I think he may be Richard Wendler instead.

J. J. Short Platen Day!!! My First Swap and Review- Test and Talk by [[The HotRod Typewriter Co.]]

@misc{gosun_2024, title={Forget Public Charging! Power Your Electric Car with the SUN | GoSun}, url={https://www.youtube.com/watch?v=GWMkb-enEV4}, journal={YouTube}, author={GoSun}, year={2024}, month={Jun} }

wait a sec he WANTED this??

now thats a fun dynamic

he looks good bruh

cursed them :(

oh!

and your betraying him now..

uh ok? i mean both asha and robb are everything theon wants to be

others

interesting...

like shireen's fool...

cuz he's out sailing to dany i think

stfu

NO YOU ARE NOT

interesting

song of ice and fire prophecy?

another sellout

new god mentioned

stfu

Tumorigenesis: conversion of a healthy cell into cancer cells.

Oncogene activation disrupts canonical cellular pathways, leading to genome instability.

Summary of Eric Normand's Talk: Building Composable Abstractions

• Introduction

  • Eric Normand introduces himself and the purpose of the talk: "The title of this talk is building composable abstractions...to develop a process to do that and also I'd like to start a discussion about how we can do that better."

• Importance of Abstractions

  • Abstractions are critical for creating complex applications from small, simple problems. "A lot of people are able to solve small problems like Fibonacci...when they finally want to create an app...they don't know how to take the tools that they've learned and turn them into software."

• Map of the Talk

  • The talk covers the importance of abstractions, the process of developing them, an example, and concluding thoughts. "Here's sort of the map of the talk: why focus on abstractions, the process, an example abstraction, and concluding thoughts."

• Why Focus on Abstractions?

  • Refactoring introduces the distinction between the behavior of the code and its implementation. "In the general industry we now have this idea that there's a difference between the behavior of the code and the actual implementation."
  • Example of Newtonian mechanics replacing Aristotelian physics illustrates that some systems can't be refactored but need to be redesigned from scratch. "You can't refactor Aristotle into Newton."

• Objectives of the Abstraction Process

  • The process should produce good, Newtonian-style abstractions, be iterative, accessible to all, and foster collaboration. "It has to consistently produce good abstractions...an iterative process...anyone can do it...fosters collaboration."

• Example of Vector Graphics System

  • Normand uses a simple vector graphics system as an example to demonstrate the process of building abstractions. "This is the example we're going to develop: a vector graphics system."

• Step 1: Physical Metaphor

  • Choose a metaphor to capture important information. "The idea behind this is to choose a metaphor that will capture the important information in your program."
  • Shapes and construction paper is chosen as the metaphor. "Shapes and construction paper...I cut out shapes like rectangles and ellipses...and then I can move them around."

• Step 2: Meaning Construction

  • Convert physical intuition into precise mathematical language, focusing on the interface. "We're going to be focusing on the interface right now...precise mathematical language."
  • Definitions in Clojure for different components like color, shape, and transformations. "We're defining two types here: cutout and shape...defining a function that takes a cutout and returns a shape."
  • Importance of preserving shape and color, overlay order, and rotation and translation independence. "Preservation of shape...preservation of color...overlay order...rotation and translation independence."

• Step 3: Implementation

  • Implement the system based on the constructed meaning, ensuring it can be refactored to different requirements like SVG output. "Implementation...we already know what to do...refactor from quill to SVG."

• Summary of Process

  • Use a physical metaphor, define the parts and their relationships in mathematical language, and refactor for implementation details. "Use a physical metaphor...define the parts and their relationships...refactor to get all the meta properties."

• Corollaries for the Process

  • Know your domain, constructs, and refactoring techniques. "Know your domain...know your constructs...know your refactoring."

• Conclusion

  • Encourages further learning and provides resources. "Please go to my site...download the slides...sign up for my newsletter."

Hi! My name is Olivia Burns. My favorite pizza toppings are pepperoni and sausage. I am group Puka Puka.

My name is Caleb, group Aitutaki, favorite toppings are plain ole beef or sausage

I recommend making a habit of looking up some background on someone before dedicating a big chunk of time ingesting their ideas. Ray Dalio is a billionaire hedge fund manager. He is somewhere between the 88th and 123rd richest person on Earth, depending on whether you believe Forbes or Bloomberg. He has no formal education or experience in history, sociology, or political science.

지지

성경

독실한

조화

a usually general or lasting direction of thought, inclination, or interest 지향

반대

freedom of action or choice

It's just a chatbot, who cares?

What happens when they are given the capacity to take actions for us in the real world? Write posts for you "in your voice"? Send out pictures? Spend your money? Pay your bills? Have medicine delivered to your house? Upgrade your car software?

There are also cases where the decision-maker believes he has an escape route for him and his descendants which increases his willingness to take risks.

According to ChatGPT: "Considering these factors, most people in large cities could face severe dehydration and health risks within 3 to 5 days if no alternative water sources or emergency measures are provided."

If the attack is not widespread, resources will come from other geographies. If it is very general, help is not coming.

Self-identified; family does not require children.

search up definition and difference

why does this need to be run with the i flag

wait how is this possible

flatbuffers::GenerateTextFile -> flatbuffers::GenTextFile

这个是指啥

• Connell emphasizes the importance of abstraction in software design, focusing on libraries over individual programs: "the main job of a software designer... is to build abstractions."

• He introduces the concept of "denotational design," inspired by the aesthetics of code poetry: "trying to formulate a semantics for functional reactive programming... led to what's now my central consciously Central Paradigm for doing Library design."

• Connell outlines three main goals in software projects: building abstractions, implementing those abstractions, and documenting them clearly: "I want my abstractions to be precise, elegant, and reusable."

• He quotes Edgar Dijkstra to highlight the need for precision in abstractions: "The purpose of abstraction is not to be vague... it's to create a whole new semantic level in which one can be absolutely precise."

• Connell suggests a three-pronged approach to software projects: focusing on precision, elegance, and reusability in abstractions; correctness, speed, and maintainability in implementations; and simplicity, clarity, and accuracy in documentation: "I want my program my implementation to be maintainable."

• He critiques the current state of software development for its emphasis on precision in "how" rather than "what": "software without a specification is an answer without a question."

• Bertrand Russell's philosophy on precision is highlighted to argue that informal understandings in software often lead to errors: "everything is vague to a degree you do not realize until you've tried to make it precise."

• Connell advocates for specifying clear semantics in software to avoid ambiguity and errors: "if you think you know what you're doing but you haven't made precise, you're probably wrong."

• He introduces "denotational design" as a methodology that ensures precise, simple, and compelling specifications: "denotational design... gives us precise, simple and compelling specifications."

• Connell discusses the advantages of continuous, infinite models in functional programming, as opposed to discrete, finite models: "locations are functions in two-dimensional continuous space infinite continuous space."

• He proposes the use of standard algebraic abstractions like functor, monoid, and applicative to simplify and generalize software design: "standard algebraic abstractions things like functor and monoid and implicative."

• The talk covers the process of refining an API for image synthesis by identifying and generalizing operations: "we're going to design functional apis... for image synthesis and manipulation."

• Connell stresses the importance of compositionality in software design, where the meaning of compositions is defined in terms of the meanings of their components: "the meaning of a composition... is defined in terms of only the meanings of the components."

• He refines the design by leveraging standard abstractions and type classes, illustrating with examples like monoids, functors, and applicatives: "standard vocabulary in Haskell means type classes... comes with laws that hold."

• Connell concludes with examples showing how denotational design can lead to precise, efficient, and maintainable software, encouraging further exploration and application of these principles: "examples of how you can calculate a correct implementation because in this case the meaning function has an inverse."

Author response:

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

In addition to our responses to reviewer suggestions below, a minor bug in the calculation of CAIS was brought to our attention by a reader of our preprint. We have corrected this bug and rerun analyses, whose results became slightly stronger as noise was removed. While we were doing that, someone pointed out to us that our equations were almost the same as Kullback-Leibler divergence, which explains why our metric performed so well. We have made the numerically trivial (see before vs. after image below) mathematical change to use Kullback-Leibler divergence instead, and now have a better story, with a solid basis in information theory, as to why CAIS works.

Author response image 1.

Unfortunately, we discovered a second bug that caused our PIC correction code to fail to perform the needed correction for phylogenetic confounding. The previously reported correlation between CAIS (or ENC) with body mass no longer survives PIC-correction. We have therefore removed this analysis from the manuscript. Our story now stands more on the theoretical basis of CAIS and ENC than on the post facto validation than it previously did. We now also present CAIS and ENC on a more equal footing. ENC results are slightly stronger, while CAIS has the complementary advantage of correcting for amino acid frequencies.

The work involved in these changes, as well as some of the responses to reviews below, justifies changing the second author into a co-first author, and adding an additional coauthor (Hanon McShea) who discovered the second bug.

Reviewer #1 (Public Review): 

In this manuscript, the authors propose a new codon adaptation metric, Codon Adaptation Index of Species (CAIS), which they present as an easily obtainable proxy for effective population size. To permit between-species comparisons, they control for both amino acid frequencies and genomic GC content, which distinguishes their approach from existing ones. Having confirmed that CAIS negatively correlates with vertebrate body mass, as would be expected if small-bodied species with larger effective populations experience more efficient selection on codon usage, they then examine the relationship between CAIS and intrinsic structural disorder in proteins. 

The idea of a robust species-level measure of codon adaptation is interesting. If CAIS is indeed a reliable proxy for the effectiveness of selection, it could be useful to analyze species without reliable life history- or mutation rate data (which will apply to many of the genomes becoming available in the near future). 

A key question is whether CAIS, in fact, measures adaptation at the codon level. Unfortunately, CAIS is only validated indirectly by confirming a negative correlation with body mass. As a result, the observations about structural disorder are difficult to evaluate. 

As discussed in the preamble above, we have replaced the body mass validation with a stronger theoretical basis in information theory.

A potential problem is that differences in GC between species are not independent of life history. Effective population size can drive compositional differences due to the effects of GC-biased gene conversion (gBGC). As noted by Galtier et al. (2018), genomic GC correlates negatively with body mass in mammals and birds. It would therefore be important to examine how gBGC might affect CAIS, and to what extent it could explain the relationship between CAIS and body mass. 

Suppose that gBGC drives an increase in GC that is most pronounced at 3rd codon positions in highrecombination regions in small-bodied species. In this case, could observed codon usage depart more strongly from expectations calculated from overall genomic GC in small vertebrates compared to large ones? The authors also report that correcting for local intergenic GC was unsuccessful, based on the lack of a significant negative relationship with body mass (Figure 3D). In principle, this could also be consistent with local GC providing a relatively more appropriate baseline in regions with high recombination rates. Considering these scenarios would clarify what exactly CAIS is capturing. 

Figure 3 (previously Supplementary Figures S5A and S5B) shows that CAIS is negligibly correlated with %GC (not robust to multiple comparisons correction), and ENC not at all. We believe this is evidence against the possibility brought up by the reviewer, i.e. that Ne might affect gBGC (and hence global %GC). This relationship, if present, could act as a confounding effect, but it is not present within our species dataset. 

Note that we expect our genomic-GC-based codon usage expectations to reflect unchecked gBGC in an average genomic region, independently of whether that species has high or low Ne. Our working model is that non-selective forces, include gBGC as well as conventional mutation biases, vary among species, and that they rather than selection determine each species’ genome-wide %GC. By correcting for genome-wide %GC, CAIS and ENC correct for both mutation bias and gBGC, in order to isolate the effects of selection.

