance to bacteria (bacterial proliferation, disease symptoms)? Is IDA genetically required for plant disease resistance immunity? Is the IDA signalling pathway genetically required for transcriptional changes induced by flg22, such as increase in FRK1 transcripts? In addition, the authors propose that the proposed function of IDA in modulating immune signalling prevents bacterial infection in tissue exposed to stress(es). Does loss of function of IDA or of its corresponding receptors leads to changes in the ability of bacteria to colonise plant root upon stress(es)?

Please see the comment above regarding pathogen assays.

(3) Several aspects of the work appear to correspond to preliminary investigation. For instance, the authors analyse loss of function mutant for genes encoding for Ca2+ permeable channels (CNGCs) which are transcriptionally active during the onset of abscission (Sup. Figure 5). None of the single mutants present an abscission defect. These observations provide no information regarding the identity of the channel(s) involved in IDA-induced calcium influx.

We agree with the reviewer that we have not been able to identify the channels responsible for the IDA-induced calcium influx. Given the redundancy for many of the members of this multigenic family a future approach to identify proteins responsible for the IDA triggered calcium response could be to create multiple KO mutants by CRISPR Cas9.

(4) Using H2DCF-DA, the authors observed a decrease in ROS accumulation in the abscission zone of rbohd/rbohf double KO line (Sup Figure 5c) but describe in the text that ROS production in this zone does not depend on RBOHD and RBOHF (L220). Please clarify.

This has now been clarified in the text.

(5) The authors describe that rbohd/rbohf double KO present a lower petal break-strength, which they describe as an indication of premature cell wall loosening, and that petals of rbohd/rbohf abscised one position earlier than in WT. Yet, the authors postulate that IDA-induced ROS production does not regulate abscission but may regulate additional responses. Instead the data seems to indicate that ROS production by RBOHD and RBOHF regulate the timing of abscission. In addition, it would have been interesting to test whether IDA signalling pathway regulate ROS production in the abscission zone.

The rbohd and rbohf double mutants show several phenotypes associated to developmental stress, the mild phenotype observed with regards to premature abscission (by one position) could be caused by the phenotype of the double mutant rather than related to ROS production. Indeed, it has been suggested that the lignified brace in the AZ dependent on ROS production by the aforementioned RBOHs in necessary for the correct concentration of cell modifying enzymes (Lee et al., 2018, https://doi.org/10.1016/j.cell.2018.03.060). The precocious abscission in this double mutant clearly shows this not to be the case. We have tried to do a ROS burst assay on AZ tissue/flowers with the mIDA peptide but have not been successful with this approach. A ROS sensor expressed in AZ tissue would be a valuable tool to address whether IDA signalling regulates ROS production in AZs.

(6) In Sup. Figure5a, it would be of interest to have a direct comparison of the transcript accumulation of the presented CNGCs and RBOHDs with other of these multigenic families.

The CNGCs and RBOH gene expression profile shown in the figure are the family members expressed during the developmental progress of floral abscission in stamen AZs. Since there is no difference in the temporal expression of the other family members (and most are either not expressed or very weakly expressed in this tissue) it is not possible to do this comparison (Cai & Lashbrook, 2008, https://doi.org/10.1104/pp.107.110908).

(7) L251-253, since IDAdeltaN69 cannot be perceived by its receptors, the absence of induction of pIDA::GUS by IDAdeltaN69 compared to flg22 cannot be seen as a sign of specificity in peptideinduced increase in IDA promotor activity.

We have rephased this in the text

(8) Please provide quantitative and statistical analysis of the calcium measurement presented in sup figure 3.

This has been addressed.

(9) L339-341; This sentence is unclear to me, please rephrase.

We have rephased this in the text

Reviewer #3 (Recommendations For The Authors):

(1) In order to assess the role of CNGCs in abscission process, it would be more interesting to see the effect on the Ca2+ pattern and ROS signaling after application of mIDA on cngc and rbohf rbohd mutants.

We agree in this statement and the studies on mIDA induced ROS and Ca2+ on these mutants will provide valuable information to the regulation of the response. We are in the process of making the lines needed to be able to perform these experiments. However, since it requires crossing of genetically encoded sensors into each mutant, and generation of higher order mutants this is a long process.

(2) With regard to the ROS production (Sup Fig. 1), the application of mIDA can trigger ROS in p35S::HAE:YFP lines, but not in the wild-type plant, which is according to the text "most likely due to the absence of HAE expression" in leaves. The experiment on callose deposition is performed in wild-type cotyledons where no callose deposition could be observed after mIDA treatment (Fig. 4a,b). The conclusion from text is that IDA "is not involved in promoting deposition of callose as a long-term defence response". It appears more likely that neither ROS nor callose can be observed in wild-type plants due to the lack of HAE expression. Therefore, the callose experiment should include the p35S::HAE:YFP lines. The experiment as it is does not allow to draw any conclusion on HAE/IDA involvement in callose formation.

We fully agree with this comment, thank you for pinpointing this out. We have now performed the callose experiment with the 35S:HAE lines. Please see our answer to reviewer #1.

(3) Between Sup Fig. 3 and Sup Fig. 5 two different systems were used to asses the floral stage. An adjustment of the floral stages would be easier to convey the levels of HAE/HSL2 expression and hence potentially with the onset of cell-wall degradation.

We now used the same system to assess floral stages throughout the whole manuscript.

(4) For the Fig. 1 and 2, it will be helpful to mention the genotype used for imaging/quantification of Ca2+.

This has been addressed.

(5) Some of the abbreviations are not introduced as full-text at their first time use in the text, such as: mIDA (Line 68), Ef-Tu (line 85), NADPH (line 77).

The abbreviations have now been introduced.

(6) In the legend of Fig. 5 (lines 897 and 898)- in the figure description, the box plots are identified as light gray and dark gray, while in the panel a of the figure the box plots are colored in red and blue.

Thank you for pointing this out, this has now been corrected.

(7) In figure 1 and 2. the authors write that the number of replicates is 10 (n=10) but data represents a single analysis. Please provide the quantitative ROI analysis, demonstrating that the observed example is representative. This is particularly important since the authors claim very specific changes in pattern of Ca signaling between mIDA and FLG22 treatments (Line 148).

(8) Figure 4: please use alternative scaling on the Y axis instead of breaks.

This has now been fixed.

(9) Figure 5: it is not clear what n=4 refers to when the authors state three independent replicates. In figure 6 they state 4 technical reps and 3 biological reps. Please ensure this is similar across all descriptions.

We have now ensured the correct information in all descriptions.

Author Response

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

eLife assessment

This study presents valuable findings on Legionella pneumophila effector proteins that target host vesicle trafficking GTPases during infection and more specifically modulate ubiquitination of the host GTPase Rab10. The evidence supporting the claims of the authors is solid, although it remains unclear how modification of the GTPase Rab10 with ubiquitin supports Legionella virulence and the impact of ubiquitination during LCV formation. The work will be of interest to colleagues studying animal pathogens as well as cell biologists in general.

We greatly appreciate the positive and valuable feedback from the editors and the reviewers. According to their suggestions, we added many new experimental data and implications of our findings in Legionella virulence in terms of the biological process of its replication niche. Please find our point-to-point responses below.

Public Reviews:

Reviewer #1 (Public Review):

In this manuscript, Kubori and colleagues characterized the manipulation of the host cell GTPase Rab10 by several Legionella effector proteins, specifically members of the SidE and SidC family. They show that Rab10 undergoes both conventional ubiquitination and noncanonical phosphoribose-ubiquitination, and that this posttranslational modification contributes to the retention of Rab10 around Legionella vacuoles.

Strengths

Legionella is an emerging pathogen of increasing importance, and dissecting its virulence mechanisms allows us to better prevent and treat infections with this organism. How Legionella and related pathogens exploit the function of host cell vesicle transport GTPases of the Rab family is a topic of great interest to the microbial pathogenesis field. This manuscript investigates the molecular processes underlying Rab10 GTPase manipulation by several Legionella effector proteins, most notably members of the SidE and SidC families. The finding that MavC conjugates ubiquitin to SdcB to regulate its function is novel, and sheds further light into the complex network of ubiquitin-related effectors from Lp. The manuscript is well written, and the experiments were performed carefully and examined meticulously.

Weaknesses

Unfortunately, in its current form this manuscript offers only little additional insight into the role of effector-mediated ubiquitination during Lp infection beyond what has already been published. The enzymatic activities of the SidC and SidE family members were already known prior to this study, as was the importance of Rab10 for optimal Lp virulence. Likewise, it had previously been shown that SidE and SidC family members ubiquitinate various host Rab GTPases, like Rab33 and Rab1. The main contribution of this study is to show that Rab10 is also a substrate of the SidE and SidC family of effectors. What remains unclear is if Rab10 is indeed the main biological target of SdcB (not just 'a' target), and how exactly Rab10 modification with ubiquitin benefits Lp infection.

Reviewer #1 (Recommendations for The Authors):

Major points of concern

(1) The authors show that SdcB increases Rab10 levels on LCVs at later times of infection and conclude that this is its main biological role. An alternative explanation may be that Rab10 is not 'the main' target of SdcB but merely 'a' target, which may explain why the effect of SdcB on Rab10 accumulation on LCV is only detectable after several hours of infection. An unbiased omics-based approach to identify the actual host target(s) of SdcB may be needed to confirm that Rab10 modification by SdcB is biologically relevant.

We totally agree with your comment that SdcB should have multiple targets considering the abundance of ubiquitin observed on the LCVs when SdcB was expressed (Figure 3). However, the effect of SdcB on Rab10 accumulation at the later time point (7 h) (current Figure 4e) was well supported by the new data showing that the SdcB-mediated ubiquitin conjugation to Rab10 was highly detected at this time point (new Figure 4c). We have tried the comprehensive search of interaction partners of the ANK domain of SdcB. This analysis is planned to be included in our on-going study. We therefore decided not to add the data in this manuscript.

(2) The authors show that Rab10 within cell lysate is ubiquitinated and conclude that ubiquitination of Rab10 is directly responsible for its retention on the LCV. What is the underlying molecular mechanism for this retention? Are GAP proteins prevented from binding and deactivating Rab10. This may be worth testing.

It would be a fantastic hypothesis that a Rab10GAP is involved in the regulation of Rab10 localization on the LCV. However, as far as we know, GAP proteins against Rab10 have not been identified yet. It should be an important issue to be addressed when a Rab10GAP will be found.

(3) Related to this, an alternative explanation would be that Rab10 retention is an indirect effect where inactivators of Rab10, such as host cell GAP proteins, are the main target of SidE/C family members and sent for degradation (see point #1). Can the authors show that Rab10 on the LCV is indeed ubiquitinated?

The possible involvement of a putative Rab10GAP is currently untestable as it is not known. To address whether Rab10 located on the LCV is ubiquitinated nor not, we conducted the critical experiments using active Rab10 (QL) and inactive Rab10 (TN) (new Figure 4a, new Figure 4-figure supplement 1). As revealed for Rab1 (Murata et al., Nature Cell Biol. 2006; Ingmundson et al., Nature 2007), Rab10 is expected to be recruited to the LCV as a GDPbound inactive form and converted to a GTP-bound active form on the LCV. The new results clearly demonstrated that GTP-locked Rab10QL is preferentially ubiquitinated upon infection, strongly supporting the model; Rab10 is ubiquitinated “on the LCV” by the SidE and SidC family ligases.

(4) Also, on what residue(s) is Rab10 ubiquitinated? Jeng et. al. (Cell Host Microbe, 2019, 26(4): 551-563)) suggested that K102, K136, and K154 of Rab10 are modified during Lp infection. How does substituting those residues affect the residency of Rab10 on LCVs? Addressing these questions may ultimately help to uncover if the growth defect of a sidE gene cluster deletion strain is due to its inability to ubiquitinate and retain Rab10 on the LCV.

Thank you for the suggestion. We conducted mutagenesis of the three Lys residues of Rab10 and applied the derivative on the ubiquitination analysis (new Figure 1-figure supplement 1). The Lys substitution to Ala residues did not abrogate the ubiquitination upon Lp infection. This result indicates that ubiquitination sites are present in the other residue(s) including the PR-ubiquitination site(s), raising possibility that disruption of sidE genes would be detrimental for intracellular growth of L. pneumophila because of failure of Rab10 retention.

(5) The authors proposed that "the SidE family primarily contributes towards ubiquitination of Rab10". In this case, what is the significance of SdcB-mediated ubiquitination of Rab10 during Lp infection?

We found that the major contribution of SdcB is retention of Rab10 until the late stage of infection. This claim was supported by our new data (new Figure 4c) as mentioned above (response to comment #1).

(6) The contribution of SdcB to ubiquitination of Rab10 relative to SidC and SdcA is unclear. SidC is shown to be unaffected by MavC. In this case, SidC can ubiquitinate Rab10 regardless of the regulatory mechanism of SdcB by MavC. This is not further being examined or discussed in the manuscript.