This argument, based on an average genomic region, is vulnerable to gene-rich genomic regions having differentially higher recombination rates and hence GC-biased gene conversion. However, we do not see the expected positive correlation between |𝐥𝐨𝐜𝐚𝐥 𝐆𝐂 - global GC| and CAIS (see new Figure 5), again suggesting that gene conversion strength is not a confounding factor acting on CAIS.

Given claims about "exquisitely adapted species", the case for using CAIS as a measure of codon adaptation would also be stronger if a relationship with gene expression could be demonstrated. RSCU is expected to be higher in highly expressed genes. Is there any evidence that the equivalent GCcontrolled measure behaves similarly? 

Correlations with gene expression are outside the scope of the current work, which is focused on producing and exploiting a single value of codon adaptation per species. It is indeed possible that our general approach of using Kullback-Leibler divergence to correct for genomic %GC could be useful in future work investigating differences among genes.  

The manuscript is overall easy to follow, though some additional context may be helpful for the general reader. A more detailed discussion of how this work compares to the approach taken by Galtier et al. (2018), which accounted for GC content and gBGC when examining codon preferences, would be appropriate, for example. In addition, it would have been useful to mention past work that has attempted to explicitly quantify selection on codon usage. 

One key difference between our work and that of Galtier et al. 2018 is that our approach does not rely on identifying specific codon preferences as a function of species. Our approach might therefore be robust to scenarios where different genes have different codon preferences (see Gingold et al. 2014 https://doi.org/10.1016/j.cell.2014.08.011). At a high level, our results are in broad agreement with those of Galtier et al., 2018, who found that gBGC affected all animal species, regardless of Ne, and who like us, found that the degree of selection on codon usage depended on Ne.

Reviewer #2 (Public Review): 

Summary 

The goal of the authors in this study is to develop a more reliable approach for quantifying codon usage such that it is more comparable across species. Specifically, the authors wish to estimate the degree of adaptive codon usage, which is potentially a general proxy for the strength of selection at the molecular level. To this end, the authors created the Codon Adaptation Index for Species (CAIS) that controls for differences in amino acid usage and GC% across species. Using their new metric, the authors find a previously unobserved negative correlation between the overall adaptiveness of codon usage and body size across 118 vertebrates. As body size is negatively correlated with effective population size and thus the general strength of natural selection, the negative correlation between CAIS and body size is expected. The authors argue this was previously unobserved due to failures of other popular metrics such as Codon Adaptation Index (CAI) and the Effective Number of Codons (ENC) to adequately control for differences in amino acid usage and GC content across species. Most surprisingly, the authors also find a positive relationship between CAIS and the overall "disorderedness" of a species protein domains. As some of these results are unexpected, which is acknowledged by the authors, I think it would be particularly beneficial to work with some simulated datasets. I think CAIS has the potential to be a valuable tool for those interested in comparing codon adaptation across species in certain situations. However, I have certain theoretical concerns about CAIS as a direct proxy for the efficiency of selection $sN_e$ when the mutation bias changes across species.  

Strengths 

(1) I appreciate that the authors recognize the potential issues of comparing CAI when amino acid usage varies and correct for this in CAIS. I think this is sometimes an under-appreciated point in the codon usage literature, as CAI is a relative measure of codon usage bias (i.e. only considers synonyms). However, the strength of natural selection on codon usage can potentially vary across amino acids, such that comparing mean CAI between protein regions with different amino acid biases may result in spurious signals of statistical significance (see Cope et al. Biochemica et Biophysica Acta - Biomembranes 2018 for a clear example of this). 

We now cite Cope et al. as an example of how amino acid composition can act as a confounding factor.

(2) The authors present numerous analysis using both ENC and mean CAI as a comparison to CAIS, helping given a sense of how CAIS corrects for some of the issues with these other metrics. I also enjoyed that they examined the previously unobserved relationship between codon usage bias and body size, which has bugged me ever since I saw Kessler and Dean 2014. The result comparing protein disorder to CAIS was particularly interesting and unexpected. 

Unfortunately, our previous PIC correction code was buggy, and in fact the relationship with body size does not survive PIC correction (although it is strong prior to PIC correction). We have therefore removed it from the paper. However, the more novel result on protein disorder remains strong.

(3) The CAIS metric presented here is generally applicable to any species that has an annotated genome with protein-coding sequences. 

Weaknesses 

(1) The main weakness of this work is that it lacks simulated data to confirm that it works as expected. This would be particularly useful for assessing the relationship between CAIS and the overall effect of protein structure disorder, which the authors acknowledge is an unexpected result. I think simulations could also allow the authors to assess how their metric performs in situations where mutation bias and natural selection act in the same direction vs. opposite directions. Additionally, although I appreciate their comparisons to ENC and mean CAI, the lack of comparison to other popular codon metrics for calculating the overall adaptiveness of a genome (e.g. dos Reis et al.'s $S$ statistic, which is a function of tRNA Adaptation Index (tAI) and ENC) may be more appropriate. Even if results are similar to $S$, CAIS has a noted advantage that it doesn't require identifying tRNA gene copy numbers or abundances, which I think are generally less readily available than genomic GC% and protein-coding sequences. 

The main limitation of dos Reis’s test in our view is that, like the better versions of CAI, it requires comparable orthologs across species. See also the discussion below re the benefits of proteome-wide approach. We now also note the advantage of not needing tRNA gene copy numbers and abundances. 

Simulated datasets would be great, but we think it a nice addition rather than must-have, in particular because we are skeptical about whether our understanding of all relevant processes is good enough such that simulations would add much to our more heuristic argument along the lines of Figure 2. E.g. the complications of Gingold et al. 2014 cited above are pertinent, but incorporating them would make simulations quite involved. Instead, we now have a stronger theoretical justification for CAIS grounded in information theory. We have significantly expanded discussion of Figure 2 to give a clearer idea of the conceptual underpinnings of CAIS and ENC.

The authors mention the selection-mutation-drift equilibrium model, which underlies the basic ideas of this work (e.g. higher $N_e$ results in stronger selection on codon usage), but a more in-depth framing of CAIS in terms of this model is not given. I think this could be valuable, particularly in addressing the question "are we really estimating what we think we're estimating?" 

Let's take a closer look at the formulation for RSCUS. From here on out, subscripts will only be used to denote the codon and it will be assumed that we are only considering the case of $r = genome$ for some species $s$. 

\begin{align*} 

RSCUS_i &= \frac{O_i}{E_i} 

\end{align*} 

I think what the authors are attempting to do is "divide out" the effects of mutation bias (as given by $E_i$), such that only the effects of natural selection remain, i.e. deviations from the expected frequency based on mutation bias alone represent adaptive codon usage. Consider Gilchrist et al. MBE 2015, which says that the expected frequency of codon $i$ at selection-mutation-drift equilibrium in gene $g$ for an amino acid with $N_a$ synonymous codons is 

\begin{align} 

E_{i,g} &= \frac{e^{-\Delta M_i-\Delta\eta_i\phi_g}}{\sum_{j=1}^{N_a}e^{-\Delta M_j\Delta\eta_j\phi_g}} 

\end{align} 

where$\Delta M$ is the mutation bias, $\Delta\eta$ is the strength of selection scaled by the strength of drift, and $\phi_g$ is the gene expression level of gene $g$. In this case, $\Delta M$ and $\Delta\eta$ reflect the strength and direction of mutation bias and natural selection relative to a reference codon, for which $\Delta M_ref, \Delta\eta_ref = 0$. Assuming the selection-mutation-drift equilibrium model is generally adequate to model the true codon usage patterns in a genome (as I do and I think the authors do, too), the $E_{i,g}$ could be considered the expected observed frequency codon $i$ in gene $g$ $E[O_{i,g}]$. 

Let's re-write the $E_i = \frac{p_i}{\sum_{j=1}^{N_a}p_j}$ in the form of Gilchrist et al., such that it is a function of mutation bias $\Delta M$. For simplicity, we will consider just the two-codon case and assume the amino acid sequence is fixed. Assuming GC% is at equilibrium, the term $g_r$ and $1 - g_r$ can be written as 

\begin{align*} 

g_r &= \frac{\mu_{AT\rightarrow GC}}{\mu_{AT\rightarrow GC} + \mu_{GC\rightarrow AT}} \1 - g_r & = \frac{\mu_{GC\rightarrow AT}}{\mu_{AT\rightarrow GC} + \mu_{GC\rightarrow AT}}  \end{align*} 

where $\mu_{x\rightarrow y}$ is the mutation rate from nucleotides $x$ to $y$. As described in Gilchrist et al. MBE 2015 and Shah and Gilchrist PNAS 2011, the mutation bias $\Delta M_{NNA,NNG} = log(\frac{\mu_{AT\rightarrow GC}}{\mu_{GC\rightarrow AT}})$. This can be expressed in terms of the equilibrium GC content by recognizing that 

\begin{align*} 

\frac{g_r}{1-g_r} &= \frac{\mu_{AT\rightarrow GC}}{\mu_{GC\rightarrow AT}} \\implies \frac{g_r}{1-g_r} &= e^{\Delta M} 

\end{align*} 

As we are assuming the amino acid sequence is fixed, the probability of observing a synonymous codon $i$ at an amino acid becomes just a Bernoulli process. 

\begin{align*} 

p_i &= g_r^x(1-g_r)^{(1-x)} 

\end{align*} 

If we do this, then 

\begin{align} 

E_{NNA} &= \frac{p_{NNA}}{p_{NNA} + p_{NNG}} \ &= \frac{1-g_r}{g_r + (1-g_r)} \ &=

\frac{1}{\frac{g_r}{1-g_r} + 1} \ &= \frac{1}{e^{\Delta M} + 1} \ & = \frac{e^{-\Delta M}}{1 + e^{-\Delta M}} 

\end{align} 

Recall that in the Gilchrist et al. framework, the reference codon has $\Delta M_{NNG,NNG} = 0 \implies e^{-\Delta M_{NNG,NNG}} = 1$. Thus, we have recovered the Gilchrist et al. model from the formulation of $E_i$ under the assumption that natural selection has no impact on codon usage and codon NNG is the pre-defined reference codon. To see this, plug in 0 for $\Delta\eta$ in equation (1). 

We can then calculate the expected RSCUS using equation (1) (using notation $E[O_i]$) and equation (6) for the two codon case. For simplicity assume, we are only considering a gene of average expression (defined as $\phi_g = 1$). 