The effect of intrinsic MavC is apparent at the later stage (9 h) of infection (Figure 7c) when SdcB gains its activity (see above). We therefore do not think that the contribution of MavC on the SidC/SdcA activities, which are effective in the early stage, would impact on Rab10 localization. However, without specific experiments addressing this issue, possible MavC effects on SidC/SdcA would be beyond the scope in this manuscript.

(7) When is Rab10 required during Lp infection? The authors showed that Rab10 levels at LCV are rather stable from 1hr to 7hr post infection. If MavC regulates the activity of SdcB, when does this occur?

While the Rab10 levels on the LCV (~40 %) are stable during 1-7 h post infection (Figure 2b), it reduced to ~20% at 9 h after infection (Figure 7c) (the description was added in lines 304-306). Rab10 seems to be required for optimal LCV biogenesis over the early to late stages, but may not be required at the maturation stage (9 h). We validated the effect of MavC on the Rab10 localization at this time point (Figure 7c). These observations allowed us to build the scheme described in Figure 7d. We revised the illustration in new Figure 7d according to the helpful suggestions from both the reviewers.

(8) Previous analyses by MS showed that ubiquitination of Rab10 in Lp-infected cells decreases over time (from 1 hpi to 8 hpi - Cell Host Microbe, 2019, 26(4): 551-563). How does this align with the findings made here that Rab10 levels on the LCV and likely its ubiquitination levels increase over time?

We carefully compared the Rab10 ubiquitination at 1 h and 7 h after infection (new Figure 1figure supplement 1b). This analysis showed that the level of its ubiquitination decreased over time in agreement with the previous report. Nevertheless, Rab10 was still significantly ubiquitinated at 7 h, which we believe to cause the sustained retention of Rab10 on the LCV at this time point. We added the observation in lines 146-148.

(9) Polyubiquitination of Rab10 was not detected in cells ectopically producing SdcB and SdeA lacking its DUB domain (Figure 7 - figure supplement 2). Does SdcB actually ubiquitinate Rab10 (see also point #5)? Along the same line, it is curious to find that the ubiquitination pattern of Rab10 is not different for LpΔsidC/ΔsdcA compared to LpΔsidC/dsdcA/dsdcB (Figure 1C). The actual contribution of SdcB to ubiquitinating Rab10 compared to SidC/SdcA thus needs to be clarified.

Thank you for the important point. We currently hypothesize that SidC/SdcA/SdcB-mediated ubiquitin conjugation can occur only in the presence of PR-ubiquitin on Rab10 (either directly on the PR-ubiquitin or on other residue(s) of Rab10). Failure to detect the polyubiquitination in the transfection condition (Figure 7-figure supplement 2) suggests that this specific ubiquitin conjugation can occur in the restricted condition, i.e. only “on the LCV”. We added this description in the discussion section (lines 334-335). No difference between the ΔsidCΔsdcA and ΔsidCΔsdcAΔsdcB strains (Figure 1C, 1h infection) can be explained by the result that SdcB gains activity at the later stages (see above).

Minor comments In Figure 4b and 7b, the authors show a quantification of "Rab10-positive LCVs/SdcBpositive LCVs". Whys this distinction? It begs the question what the percentile of Rab10positive/SdcB-negative LCVs might be?

We took this way of quantification as we just wanted to see the effect of SdcB on the Rab10 localization. To distinguish between SdcB-positive and negative LCVs, we would need to rely on the blue color signals of DAPI to visualize internal bacteria, which we thought to be technically difficult in this specific analysis.

The band of FLAG-tagged SdcB was not detected by immunoblot using anti-FLAG antibody (Figure 5). The authors hypothesized that "disappearance of the SdcB band can be caused by auto-ubiquitination, as SdcB has an ability to catalyze auto-ubiquitination with a diverse repertoire of E2 enzymes. This can be easily confirmed by using MG-132 to inhibit proteasomal degradation of polyubiquitinated substrates.

We conducted the experiment using MG-132 as suggested and found that proteasomal degradation is not the cause of the disappearance of the band (new Figure 5-figure supplement 2, added description in lines 228-233). SdcB is actually not degraded. Instead, its polyubiquitination causes its apparent loss by distributing the SdcB bands in the gel.

In Figure 5F, the authors mentioned that "HA-UbAA did not conjugate to SdcB", whereas "shifted band detected by FLAG probing plausibly represents conjugation of cellular intrinsic Ub". The same argument was made in Figure 6B. These claims should be confirmed by immunoblot using anti-Ub antibody.

Thank you. We added the data using anti-Ub antibody (P4D1) (Figure 6f, new third panel).

Figure 7A: In cell producing MavC, SdcB is clearly present on LCV. However, in Figure 5A, SdcB was not detected by immunoblot in cells ectopically expressing MavC-C74A. What is the interpretation for these results?

SdcB was not degraded in the cells, but just its apparent molecular weight shift occurred by polyubiquitination (see above). The detection of SdcB in the IF images (Figure 7a) supported this claim.

Reviewer #2 (Public Review):

This manuscript explores the interplay between Legionella Dot/Icm effectors that modulate ubiquitination of the host GTPase Rab10. Rab10 undergoes phosphoribosyl-ubiquitination (PR-Ub) by the SidE family of effectors which is required for its recruitment to the Legionella containing vacuole (LCV). Through a series of elegant experiments using effector gene knockouts, co-transfection studies and careful biochemistry, Kubori et al further demonstrate that:

(1) The SidC family member SdcB contributes to the polyubiquitination (poly-Ub) of Rab10 and its retention at the LCV membrane.

(2) The transglutaminase effector, MavC acts as an inhibitor of SdcB by crosslinking ubiquitin at Gln41 to lysine residues in SdcB.

Some further comments and questions are provided below.

(1) From the data in Figure 1, it appears that the PR-Ub of Rab10 precedes and in fact is a prerequisite for poly-Ub of Rab10. The authors imply this but there's no explicit statement but isn't this the case?

Yes, we think that it is the case. We revised the description in the text accordingly (lines 326327).

(2) The complex interplay of Legionella effectors and their meta-effectors targeting a single host protein (as shown previously for Rab1) suggests the timing and duration of Rab10 activity on the LCV is tightly regulated. How does the association of Rab10 with the LCV early during infection and then its loss from the LCV at later time points impact LCV biogenesis or stability? This could be clearer in the manuscript and the summary figure does not illustrate this aspect.

Thank you for pointing the important issue. Association of Rab10 with the LCV is thought to be beneficial for L. pneumophila as it is the identified factor which supports bacterial growth in cells (Jeng et al., 2019). We speculate that its loss from the LCV at the later stage of infection would also be beneficial, since the LCV may need to move on to the maturation stage in which a different membrane-fusion process may proceed. As this is too speculative, we gave a simple modification on the part of discussion section (lines 356-358). We also modified the summary figure (revised Figure 7d) as illustrated with the time course.

(3) How do the activities of the SidE and SidC effectors influence the amount of active Rab10 on the LCV (not just its localisation and ubiquitination)

We agree that it is an important point. We tested the active Rab10 (QL) and inactive Rab10 (TN) for their ubiquitination and LCV-localization profiles (new Figure 4ab, new Figure 4figure supplement 1 and 2). These analyses led us to the unexpected finding that the active form of Rab10 is the preferential target of the effector-mediated manipulation. See also our response to Reviewer 1’s comment #3. Thank you very much for your insightful suggestion.

(4) What is the fate of PR-Ub and then poly-Ub Rab10? How does poly-Ub of Rab10 result in its persistence at the LCV membrane rather than its degradation by the proteosome?

We have not revealed the molecular mechanism in this study. We believe that it is an important question to be solved in future. We added the sentence in the discussion section (lines 376378).

(5) Mutation of Lys518, the amino acid in SdcB identified by mass spec as modified by MavC, did not abrogate SdcB Ub-crosslinking, which leaves open the question of how MavC does inhibit SdcB. Is there any evidence of MavC mediated modification to the active site of SdcB?

The active site of SdcB (C57) is required for the modification (Figure 5b), but it is not likely to be the target residue, as the MavC transglutaminase activity restricts the target residues to Lys. It would be expected that multiple Lys residues on SdcB can be modified by MavC to disturb the catalytic activity.

(6) I found it difficult to understand the role of the ubiquitin glycine residues and the transglutaminase activity of MavC on the inhibition of SdcB function. Is structural modelling using Alphafold for example helpful to explain this?

We conducted the Alphafold analysis of SdcB-Ub. Unfortunately, when the Glycine residues of Ub was placed to the catalytic pocket of SdcB, Q41 of Ub did not fit to the expected position of SdcB (K518). Probably, the ternary complex (MavC-Ub-SdcB) would cause the change of their entire conformation. A crystal structure analysis or more detailed molecular modeling would be required to resolve the issue.

(7) Are the lys mutants of SdbB still active in poly-Ub of Rab10?

We performed the experiment and found that K518R K891R mutant of SdcB still has the E3 ligase activity of similar level with the wild-type upon infection (new Figure 6-figure supplement 2) (lines 283-284). The level was actually slightly higher than that of the wildtype. This result may suggest that the blocking of the modification sites can rescue SdcB from MavC-mediated down regulation.

Reviewer #2 (Recommendations For The Authors):

see above

Author Response

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

eLife assessment

This valuable study applies voltage clamp fluorometry to provide new information about the function of serotonin-gated ion channels 5-HT3AR. The authors convincingly investigate structural changes inside and outside the orthosteric site elicited by agonists, partial agonists, and antagonists, helping to annotate existing cryo-EM structures. This work confirms that the activation of 5-HT3 receptors is similar to other members of this well-studied receptor superfamily. The work will be of interest to scientists working on channel biophysics but also drug development targeting ligand-gated ion channels.

Public Reviews:

All reviewers agreed that these results are solid and interesting. However, reviewers also raised several concerns about the interpretation of the data and some other aspects related to data analysis and discussion that should be addressed by the authors. Essential revisions should include:

(1) Please try to explicitly distinguish between a closed pore and a resting or desensitized state of the pore, to help in clarity.

(2) Add quantification of VCF data (e.g. sensor current kinetics, as suggested by reviewer #2) or better clarify/discuss the VCF quantitative aspects that are taken into account to reach some conclusions (reviewer #3).

(3) Review and add relevant foundational work relevant to this study that is not adequately cited.

(4) Revise the text according to all recommendations raised by the reviewers and listed in the individual reviews below.

We have revised the text to address all four points. See the answers to referees’ recommendations.

Reviewer #1 (Public Review):

Summary:

This study brings new information about the function of serotonin-gated ion channels 5-HT3AR, by describing the conformational changes undergoing during ligands binding. These results can be potentially extrapolated to other members of the Cys-loop ligand-gated ion channels. By combining fluorescence microscopy with electrophysiological recordings, the authors investigate structural changes inside and outside the orthosteric site elicited by agonists, partial agonists, and antagonists. The results are convincing and correlate well with the observations from cryo-EM structures. The work will be of important significance and broad interest to scientists working on channel biophysics but also drug development targeting ligand-gated ion channels.

Strengths:

The authors present an elegant and well-designed study to investigate the conformational changes on 5-HT3AR where they combine electrophysiological and fluorometry recordings. They determined four positions suitable to act as sensors for the conformational changes of the receptor: two inside and two outside the agonist binding site. They make a strong point showing how antagonists produce conformational changes inside the orthosteric site similarly as agonists do but they failed to spread to the lower part of the ECD, in agreement with previous studies and Cryo-EM structures. They also show how some loss-of-function mutant receptors elicit conformational changes (changes in fluorescence) after partial agonist binding but failed to produce measurable ionic currents, pointing to intermediate states that are stabilized in these conditions. The four fluorescence sensors developed in this study may be good tools for further studies on characterizing drugs targeting the 5-HT3R.

Weaknesses:

Although the major conclusions of the manuscript seem well justified, some of the comparison with the structural data may be vague. The claim that monitoring these silent conformational changes can offer insights into the allosteric mechanisms contributing to signal transduction is not unique to this study and has been previously demonstrated by using similar techniques with other ion channels.

The referee emphasizes that “some of the comparison with the structural data may be vague”. To better illustrate the structural reorganizations seen in the cryo-EM structures and that are used for VCF data interpretation, we added a new supplementary figure 3. It shows a superimposition of Apo, setron and 5-HT bond structures, with reorganization of loop C and Cys-loop consistent with VCF data.

Reviewer #2 (Public Review):

Summary:

This study focuses on the 5-HT3 serotonin receptor, a pentameric ligand-gated ion channel important in chemical neurotransmission. There are many cryo-EM structures of this receptor with diverse ligands bound, however assignment of functional states to the structures remains incomplete. The team applies voltage-clamp fluorometry to measure, at once, both changes in ion channel activity, and changes in fluorescence. Four cysteine mutants were selected for fluorophore labeling, two near the neurotransmitter site, one in the ECD vestibule, and one at the ECD-TMD junction. Agonists, partial agonists, and antagonists were all found to yield similar changes in fluorescence, a proxy for conformational change, near the neurotransmitter site. The strength of the agonist correlated to a degree with propagation of this fluorescence change beyond the local site of neurotransmitter binding. Antagonists failed to elicit a change in fluorescence in the vestibular the ECD-TMD junction sites. The VCF results further turned up evidence supporting intermediate (likely pre-active) states.

Strengths:

The experiments appear rigorous, the problem the team tackles is timely and important, the writing and the figures are for the most part very clear. We sorely need approaches orthogonal to structural biology to annotate conformational states and observe conformational transitions in real membranes- this approach, and this study, get right to the heart of what is missing.

Weaknesses:

The weaknesses in the study itself are overall minor, I only suggest improvements geared toward clarity. What we are still missing is application of an approach like this to annotate the conformation of the part of the receptor buried in the membrane; there is important debate about which structure represents which state, and that is not addressed in the current study.

Reviewer #3 (Public Review):

Summary:

The authors have examined the 5-HT3 receptor using voltage clamp fluorometry, which enables them to detect structural changes at the same time as the state of receptor activation. These are ensemble measurements, but they enable a picture of the action of different agonists and antagonists to be built up.

Strengths:

The combination of rigorously tested fluorescence reporters with oocyte electrophysiology is a solid development for this receptor class.

Weaknesses:

The interpretation of the data is solid but relevant foundational work is ignored. Although the data represent a new way of examining the 5-HT3 receptor, nothing that is found is original in the context of the superfamily. Quantitative information is discussed but not presented.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Here are some suggestions that may help to improve the manuscript: - Page 6, point 2), typo: "L131W is positioned more profound in each ECD, its side chain (...)"

“profound” have been corrected into “profoundly”

• Fig 1C: Why not compare 5-HT responses for the four sensors studied? If the reason is the low currents elicited by 5-HT on I160C/Y207W sensor, could you comment on this effect that is not observed for the other full agonist tested (mCPBG)?

The point of this figure (Fig 1G) is to show currents that desensitize to follow the evolution of the fluorescence signal during desensitization, that’s why for the I160C/Y207W sensor where 5-HT become a partial agonist we have judge more appropriate to use mCPBG acting as a more potent agonist to elicit currents with clear desensitization component. We have added a sentence in the legend of the figure to explain this choice more clearly.

• Page 9, paragraph 2: "However, concentration-response curves on V106C/L131W show a small yet visible decorrelation of fluorescence and current (...)" Statistical analysis on EC50c and EC50f will help to see this decorrelation.

Statistical analysis (unpaired t test) has been added to figure 3 panel A.

• Page 10, paragraph 1: the authors describe how "different antagonists promote different degrees of local conformational changes". Does it have any relation to the efficacy or potency of these antagonists? Is there any interpretation for this result?

Since setrons are competitive antagonists, the concept of efficacy of these molecules is unclear. Concerning potency, no correlation between affinity and fluorescence variation is observed. For instance, ondansetron and alosetron bind with similar nanomolar affinity to the 5-HT3R (Thompson & Lummis Curr Pharm Des. 2006;12(28):3615-30) but elicit different fluorescence variations on both S204C and I160C/Y207W sensors.

• Fig. 1 panel A, graph to far right: axis label is cut ("current (uA)/..."). Colors of graph A - right are not clearly distinguishable e.g. cyan from green.

The fluorescent green color that describes the mutant has been changed into limon color which is more clearly distinguishable from cyan.

• Why is R219C/F142W not selected in the study? Are the signals comparable to the chosen R219C/F142W?

We have chosen not to select R219C/F142W because the current elicited by this construct was lower than the current elicited by the construct R219C/Y140W. Moreover, the residue F142 belongs to the FPF motif from Cys-loop that is essential for gating (Polovinkin et al, 2018, Nature).

• Fig. 1 legend typo: "mutated in tryptophan”

“in” has been changed by “into”

• Fig. 2: yellow color (graphs in panel B) is very hard to read.

Yellow color has been darkened to yellow/brown to allow easy reading.

• Fig. 4 is too descriptive and undermines the information of the study. It could be improved e.g. by representing specific structures or partial structures involved. As an additional minor comment, some colors in the figure are hard to differentiate, e.g. magenta and purple.

We have added relevant specific structures involved, namely loop C, the Cys-loop and pre-M1 loop to clarify. The intensity of magenta and purple has been increased to help differentiate the two sensor positions.

• Fig S1C: it is confusing to see the same color pattern for the single mutants without the W. I would recommend to label each trace to make it clearer.

Labelling of the traces corresponding to the single mutants has been added.

• Fig S2: Indicating the statistical significance in the graph for the mutants with different desensitization properties compared to the WT receptor will help its interpretation.

The statistical significance of the difference in the desensitization properties has been added to Figure S2.

Reviewer #2 (Recommendations For The Authors):

Overall comments for the authors:

Selection of cysteine mutants and engineered Trp sites is clear and logical. VCF approach with controls for comparing the functionality of WT vs. mutants, and labeled with unlabeled receptor, is well explained and satisfying. The finding that desensitization involves little change in ECD conformation makes sense. It is somewhat surprising, at least superficially, to find that competitive antagonists promote changes in fluorescence in the same 'direction' and amplitude as strong agonists, however, this is indeed consistent with the structural biology, and with findings from other groups testing different labeling sites. Importantly, the team finds that antagonist-binding changes in deltaF do not spread beyond the region near the neurotransmitter site. The finding that most labeling sites in the ECD, in particular those not in/near the neurotransmitter site, fail to report measurable fluorescence changes, is noteworthy. It contrasts with findings in GlyR, as noted by the authors, and supports a mechanism where most of each subunit's ECD behaves as a rigid body.

Specific questions/comments:

I am confused about the sensor current kinetics. Results section 2) states that all sensors share the same current desensitization kinetics, while Results section 5) states that the ECD-TMD site and the vestibule site sensors exhibit faster desensitization. SF1C, right-most panel of R219C suggests the mutation and/or labeling here dramatically changes apparent activation and deactivation rates measured by TEVC. Both activation and deactivation upon washout appear faster in this one example. Data for desensitization are not shown here but are shown in aggregate in earlier panels. It is a bit surprising that activation and deactivation would both change but no effect on desensitization. Indeed, it looks like, in Fig. 1G, that desensitization rate is not consistent across all constructs. Can you please confirm/clarify?

TEVC and VCF recordings in this study show a significant variability concerning both the apparent desensitization and desactivation kinetics. This is illustrated concerning desensitization in TEVC experiments in figure S2, where the remaining currents after 45 secondes of 5-HT perfusion and the rate constants of desensitization are measured on different oocytes from different batches. Therefore, the differences in desensitization kinetics shown in fig 1.G are not significant, the aim of the figure being solely to illustrate that no variation of fluorescence is observed during the desensitization phase. A sentence in the legend of fig 1.G has been added to precise this point. We also revised the first paragraph of result section 5, clearly stating that the slight tendency of faster desensitization of V106C/L131W and R219C/Y140W sensors is not significant.

An alternative to the conclusion-like title of Results section 2) is that the ECD (and its labels) does not undergo notable conformational changes between activated and desensitized states.

This is a good point and we have added a sentence at the end of results section 2 to present this idea.

I find the discussion paragraph on partial agonist mechanisms, starting with "However," to be particularly important but at times hard to follow. Please try to revise for clarity. I am particularly excited to understand how we can understand/improve assignments of cryo-EM structures using the VCF (or other) approaches. As examples of where I struggled, near the top of p. 11, related to the partial agonist discussion, there is an assumption about the pore being either activated, or resting. Is it not also possible that partial agonists could stabilize a desensitized state of the pore? Strictly speaking, the labeling sites and current measurements do not distinguish between pre-active resting and desensitized channel conformations/states. However, the cryo-EM structures can likely help fill in the missing information there- with all the normal caveats. Please try to explicitly distinguish between a closed pore and a resting or desensitized state of the pore, to help in clarity.

We have revised the section, and hope it is clearer now. We notably state more explicitly the argument for annotation of partial agonist bound closed structures as pre-active, mainly from kinetic consideration of VCF experiments. We also mention and cite a paper by the Chakrapani group published the 4th of January 2024 (Felt et al, Nature Communication), where they present the structures of the m5HT3AR bound to partial agonists, with a set of conformations fully consistent with our VCF data.

This statement likely needs references: "...indirect experiments of substituted cysteine accessibility method (SCAM) and VCF experiments suggested that desensitization involves weak reorganizations of the upper part of the channel that holds the activation gate, arguing for the former hypothesis."

Reference Polovinkin et al, Nature, 2018, has been added.

I respectfully suggest toning down this language a little bit: "VCF allowed to characterize at an unprecedented resolution the mechanisms of action of allosteric effectors and allosteric mutations, to identify new intermediate conformations and to propose a structure-based functional annotation of known high-resolution structures." This VCF stands strongly without unclear claims about unprecedented resolution. What impresses me most are the findings distinguishing how agonists/partial agonists/antagonists share a conserved action in one area and not in another, the observations consistent with intermediate states, and the efforts to integrate these simultaneous current and conformation measurements with the intimidating array of EM structures.

We thank the referee for his positive comments. We have removed “unprecedented resolution” and revised the sentences.

It is beyond the scope of the current study, but I am curious what the authors think the hurdles will be to tracking conformation of the pore domain- an area where non-cryo-EM based conformational measurements are sorely needed to help annotate the EM structures.

We fully agree with the referee that structures of the TMD are very divergent between the various conditions depending on the membrane surrogate. We are at the moment working on this region by VCF, incorporating the fluorescent unnatural amino acid ANAP.

Minor:

(1) P. 5, m5-HT3R: Please clarify that this refers to the mouse receptor, if that is correct.

OK, “mouse” has been added.

(2) Fig. 1D, I suggest moving the 180-degree arrow to the right so it is below but between the two exterior and vestibular views.

Ok, it has been done.

(3) Please add a standard 2D chemical structure of MTS-TAMRA, and TAMRA attached to a cysteine, to Fig 1.

A standard chemical structure has been added for the two isomers of MTS-TAMRA.

(4) Please label subpanels in Fig. 1G with the identity of the label site.

The subpanels have been labelled.

Reviewer #3 (Recommendations For The Authors):

This is solid work but I mainly have suggestions about placing it in context.

(1) Abstract "Data show that strong agonists promote a concerted motion of all sensors during activation, "

The concept of sensors here is the fluorescent labels? I did not find this meaningful until I read the significance statement.

We have specified “fluorescently-labelled” before sensors in the abstract.

(2) p4 "each subunit in the 5-HT3A pentamer...." this description would be identical for any pentameric LGIC so the authors should beware of a misleading specificity. This goes for other phrases in this paragraph. However, the summary of the 5HT specific results is very good.

About the description of the structure, we added “The 5-HT3AR displays a typical pLGIC structure, where….”.

(3) This paper is very nicely put together and generally explains itself well. The work is rigorous and comprehensive. But the meaning of quenching (by local Trp) seems straightforward, but it is not made explicit in the paper. Why doesn't simple labelling (single Cys) at this site work? And can we have a more direct demonstration of the advantage of including the Trp (not in the supplementary figure?) All this information is condensed into the first part of figure 1 (the graph in Figure 1A). Figure 1 could be split and the principle of the introduced quenching could be more clearly shown

detailed in a few more sentences the principle of the TrIQ approach. In addition, to be more explicit, the significative differences of fluorescence comparing sensors with and without tryptophan have been added in Figure 1, panel screening and a sentence have been added in the legend of this figure.

(4) p10 "VCF measurements are also remarkably coherent with the atomic structures showing an open pore (so called F, State 2 and 5-HT asymmetric states), "

This statement is intriguing. What do these names or concepts represent? Are they all the same thing? Where do the names come from? What is meant here? Three different concepts, all consistent? Or three names for the same concept?

We have tried to clarify the statement by making reference to the PDB of the structures.

(5) "Fluorescence and VCF studies identified similar intermediate conformations for nAChRs, ⍺1-GlyRs and the bacterial homolog GLIC(21,32-35). "

Whilst this is true, the motivation for such ideas came from earlier work identifying intermediates from electrophysiology alone (such as the flip state (Burzomato et al 2004), the priming state (Mukhatsimova 2009) and the conformational wave in ACh channels grosman et al 2000). It would be appropriate to mention some of this earlier work.

We have incorporated and described these references in the discussion. Of note, we fully quoted these references in our previous papers on the subject (Menny 2017, Lefebvre 2021, Shi 2023), but the referee is right in asking to quote them again.

(6) "A key finding of the study is the identification of pre-active intermediates that are favored upon binding of partial agonists and/or in the presence of loss-of-function mutations. "

Even more fundamental, the idea of a two-state equilibrium for neurotransmitter receptors was discarded in 1957 according to the action of partial agonists.