Assume in this case that NNG is the reference codon ($\Delta M_{NNG},\Delta\eta_{NNG} = 0$). 

\begin{align} 

E[RSCUS_{NNA}] &= \frac{E[O_{NNA}]}{E_{NNA}} \ &= \frac{e^{-\Delta\eta_{NNA}}(e^{-\Delta

M_{NNA}}+e^{-\Delta M_{NNG}})}{e^{-\Delta M_{NNA}-\Delta\eta_{NNA}} + e^{-\Delta M_{NNG}\Delta\eta_{NNG}}} \ & = \frac{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{-\Delta M_{NNG} -

\Delta\eta_{NNA}}}{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{-\Delta M_{NNG} -

\Delta\eta_{NNG}}} \ &= \frac{e^{-\Delta M_{NNA} - \Delta\eta_{NNA}} + e^{- \Delta\eta_{NNA}}}{e^{\Delta M_{NNA} - \Delta\eta_{NNA}} + 1} 

\end{align} 

This shows that the expected value of RSCUS for a two-codon amino acid is expected to increase as the strength of selection $\Delta\eta$ increases, which is desired. Note that $\Delta\eta$ in Gilchrist et al. is formulated in terms of selection *against* a codon relative to the reference, such that a negative value represents that a codon is favored relative to the reference. If $\Delta\eta = 0$ (i.e. selection does not favor either codon), then $E[RSCUS] = 1$. Also note that the expected RSCUS does not remain independent of the mutation bias. This means that even if $sN_e$ (i.e. the strength of natural selection) does not change between species, changes to the strength and direction of mutation bias across species could impact RSCUS. Assuming my math is right, I think one needs to be cautious when interpreting CAIS as representative of the differences in the efficiency of selection across species except under very particular circumstances. One such case could be when it is known that mutation bias varies little across the species of interest. Looking at the species used in this manuscript, most of them have a GC content ranging around 0.41, so I suspect their results are okay. 

Although I have not done so, I am sure this could be extended to the 4 and 6 codon amino acids. 

We thank Reviewer 2 for explicitly laying out the math that was implicit in our Figures 1 and 2. While we keep our more heuristic presentation, our revised manuscript now more clearly acknowledges that the per-site codon adaptation bias depicted in Figure 1 has limited sensitivity to s*Ne. The reason that we believe our approach worked despite this, is that we think the phenomenon is driven by what is shown in Figure 2. I.e., where Ne makes a difference is by determining the proteome-wide fraction of codons subject to significant codon adaptation, rather than by determining the strength of codon adaptation at any particular site or gene. We have made multiple changes to the texts to make this point clearer.

Another minor weakness of this work is that although the method is generally applicable to any species with an annotated genome and the code is publicly available, the code itself contains hard-coded values for GC% and amino acid frequencies across the 118 vertebrates. The lack of a more flexible tool may make it difficult for less computationally-experienced researchers to take advantage of this method. 

Genome-wide %GC values are hard-coded because they were taken from the previous study of James et al. (2023) https://doi.org/10.1093/molbev/msad073. As summarized in the manuscript, genome-wide %GC was a byproduct of a scan of all six reading frames across genic and intergenic sequences available from NCBI with access dates between May and July 2019. The more complicated code used to calculate the intergenic %GC, and the code used to calculate amino acid frequencies is located at https://github.com/MaselLab/CodonAdaptation-Index-of-Species. Luckily, someone else just wrote a simpler end to end pipeline for us, on the basis of our preprint. We now note this in the Acknowledgements, and link to it: https://github.com/gavinmdouglas/handy_pop_gen/blob/main/CAIS.py.

eLife assessment

This study develops a useful metric for quantifying codon usage adaptation - the Codon Adaptation Index of Species (CAIS). This metric permits direct comparisons of the strength of selection at the molecular level across species. The study is based on solid evidence, and the authors identify relationships between CAIS and the presence of disordered protein domains. Other correlations, such as the one between CAIS and body size, are weak and non-significant. In summary, the study introduces an interesting new approach to quantifying codon usage across species, which may be helpful in attempts to measure selection at the molecular level.

Reviewer #2 (Public Review):

MotorNet aims to provide a unified interface where the trained RNN controller exists within the same TensorFlow environment as the end effectors being controlled. This architecture provides a much simpler interface for the researcher to develop and iterate through computational hypotheses. In addition, the authors have built a set of biomechanically realistic end effectors (e.g., a 2 joint arm model with realistic muscles) within TensorFlow that are fully differentiable.

MotorNet will prove a highly useful starting point for researchers interested in exploring the challenges of controlling movement with realistic muscle and joint dynamics. The architecture features a conveniently modular design and the inclusion of simpler arm models provides an approachable learning curve. Other state-of-the-art simulation engines offer realistic models of muscles and multi-joint arms and afford more complex object manipulation and contact dynamics than MotorNet. However, MotorNet's approach allows for direct optimization of the controller network via gradient descent rather than reinforcement learning, which is a compromise currently required when other simulation engines (as these engines' code cannot be differentiated through).

The paper has been reorganized to provide clearer signposts to guide the reader. Importantly, the software has been rewritten atop PyTorch which is increasingly popular in ML and computational neuroscience research.

eLife assessment

This work will be of interest to the motor control community as well as neuroAI researchers interested in how bodies constrain neural circuit function. The authors present "MotorNet", a useful software package to train artificial neural networks to control a biomechanical model of an effector. The manuscript provides solid evidence that MotorNet is easy to use and can reproduce past results in the field, both at the neural and behavioural levels. Validation is limited to planar arm-like plants or point-masses, so future work exploring three-dimensional movements and other types of plants would strengthen the impact of the tool.

Reviewer #1 (Public Review):

Summary:

Codol et al. present a toolbox that allows simulating biomechanically realistic effectors and training Artificial Neural Networks (ANNs) to control them. The paper provides a detailed explanation of how the toolbox is structured and several examples demonstrating its utility.

Main comments:

(1) The paper is well-written and easy to follow. The schematics facilitate understanding of the toolbox's functionality, and the examples give insight into the potential results users can achieve.

(2) The toolbox's latest version, developed in PyTorch, is expected to offer greater benefits to the community.

(3) The new API, being compatible with Gymnasium, broadens the toolbox's application scope, enabling the use of Reinforcement Learning for training the ANNs.

Impact:

MotorNet is designed to simplify the process of simulating complex experimental setups, enabling the rapid testing of hypotheses on how the brain generates specific movements. Implemented in PyTorch and compatible with widely-used machine learning toolboxes, including Gymnasium, it offers an end-to-end pipeline for training ANNs on simulated setups. This can greatly assist experimenters in determining the focus of their subsequent efforts.

Additional context:

The main outcome of the work, a toolbox, is supplemented by a GitHub repository and a documentation webpage. Both the repository and the webpage are well-organized and user-friendly. The webpage guides users through the toolbox installation process, as well as the construction of effectors and Artificial Neural Networks (ANNs).

Author response:

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

Reviewer #1 (Public Review):

Summary:

Codol et al. present a toolbox that allows simulating biomechanically realistic effectors and training Artificial Neural Networks (ANNs) to control them. The paper provides a detailed explanation of how the toolbox is structured and several examples that demonstrate its usefulness.

Main comments:

(1) The paper is well written and easy to follow. The schematics help in understanding how the toolbox works and the examples provide an idea of the results that the user can obtain.

We thank the reviewer for this comment.

(2) As I understand it, the main purpose of the paper should be to facilitate the usage of the toolbox. For this reason, I have missed a more explicit link to the actual code. As I see it, researchers will read this paper to figure out whether they can use MotorNet to simulate their experiments, and how they should proceed if they decide to use it. I'd say the paper provides an answer to the first question and assures that the toolbox is very easy to install and use. Maybe the authors could support this claim by adding "snippets" of code that show the key steps in building an actual example.

This is an important point, which we also considered when writing this paper. We instead decided to focus on the first approach, because it is easier to illustrate the scientific use of the toolbox using code or interactive (Jupyter) notebooks than a publication format. We find the “how to proceed” aspect of the toolbox can more easily and comprehensively be covered using online, interactive tutorials. Additionally, this allows us to update these tutorials as the toolbox evolves over different versions, while it is more difficult to update a scientific article. Consequently, we explicitly avoided code snippets on the article itself. However, we appreciate that the paper would gain in clarity if this was more explicitly stated early. We have modified the paper to include a pointer to where to find tutorials online. We added this at the last paragraph of the introduction section:

The interested reader may consult the full API documentation, including interactive tutorials on the toolbox website at https://motornet.org.

(3) The results provided in Figures 1, 4, 5 and 6 are useful, because they provide examples of the type of things one can do with the toolbox. I have a few comments that might help improving them:

a. The examples in Figures 1 and 5 seem a bit redundant (same effector, similar task). Maybe the authors could show an example with a different effector or task? (see point 4).

The effectors from figures 1 and 5 are indeed very similar. However, the tasks in figure 1 and 5 present some important differences. The training procedure in figure 1 never includes any perturbations, while the one from figure 5 includes a wide range of perturbations of different magnitudes, timing and directions. The evaluation procedure of figure 1 includes center-out reaches with permanent viscous (proportional to velocity) external dynamics, while that of figure 5 are fixed, transient, square-shaped perturbation orthogonal to the reach direction. Finally, the networks in figure 1 undergo a second training procedure after evaluation while the network of figure 5 do not.

While we agree that some variation of effectors would be beneficial, we do show examples of a point-mass effector in figure 6. Overall, figure 5 shows a task that is quite different from that of figure 1 with a similar effector, while the opposite is true for figure 6. We have modified the text to clarify this for the reader, by adding the following.

End of 1st paragraph, section 2.4.

Therefore, the training protocol used for this task largely differed from section 2.1 in that the networks are exposed to a wide range of mechanical perturbations with varying characteristics.

1st paragraph of section 2.5

[…] this asymmetrical representation of PMDs during reaching movements did not occur when RNNs were trained to control an effector that lacked the geometrical properties of an arm such as illustrated in Figure 4c-e and section 2.1.

b. I missed a discussion on the relevance of the results shown in Figure 4. The moment arms are barely mentioned outside section 2.3. Are these results new? How can they help with motor control research?

We thank the reviewer for this comment. This relates to a point from reviewer 2 indicating that the purpose of each section was sometimes difficult to grasp as one reads. Section 2.3 explains the biomechanical properties that the toolbox implements to improve realism of the effector. They are not new results in the sense that other toolboxes implement these features (though not in differentiable formats) and these properties of biological muscles are empirically well-established. However, they are important to understand what the toolbox provides, and consequently what constraints networks must accommodate to learn efficient control policies. An example of this is the results in figure 6, where a simple effector versus a more biomechanically complex effector will yield different neural representations.

Regarding the manuscript itself, we agree that more clarity on the goal of every paragraph may improve the reader’s experience. Consequently, we ensured to specify such goals at the start of each section. Particularly, we clarify the purpose of section 2.3 by adding several sentences on this at the end of the first paragraph in that section. We also now clearly state the purpose of section 2.3 with the results of figure 6 and reference figure 4 in that section.

c. The results in Figure 6 are important, since one key asset of ANNs is that they provide access to the activity of the whole population of units that produces a given behavior. For this reason, I think it would be interesting to show the actual "empirical observations" that the results shown in Fig. 6 are replicating, hence allowing a direct comparison between the results obtained for biological and simulated neurons.

These empirical observations are available from previous electrophysiological and modelling work. Particularly, polar histograms across reaching directions like panel C are displayed in figures 2 and 3 of Scott, Gribble, Graham, Cabel (2001, Nature). Colormaps of modelled unit activity across time and reaching directions like panel F are also displayed in figure 2 of Lillicrap, Scott (2013, Neuron). Electrophysiological recordings of M1 neurons during a similar task in non-human primates can also be seen on “Preserved neural population dynamics across animals performing similar behaviour” figure 2 B (https://doi.org/10.1101/2022.09.26.509498) and “Nonlinear manifolds underlie neural population activity during behaviour” figure 2 B as well (https://doi.org/10.1101/2023.07.18.549575). Note that these two pre-prints use the same dataset.