DEL CASTILLO J, KATZ B (1957) Interaction at end-plate receptors between different choline derivatives. Proc R Soc Lond B Biol Sci

So to discover this "intermediate" - that is, bound but minimal activity - in the present context seems a bit much. It is a big positive of this paper that the results are congruent with our expectations, but I cannot see value in posing the results as an extension of the 2-state equilibrium (for which there are anyway other objections).

As for intermediates being favoured by loss of function mutations, this concept is already well established in glycine receptors (Plested et al 2007, Lape et al 2012) and doubtless in other cases too.

I do get the point that the authors want to establish a basis in 5-HT3 receptors, but these previous works suggest the results are somewhat expected. This should be commented on.

We also agree. We replace “key finding” by “key observation”, quote most of the references proposed, and explicitly conclude that “The present work thus extends this idea to the 5HT3AR, together with providing structural blueprints for cryo-EM structure annotation”.

(7) "In addition, VCF data allow a quantitative estimate of the complex allosteric action of partial agonists, that do not exclusively stabilize the active state and document the detailed phenotypes of various allosteric mutations."

Where is this provided? If the authors are not motivated to do this, I have some doubts that others will step in. If it is not worth doing, it's probably not worth mentioning either.

Language has been toned down by “In addition, VCF data give insights in the action of partial agonists, that do not exclusively stabilize the active state and document the phenotypes of various allosteric mutations."

(8) Figure 1G please mark which construct is which.

This has been added into Figure 1G

Author Response

Provisional response

We would like to thank the reviewers for taking the time to review our manuscript, for providing useful suggestions for improvement, and for highlighting the significance of our approach.

Reviewer #1 (Public Review):

Summary:

The authors demonstrate that it is possible to carry out eQTL experiments for the model eukaryote S. cerevisiae, in "one pot" preparations, by using single-cell sequencing technologies to simultaneously genotype and measure expression. This is a very appealing approach for investigators studying genetic variation in single-celled and other microbial systems, and will likely inspire similar approaches in non-microbial systems where comparable cell mixtures of genetically heterogeneous individuals could be achieved.

Strengths:

While eQTL experiments have been done for nearly two decades (the corresponding author's lab are pioneers in this field), this single-cell approach creates the possibility for new insights about cell biology that would be extremely challenging to infer using bulk sequencing approaches. The major motivating application shown here is to discover cell occupancy QTL, i.e. loci where genetic variation contributes to differences in the relative occupancy of different cell cycle stages. The authors dissect and validate one such cell cycle occupancy QTL, involving the gene GPA1, a G-protein subunit that plays a role in regulating the mating response MAPK pathway. They show that variation at GPA1 is associated with proportional differences in the fraction of cells in the G1 stage of the cell cycle. Furthermore, they show that this bias is associated with differences in mating efficiency.

We thank the reviewer for recognizing the strengths of our overall approach and our dissection of the functional consequences of the W82R variant of GPA1.

Weaknesses:

While the experimental validation of the role of GPA1 variation is well done, the novel cell cycle occupancy QTL aspect of the study is somewhat underexploited. The cell occupancy QTLs that are mentioned all involve loci that the authors have identified in prior studies that involved the same yeast crosses used here. It would be interesting to know what new insights, besides the "usual suspects", the analysis reveals. For example, in Cross B there is another large effect cell occupancy QTL on Chr XI that affects the G1/S stage. What candidate genes and alleles are at this locus?

We thank the reviewer for this suggestion. We plan to expand the section on cell cycle occupancy QTL in our revision.

And since cell cycle stages are not biologically independent (a delay in G1, could have a knock-on effect on the frequency of cells with that genotype in G1/S), it would seem important to consider the set of QTLs in concert.

We thank the reviewer for this suggested clarification. In our revision, we will clarify that the cell cycle occupancy phenotype represents the proportion of cells assigned to a given stage. As the reviewer correctly notes, a change in the proportion of cells in one stage may alter the proportion of cells in other stages, and this could result in cell cycle occupancy QTL for multiple stages. We will make efforts to consider the cell cycle occupancy QTLs in concert in the revised manuscript.

Reviewer #2 (Public Review):

Boocock and colleagues present an approach whereby eQTL analysis can be carried out by scRNA-Seq alone, in a one-pot-shot experiment, due to genotypes being able to be inferred from SNPs identified in RNA-Seq reads. This approach obviates the need to isolate individual spores, genotype them separately by low-coverage sequencing, and then perform RNA-Seq on each spore separately. This is a substantial advance and opens up the possibility to straightforwardly identify eQTLs over many conditions in a cost-efficient manner. Overall, I found the paper to be well-written and well-motivated, and have no issues with either the methodological/analytical approach (though eQTL analysis is not my expertise), or with the manuscript's conclusions.

We thank the reviewer for recognizing the significant contributions our work makes to the field.

393 segregant experiment:

For the experiment with the 393 previously genotyped segregants, did the authors examine whether averaging the expression by genotype for single cells gave expression profiles similar to the bulk RNA-Seq data generated from those genotypes? Also, is it possible (and maybe not, due to the asynchronous nature of the cell culture) to use the expression data to aid in genotyping for those cells whose genotypes are ambiguous? I presume it might be if one has a sufficient number of cells for each genotype, though, for the subsequent one-pot experiments, this is a moot point.

We thank the reviewer for this comment. While we could expand the analysis along these lines, this is not relevant for the subsequent one-pot eQTL experiments, as the reviewer notes, and is therefore beyond the scope of the manuscript. We will make the data available so that anyone interested can try these analyses.

Figure 1B:

Is UMAP necessary to observe an ellipse/circle - I wouldn't be surprised if a simple PCA would have sufficed, and given the current discussion about whether UMAP is ever appropriate for interpreting scRNA-Seq (or ancestry) data, it seems the PCA would be a preferable approach. I would expect that the periodic elements are contained in 2 of the first 3 principal components. Also, it would be nice if there were a supplementary figure similar to Figure 4 of Macosko et al (PMID 26000488) to indeed show the cell cycle dependent expression.

We thank the reviewer for this comment. We too have been following the debate on the utility of UMAP for scRNA-seq, and in our revision we will provide an alternative visualization of the cell cycle. We will also generate a supplementary figure similar to Figure 4 of Macosko et al. to visualize cell-cycle-dependent gene expression.

Aging, growth rate, and bet-hedging:

The mention of bet-hedging reminded me of Levy et al (PMID 22589700), where they saw that Tsl1 expression changed as cells aged and that this impacted a cell's ability to survive heat stress. This bet-hedging strategy meant that the older, slower-growing cells were more likely to survive, so I wondered a couple of things. It is possible from single-cell data to identify either an aging, or a growth rate signature? A number of papers from David Botstein's group culminated in a paper that showed that they could use a gene expression signature to predict instantaneous growth rate (PMID 19119411) and I wondered if a) this is possible from single-cell data, and b) whether in the slower growing cells, they see markers of aging, whether these two signatures might impact the ability to detect eQTLs, and if they are detected, whether they could in some way be accounted for to improve detection.

We thank the reviewer for this comment and suggested analyses. We are not sure whether one can see gene expression signatures of aging in yeast scRNA-seq data. We believe that such analyses are beyond the scope of this work, but we will make the data available so that anyone interested can try them.

AIL vs. F2 segregants:

I'm curious if the authors have given thought to the trade-offs of developing advanced intercross lines for scRNA-Seq eQTL analysis. My impression is that AIL provides better mapping resolution, but at the expense of having to generate the lines. It might be useful to see some discussion on that.

We thank the reviewer for their comment. We will include some discussion of the trade-offs of different experimental designs in our revision.

10x vs SPLit-Seq

10x is a well established, but fairly expensive approach for scRNA-Seq - I wondered how the cost of the 10x approach compares to the previously used approach of genotyping segregants and performing bulk RNA-Seq, and how those costs would change if one used SPLiT-Seq (see PMID 38282330).

We will provide some ballpark estimates of the costs, and we will discuss the trade-offs of different scRNA-seq technologies in our revision

Author Response

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

We would like to thank the reviewers for their insightful comments and recommendations. We have extensively revised the manuscript in response to the valuable feedback. We believe the results is a more rigorous and thoughtful analysis of the data. Furthermore, our interpretation and discussion of the findings is more focused and highlights the importance of the circuit and its role in the response to stress. Thank you for helping to improve the presented science.

Key changes made in response to the reviewers comments include:

• Revision of statistical analyses for nearly all figures, with the addition of a new table of summary statistics to include F and/or t values alongside p-values.

• Addition of statistical analyses for all fiber photometry data.

• Examination of data for possible sex dependent effects.

• Clarification of breeding strategies and genotype differences, with added details to methods to improve clarity.

• Addressing concerns about the specificity of virus injections and the spread, with additional details added to methods.

• Modification of terminology related to goal-directed behavior based on reviewer feedback, including removal of the term from the manuscript.

• Clarification and additional data on the use of photostimulation and its effects, including efforts to inactivate neurons for further insight, despite technical challenges.

• Correction of grammatical errors throughout the manuscript.

Reviewer 1:

Despite the manuscript being generally well-written and easy to follow, there are several grammatical errors throughout that need to be addressed.

Thank you for highlighting this issue. Grammatical errors have been fixed in the revised version of the manuscript.

Only p values are given in the text to support statistical differences. This is not sufficient. F and/or t values should be given as well.

In response to this critique and similar comments from Reviewer 2, we re-evaluated our approach to statistical analyses and extensively revised analyses for nearly all figures. We also added a new table of summary statistics (Supplemental Table 1) containing the type of analysis, statistic, comparison, multiple comparisons, and p value(s). For Figures 4C-E, 5C, 6C-E, 7H-I, and 8H we analyzed these data using two-way repeated measures (RM) ANOVA that examined the main effect of time (either number of sessions or stimulation period) in the same animal and compared that to the main effect of genotype of the animal (Cre+ vs Cre-), and if there was an interaction. For Supplemental Figure 7A we also conducted a two-way RM ANOVA with time as a factor and activity state (number of port activations in active vs inactive nose port) as the other in Cre+ mice. For Figures 5D-E we conducted a two-way mixed model ANOVA that accounted and corrected for missing data. In figures that only compared two groups of data (Figures 5F-L, 6F, 8C-D, 8I, and Supp 6F-G) we used two-tailed t-test for the analysis. If our question and/or hypothesis required us to conduct multiple comparisons between or within treatments, we conducted Bonferroni’s multiple comparisons test for post hoc analysis (we note which groups we compared in Supplemental Table 1). For figures that did or did not show a change in calcium activity (Figure 3G, 3I-K, 7B, 7D-E, 8E-F), we compared waveform confidence intervals (Jean-Richard-Dit-Bressel, Clifford, McNally, 2020). The time windows we used as comparison are noted in Supplemental Table 1, and if the comparisons were significant at 95%, 99%, and 99.9% thresholds.

None of prior comparisons in prior analyses that were significant were found to have fallen below thresh holds for significance. Of those found to be not significantly different, only one change was noted. In Figure 6E there was now a significant baseline difference between Cre+ and Cre- mice with Cre- mice taking longer to first engage the port compared to Cre+ mice (p=0.045). Although the more rigorous approach the statistical analyses did not change our interpretations we feel the enhanced the paper and thank the reviewer for pushing this improvement.

Moreover, the fibre photometry data does not appear to have any statistical analyses reported - only confidence intervals represented in the figures without any mention of whether the null hypothesis that the elevations in activity observed are different from the baseline.

This is particularly important where there is ambiguity, such as in Figure 3K, where the spontaneous activity of the animal appears to correlate with a spike in activity but the text mentions that there is no such difference. Without statistics, this is difficult to judge.

Thank you for highlighting this critical point and providing an opportunity to strengthen our manuscript. We added statistical analyses of all fiber photometry data using a recently described approach based on waveform confidence intervals (Jean-Richard-Dit-Bressel, Clifford, McNally, 2020). In the statistical summary (Supplemental Table 1) we note the time window that we used for comparison in each analysis and if the comparisons were significant at 95%, 99%, and 99.9% thresholds. Thank you from highlighting this and helping make the manuscript stronger.

With respect to Figure 3K, we are not certain we understood the spike in activity the reviewer referred to. Figure 3J and K include both velocity data (gold) and Ca2+ dependent signal (blue). We used episodes of velocity that were comparable to the avoidance respond during the ambush test and no significant differences in the Ca2+ signal when gating around changes in velocity in the absence of stressor (Supplemental Table1). This is in contrast to the significant change in Ca2+ signal following a mock predator ambush (Figure 3J). We interpret these data together to indicate that locomotion does not correlate with an increase in calcium activity in SuMVGLUT2+::POA neurons, but that coping to a stressor does. This conclusion is further examined in supplemental Figure 5, including examining cross-correlation to test for temporally offset relationship between velocity and Ca2+ signal in SUMVGLUT2+::POA neurons.