We have added these citations to the text and made it explicit that they contain visualizations of similar modelling and empirical data for comparison:

This heterogeneous set of responses matches empirical observations in non-human primate primary motor cortex recordings (Churchland & Shenoy, 2007; Michaels et al., 2016) and replicate similar visualizations from previously published work (Fortunato et al., 2023; Lillicrap & Scott, 2013; Safaie et al., 2023).

(4) All examples in the paper use the arm26 plant as effector. Although the authors say that "users can easily declare their own custom-made effector and task objects if desired by subclassing the base Plant and Task class, respectively", this does not sound straightforward. Table 1 does not really clarify how to do it. Maybe an example that shows the actual code (see point 2) that creates a new plant (e.g. the 3-joint arm in Figure 7) would be useful.

Subclassing is a Python process more than a MotorNet process, as python is an object-oriented language. Therefore, there are many Python tutorials on subclassing in the general sense that would be beneficial for that purpose. We have amended the main text to ensure that this is clearer to the reader.

Subclassing a MotorNet object, in a more specific sense, requires overwriting some methods from the base MotorNet classes (e.g., Effector or Environment classes, which correspond to the original Plant and Task object, respectively). Since we made the decision (mentioned above) to not include code in the main text, we added tutorials to the online documentation, which include dedicated tutorials for MotorNet class subclassing. For instance, this tutorial showcases how to subclass Environment classes:

https://colab.research.google.com/github/OlivierCodol/MotorNet/blob/master/examples/3-environments.ipynb

(5) One potential limitation of the toolbox is that it is based on Tensorflow, when the field of Computational Neuroscience seems to be, or at least that's my impression, transitioning to pyTorch. How easy would it be to translate MotorNet to pyTorch? Maybe the authors could comment on this in the discussion.

We have received a significant amount of feedback asking for a PyTorch implementation of the toolbox. Consequently, we decided to enact this, and the next version of the toolbox will be exclusively in PyTorch. We will maintain the Application Programming Interface (API) and tutorial documentation for the TensorFlow version of the toolbox on the online website. However, going forward we will focus exclusively on bug-fixing and expanding from the latest version of MotorNet, which will be in PyTorch. We now believe that the greater popularity of PyTorch in the academic community makes that choice more sustainable while helping a greater proportion of research projects.

These changes led to a significant alteration of the MotorNet structure, which are reflected by changes made throughout the manuscript, notably in Figure 3 and Table 1.

(6) Supervised learning (SL) is widely used in Systems Neuroscience, especially because it is faster than reinforcement learning (RL). Thus providing the possibility of training the ANNs with SL is an important asset of the toolbox. However, SL is not always ideal, especially when the optimal strategy is not known or when there are different alternative strategies and we want to know which is the one preferred by the subject. For instance, would it be possible to implement a setup in which the ANN has to choose between 2 different paths to reach a target? (e.g. Kaufman et al. 2015 eLife). In such a scenario, RL seems to be a more natural option Would it be easy to extend MotorNet so it allows training with RL? Maybe the authors could comment on this in the discussion.

The new implementation of MotorNet that relies on PyTorch is already standardized to use an API that is compatible with Gymnasium. Gymnasium is a standard and popular interfacing toolbox used to link RL agents to environments. It is very well-documented and widely used, which will ensure that users who wish to employ RL to control MotorNet environments will be able to do so relatively effortlessly. We have added this point to accurately reflect the updated implementation, so users are aware that it is now a feature of the toolbox (new section 3.2.4.).

Impact:

MotorNet aims at simplifying the process of simulating complex experimental setups to rapidly test hypotheses about how the brain produces a specific movement. By providing an end-to-end pipeline to train ANNs on the simulated setup, it can greatly help guide experimenters to decide where to focus their experimental efforts.

Additional context:

Being the main result a toolbox, the paper is complemented by a GitHub repository and a documentation webpage. Both the repository and the webpage are well organized and easy to navigate. The webpage walks the user through the installation of the toolbox and the building of the effectors and the ANNs.

Reviewer #2 (Public Review):

MotorNet aims to provide a unified interface where the trained RNN controller exists within the same TensorFlow environment as the end effectors being controlled. This architecture provides a much simpler interface for the researcher to develop and iterate through computational hypotheses. In addition, the authors have built a set of biomechanically realistic end effectors (e.g., an 2 joint arm model with realistic muscles) within TensorFlow that are fully differentiable.

MotorNet will prove a highly useful starting point for researchers interested in exploring the challenges of controlling movement with realistic muscle and joint dynamics. The architecture features a conveniently modular design and the inclusion of simpler arm models provides an approachable learning curve. Other state-of-the-art simulation engines offer realistic models of muscles and multi-joint arms and afford more complex object manipulation and contact dynamics than MotorNet. However, MotorNet's approach allows for direct optimization of the controller network via gradient descent rather than reinforcement learning, which is a compromise currently required when other simulation engines (as these engines' code cannot be differentiated through).

The paper could be reorganized to provide clearer signposts as to what role each section plays (e.g., that the explanation of the moment arms of different joint models serves to illustrate the complexity of realistic biomechanics, rather than a novel discovery/exposition of this manuscript). Also, if possible, it would be valuable if the authors could provide more insight into whether gradient descent finds qualitatively different solutions to RL or other non gradient-based methods. This would strengthen the argument that a fully differentiable plant is useful beyond improving training time / computational power required (although this is a sufficiently important rationale per se).

We thank the reviewer for these comments. We agree that more clarity on the section goals may improve the reader’s experience and ensured this is the case throughout the manuscript. Particularly, we added the following on the first paragraph of section 2.3, for which an explicit goal was most missing:

In this section we illustrate some of these biomechanical properties displayed by MotorNet effectors using specific examples. These properties are well-characterised in the biology and are often implemented in realistic biomechanical simulation software.

Regarding the potential difference in solutions obtained from reinforcement or supervised learning, this would represent a non-trivial amount of work to do so conclusively and so may not be within the scope of the current article. We do appreciate however that in some situations RL may be a more fitting approach to a given task design. In relation to this point we now specify in the discussion that the new API can accommodate interfacing with reinforcement learning toolboxes for those who may want to pursue this type of policy training approach when appropriate (new section 3.2.4.).

Reviewer #3 (Public Review):

Artificial neural networks have developed into a new research tool across various disciplines of neuroscience. However, specifically for studying neural control of movement it was extremely difficult to train those models, as they require not only simulating the neural network, but also the body parts one is interested in studying. The authors provide a solution to this problem which is built upon one of the main software packages used for deep learning (Tensorflow). This allows them to make use of state-of-the-art tools for training neural networks.

They show that their toolbox is able to (re-)produce several commonly studied experiments e.g., planar reaching with and without loads. The toolbox is described in sufficient detail to get an overview of the functionality and the current state of what can be done with it. Although the authors state that only a few lines of code can reproduce such an experiment, they unfortunately don't provide any source code to reproduce their results (nor is it given in the respective repository).

The possibility of adding code snippets to the article is something we originally considered, and which aligns with comment two from reviewer one (see above). Hopefully this provides a good overview of the motivation behind our choice not to add code to the article.

The modularity of the presented toolbox makes it easy to exchange or modify single parts of an experiment e.g., the task or the neural network used as a controller. Together with the open-source nature of the toolbox, this will facilitate sharing and reproducibility across research labs.

I can see how this paper can enable a whole set of new studies on neural control of movement and accelerate the turnover time for new ideas or hypotheses, as stated in the first paragraph of the Discussion section. Having such a low effort to run computational experiments will be definitely beneficial for the field of neural control of movement.

We thank the reviewer for these comments.

Author response:

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

Reviewer #1 (Public Review):

Summary: The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections is largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd. 

Strengths: This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens. 

We thank the reviewer for recognizing the significance and the novelty of our work.

Weaknesses: The main weakness of the study is lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathway. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides the protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cell release any contents mentioned above. 

We would like to thank the reviewer for taking the time to review our work and providing us with valuable feedback.

Please note, that as indicated in our previous rebuttal to reviewers, amphibians do not possess the IgE antibody isotype1.

To our knowledge, there are no published works describing the approaches used in studying mammalian mast cell degranulation towards examining amphibian mast cells. While there are commercially available kits and reagents for examining mammalian mast cell granule content, most of these do not cross-react with amphibian counterparts. This is especially true of cytokines and chemokines, which diverged quickly with evolution and thus do not share substantial protein sequence identity across species as diverged as frogs and mammals. We would also like to highlight the fact that several studies suggest that amphibian mast cells lack histamine2, 3, 4, 5 and serotonin2, 6. While following up on these findings would be possible, we would like to respectfully emphasize that adopting approaches used in mammalian research to comparative immunology work is not always straightforward.

As we highlight in our manuscript, frog mast cells upregulate their expression of interleukin-4 (IL4), a hallmark cytokine associated with mammalian mast cells7. The additional findings presented in our revised manuscript indicate that mast cells respond to Bd by upregulating IL4 expression in vitro and in vivo. Together, this suggests that IL4 may be a central means by which frog mast cells confer protection against Bd, by counteracting Bd-elicited inflammation, including minimizing neutrophil infiltration, maintaining skin integrity, and promoting cutaneous mucus production. Please find that these additional results are presented in Figure 8 and are described in the results and discussion sections of our revised manuscript.

Our attempts to elicit degranulation of frog mast cells using compound 48/80 have so far not been successful. This may reflect technical issues with assays optimized for mammalian mast cells or biological difference between frog and mammalian mast cells, such as species differences in mas-related G-protein coupled receptors, through which compound 48/80 acts8. We will continue to explore means to study frog mast cell degranulation both in vitro and in vivo but also respectfully point out that while degranulation is a feature commonly associated with mammalian mast cells, this is not the only means by which the mammalian mast cells confer their immunological effects. Indeed, our studies suggest that frog mast cell IL4 production may be a key means by which these cells offer anti-Bd protection.

Please note that we successfully adopted an avidin staining approach to visualize mast cell heparin content in vitro and to evaluate cutaneous mast cell numbers in vivo in control and mast cell-enriched, mock- and Bd-infected animals. This additional work is depicted in Figure 4 and addressed in the results and discussion sections of our revised manuscript.

Reviewer #2 (Public Review):

Summary: In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent in the skin of X. laevis and find that this stimulates expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. Combining these two observations, they demonstrate that mast cell expansion using rSCF attenuates cutaneous neutrophilic infiltration. They further show that mast cell expansion correlates to cutaneous IL-4 expression, and that treatment with exogenous rIL-4 reduces neutrophilic infiltration and restores markers of epithelial health, offering a mechanism by which mast cell expansion protects from Bd infection. 

Strengths: The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. Building on prior work, they are able to contrast mast cell expansion with their neutrophil expansion model, allowing them to infer a mechanistic link between mast cell expansion and IL-4 production and subsequent suppression of neutrophil infiltration and cutaneous dysbiosis. 

We thank the reviewer for recognizing the rigorousness and utility of the studies presented in our manuscript.

Weaknesses: The main weaknesses derive from technical limitations inherent to the Xenopus model at this time. For example, in mice a mechanistic study would be expected to use IL-4 knockouts, preferably mast cell-specific, to prove the link between mast cell expansion and IL-4 production being necessary and sufficient to suppress neutrophils. However, the novel reagents in this manuscript present a compelling technical advance and a step forward in the tools available to study amphibian biology. 

We agree with the reviewer that an IL4 knock-out animal model would be a great way to support our findings. Unfortunately, working with a non-mammalian model such as X. laevis poses limitations that include lack of knock-out lines for immunology research. Moreover, as mentioned in our manuscript, we do not believe that IL4 is the sole mast cell-produced component responsible for the conferred antifungal protection. We thank the reviewer for acknowledging the limitations of our model system and recognizing the novelty, technical advances, and merits of the work presented in our manuscript.

In addition to their discussion, one open question from the revised manuscript is how a single treatment with rSCF leads to a peak in mast cell numbers and then decline to baseline in mock-infected frogs, while Bd infection either sustains rSCF-boosted mast cells or leads to steady mast cell increase over time in control-treated frogs. Whether this is mediated by endogenous SCF or some other factor remains unexplored.

This is an interesting question that we hope to explore in future studies. We did not see significant differences in skin SCF gene expression at 21 days post Bd infection. This does not rule out the possibility that the observed Bd-mediated effects to frog skin mast cell composition are not due to changes in skin SCF gene expression at earlier infection times, alone or in combination with other host or pathogen derived factors. We know that other factors are responsible for homing/retention of antimicrobial and immunosuppressive granulocyte subsets within frog skin9 and we postulate that some of these may be distinct mast cell types. Additionally, Bd is known to produce a myriad of immunomodulatory factors10, which may well also directly affect frog skin mast cell composition. Mammalian mast cells are heterogenous and are homed or recruited into tissues by an extensive array of host as well as microbiome-derived components11, 12. Undoubtedly, the frog skin mast cell composition is likewise complex, dynamic, and contingent on a plethora of host, cutaneous microbial flora- and in this case also Bd-produced factors.

References

(1) Flajnik, M.F. A cold-blooded view of adaptive immunity. Nat Rev Immunol 18, 438-453 (2018).

(2) Mulero, I., Sepulcre, M.P., Meseguer, J., Garcia-Ayala, A. & Mulero, V. Histamine is stored in mast cells of most evolutionarily advanced fish and regulates the fish inflammatory response. Proc Natl Acad Sci U S A 104, 19434-19439 (2007).

(3) Reite, O.B. A phylogenetical approach to the functional significance of tissue mast cell histamine. Nature 206, 1334-1336 (1965).

(4) Reite, O.B. Comparative physiology of histamine. Physiol Rev 52, 778-819 (1972).

(5) Takaya, K., Fujita, T. & Endo, K. Mast cells free of histamine in Rana catasbiana. Nature 215, 776-777 (1967).

(6) Galli, S.J. New insights into "the riddle of the mast cells": microenvironmental regulation of mast cell development and phenotypic heterogeneity. Lab Invest 62, 5-33 (1990).

(7) Babina, M., Guhl, S., Artuc, M. & Zuberbier, T. IL-4 and human skin mast cells revisited: reinforcement of a pro-allergic phenotype upon prolonged exposure. Archives of dermatological research 308, 665-670 (2016).

(8) Hermans, M.A.W. et al. Human Mast Cell Line HMC1 Expresses Functional Mas-Related G-Protein Coupled Receptor 2. Front Immunol 12, 625284 (2021).

(9) Hauser, K. et al. Discovery of granulocyte-lineage cells in the skin of the amphibian Xenopus laevis. FACETS 5, 571 (2020).

(10) Rollins-Smith, L.A. & Le Sage, E.H. Batrachochytrium fungi: stealth invaders in amphibian skin. Curr Opin Microbiol 61, 124-132 (2021).

(11) Halova, I., Draberova, L. & Draber, P. Mast cell chemotaxis - chemoattractants and signaling pathways. Front Immunol 3, 119 (2012).

(12) West, P.W. & Bulfone-Paus, S. Mast cell tissue heterogeneity and specificity of immune cell recruitment. Front Immunol 13, 932090 (2022).

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

Reviewer #1 (Public Review):

Summary:

The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in the immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections are largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd.

Strengths:

This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens. 

We thank the reviewer for acknowledging the novelty and importance of the work presented in our manuscript.

Weaknesses:

The main weakness of the study is the lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathways. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cells release any contents mentioned above. 

We would like to thank the reviewer for taking the time to review our work and providing us with valuable feedback. We feel that we have successfully incorporated the reviewer’s suggestions into our revised manuscript, thereby improving this work.

Please note that amphibians do not possess the IgE antibody isotype1.

To our knowledge there have been no published work assimilating approaches used when studying mammalian mast cell degranulation towards examining amphibian mast cells. While there are commercially available kits and reagents for examining mammalian mast cell granule content, most of these reagents do not cross-react with amphibian counterparts. This is especially true of cytokines and chemokines, which diverged quickly with evolution and thus do not share substantial protein sequence identity across species as diverged as frogs and mammals. Additionally, several studies suggest that amphibian mast cells lack histamine2, 3, 4, 5 and serotonin2, 6. Respectfully, while following up on these findings is possible, we would not consider adopting approaches used in mammalian research to comparative immunology work as easy.

As noted in our manuscript, frog mast cells upregulate their expression of interleukin-4 (IL4), which is a hallmark cytokine associated with mammalian mast cells7. The additional findings, presented in our revised manuscript indicate that mast cells respond to Bd by upregulating IL4 expression in vitro and in vivo. In turn, our work indicates that IL4 may be a central means by which frog mast cells confer protection against Bd, by counteracting Bd-elicited inflammation, including minimizing neutrophil infiltration, maintaining skin integrity, and promoting mucus production by skin mucus glands. Please find that these additional findings are presented in Figure 8 of our revised manuscript and are described in the results and discussion sections of the paper.

Our attempts to elicit degranulation of frog mast cells using compound 48/80 have so far not been successful. This may reflect technical issues with assays optimized for mammalian mast cells or biological difference between frog and mammalian mast cells, such as species differences in mas-related G-protein coupled receptors, through which compound 48/80 acts8. We will continue explore means to study frog mast cell degranulation both in vitro and in vivo but would also like to respectfully point out that while mast cell degranulation is a feature most associated with mammalian mast cells, this is not the only means by which the mammalian mast cells confer their immunological effects. Indeed, our additional studies suggest that mast cell IL4 production may be a key means by which these cells offer anti-Bd protection.

Please find that we have adopted an avidin-staining approach to visualize mast cell heparin content in vitro and to evaluate mast cell numbers in vivo in the skins of control and mast cell-enriched, mock- and Bd-infected animals. This additional work is depicted in Figure 4 of our revised manuscript and addressed in the results and discussion sections of our revised paper.

Reviewer #2 (Public Review):

Summary:

In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent into the skin of X. laevis and find that this stimulates the expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates the skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony-stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. 

Strengths:

The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of the exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. 

We thank the reviewer for recognizing the breadth and extent of the undertaking that culminated in this manuscript. Indeed, this manuscript would not have been possible without considerable reagent development and adaptation of techniques that had previously not been used for amphibian immunity research. In line with the reviewer’s sentiment, to our knowledge this is the first report of using molecular approaches to augment amphibian mast cells, which we hope will pave the way for new areas of research within the fields of comparative immunology and amphibian disease biology.

Weaknesses:

The conclusions regarding the role of mast cell expansion in controlling Bd infection would be stronger with a more rigorous evaluation of the model, as there are some key gaps and remaining questions regarding the data. For example: 

(1) Granulocyte expansion is carefully quantified in the initial time courses of rSCF and rCSF3 injections, but similar quantification is not provided in the disease models (Figures 3E, 4G, 5D-G). A key implication of the opposing effects of mast cell vs neutrophil expansion is that mast cells may suppress neutrophil recruitment or function. Alternatively, mast cells also express notable levels of csfr3 (Figure 2) and previous work from this group (Hauser et al, Facets 2020) showed rG-CSF-stimulated peritoneal granulocytes express mast cell markers including kit and tpsab1, raising the question of what effect rCSF3 might have on mast cell populations in the skin. Considering these points, it would be helpful if both mast cells and neutrophils were quantified histologically (based on Figure 1, they can be readily distinguished by SE or Giemsa stain) in the Bd infection models. 

We thank the reviewer for this insightful suggestion. Please find that we successfully adopted an in situ hybridization approach to evaluate neutrophil numbers in the skins of control and mast cell-enriched, mock- and Bd-infected animals based on expression of the neutrophil marker, myeloperoxidase (MPO9).  Please find these results are presented in Figures 6 and 8 of our revised manuscript and addressed in the appropriate sections of our revised paper.

Our findings suggest that rSCF administration results in the accumulation of mast cells that are polarized such, that they ablate the inflammatory response elicited by Bd infection, such as through mechanisms like IL4 production. Mammalian mast cells, including peritonea-resident mast cells, express csf3r10, 11. For this reason, we used MPO expression to visualize neutrophil skin infiltration in Figures 6 and 8 of our revised work. While the X. laevis animal model does not permit nearly the degree of immune cell resolution afforded by mammalian animal models, we do know that the adult X. laevis peritonea contain a myriad of immune cell subsets. We anticipate that the high kit expression reported by Hauser et al., 2020 in the rCSF3-recruited peritoneal leukocytes reflects the presence of mast cells therein.

Please find that we have used avidin-staining and MPO in situ hybridization to respectively visualize and enumerate mast cells and neutrophils in the skin of control and mast cell-enriched, mock- and Bd-infected animals. Indeed, our results show interesting, experimental condition-dependent changes in both the skin neutrophil and mast cell numbers. The results of these additional studies are presented in Figures 4, 6 and 8 of the revised manuscript and addressed in the results and discussions sections of our revised paper.

(2) Epithelial thickness and inflammation in Bd infection are reported to be reduced by rSCF treatment (Figure 3E, 5A-B) or increased by rCSF3 treatment (Figure 4G) but quantification of these critical readouts is not shown.

We thank the reviewer for this suggestion. We scored epithelial thickness under the distinct conditions described in our manuscript and presented the quantified data in Figures 5 and 8 of the revised paper.

(3) Critical time points in the Bd model are incompletely characterized. Mast cell expansion decreases zoospore burden at 21 dpi, while there is no difference at 7 dpi (Figure 3E). Conversely, neutrophil expansion increases zoospore burden at 7 dpi, but no corresponding 21 dpi data is shown for comparison (Figure 4G). Microbiota analysis is performed at a third time point,10 dpi (Figure 5D-G), making it difficult to compare with the data from the 7 dpi and 21 dpi time points. Reporting consistent readouts at these three time points is important to draw solid conclusions about the relationship of mast cell expansion to Bd infection and shifts in microbiota.

We thank reviewer for noting this discrepancy. Please find that we have repeated our mast cell-enrichment, Bd-challenge studies, examining days 10 and 21 post infection. Our new findings indicate that compared to control animals, mast cell-enrichment does result in significant reduction in Bd loads at both 10 and 21 dpi. The difference in Bd loads between r-ctrl and rSCF-treated animals at 10 dpi corroborates the other parameters that are altered between the two treatment groups at this experimental time point.

Our question regarding the roles of inflammatory granulocytes/neutrophils during Bd infections was that of ‘how’ rather ‘when’ these cells affect Bd infections.  Thus, and because the central focus of this work was mast cells and not other granulocyte subsets; when we saw that rCSF3-recruited granulocytes adversely affect Bd infections at 7 days, we did not pursue the kinetics of these observations further. We plan to explore the roles of inflammatory mediators and immune cell subsets during the course of Bd infections but feel that these future studies are more peripheral to the central thesis of the present manuscript regarding the roles of frog mast cells during Bd infections.