The use of photostimulation only is unfortunate, it would have been really nice to see some inactivation of these neurons as well. This is because of the well-documented issues with being able to determine whether photostimulation is occurring in a physiological manner, and therefore makes certain data difficult to interpret. For instance, with regards to the 'active coping' behaviours - is this really the correct characterisation of what's going on? I wonder if the mice simply had developed immobile responding as a coping strategy but when they experience stimulation of these neurons that they find aversive, immobility is not sufficient to deal with the summative effects of the aversion from the swimming task as well as from the neuronal activation? An inactivation study would be more convincing.

We agree with the point of the reviewer, experiments demonstrating necessity of SUMVGLUT2+::POA neurons would have added to the story here. We carried out multiple experiments aimed at addressing questions about necessity of SuMVGLUT2+::POA neurons in stress coping behaviors, specifically the forced swim assay. Efforts included employing chemogenetic, optogenetic, and tetanus toxin-based methods. We observed no effects on locomotor activity or stress coping. These experiments are both technically difficult and challenging to interpret. Interpretation of negative results, as we obtained, is particularly difficult because of potential technical confounds. Selective targeting of SuMVGLUT2+::POA neurons for inhibition requires a process requiring three viral injections and two recombination steps, increasing variability and reducing the number of neurons impacted. Alternatively, photoinhibition targeting SuMVGLUT2+::POA cells can be done using Retro-AAV injected into POA and a fiber implant over SuM. We tried both approaches. Data obtained were difficult to interpret because of questions about adequate coverage of SuMVGLUT2+::POA population by virally expressed constructs and/or light spread arose. The challenge of adequate coverage to effectively prevent output from the targeted population is further confounded by challenges inherent in neural inhibition, specifically determining if the inhibition created at the cellular level is adequate to block output in the context of excitatory inputs or if neurons must be first engaged in a particular manner for inhibition to be effective. Baseline neural activity, release probability, and post-synaptic effects could all be relevant, which photo-inhibition will potentially not resolve. So, while the trend is to always show “necessary and sufficient” effects, we’ve tried nearly everything, and we simply cannot conclude much from our mixed results. There are also wellestablished problems with existing photo-inhibition methods, which while people use them and tout them, are often ignored. We have a lot of expertise in photo-inhibition optogenetics, and indeed have used it with some success, developed new methods, yet in this particular case we are unable to draw conclusions related to inhibition. People have experienced similar challenges in locus coeruleus neurons, which have very low basal activity, and inhibition with chemogenetics is very hard, as well as with optogenetic pump-based approaches, because the neurons fire robust rebound APs. We have spent almost 2.5 years trying to get this to work in this circuit because reviews have been insistent on this result for the paper to be conclusive. Unfortunately, it simply isn’t possible in our view until we know more about the cell types involved. This is all in spite of experience using the approach in many other publications.

We also employed less selective approaches, such as injecting AAV-DIO-tetanus toxin light chain (Tettox) constructs directly into SuM VGLUT2-Cre mice but found off target effects impacting animal wellbeing and impeding behavioral testing due viral spread to surrounding areas.

While we are disappointed for being unable to directly address questions about necessity of SuMVGLUT2+::POA neurons in active coping with experimental data, we were unable to obtain results allowing for clear interpretation across numerous other domains the reviewers requested. We also feel strongly that until we have a clear picture of the molecular cell type architecture in the SuM, and Cre-drivers to target subsets of neurons, this question will be difficult to resolve for any group. We are working now on RNAseq and related spatial transcriptomics efforts in the SuM and examining additional behavioral paradigm to resolve these issues, so stay tuned for future publications.

Accordingly, we avoid making statements relating to necessity in the manuscript. In spite of having several lines of physiological data with strong robust correlations behavior related to the SuMVGLUT2+::POA circuit.

Nose poke is only nominally instrumental as it cannot be shown to have a unique relationship with the outcome that is independent of the stimuli-outcome relationships (in the same way that a lever press can, for example). Moreover, there is nothing here to show that the behaviours are goal-directed.

Thank you for highlighting this point. Regarding goal-direct terminology, we removed this terminology from the manuscript. Since the mice perform highly selective (active vs inactive) port activation robustly across multiple days of training the behavior likely transitions to habitual behavior. We only tested the valuation of stimuli termination of the final day of training with time limited progressive ratio test. With respect to lever press versus active port activation, we are unclear how using a lever in this context would offer a different interpretation. Lever pressing may be more sensitive to changes in valuation when compared to nose poke port activation (Atalayer and Rowland 2008); however, in this study the focus of the operant behavior is separating innate behaviors for learned action–outcome instrumental learned behaviors for threat response (LeDoux and Daw 2018). The robust highly selective activation of the active port illustrated in Figure 6 fits as an action–outcome instrumental behavior wherein mice learn to engage the active but not inactive port to terminate photostimulation. The first activation of the port occurs through exploration of the arena but as demonstrated by the number of active port activations and the decline in time of the first active port engagement, mice expressing ChR2eYFP learn to engage the port to terminate the stimulation. To aid in illustrating this point we have added Supplemental Figure 7 showing active and inactive port activations for both Cre+ and Cre- mice. This adds clarity to high rate of selective port activation driven my stimulation of SUMVGLUT2+::POA neurons compared to controls. The elimination of goal directed and providing additional data narrows and supports one of the key points of the operant experiment.

With regards to Figure 1: This is a nice figure, but I wonder if some quantification of the pathways and their density might be helpful, perhaps by measuring the intensity of fluorescence in image J (as these are processes, not cell bodies that can be counted)? Mind you, they all look pretty dense so perhaps this is not necessary! However, because the authors are looking at projections in so-called 'stress-engaged regions', the amygdala seems conspicuous by its absence. Did the authors look in the amygdala and find no projections? If so it seems that this would be worth noting.

This is an interesting question but has proven to be a very technically challenging question. We consulted with several leaders who routinely use complimentary viral tracing methods in the field. We were unable to devise a method to provide a satisfactorily meaningful quantitative (as opposed to qualitative) approach to compare SUMVGLUT2+::POA to SuMVGLUT2+ projections. A few limitations are present that hinder a meaningful quantitative approach. One limitation was the need for different viral strategies to label the two populations. Labeling SuMVGLUT2+::POA neurons requires using VGLUT2-Flp mice with two injections into the POA and one into SuM. Two recombinase steps were required, reducing efficiency of overlap. This combination of viral injections, particularly the injections of RetroAAVs in the POA, can induce significant quantitative variability due to tropism, efficacy, and variability of retro-viral methods, and viral infection generally. These issues are often totally ignored in similar studies across the “neural circuit” landscape, but it doesn’t make them less relevant here.

Although people do this in the field, and show quantification, we actually believe that it can be a quite misleading read-out of functionally relevant circuitry, given that neurotransmitter release ultimately is amplified by receptors post-synaptically, and many examples of robust behavioral effects have been observed with low fiber tracing complimentary methods (McCall, Siuda et al. 2017). In contrast, the broader SuMVGLUT2+ population was labeled using a single injection into the SuM. This means there like more efficient expression of the fluorophore. Additionally, in areas that contain terminals and passing fibers understanding and interpreting fluorescent signal is challenging. Together, these factors limit a meaningful quantitative comparison and make an interpretation difficult to make. In this context, we focused on a conservative qualitative presentation to demonstrate two central points. That 1) SuMVGLUT2+::POA neurons are subset of SuMVGLUT2+ neurons that project to specific areas and that exclude dentate gyrus, and they 2) arborize extensively to multiple areas which have be linked to threat responses. We agree that there is much to be learned about how different populations in SuM connect to targets in different regions of the brain and to continue to examine this question with different techniques. A meaningful quantitative study comparing projections is technically complex and, we feel, beyond our ability for this study.

Also, for the reasons above we do not believe that quantification provides exceptional clarity with respect to the putative function of the circuit, glutamate released, or other cotransmitters given known amplification at the post-synaptic side of the circuit.

With regard to the amygdala, other studies on SuM projections have found efferent projections to amygdala (Ottersen, 1980; Vertes, 1992). In our study we were unable to definitively determine projections from SuMVGLUT2+::POA neurons to amygdala, which if present are not particularly dense. For this reason we were conservative and do not comment on this particular structure.

I would suggest removing the term goal-directed from the manuscript and just focusing on the active vs. passive distinction.

We removed the use of goal-directed. Thank you for helping us clarify our terminology.

The effect observed in Figure 7I is interesting, and I'm wondering if a rebound effect is the most likely explanation for this. Did the authors inhibit the VGAT neurons in this region at any other times and observe a similar rebound? If such a rebound was not observed it would suggest that it is something specific about this task that is producing the behaviour. I would like it if the authors could comment on this.

We agree that results showing the change in coping strategy (passive to active) in forced swim after but not during stimulation of SuMVGAT+ neurons is quite interesting (Figure 7I). This experiment activated SuMVGAT+ neurons during a section of the forced swim assay and mice showed a robust shift to mobility after the stimulation of SuMVGAT+ neurons stopped. We did not carry out inhibition of SuMVGAT+ neurons in this manuscript. As the reviewer suggested, strong inhibition of local SuM neurons, including SUMVGLUT2+::POA neurons, could lead to rebound activity that may shift coping behaviors in confusing ways. We agree this is an interesting idea but do not have data to support the hypothesis further at this time.

Reviewer 2

(1) These are very difficult, small brain regions to hit, and it is commendable to take on the circuit under investigation here. However, there is no evidence throughout the manuscript that the authors are reliably hitting the targets and the spread is comparable across experiments, groups, etc., decreasing the significance of the current findings. There are no hit/virus spread maps presented for any data, and the representative images are cropped to avoid showing the brain regions lateral and dorsal to the target regions. In images where you can see the adjacent regions, there appears expression of cell bodies (such as Supp 6B), suggesting a lack of SuM specificity to the injections.

We agree with the reviewer that the areas studied are small and technically challenging to hit. This was one of driving motivations for using multiple tools in tandem to restrict the area targeted for stimulation. Approaches included using a retrograde AAVs to express ChR2eFYP in SUMVGLUT2+::POA neurons; thereby, restricting expression to VGLUT2+ neurons that project to the POA. Targeting was further limited by placement of the optic fiber over cell bodies on SuM. Thus, only neurons that are VGLUT2+, project to the POA, and were close enough to the fiber were active by photostimulation. Regrettably, we were not able to compile images from mice where the fiber was misplaced leading to loss of behavioral effects. We would have liked to provide that here to address this comment. Unfortunately, generating heat maps for injections is not possible for anatomic studies that use unlabeled recombinase as part of an intersectional approach. Also determining the point of injection of a retroAAV can be difficult to accurately determine its location because neurons remote to injection site and their processes are labeled.

Experiments described in Supplemental Figure 6B on VGAT neurons in SuM were designed and interpreted to support the point that SUMVGLUT2+::POA neurons are a distinct population that does not overlap with GABAergic neurons. For this point it is important that we targeted SuM, but highly confined targeting is not needed to support the central interpretation of the data. We do see labeling in SuM in VGAT-Cre mice but photo stimulation of SuMVGAT+ neurons does not generate the behavioral changes seen with activation of SUMVGLUT2+::POA neurons. As the reviewer points out, SuM is small target and viral injection is likely to spread beyond the anatomic boundaries to other VGAT+ neurons in the region, which are not the focus here. The activation would be restricted by the spread of light from the fiber over SuM (estimated to be about a 200um sphere in all directions). We did not further examine projections or localization of VGAT+ neurons in this study but focused on the differential behavioral effects of SUMVGLUT2+::POA neurons.

(2) In addition, the whole brain tracing is very valuable, but there is very little quantification of the tracing. As the tracing is the first several figures and supp figure and the basis for the interpretation of the behavior results, it is important to understand things including how robust the POA projection is compared to the collateral regions, etc. Just a rep image for each of the first two figures is insufficient, especially given the above issue raised. The combination of validation of the restricted expression of viruses, rep images, and quantified tracing would add rigor that made the behavioral effects have more significance.

For example, in Fig 2, how can one be sure that the nature of the difference between the nonspecific anterograde glutamate neuron tracing and the Sum-POA glutamate neuron tracing is real when there is no quantification or validation of the hits and expression, nor any quantification showing the effects replicate across mice? It could be due to many factors, such as the spread up the tract of the injection in the nonspecific experiment resulting in the labeling of additional regions, etc.

Relatedly, in Supp 4, why isn’t C normalized to DAPI, which they show, or area? Similar for G what is the mcherry coverage/expression, and why isn’t Fos normalized to that?