(4) Although the effect of rSCF treatment on Bd zoospores is significant at 21 dpi (Figure 3E), bacterial microbiota changes at 21 dpi are not (Figure S3B-C). This discrepancy, how it relates to the bacterial microbiota changes at 10 dpi, and why 7, 10, and 21 dpi time points were chosen for these different readouts (Figure 5F-G), is not discussed.

Please find that our additional studies indicate that compared to control animals, frog skin mast cell-enrichment results in significant reduction in Bd loads at 10 dpi. This corroborate our other findings including the observation that at 10 dpi, control animals exhibit reduced microbial richness whereas mast cell-enriched frogs were protected from this disruption of their microbiome. The amphibian microbiome serves as a major barrier to these fungal infections12 and we anticipate that Bd-mediated disruption of microbial richness facilitates host skin colonization by this pathogen. In turn, we anticipate that frog mast cells are conferring the observed anti-Bd protection in part by preventing microbial disassembly and thus interfering with optimal Bd colonization and growth on frog skins. Please find that we acknowledge and discuss these notions in our revised manuscript.

(5) The time course of rSCF or rCSF3 treatments relative to Bd infection in the experiments is not clear. Were the treatments given 12 hours prior to the final analysis point to maximize the effect? For example, in Figure 3E, were rSCF injections given at 6.5 dpi and 20.5 dpi? Or were treatments administered on day 0 of the infection model? If the latter, how do the authors explain the effects at 7 dpi or 21 dpi given mast cell and neutrophil numbers return to baseline within 24 hours after rSCF or rCSF3 treatment, respectively?

Please find that in our revised manuscript, we underlined the kinetics of our animal treatments and Bd-infections. In brief, for mast cell-enrichment, animals were injected with r-ctrl or rSCF, challenged 12 hours later with Bd and examined after 10 (per reviewers’ suggestions) and 21 days of infection. For neutrophil enrichment, animals were injected with r-ctrl or rCSF3, challenged 12 hours later with Bd and examined after 7 days of infection.

The title of the manuscript may be mildly overstated. Although Bd infection can indeed be deadly, mortality was not a readout in this study, and it is not clear from the data reported that expanding skin mast cells would ultimately prevent progression to death in Bd infections.

We acknowledge this point. The revised manuscript will be titled: “Amphibian mast cells: barriers to chytrid fungus infections”.

Reviewer #3 (Public Review):

Summary:

Hauser et al. provide an exceptional study describing the role of resident mast cells in amphibian epidermis that produce anti-inflammatory cytokines that prevent Batrachochytrium dendrobatidis (Bd) infection from causing harmful inflammation, and also protect frogs from changes in skin microbiomes and loss of mucin in glands and loss of mucus integrity that otherwise cause changes to their skin microbiomes. Neutrophils, in contrast, were not protective against Bd infection. Beyond the beautiful cytology and transcriptional profiling, the authors utilized elegant cell enrichment experiments to enrich mast cells by recombinant stem cell factor, or to enrich neutrophils by recombinant colony-stimulating factor-3, and examined respective infection outcomes in Xenopus.

Strengths:

Through the use of recombinant IL4, the authors were able to test and eliminate the hypothesis that mast cell production of IL4 was the mechanism of host protection from Bd infection. Instead, impacts on the mucus glands and interaction with the skin microbiome are implicated as the protective mechanism. These results will press disease ecologists to examine the relative importance of this immune defense among species, the influence of mast cells on the skin microbiome and mucosal function, and open the potential for modulating mucosal defense.

We thank the reviewer for recognizing the utility of the work presented in our manuscript.

Weaknesses:

A reduction of bacterial diversity upon infection, as described at the end of the results section, may not always be an "adverse effect," particularly given that anti-Bd function of the microbiome increased. Some authors (see Letourneau et al. 2022 ISME, or Woodhams et al. 2023 DCI) consider these short-term alterations as encoding ecological memory, such that continued exposure to a pathogen would encounter an enriched microbial defense. Regardless, mast cell-initiated protection of the mucus layer may negate the need for this microbial memory defense.

We thank the reviewer their insightful comment. We have revised our discussion to include this notion.

While the description of the mast cell location in the epidermal skin layer in amphibians is novel, it is not known how representative these results are across species ranging in chytridiomycosis susceptibility. No management applications are provided such as methods to increase this defense without the use of recombinant stem cell factor, and more discussion is needed on how the mast cell component (abundance, distribution in the skin) of the epidermis develops or is regulated.

We thank the reviewer for this suggestion. Please find that we have added a paragraph to our revised manuscripts to address possible source(s) of skin mast cells and a statement acknowledging that greater understanding of mast cell biology across distinct amphibian species may be used to develop future strategies for management of amphibian diseases.

We are very thankful to the reviewer for this excellent suggestion but would like to point out that the work presented in our manuscript was driven by comparative immunology questions more than by conservation biology. As such and considering just how little is known about mast cells outside of mammals; we chose not to speculate too much into possible utilities of altering amphibian skin mast cell composition and instead to focus our discussion on the immediate takeaways of the work presented by our paper.

References

(1) Flajnik, M.F. A cold-blooded view of adaptive immunity. Nat Rev Immunol 18, 438-453 (2018).

(2) Mulero, I., Sepulcre, M.P., Meseguer, J., Garcia-Ayala, A. & Mulero, V. Histamine is stored in mast cells of most evolutionarily advanced fish and regulates the fish inflammatory response. Proc Natl Acad Sci U S A 104, 19434-19439 (2007).

(3) Reite, O.B. A phylogenetical approach to the functional significance of tissue mast cell histamine. Nature 206, 1334-1336 (1965).

(4) Reite, O.B. Comparative physiology of histamine. Physiol Rev 52, 778-819 (1972).

(5) Takaya, K., Fujita, T. & Endo, K. Mast cells free of histamine in Rana catasbiana. Nature 215, 776-777 (1967).

(6) Galli, S.J. New insights into "the riddle of the mast cells": microenvironmental regulation of mast cell development and phenotypic heterogeneity. Lab Invest 62, 5-33 (1990).

(7) Babina, M., Guhl, S., Artuc, M. & Zuberbier, T. IL-4 and human skin mast cells revisited: reinforcement of a pro-allergic phenotype upon prolonged exposure. Archives of dermatological research 308, 665-670 (2016).

(8) Hermans, M.A.W. et al. Human Mast Cell Line HMC1 Expresses Functional Mas-Related G-Protein Coupled Receptor 2. Front Immunol 12, 625284 (2021).

(9) Buchan, K.D. et al. A transgenic zebrafish line for in vivo visualisation of neutrophil myeloperoxidase. PLoS One 14, e0215592 (2019).

(10) Aponte-Lopez, A., Enciso, J., Munoz-Cruz, S. & Fuentes-Panana, E.M. An In Vitro Model of Mast Cell Recruitment and Activation by Breast Cancer Cells Supports Anti-Tumoral Responses. Int J Mol Sci 21 (2020).

(11) Jamur, M.C. et al. Mast cell repopulation of the peritoneal cavity: contribution of mast cell progenitors versus bone marrow derived committed mast cell precursors. BMC Immunol 11, 32 (2010).

(12) Walke, J.B. & Belden, L.K. Harnessing the Microbiome to Prevent Fungal Infections: Lessons from Amphibians. PLoS Pathog 12, e1005796 (2016).

Reviewer #2: (Recommendations For The Authors): 

We thank the reviewer for their excellent suggestions, their time reviewing this work and their help with this manuscript.

While we were not able to incorporate some of these changes, please find that we have significantly altered our manuscript in accordance with the reviewer’s suggestions from their public review. We feel that we have substantially altered our paper, including providing considerable additional data, supporting the key findings therein.

(1) The heatmap in Figure 1I appears to be scaled data, similar to Figure 4A, in which case the indicated scale numbers are not correct (e.g. they should be -2 to 2, or -3 to 3) 

Thank you for the suggestion. Please find that we have changed this figure accordingly.

(2) For Figure 1, additional curated gene lists might better illustrate the difference in cell types, e.g. include the data for a panel of mast cell genes in a heatmap (mcpt1, tpsab1, etc.) and another panel of curated neutrophil genes (e.g. lyz) in a heatmap. If the authors still have leftover RNA, qPCR verification of some of the critical genes (e.g. kit) would add to the rigor of the analysis, as this study is the foundation of a new method for culturing amphibian mast cells. 

We thank the reviewer for this suggestion. Unfortunately, we do not have leftover RNA/cDNA and we have not been able to locate mcpt1 or tpsab1 in our DEGs. We anticipate that this issue may stem from the suboptimal annotation of the Xenopus laevis genome. We agree that curating more mast cell/neutrophil genes would be ideal but feel that we have adequately highlighted those genes that are differentially expressed between the two populations in our analysis.

(3) The presentation of counts in Figure 2 is a bit hard to interpret. Although it is mentioned that everything is statistically significant, explicitly showing statistics for each gene would be better. One possibility would be to use a volcano plot (p-value vs log2 fold change) and highlight the genes shown in Figure 2, potentially with an accompanying heat map to show replicate variability. 

We thank the reviewer for this suggestion. We entertained presenting the data as volcano plots or heat maps, but in the end felt that the bar graphs better conveyed the information that we are hoping to get across. Please note that the error bars in the bar graph depict the replicate variability. Please also note that to highlight that all the depicted genes were differentially expressed, we italicized the statement in the corresponding figure legend: “All depicted genes were significantly differentially expressed between the two populations”.

(4) Narratively, it might make more sense to put Figure 4A-C with Figure 3. 

We thank the reviewer for this suggestion. Please find that we significantly revised most of our figures to better convey the content therein. We combined the content of Figure 4A-C with Figure 5A-C and added data on epidermal thickness under different conditions into this figure; Figure 5 of our revised manuscript.

(5) If possible, complementing the skin RNA-seq from rSCF treatment in Bd infection with skin RNA-seq from rCSF3 treatment to compare effects on transcriptional programs of barrier function, etc would elevate this study and add additional insights into cutaneous inflammation in the setting of Bd infection. 

We thank the reviewer for this suggestion. We anticipate that the skin inflammation caused by Bd infection is not due solely to neutrophil infiltration and artificially altering the frog skin neutrophil content would thus not recapitulate chytridiomycosis progression. We completely agree that it would be valuable to examine barrier functions in control and mast cell-enriched, Bd-infected frogs. This is something that we hope to pursue further in future studies but feel that together with our additional findings, we are presenting a significant amount of data to constitute a stand-alone story.

(6) In Figure S1A, analyzing only 3 AMP genes by qPCR is perhaps too focused. As a control, it would be useful to also test some genes known to be functionally important in neutrophil anti-microbial responses, e.g. lyz. Expanding on this experiment by performing RNA-seq on Bd-treated, bone-marrow-derived mast cells and neutrophils would be a great addition to the manuscript and an important resource for future studies in the field. The fact that the use of rSCF (or rCSF3) enables the differentiation of these cells in large numbers of pure populations presents this unique opportunity. Although IL-4 did not end up affecting mucus production, clues to the mediator(s) of this mast cell-dependent effect may be found with unbiased RNA-seq after exposure to Bd. 

We thank the reviewer for this suggestion but would like to point out that our manuscript is focused on mast cells rather than neutrophils. We also believe that in vitro exposure of leukocytes to Bd is not the most physiologically relevant model of what would happen to skin-resident and incoming immune cell subsets, since Bd primarily infects top-most keratinocytes. We anticipate that rather than coming into direct contact with the fungus, cells like mast cells and neutrophils are responding to Bd-produced and infected cell-produced products. For this reason, we did not perform RNA-seq analysis of in vitro derived mast cells or neutrophils stimulated with Bd. As we develop more X. laevis-specific reagents, we hope to revisit the question of infected skin mast cell and neutrophil gene expression profiles but are not in a position to ask these questions at this time.

This work is also guided by a finite budget, and we feel that together with our significant additional findings described in our revised manuscript, we are presenting a substantial amount of work to constitute a stand-alone story and manuscript.

Reviewer #3 (Recommendations For The Authors): 

The following are minor edits needed in the text and figure legends: 

Standardize terms such as IL4 instead of il4 or ril4 vs rIL4 throughout. Also, r-SCF vs rSCF. 

Thank you. Please find that we have standardized such terms throughout our revised manuscript. Please note that we are adhering to the convention that gene names are in lower case, protein names are in upper case and recombinant protein names are preceded by an ‘r’.

Pg 9 Change "In contract" to "In contrast". 

Thank you and changed accordingly.

Fig 4 - Perhaps indicate if results in addition to 7dpi are also available. 

Please find that we analyzed Bd loads in control and mast cell-enriched, infected frogs after 10 dpi. This data is presented in Figures 3 and 4 of our revised manuscript.

Similarly in Fig. 5, are results other than 10dpi available in the supplement? 

Please find that the results from the microbiome studies are presented in supplemental figure 3 (Fig. S3). Please note that the results presented in original manuscript Fig. 5A-C - revised manuscript Fig. 5B-E depict data for 21 dpi, which is the longest examined infection timepoint. We present data from 1 and 10 dpi in Fig. 4 of our revised manuscript.

Indicate why these days were chosen in the methods. 

Please find that we indicated why the experimental timepoints were chosen, in the methods section of our revised manuscript.

Fig S1 legend has errors in describing which panels are for which asterisks. 

Fig. S3 legend indicates panels F and G. 

Thank you. Please find that we revised our supplemental figures and amended the corresponding figure legends.

eLife assessment

This important study reveals the role of skin-resident mast cells in amphibians in mediating antimicrobial responses. The data are compelling and highlight species-specific biology that can cross-inform human mast cell biology in a species that does not rely on IgE as a primary mechanism for antimicrobial skin responses.

Reviewer #1 (Public Review):

Summary:

The global decline of amphibians is primarily attributed to deadly disease outbreaks caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). It is unclear whether and how skin-resident immune cells defend against Bd. Although it is well known that mammalian mast cells are crucial immune sentinels in the skin and play a pivotal role in immune recognition of pathogens and orchestrating subsequent immune responses, the roles of amphibian mast cells during Bd infections is largely unknown. The current study developed a novel way to enrich X. laevis skin mast cells by injecting the skin with recombinant stem cell factor (SCF), a KIT ligand required for mast cell differentiation and survival. The investigators found an enrichment of skin mast cells provides X. laevis substantial protection against Bd and mitigates the inflammation-related skin damage resulting from Bd infection. Additionally, the augmentation of mast cells leads to increased mucin content within cutaneous mucus glands and shields frogs from the alterations to their skin microbiomes caused by Bd.

Strengths:

This study underscores the significance of amphibian skin-resident immune cells in defenses against Bd and introduces a novel approach to examining interactions between amphibian hosts and fungal pathogens.

Weaknesses:

The main weakness of the study is lack of functional analysis of X. laevis mast cells. Upon activation, mast cells have the characteristic feature of degranulation to release histamine, serotonin, proteases, cytokines, and chemokines, etc. The study should determine whether X. laevis mast cells can be degranulated by two commonly used mast cell activators IgE and compound 48/80 for IgE-dependent and independent pathway. This can be easily done in vitro. It is also important to assess whether in vivo these mast cells are degranulated upon Bd infection using avidin staining to visualize vesicle releases from mast cells. Figure 3 only showed rSCF injection caused an increase in mast cells in naïve skin. They need to present whether Bd infection can induce mast cell increase and rSCF injection under Bd infection causes a mast cell increase in the skin. In addition, it is unclear how the enrichment of mast cells provides the protection against Bd infection and alternations to skin microbiomes after infection. It is important to determine whether skin mast cell release any contents mentioned above.

Reviewer #2 (Public Review):

Summary:

In this study, Hauser et al investigate the role of amphibian (Xenopus laevis) mast cells in cutaneous immune responses to the ecologically important pathogen Batrachochytrium dendrobatidis (Bd) using novel methods of in vitro differentiation of bone marrow-derived mast cells and in vivo expansion of skin mast cell populations. They find that bone marrow-derived myeloid precursors cultured in the presence of recombinant X. laevis Stem Cell Factor (rSCF) differentiate into cells that display hallmark characteristics of mast cells. They inject their novel (r)SCF reagent in the skin of X. laevis and find that this stimulates expansion of cutaneous mast cell populations in vivo. They then apply this model of cutaneous mast cell expansion in the setting of Bd infection and find that mast cell expansion attenuates skin burden of Bd zoospores and pathologic features including epithelial thickness and improves protective mucus production and transcriptional markers of barrier function. Utilizing their prior expertise with expanding neutrophil populations in X. laevis, the authors compare mast cell expansion using (r)SCF to neutrophil expansion using recombinant colony stimulating factor 3 (rCSF3) and find that neutrophil expansion in Bd infection leads to greater burden of zoospores and worse skin pathology. Combining these two observations, they demonstrate that mast cell expansion using rSCF attenuates cutaneous neutrophilic infiltration. They further show that mast cell expansion correlates to cutaneous IL-4 expression, and that treatment with exogenous rIL-4 reduces neutrophilic infiltration and restores markers of epithelial health, offering a mechanism by which mast cell expansion protects from Bd infection.

Strengths:

The authors report a novel method of expanding amphibian mast cells utilizing their custom-made rSCF reagent. They rigorously characterize expanded mast cells in vitro and in vivo using histologic, morphologic, transcriptional, and functional assays. This establishes solid footing with which to then study the role of rSCF-stimulated mast cell expansion in the Bd infection model. This appears to be the first demonstration of exogenous use of rSCF in amphibians to expand mast cell populations and may set a foundation for future mechanistic studies of mast cells in the X. laevis model organism. Building on prior work, they are able to contrast mast cell expansion with their neutrophil expansion model, allowing them to infer a mechanistic link between mast cell expansion and IL-4 production and subsequent suppression of neutrophil infiltration and cutaneous dysbiosis.

Weaknesses:

The main weaknesses derive from technical limitations inherent to the Xenopus model at this time. For example, in mice a mechanistic study would be expected to use IL-4 knockouts, preferably mast cell-specific, to prove the link between mast cell expansion and IL-4 production being necessary and sufficient to suppress neutrophils. However, the novel reagents in this manuscript present a compelling technical advance and a step forward in the tools available to study amphibian biology.

In addition to their discussion, one open question from the revised manuscript is how a single treatment with rSCF leads to a peak in mast cell numbers and then decline to baseline in mock-infected frogs, while Bd infection either sustains rSCF-boosted mast cells or leads to steady mast cell increase over time in control-treated frogs. Whether this is mediated by endogenous SCF or some other factor remains unexplored.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The study entitled "Rifampicin tolerance and growth fitness among isoniazid-resistant clinical Mycobacterium tuberculosis isolates: an in-vitro longitudinal study" by Vijay et al. provides valuable insights into the association of rifampicin tolerance and growth fitness with isoniazid resistance among clinical isolates of M. tuberculosis. Antibiotic tolerance in M. tuberculosis is an important topic since it contributes to the lengthy and complicated treatment required to cure tuberculosis disease and may portend the emergence of antibiotic resistance. The authors found that rifampicin tolerance was correlated with bacterial growth, rifampicin minimum inhibitory concentrations, and isoniazid-resistance mutations.

Strengths:

The large number of clinical isolates evaluated and their longitudinal nature during treatment for TB (including exposure to rifampin) are strengths of the study.

Weaknesses:

Some of the methodologies are not well explained or justified and the association of antibiotic tolerance with growth rate is not a novel finding. In addition, the molecular mechanisms underlying rifampicin tolerance only in rapidly growing isoniazid-resistant isolates have not been elucidated and the potential implications of these findings for clinical management are not immediately apparent.

We thank the reviewer for the comments, we have modified the method section and figure 1 to clarify the method as suggested by the reviewer.

Although we agree that previous studies have shown the association of slow growth rate with antibiotic tolerance, ours is the most comprehensive assessment of rifampicin tolerance among clinical isolates, to our knowledge. In particular, we show that the degree of tolerance in clinical isolates can vary over several orders of magnitude: which had not been previously documented or appreciated. Furthermore, the association of high tolerance among IR isolates is a new finding, and given the potential for tolerance to increase risk of de novo drug resistance, our study suggests that IR isolates with high rifampicin tolerance may present a risk for development of MDR-TB.

In addition, we have also analysed the longitudinal isolates and the genetic variants emerging in them associated with increase in rifampicin tolerance. This analysis reveals possible multiple pathways to increase in rifampicin tolerance among clinical M. tuberculosis isolates. Possible clinical implication includes associating high rifampicin tolerance and isoniazid resistance as a risk factor for tuberculosis treatment failure. This study helps to develop further clinical studies to evaluate the role of rifampicin tolerance in IR isolates and treatment outcome. We have focused on these aspects in the discussion of the revised manuscript.

Reviewer #2 (Public Review):

Summary:

This study by Vijay and colleagues addresses a clinically important, and often overlooked aspect of Tb treatment. Detecting for variations in the level of antibiotic tolerance amongst otherwise antibiotic-susceptible isolates is difficult to routinely screen for, and consequently not performed. The authors, present a convincing argument that indeed, there is significant variation in the susceptibility of isoniazid-resistant strains to killing by rifampicin, in some cases at the same tolerance levels as bona fide resistant strains. On the whole, the study is easy to follow and the results are justified. This work should be of interest to the wider TB community at both a clinical and basic level.

Weaknesses:

The manuscript is long, repetitive in places, and the figures could use some amending to improve clarity (this could be a me-specific issue as they look ok on my screen, yet the colour is poor when printed).

We thank the reviewer for the comments, we have modified the revised manuscript as per the reviewer suggestions.

It would have been great to have seen some correlation between increased rifampicin tolerance and treatment outcome, although I'm not sure if this data is available to the researchers. I agree with the researchers the use of a single media condition is a limitation. However, this is true of a lot of studies. Rifampicin tolerance and treatment outcome analysis.

We agree with the reviewer that correlation between rifampicin tolerance and treatment outcome is important. This needs to be performed in future studies with better design to correlate rifampicin tolerance with treatment progression or outcome data.  

Reviewer #3 (Public Review):

Summary:

The authors have initiated studies to understand the molecular mechanisms underlying the devolvement of multi-drug resistance in clinical Mtb strains. They demonstrate the association of isoniazid-resistant isolates by rifampicin treatment supporting the idea that selection of MDR is a microenvironment phenomenon and involves a group of isolates.

Strengths:

The methods used in this study are robust and the results support the authors' claims to a major extent.