Thank you for highlighting the importance of anatomy and the value of anatomy. Two points based on the anatomic studies are central to our interpretation of the experimental data. First, SUMVGLUT2+::POA are a distinct population within the SuM. We show this by demonstrating they are not GABAergic and that they do not project to dentate gyrus. Projections from SuM to dentate gyrus have been described in multiple studies (Boulland et al., 2009; Haglund et al., 1987; Hashimotodani et al., 2018; Vertes, 1992) and we demonstrate them here for SuMVGLUT2+ cells. Using an intersectional approach in VGLUT2-Flp mice we show SUMVGLUT2+::POA neurons do not project to dentate gyrus. We show cell bodies of SUMVGLUT2+::POA neurons located in SuM across multiple figures including clear brain images. Thus, SUMVGLUT2+::POA neurons are SuM neurons that do not project to dentate gyrus, are not GABAergic, send projections to a distinct subset of targets, most notably excluding dentate gyrus. Second, SUMVGLUT2+::POA neurons arborize sending projections to multiple regions. We show this using a combinatorial genetic and viral approach to restrict expression of eYFP to only neurons that are in SuM (based on viral injection), project to the POA (based on retrograde AAV injection in POA), and VGLUT2+ (VGLUT2-Flp mice). Thus, any eYFP labeled projection comes from SUMVGLUT2+::POA neurons. We further confirmed projections using retroAAV injection into areas identified using anterograde approaches (Supplemental Figure 2). As discussed above in replies to Reviewer 1, we feel limitations are present that preclude meaningful quantitative analysis. We thus opted for a conservative interpretation as outlined.

Prior studies have shown efferent projections from SuM to many areas, and projections to dentate gyrus have received substantial attention (Bouland et al., 2009; Haglund, Swanson, and Kohler, 1984; Hashimotodani et al., 2018; Soussi et al., 2010; Vertes, 1992; Pan and McNaugton, 2004). We saw many of the same projections from SuMVGLUT2+ neurons. We found no projections from SUMVGLUT2+::POA neurons to dentate gyrus (Figure 2). Our description of SuM projection to dentate gyrus is not new but finding a population of neurons in SuM that does not project to dentate gyrus but does project to other regions in hippocampus is new. This finding cannot be explained by spread of the virus in the tract or non-selective labeling.

(3) The authors state that they use male and female mice, but they do not describe the n’s for each experiment or address sex as a biological variable in the design here. As there are baseline sex differences in locomotion, stress responses, etc., these could easily factor into behavioral effects observed here.

Sex specific effects are possible; however, the studies presented here were not designed or powered to directly examine them. A point about experimental design that helps mitigate against strong sex dependent effect is that often the paradigm we used examined baseline (pre-stimulation) behavior, how behavior changed during stimulation, and how behavior returned (or not) to baseline after stimulation. Thus, we test changes in individual behaviors. Although we had limited statistical power, we conducted analyses to examine the effects of sex as variable in the experiments and found no differences among males and females.

(4) In a similar vein as the above, the authors appear to use mice of different genotypes (however the exact genotypes and breeding strategy are not described) for their circuit manipulation studies without first validating that baseline behavioral expression, habituation, stress responses are not different. Therefore, it is unclear how to interpret the behavioral effects of circuit manipulation. For example in 7H, what would the VGLUT2-Cre mouse with control virus look like over time? Time is a confound for these behaviors, as mice often habituate to the task, and this varies from genotype to genotype. In Fig 8H, it looks like there may be some baseline differences between genotypes- what is normal food consumption like in these mice compared to each other? Do Cre+ mice just locomote and/or eat less? This issue exists across the figures and is related to issues of statistics, potential genotype differences, and other experimental design issues as described, as well as the question about the possibility of a general locomotor difference (vs only stress-induced). In addition, the authors use a control virus for the control groups in VGAT-Cre manipulation studies but do not explain the reasoning for the difference in approach.

Thank you for highlighting the need for greater clarity about the breeding strategies used and for these related questions. We address the breeding strategy and then move to address the additional concerns raised. We have added details to the methods section to address this point. For VGLUT2-Cre mice we use litter mates controls from Cre/WT x WT/WT cross. The VGLUT2-Cre line (RRID:IMSR_JAX:028863) (Vong L , et al. 2011) used here been used in many other reports. We are not aware of any reports indicating a phenotype associated with the addition of the IRES-Cre to the Slc17a6 loci and there is no expected impact of expression of VGLUT2. Also, we see in many of the experiments here that the baseline (Figures 4, 5, and 7) behaviors are not different between the Cre+ and Cre- mice. For VGAT-Cre mice we used a different breeding strategy that allowed us to achieve greater control of the composition of litters and more efficient cohorts cohort. A Cre/Cre x WT/WT cross yielded all Cre/WT litters. The AAV injected, ChR2eYFP or eYFP, allowed us to balance the cohort.

Regarding Figure 7H, which shows time immobile on the second day of a swim test, data from the Cre- mice demonstrate the natural course of progression during the second day of the test. The control mice in the VGAT-Cre cohort (Figure 7I) have similar trend. The change in behavior during the stimulation period in the Cre+ mice is caused by the activation of SUMVGLUT2+::POA neurons. The behavioral shift largely, but not completely, returns to baseline when the photostimulation stops. We have no reason to believe a VGLUT2-Cre+ mouse injected with control AAV to express eYFP would be different from WT littermate injected with AVV expressing ChR2eYFP in a Cre dependent manner.

Turning to concerns related to 8H, which shows data from fasted mice quantify time spent interacting with food pellet immediately after presentation of a chow pellet, we found no significant difference between the control and Cre+ mice. We unaware of any evidence indicating that the two groups should have a different baseline since the Cre insertion is not expected to alter gene expression and we are unaware of reports of a phenotype relating to feeding and the presence of the transgene in this mouse line. Even if there were a small baseline shift this would not explain the large abrupt shift induced by the photostimulation. As noted above, we saw shifts in behavior abruptly induced by the initiation of photostimulation when compared to baseline in multiple experiments. This shift would not be explained by a hypothetical difference in the baseline behaviors of litter mates.

(5) The statistics used throughout are inappropriate. The authors use serial Mann-Whitney U tests without a description of data distributions within and across groups. Further, they do not use any overall F tests even though most of the data are presented with more than two bars on the same graph. Stats should be employed according to how the data are presented together on a graph. For example, stats for pre-stim, stim, and post-stim behavior X between Cre+ and Cre- groups should employ something like a two-way repeated measures ANOVA, with post-hoc comparisons following up on those effects and interactions. There are many instances in which one group changes over time or there could be overall main effects of genotype. Not only is serially using Mann-Whitney tests within the same panel misleading and statistically inaccurate, but it cherry-picks the comparisons to be made to avoid more complex results. It is difficult to comprehend the effects of the manipulations presented without more careful consideration of the appropriate options for statistical analysis.

We thank the reviewer for pointing this out and suggesting alterative analyses, we agree with the assessment on this topic. Therefore, we have extensively revised the statical approach to our data using the suggested approach. Reviewer 1 also made a similar comment, and we would like to point to our reply to reviewer 1’s second point in regard to what we changed and added to the new statistical analyses. Further, we have added a full table detailing the statical values for each figure to the paper.

Conceptual:

(6) What does the signal look like at the terminals in the POA? Any suggestion from the data that the projection to the POA is important?

This is an interesting question that we will pursue in future investigations into the roles of the POA. We used the projection to the POA from SuM to identify a subpopulation in SuM and we were surprised to find the extensive arborization of these neurons to many areas associated with threat responses. We focused on the cell bodies as “hubs” with many “spokes”. Extensive studies are needed to understand the roles of individual projections and their targets. There is also the hypothetical technical challenge of manipulating one projection without activating retrograde propagation of action potentials to the soma. At the current time we have no specific insights into the roles of the isolated projection to POA. Interpretation of experiments activating only “spoke” of the hub would be challenging. Simple terminal stimulation experiments are challenged by the need to separate POA projections from activation of passing fibers targeting more anterior structures of the accumbens and septum.

(7) Is this distinguishing active coping behavior without a locomotor phenotype? For example, Fig. 5I and other figure panels show a distance effect of stimulation (but see issues raised about the genotype of comparison groups). In addition, locomotor behavior is not included for many behaviors, so it is hard to completely buy the interpretation presented.

We agree with the reviewer and thank them for highlighting this fundamental challenge in studies examining active coping behaviors in rodents, which requires movement. Additionally, actively responding to threatening stressors would include increased locomotor activity. Separation of movement alone from active coping can be challenging. Because of these concerns we undertook experiments using diverse behavioral paradigms to examine the elicited behaviors and the recruitment of SuMVGLUT2+::POA neurons to stressors. We conducted experiments to directly examine behaviors evoked by photoactivation of SuMVGLUT2+::POA. In these experiments we observed a diversity of behaviors including increased locomotion and jumping but also treading/digging (Figure 4). These are behaviors elicited in mice by threatening and noxious stimuli. An Increase of running or only jumping could signify a specific locomotor effect, but this is not what was observed. Based on these behaviors, we expected to find evidence of increase movement in open field (Figure 5G-I) and light dark choice (Figure 5J-L) assays. For many of the assays, reporting distance traveled is not practical. An important set of experiments that argues against a generic increase in locomotion is the operant behavior experiments, which require the animal to engage in a learned behavior while receiving photostimulation of SuMVGLUT2+::POA neurons (Figure 6). This is particularly true for testing using a progressive ratio when the time of ongoing photostimulation is longer, yet animals actively and selectively engage the active port (Figure 6G-H). Further, we saw a shift in behavioral strategy induce by photoactivation in forced swim test (Figure 7H). Thus, activation of SUMVGLUT2+::POA neurons elicited a range of behaviors that included swimming, jumping, treading, and learned response, not just increased movement. Together these data strongly argue that SuMVGLUT2+::POA neurons do not only promote increased locomotor behavior. We interpret these data together with the data from fiber photometry studies to show SuMVGLUT2+::POA neurons are recruited during acute stressors, contribute to aversive affective component of stress, and promote active behaviors without constraining the behavioral pattern.

Regarding genotype, we address this in comments above as well but believe that clarifying the use of litter mates, the extensive use of the VGLUT2-Cre line by multiple groups, and experimental design allowing for comparison to baseline, stimulation evoked, and post stimulation behaviors within and across genotypes mitigate possible concerns relating to the genotype.

(8) What is the role of GABA neurons in the SuM and how does this relate to their function and interaction with glutamate neurons? In Supp 8, GABA neuron activation also modulates locomotion and in Fig 7 there is an effect on immobility, so this seems pretty important for the overall interpretation and should probably be mentioned in the abstract.

Thank you for noting these interesting findings. We added text to highlight these findings to the abstract. Possible roles of GABAergic neurons in SuM extend beyond the scope of the current study particularly since SuM neurons have been shown to release both GABA and glutamate (Li Y, Bao H, Luo Y, et al. 2020, Root DH, Zhang S, Barker DJ et al. 2018). GABAergic neurons regulate dentate gyrus (Ajibola MI, Wu JW, Abdulmajeed WI, Lien CC 2021), REM sleep (Billwiller F, Renouard L, Clement O, Fort P, Luppi PH 2017), and novelty processing Chen S, He L, Huang AJY, Boehringer R et al. 2020). The population of exclusively GABAergic vs dual neurotransmitter neurons in SuM requires further dissection to be understood. How they may relate to SUMVGLUT2+::POA neurons require further investigation.

Questions about figure presentation:

(9) In Fig 3, why are heat maps shown as a single animal for the first couple and a group average for the others?

Thank you for highlighting this point for further clarification. We modified the labels in the figure to help make clear which figures are from one animal across multiple trials and those that are from multiple animals. In the ambush assay each animal one had one trial, to avoid habituation to the mock predator. Accordingly, we do not have multiple trials for each animal in this test. In contrast, the dunk assay (10 trial/animal) and the shock (5 trials/animal) had multiple trials for each animal. We present data from a representative animal when there are multiple trials per animal and the aggerate data.

Why is the temporal resolution for J and K different even though the time scale shown is the same?

Thank you for noticing this error carried forward from a prior draft of the figure so we could correct it. We replaced the image in 3J with a more correctly scaled heatmap.

What is the evidence that these signal changes are not due to movement per se?

Thank you for the question. There are two points of evidence. First, all the 465 nm excitation (Ca2+ dependent) data was collected in interleaved fashion with 415 nm (isosbestic) excitation data. The isosbestic signal is derived from GCaMP emission but is independent of Ca2+ binding (Martianova E, Aronson S, Proulx CD. 2019). This approach, time-division multiplexing, can correct calcium-dependent for changes in signal most often due to mechanical change. The second piece of evidence is experimental. Using multiple cohorts of mice, we examined if the change in Ca2+ signal was correlated with movement. We used the threshold of velocity of movement seen following the ambush. We found no correlation between high velocity movements and Ca2+ signal (Figure 3K) including cross correlational analysis (Supplemental figure 5). Based on these points together we conclude the change in the Ca2+ signal in SUMVGLUT2+::POA neurons is not due to movement induced mechanical changes and we find no correlation to movement unless a stressor is present, i.e. mock predator ambush or forced swim. Further, the stressors evoke very different locomotor responses fleeing, jumping, or swimming.

(10) In Fig 4, the authors carefully code various behaviors in mice. While they pick a few and show them as bars, they do not show the distribution of behaviors in Cre- vs Cre+ mice before manipulation (to show they have similar behaviors) or how these behaviors shift categories in each group with stimulation. Which behaviors in each group are shifting to others across the stim and post-stim periods compared to pre-stim?

This is an important point. We selected behaviors to highlight in Figure4 C-E because these behaviors are exhibited in response to stress (De Boer & Koolhaas, 2003; van Erp et al., 1994). For the highlighted behaviors, jumping, treading/digging, grooming, we show baseline (pre photostimulation), stimulation, and post stimulation for Cre+ and Cre- mice with the values for each animal plotted. We show all nine behaviors as a heat map in Figure 4B. The panels show changes that may occur as a function of time and show changes induced by photostimulation.

The heatmaps demonstrate that photostimulation of SUMVGLUT2+::POA neurons causes a suppression of walking, grooming, and immobile behaviors with an increase in jumping, digging/treading, and rapid locomotion. After stimulation stops, there is an increase in grooming and time immobile. The control mice show a range of behaviors with no shifts noted with the onset or termination of photostimulation.

Of note, issues of statistics, genotype, and SABV are important here. For example, the hint that treading/digging may have a slightly different pre-stim basal expression, it seems important to first evaluate strain and sex differences before interpreting these data.

We examined the effects of sex as a biological variable in the experiments reported in the manuscript and found no differences among males and females in any of the experiments where we had enough animals in each sex (minimum of 5 mice) for meaningful comparisons. We did this by comparing means and SEM of males and females within each group (e.g. Cre+ males vs Cre+ female, Cre- males vs Cre- females) and then conducted a t-test to see if there was a difference. For figures that show time as a variable (e.g Figure 6C-E), we compared males and females with time x sex as main factors and compared them (including multiple comparisons if needed). We found no significant main effects or interactions between males and females. Because of this, and to maximize statistical power, we decided to move forward to keep males and females together in all the analyses presented in the manuscript. It is worth noting also that the core of the experimental design employed is a change in behavior caused by photostimulation. The mice are also the same strain with only difference being the modification to add an IRES and sequence for Cre behind the coding sequence of the Slc17A6 (VGLUT2) gene.

(11) Why do the authors use 10 Hz stimulation primarily? is this a physiologically relevant stim frequency? They show that they get effects with 1 Hz, which can be quite different in terms of plasticity compared to 10 Hz.

Thank you for the raising this important question. Because tests like open field and forced swim are subject to habituation and cannot be run multiple times per animal a test frequency was needed to use across multiple experiments for consistency. The frequency of 10Hz was selected because it falls within the rate of reported firing rates for SuM neurons (Farrel et al., 2021; Pedersen et al., 2017) and based on the robust but sub maximal effects seen in the real-time place preference assays. Identification of the native firing rates during stress response would be ideal but gathering this data for the identified population remains a dauting task.

(12) In Fig 5A-F, it is unclear whether locomotion differences are playing a role. Entrances (which are low for both groups) are shown but distance traveled or velocity are not.

In B, there is no color in the lower left panel. where are these mice spending their time? How is the entirety of the upper left panel brighter than the lower left? If the heat map is based on time distribution during the session, there should be more color in between blue and red in the lower left when you start to lose the red hot spots in the upper left, for example. That is, the mice have to be somewhere in apparatus. If the heat map is based on distance, it would seem the Cre- mice move less during the stim.

We appreciate the opportunity to address this question, and the attention to detail the reviewer applied to our paper. In the real time place preference test (RTPP) stimulation would only be provided while the animal was on the stimulation side. Mice quickly leave the stimulation side of the arena, as seen in the supplemental video, particularly at the higher frequencies. Thus, the time stimulation is applied is quite low. The mice often retreat to a corner from entering the stimulation side during trials using higher frequency stimulation. Changing locomotor activity along could drive changes in the number entrances but we did not find this. In regard to the heat map, the color scale is dynamically set for each of the paired examples that are pulled from a single trial. To maximize the visibility between the paired examples the color scale does not transfer between the trials. As a result, in the example for 10 Hz the mouse spent a larger amount of time in the in the area corresponding to the lower right corner of the image and the maximum value of the color scale is assigned to that region. As seen in the supplemental video, mice often retreated to the corner of the non-stimulation side after entering the stimulation side. The control animal did not spend a concentrated amount of time in any one region, thus there is a lack of warmer colors. In contrast the baseline condition both Cre+ and Cre- mice spent time in areas disturbed on both sides of arena, as expected. As a result, the maximum value in the heat map is lower and more area are coded in warmer colors allowing for easier visual comparison between the pair. Using the scale for the 10 Hz pair across all leads to mostly dark images. We considered ways to optimized visualization across and within pairs and focused on the within pair comparison for visualization.

(13) By starting with 1 hz, are the experimenters inducing LTD in the circuit? what would happen if you stop stimming after the first epoch? Would the behavioral effect continue? What does the heat map for the 1 hz stim look like?

Relatedly, it is a lot of consistent stimulation over time and you likely would get glutamate depletion without a break in the stim for that long.

Thank you for the opportunity to add clarity around this point regarding the trials in RTPP testing. Importantly, the trials were not carried out in order of increasing frequency of stimulation, as plotted. Rather, the order of trials was, to the extent possible with the number of mice, counterbalanced across the five conditions. Thus, possible contribution of effects of one trial on the next were minimized by altering the order of the trials.

We have added a heat map for the 1 Hz condition to figure 5B.

For experiments on RTPP the average stimulation time at 10Hz was less than 10 seconds per event. As a result, the data are unlikely to be affected by possible depletion of synaptic glutamate. For experiments using sustained stimulation (open field or light dark choice assays) we have no clear data to address if this might be a factor where 10Hz stimulation was applied for the entire trial.

(14) In Fig 6, the authors show that the Cre- mice just don't do the task, so it is unclear what the utility of the rest of the figure is (such as the PR part). Relatedly, the pause is dependent on the activation, so isn't C just the same as D? In G and H, why ids a subset of Cre+ mice shown?

Why not all mice, including Cre- mice?

Thank you for the opportunity to improve the clarity of this section. A central aspect of the experiments in Figure 6 is the aversiveness of SUMVGLUT2+::POA neuron photostimulation, as shown in Figure 5B-F. The aversion to photostimulation drives task performance in the negative reinforcer paradigm. The mice perform a task (active port activation) to terminate the negative reinforcer (photostimulation of SuMVGLUT2+::POA neurons). Accordingly, control mice are not expected to perform the task because SuMVGLUT2+::POA neurons are not activated and, thus the mice are not motivated to perform the task.

A central point we aim to covey in this figure is that while SuMVGLUT2+::POA neurons are being stimulated, mice perform the operant task. They selectively activated the active port (Supplemental Figure 7). As expected, control mice activate the active port at a low level in the process of exploring the arena. This diminishes on subsequent trials as mice habituate to the arena (Figure 6D). The data in Figures 6 C and D are related but can be divergent. Each pause in stimulation requires a port activation of a FR1 test but the number of port activations can exceed the pauses, which are 10 seconds long, if the animal continues to activate the port. Comparing data in Figures 6 C and D revels that mice generally activated the port two to three times for each pause earned with a trend towards greater efficiency on day 4 with more rewards and fewer activations.

The purpose of the progressive ratio test is to examine if photostimulation of SuMVGLUT2+::POA continues to drive behavior as the effort required to terminate the negative stimuli increases. As seen in Figures 6 G and H, the stimulation of SuMVGLUT2+::POA neurons remains highly motivating. In the 20-minute trial we did not find a break point even as the number of port activations required to pause the stimulation exceed 50. We do not show the Cre- mice is Figure 6G and H because they did not perform the task, as seen in Figure 6F. For technical reasons in early trials, we have fully timely time stamped data for rewards and port activations from a subset of the Cre+ mice. Of note, this contains both the highest and lowest performing mice from the entire data set.

Taken together, we interpret the results of the operant behavioral testing as demonstrating that SuMVGLUT2+::POA neuron activation is aversive, can drive performance of an operant tasks (as opposed to fixed escape behaviors), and is highly motivating.

(15) In Fig 7, what does the GCaMP signal look like if aligned to the onset of immobility? It looks like since the hindpaw swimming is short and seems to precede immobility, and the increase in the signal is ramping up at the onset of hindpaw swimming, it may be that the calcium signal is aligned with the onset of immobility.

What does it look like for swimming onset?

In I, what is the temporal resolution for the decrease in immobility? Does it start prior to the termination of the stim, or does it require some elapsed time after the termination, etc?

Thank for the opportunity to addresses these points and improve that clarity of our interpretation of the data. Regarding aligning the Ca2+ signal from fiber photometry recordings to swimming onset and offset, it is important to note that the swimming bouts are not the same length. As a result, in the time prior to alignment to offset of behaviors animals will have been swimming for different lengths of time. In Figure 7 C, we use the behavioral heat map to convey the behavioral average. Below we show the Ca2+ dependent signal aligned at the offset of hindpaw swim for an individual mouse (A) and for the total cohort (B). This alignment shows that the Ca2+ dependent signal declines corresponding to the termination of hindpaw swimming. Because these bouts last less than the total the widow shown, the data is largely included in Figure 7 C and D, which is aligned to onset. Due to the nuance of the difference is the alignment and the partial redundancy, we elected to include the requested alignment to swimming offset in the reply rather in primary figure.

Author response image 1.

Turning to the question regarding swimming onset, the animals started swimming immediately when placed in the water and maintained swimming and climbing behaviors until shifting behaviors as illustrated in Figure 7A and B. During this time the Ca2+-dependent signal was elevated but there is only one trial per animal. This question can perhaps be better addressed in the dunk assay presented in Figure 3C, F and G and Supplemental Figure 4 H and I. Here swimming started with each dunk and the Ca2+ signal increased.

Regarding the question for about figure 7I. We scored for entire periods (2 mins) in aggerate. We noted in videos of the behavior test that there was an abrupt decrease in immobility tightly corresponding to the end of stimulation. In a few animals this shift occurred approximately 15-20s before the end of stimulation. This may relate to the depletion of neurotransmitter as suggested by the reviewer.

Reviewer 3

Major points

(1) Results in Figure 1 suggested that SuM-Vglu2::POA projected not only POA but also to the diverse brain regions. We can think of two models which account for this. One is that homogeneous populations of neurons in SuM-Vglu2::POA have collaterals and innervated all the efferent targets shown in Figure 1. Another is to think of distinct subpopulations of neurons projecting subsets of efferent targets shown in Figure 1 as well as POA. It is suggested to address this by combining approaches taken in experiments for Figure 1 and Supplemental Figure 2.

Thank you for raising this interesting point. We have attempted combining retroAAV injections into multiple areas that receive projections from SUMVGLUT2+::POA neurons. However, we have found the results unsatisfactory for separating the two models proposed. Using eYFP and tdTomato expressing we saw some overlapping expressing in SuM. We are not able to conclude if this indicates separate populations or partial labeling of a homogenous populations. A third option seems possible as well. There could be a mix of neurons projecting to different combinations of downstream targets. This seems particularly difficult to address using fluorophores. We are preparing to apply additional methodologies to this question, but it extends beyond the scope of this manuscript.

(2) Since the authors drew a hypothetical model in which the diverse brain regions mediate the effect of SuM-Vglu2::POA activation in behavioral alterations at least in part, examination of the concurrent activation of those brain regions upon photoactivation of SuM-Vglu2::POA. This must help the readers to understand which neural circuits act upon the induction of active coping behavior under stress.

Thank you for raising this important point. We agree that activating glutamatergic neurons should lead to activation of post synaptic neurons in the target regions. Delineating this in vivo is less straight forward. Doing so requires much greater knowledge of post synaptic partners of SUMVGLUT2+::POA neurons. There are a number of issues that would need to be accounted for. Undertaking two color photo stimulation plus fiber photometry is possible but not a technical triviality. Further, it is possible that we would measure Ca2+ signals in neurons that have no relevant input or that local circuits in a region may shape the signal. We would also lack temporal resolution to identify mono-postsynaptic vs polysynaptic connections. Thus, we would struggle to know if the change in signal was due to the excitatory input from SuM or from a second region. At present, we remain unclear on how to pursue this question experimentally in a manner that is likely to generate clearly interpretable results.

(3) In Figure 4, "active coping behaviors" must be called "behaviors relevant to the active behaviors" or "active coping-like behaviors", since those behaviors were in the absence of stressors to cope with.

Thank you for the suggestion on how to clarify our terminology. We have adopted the active coping-like term.

(4) For the Dunk test, it is suggested to describe the results and methods more in detail, since the readers would be new to it. In particular, the mice could change their behavior between dunks under this test, although they still showed immobility across trials as in Supplemental Figure 4I. Since neural activity during the test was summarized across trials as in Figure 3, it is critical to examine whether the behavior changes according to time.

Thank you for identifying this opportunity to improve our manuscript. We have expanded and added a detailed description of the dunk test in the methods section.

As for Supplemental Figure 4I, we apologize for the confusion because the purpose of this figure is to show that mice remained mobile for the entire 30-second dunk trial. This did not appreciably change over the 10 trials. We have revised this figure to plot both immobile and mobile time to achieve greater clarity on this point.

Minor points

Typos

In Figure 1, please add a serotype of AAVs to make it compatible with other figures and their legends.

In the main text and Figure 2K, the authors used MHb/LHb and mHb/lHb in a mixed fashion. Please make them unified.

In the figure legend of Figure 6, change "SuMVGLUT2+::POA neurons drive" to "SuMVGLUT2+::POA neurons " in the title.

In line 86, please change "Retro-AAV2-Nuc-flox(mCherry)-eGFP" to "AAV5-Nuc-flox(mCherry)eGFP".

In line 80, please change "Positive controls" to "As positive controls, ".

Thank you for taking the time and making the effort to identify and call these out. We have corrected them.

Author Response

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

We are very grateful to both Reviewers, the Reviewing Editor and the Senior Editor for carefully reviewing our manuscript and for providing useful comments and suggestions that further improved the quality of our work. We appreciate that our work is perceived to substantially advance the understanding of osteoblast migration and that the experiments are found to be rigorous and to provide conclusive evidence. We also look forward to reaching a broad audience in the field. Below we provide a point-by-point response to each suggestion made by the reviewers and explain how we included their recommendations in the revised manuscript.

Public Reviews

Reviewer #1 (Public Review):

Summary:

The authors were trying to achieve that Tgif1 expression is regulated by EAK1/2 and PTH in a timedependent manner, and its roles in suppressing Pak3 for facilitating osteoblast adhesion. The authors further tried to show that the Tgif1- Pak3 signaling plays a significant role in osteoblast migration to the site of bone repair and bone remodeling.

Strengths:

• In a previous study, it was demonstrated that Tgif1 is a target gene of PTH, and the absence of Tgif1 failed to increase bone mass by PTH treatment (Saito et al., Nat Commun., 2019). In this study, the authors found that Tgif1-Pak3 signaling prompts osteoblast migration through osteoblast adhesion to prompt bone regeneration. This novel finding provides a better understanding of how Tgif1 expression in osteoblasts regulates adherence, spreading, and migration during bone healing and bone remodeling.

• The authors demonstrated that ERK1/2 and PTH regulate Tgif1 expression in a time-dependent manner and its role in suppressing Pak3 through various experimental approaches such as luciferase assay, ChIP assay, and gene silencing. These results contribute to the overall strength of the article.

We thank the reviewer for acknowledging the novelty of our findings as well as the strength of the manuscript.

Weaknesses:

• The authors need to further justify why they focused on Pak3 in the introduction by mentioning its known function for cell adhesion.

We thank the reviewer for this suggestion. We mention in the introduction that we further investigated Pak3 due to its implication in cell adhesion (page 6, lines 7-8).

• Some results indicated statistically significant but small changes. The authors need to explain in the discussion part why they believe this is the major mechanism or why there may be some other possible mechanisms.

We agree with this comment. We are confident that our work identified an important mechanism by which Tgif1 regulates cellular features of osteoblasts. However, it is certainly possible that other mechanisms may exist as well. We discuss this point in the revised manuscript (page 18, lines 16-17).

• The study does not include enough in vivo data to claim that this mechanism is crucial for bone healing and bone remodeling in vivo.

Re: We agree with this point and have modified the abstract accordingly by replacing “crucial” with “implicated in” as well as the text by changing “crucial” to “important” (page 2, line 9). Furthermore, we discuss this limitation in the revised manuscript (page 18, lines 9-14).

Reviewer #2 (Public Review):

Summary:

Bolamperti S. et al. 2023 investigate whether the expression of TG-interacting factor (Tgif1) is essential for osteoblastic cellular activity regarding morphology, adherence, migration/recruitment, and repair. Towards this end, germ-line Tgif1 deletion (Tgif1-/-) mice or male mice lacking expression of Tgif1 in mature osteoblastic and osteocytic cells (Dmp1-Cre+; Tgif1fl/fl) and corresponding controls were studied in physiological, bone anabolic, and bone fracture-repair conditions. Both Tgif1-/- and Dmp1-Cre+; Tgif1fl/fl exhibited decreased osteoblasts on cancellous bone surfaces and adherent to collagen I-coated plates. Tgif1-/- mice exhibit impaired healing in the tibial midshaft fracture model, as indicated by decreased bone volume (BV/Cal.V), osteoid (OS/BS), and low osteoblasts (number and surface). Likewise, both Tgif1-/- and Dmp1-Cre+; Tgif1fl/fl show impaired PTH 1-34, (100µg/kg, 5x/wk for 3 wks) osteoblast activation in vivo, as detected by increases in quiescent bone surfaces. Mechanistic in vitro studies then utilized primary osteoblasts isolated from Tgif1-/- mice and siRNA Tgif1 knockdown OCY454 cells to further investigate and identify the downstream Tgif1 target driving these osteoblastic impairments. In vitro, Tgif1-/- osteoblastic and Tgif1 knockdown OCY454 cells exhibit decreased migration, abnormal morphology, and decreased focal adhesions/cells. Unexpectantly though, localization assays revealed Tgif1 to primarily concentrate in the nucleus and not to co-localize with focal adhesions (paxillin, talin). Also, the expression of major focal adhesion components (paxillin, talin, FAK, Src, etc.) or the Cdc42 family was not altered by loss of Tgif1 expression. In contrast, PAK3 expression is markedly upregulated by loss of Tgif1. In silico analysis followed by mechanistic molecular assays involving ChIP, siRNA (Tgif1, PAK3), and transfection (rat PAK3 promoter) techniques show that Tgif1 physically binds to a specific site in the PAK3 promoter region. Further, the knockdown of PAK3 rescues the Tgif1-deficient abnormal morphology in OCY454 cells. This is the first study to identify the novel transcriptional repression of PAK3 by Tgif1 as well as the specific Tgif1 binding site within the PAK3 promoter.

Strengths:

This work has a plethora of strengths. The co-authors achieved their aim of eliciting the role of Tgif1 expression in osteoblastic cellular functions (morphology, spreading/attachment, migration).

Further, this work is the first to depict the novel mechanism of Tgif1 transcriptional repression of PAK3 by a thorough usage of mechanistic molecular assays (in silico analysis, ChIP, siRNA, transfection etc.). The conclusions are well supported and justified by these findings, as the appropriate controls, sample sizes (statistical power), statistics, and assays were fully utilized. The claims and conclusions are justified by the data.

Re: We are grateful to this reviewer for recognizing the novelty, strengths, and rigor of our study and for acknowledging that the data convincingly support the conclusions drawn.

Weaknesses:

The discussion section could be expanded with a few sentences regarding limitations to the current study and potential future directions.

Re: In the revised manuscript, we are discussing limitations of the work and describe possible future directions (page 18, line 9-14).

Recommendations For The Authors:

Reviewer #1 (Recommendations For The Authors):

(1) The cell spreading and migration assay is quite artificial. Trypsinized osteoblasts and quiescent osteoblasts are totally different. The authors need to cite papers from other groups to justify whether the cell spreading and migration assay is appropriate to achieve the goals of this study.

Re: The reviewer is right that in vitro assays are often artificial and do not necessarily fully reflect in vivo situations. We have taken this aspect into account and discuss it in the revised manuscript (page 18, lines 9-10). In addition, we have included references from other groups who have used similar assays to study cell spreading and migration (Dejaeger M et al., 2017 and Dang et al., 2018).

(2) Page 13 Line 15: The statement "Osteoblasts are greatly impaired in the ability to migrate into the repair zone" is an overstatement. The experiments in Figure 5 do not necessarily reflect osteoblast migration activities. The authors need to rephrase the sentence or need to show observation of earlier time points (e.g., 1 week after fracture) in their bone healing experiments. The number of osteoblasts/surface in Tgif1+/+ and Tgif1-/- mice at different time points during bone healing should be a good indicator for the migration of osteoblasts to the repair site.

Re: We understand the critique that a time course or lineage tracing experiments would provide better evidence for the statement of osteoblast migration into the repair zone. To avoid overinterpretations we have removed the sentence from the revised manuscript.

(3) Page 14, Line 24: Regarding the sentence "The observation that Tgif1 is crucial for osteoblast adherence, spreading, and migration", the authors need to clearly mention this statement is based on the in vitro experiments. The animal studies are not enough to claim that the mechanism is crucial for adherence, spreading, and migration.

Re: We thank the reviewer for pointing out this limitation. We have clarified that the finding that Tgif1 is crucial for osteoblast adherence, spreading and migration was made in vitro (page 14, line 22).

(4) The authors need to demonstrate the suppression of Pak3 expression in PTH-treated mice in vivo, in addition to the in vitro culture system (Fig. 7C and 7D).

Re: We agree with the reviewer that this experiment would be very insightful. However, this is beyond the scope of the current work. Nevertheless, to take this valid point into consideration, we mention it in the discussion as potential future direction (page 18, lines 11-14).

(5) The authors need to demonstrate that the pharmacologic suppression of Pak3 in Tgif1-/- mice reduces the % of quiescent surface/BS in vivo.

Re: This point is also well taken, and we agree that a suppression of Pak3 in Tgif1-deficient mice would be very informative to support our in vitro findings. However, this may also be part of future investigations. This is emphasized in the discussion of the revised manuscript (page 18, lines 11-14).

Figures (Minor)

Fig. 1:

Fig. 1A

Arrows need to indicate a more precise position.

Re: The position of the arrows has been optimized.

Fig. 1DE

What are blue/red bars (genotypes)?

Re: The colors indicate the genotypes. A legend has been added to the revised figure.

Fig. 1K

Quantification data is needed.

Re: Thank you for this suggestion. We added a quantification of the data (Fig. 1L, M; page 8, lines 3-4; page 21, lines 5-6)

Fig. 2A

Show the representative high-magnification image of round (non-spread) cells.

Re: Representative high-magnification images (insets) are provided in the revised figure 2A.

Fig. 5

Red arrows need to indicate a more precise position.

Re: The arrows have been repositioned.

Fig. 6A, C

Red arrows need to indicate a more precise position.

Re: The arrows have been repositioned.

Reviewer #2 (Recommendations For The Authors):

(1) The microscopy images and analyses are excellent.

Re: We thank the reviewer for acknowledging the quality of our microscopy studies.

(2) Since the Tgif1-/- mouse has low osteoclast numbers, is it possible that this is a contributing factor to the delays/impairment in bone healing, given that resorption also has a role in fracture repair? Since the focus of these studies is on osteoblastic cells, this point is a little out of scope. However, would the authors consider exploring this further in the discussion section?

Re: This point is well taken by the reviewer, and we agree that osteoclasts could certainly play a role in the impaired fracture healing. To acknowledge this aspect, we followed the recommendation and discuss this aspect in the revised manuscript (page 16, lines 22-24).

Revisions

Would the authors consider slightly re-wording the title? Tgif1 suppresses PAK3 expression; however, Tgif1-deficiency leads to the unregulated elevation of PAK3 expression.

Re: Thank you for pointing this out. We agree with the reviewer and adapted the title accordingly.

Suggestions

(1) Is it possible that apoptosis and/or anoikis is being induced by Tgif1 deficiency in osteoblastic cells?

Re: We do not have data towards this direction and although Tgif1-deficient osteoblasts are overall viable and well expanding, we cannot fully exclude this possibility.

(2) For the fracture study, any differences in overall callus size? Would it be possible to perform micro-CT imaging with some of these samples?

Re: There is no difference in non-mineralized callus size between Tgif1+/+ and Tgif1-/- mice. However, there is less mineralized bone per callus area in Tgif1-/- mice, confirming an impaired osteoblast phenotype. As suggested by the reviewer, we added representative micro-CT images and the respective information to the revised manuscript (Fig 5F; pages 19-20).

(3) Fracture repair experiment-is PAK3 expression downregulated with fracture injury; and/or, is PAK3 upregulated by loss of Tgif1 expression?

Re: Unfortunately, we do not have data to answer this very interesting question and it would need to be addressed in future studies. This is mentioned in the revised discussion (page 18, lines 12-14).

(4) Fig 7F. within PTH treated cells, is the light blue SCR sphericity statistically different than the light green siTgif1 + siPAK3 ? While the statement of the "lack of both, Tgif1 and PAK3 prevented PTH-induced decrease in cell sphericity" is supported by the lack of differences between dark green vs. light green; is it also possible that this is due to the siPAK3 returning sphericity to control (scr) levels? (i.e. hitting a floor limit of detection).

Re: We thank the reviewer for this thoughtful question. There is no statistically significant difference between light blue and light green. Silencing PAK3 restores the impaired capacity to spread that occurs in the absence of Tgif1 to the level of scr controls (significant difference between dark and light red vs. dark and light green and no difference between either dark or light blue vs. dark or light green). However, unlike in the (scr) controls, in the absence of both Tgif1 and PAK3, the cells do not respond to PTH (statistically significant difference between dark and light blue, no difference between dark and light green). Based on the data, cells can reach sphericity of less than 0.2 and thus it is unlikely that sphericity is “hitting the floor level of detection” in these groups.