Weaknesses:

The manuscript needs a thorough vetting of the language. At present, the language makes it very difficult to comprehend the methodology and results.

We thank the reviewer for the comments, we have revised the manuscript as per the reviewer’s suggestions.

Reviewer #1 (Recommendations For The Authors):

Major comments:

(1) Methods: The authors attempt to differentiate between "fast"- and "slow"-growing bacteria in order to determine if the growth rate is associated with rifampicin tolerance. This is accomplished by assessing growth on solid agar at 15 and 60 days post-incubation, respectively. However, mycobacterial growth rate is not a binary phenomenon but rather a continuous variable. Moreover, it is not clear why 15 and 60 days were selected. Also, instead of a "slow growth" phenotype, the 60-day time point might simply reflect a longer lag phase. Were the plates examined at any interval time points? It would be interesting to know whether colony growth was delayed overall in the populations observed only at 60 days, or simply if the appearance of microcolonies visible to the naked eye was delayed (with normal growth afterwards).

We thank the reviewer for the comments, we want to clarify that we have not used agar plates but most-probable number method to determine the survival fraction post antibiotic treatment. We have clarified this in the revised manuscript and revised figure 1. The MPN method is a binary measure (growth/ no growth) and therefore cannot differentiate between long lag time and other mechanisms. In our original analysis, we included an intermediate time point of 30 days, but these data (included as supp fig. 1) cannot address the issue of lag phase directly. Since the 30-day time point did not add to the overall analysis and interpretation, we had not included them in the original submission.

(2) Methods/Results/Discussion: Some important clinical information is missing-how were the patients treated who had IR isolates? Did they receive the standard regimen for DS TB or was another drug substituted for isoniazid? Exposure to different drugs could affect the rifampicin-tolerant populations during the intensive phase (Figure 5).

Thank you for this comment, we have included the information regarding the treatment regimen in the revised manuscript.

Were there differences in microbiological (sputum culture conversion rate at 8 weeks or time to culture negativity) or clinical outcomes based on isoniazid susceptibility? Perhaps more importantly, were there differences in microbiological/clinical outcomes based on the proportion of bacterial subpopulations with rifampicin tolerance for a particular isolate? There should be more discussion on the potential clinical implications of the study's findings.

We agree with the reviewer that correlation between rifampicin tolerance and treatment progression or outcome is important. This needs to be performed in future studies with better design to correlate rifampicin tolerance with treatment progression or outcome data.  

(3) Results (Figure 3A): Although an interesting finding, the increased rifampicin tolerance observed only in the "rapidly" growing populations of isoniazid-resistant isolates (IR) vs. isoniazid-susceptible (IS) isolates is not explained. In contrast, equally, increased rifampicin tolerance is seen in the "slowly" growing populations of both IR and IS isolates. It would be interesting to know if these slowly growing populations show specific tolerance to rifampicin or if, as expected, slow growth confers tolerance to a range of different bactericidal antibiotics.

We thank the reviewer for the suggestions. we agree these will be interesting to investigate in a future study but are outside the scope of the current study.

(4) Results (Figure 3B): The basis for the classification into tertiles is not clear and appears somewhat arbitrary-does this represent the survival of a particular isolate following rifampicin exposure relative to the other isolates based on isoniazid susceptibility (IS or IR) or the % growth relative to other populations for the same isolate? Figure 3B is missing a y-axis label. Is it a log10 MPN ratio?

We thank the reviewer for pointing this, we want to clarify that for the classification into tertiles, first we pooled both group of isolates isoniazid susceptible (IS) and isoniazid resistant (IR) into a single population. Subsequently, we categorized this unified population into three distinct groups: low, medium, and high, based on their survival fraction following rifampicin treatment. Consequently, the 'low,' 'medium,' and 'high' tertiles represent the survival of each isolate following rifampicin exposure relative to the total number of isolates  combing both IS and IR isolates.

For clarity, we provide a breakdown of the criteria for each tertile:

+Low tertile: Consists of isolates with the lowest survival fraction (bottom 25%).

+Medium tertile: Encompasses isolates with survival fractions that fall between the bottom 25% and the top 25%.

+High tertile: Comprises isolates with the highest survival fractions (top 25%). This we have modified in the revised manuscript to clarify.

We have also modified the Figure 3B to correct the y-axis label.

(5) Results (lines 185-186): For correlating relative growth in the absence of antibiotics, 19 clinical isolates "outliers" were removed without explanation.

We have added explanation for the “outliers” which were removed earlier due to deviation from normal distribution, we have also provided the supplementary figure 3 which includes these outliers.

(6) Results (lines 203-211): The authors attempted to investigate a potential association between the mechanism of M. tuberculosis isoniazid resistance and the degree of rifampicin tolerance. However, the vast majority of IR clinical isolates (n=71) had a katG_S315X mutation and only 8 isolates had alternative mutations (inhA_I21T and fabG1_C-15X). Given the wide range of rifampicin tolerance observed within these isoniazid-resistant isolates, they concluded that other genetic or epigenetic determinants must be playing a role. WGS of longitudinally collected isolates from the same patients during TB treatment yielded non-synonymous SNPs in a list of genes previously reported to be associated with persistence, tolerance, and mycobacterial survival. However, precise mechanisms (including, e.g., expression of efflux pumps) are not investigated.

We thank the reviewer for summarising the findings. Yes, we agree that investigating the precise mechanism of rifampicin tolerance is beyond the scope of the current work.

Minor comments:

(1) Abstract (line 41): The nonstandard abbreviations "IR" and "IS" have not been introduced prior to this usage.

We have modified this in the abstract.

(2) Introduction (line 60): Insert "phenomena" or "mechanisms" after "two".

We have modified this in the introduction.

(3) Introduction (lines 66-69): This sentence is confusing, especially the second part ("supporting this studies...").

We have modified the lines to clarify.

(4) Introduction (line 84): In the current text, it appears as if "IR" is the abbreviation for "isoniazid". Therefore, I recommend changing "resistance to isoniazid" to "isoniazid resistance".

We have modified this in the revised manuscript.

(5) Results (line 141): Insert "the" before "rest".

We have modified this in the revised manuscript.

(6) Results (line 187): Replace "did not had" with "did not have".

We have modified this in the revised manuscript.

Reviewer #2 (Recommendations For The Authors):

Abstract:

The abstract is long and repetitive. It needs reworking and shortening to improve clarity and highlight the main takeaway message.

We thanks the reviewer for the suggestions and have modified this in the revised manuscript.

The introduction is interesting and contains relevant information. However, it is long and takes a while to get to the point of the study. It needs re-writing to emphasise key prior results and the purpose of this study.

We thanks the reviewer for the suggestions and we have modified this in the revised manuscript.

Results:

As the study relies predominately on the use of MPN, I think a simple schematic of how the experiment is performed would be informative. Could this be added to Figure 1?

We have revised the figure 1 in the manuscript to include the schematic representation.

Some of the differences in MKD90, whilst they may be significant, are small so it would at least provide context as to the relevance of these differences. This may also alleviate my confusion as to how the authors can measure the time required to achieve MDK90 as 1.23-1.31 days when the first time point that is taken is day 2 (the data in Figure 2). They have FigS6 but this is small and hard to follow.

We thank the reviewer for this suggestion, we have modified this in the revised manuscript and figureS6.

Figure 2:

Would be helpful to have -1 on the Y axis.

The grey dots don't print very well (Might be my printer)

We have modified this in the revised manuscript, figure 2.

Line 142: The authors note a difference in RIF tolerance at day 15 that disappeared by day 60. I assume they are referring to the day 5 timepoint although this isn't clear as written.

Yes, it is referring to the day 5 time point and we have clarified this in the revised manuscript.

The section starting at line 148 (fig 3) is interesting, but it is difficult to read and follow what the difference is between this data and the prior data in Figure 2. It also wasn't until about line 165 that the purpose became clear. Overall the conclusions are sound and interesting.

We have modified this in the revised manuscript.

Line 154: What are the early and late time recovery time points?

Is Figure 3A the same data as Figure 2?

We have clarified this in the revised manuscript, the figure 3A is the same data as Figure 2.

I found Figure 6 hard to follow. I'm not sure how better to present this data, but it should be improved. Some further clarification in the text would be helpful.

We thank the reviewer for the suggestions. We have added more explanation in the text to clarify figure 6.

Conclusions:

The conclusions are sound, based on the data presented. The clinical relevance is highlighted, yet appropriately phrased to not be too far-reaching.

Again, I think the conclusions could be condensed considerably. It is repetitive in places, which distills the main outcomes of this otherwise interesting and important study. The authors appropriately highlight some of the limitations of their study.

We thank the reviewer for these comments and have modified this in the revised manuscript.

Reviewer #3 (Recommendations For The Authors):

The manuscript "Rifampicin tolerance and growth fitness among isoniazid-resistant clinical Mycobacterium tuberculosis isolates: an in-vitro longitudinal study" by Srinivasan et.al., details the identification/ development of isoniazid-resistant strains in clinical isolates following testament with rifampicin. This is an important aspect of understanding MDR development in TB strains. the results are promising and gel well with the hypothesis. However, the manuscript requires a thorough language modification. While the overall idea is clear the methodology does not come out clearly.

Specific comments:

(1) It is not clear whether rifampicin treatments were given for 2 and 5 days before kill curves or for 15 and 60 days? The methodology needs to be phased clearly. Why was this time interval of 15 days and 60 days taken? is there a rationale for this?

We thank the reviewer for the suggestions, we have modified the method and figure 1 to clarify this in the revised manuscript.

(2) A concentration of 2ug/ml was used for in vitro culture in this study. While the authors themselves indicate that this is well above the MIC, this might represent a non- natural dose and hence may force the evolution of strains. What will be the scenario in the natural course of antibiotic treatment (dose at MIC or less than MIC)?

We have observed that till 5 days there is no significant resistant emergence but after 5 days only resistance emerges, therefore we avoided determining the survival fraction after resistance emergence, the kill curve represents mostly tolerant sub population. ADD: Pharmacokinetic studies of rifampicin dosing suggest that peak concentrations of >2-32 µg/mL are typical for standard doses of the drug, therefore we believe the chosen concentration of 2 µg/mL to be physiologically relevant.

(3) As described in line 155, the survival spanned a broad distribution, across a million times in difference. This is rather surprising that 5 days of rifampicin treatment would lead to such a spread in resistance patterns. Did the authors study the different populations to understand this phenomenon? This is important given the scale of resistance developed in this short time.

We want to clarify that the broad range of survival fraction reflect the difference in tolerant sub-populations but not resistant sub-population to rifampicin as they are determined post rifampicin treatment in rifampicin free media, this has been clarified in the revised figure 1.

Overall, the manuscript is a detailed study with new insights into the development of multi-drug resistance by Mtb. A thorough vetting for language is essential for a greater impact of the study.

We thank the reviewer and have attempted to improve the clarity of the language to increase the potential impact of our findings.

eLife assessment

This valuable study demonstrates that there is significant variation in the susceptibility of isoniazid-resistant Mycobacterium tuberculosis clinical isolates to killing by rifampicin, in some cases at the same tolerance levels as bona fide resistant strains. The evidence provided is solid, with no clear genetic marker for increased tolerance, suggesting that there may be multiple routes to achieving this phenotype. The work will be of interest to infectious disease researchers.

Reviewer #3 (Public Review):

Summary:

The authors have initiated studies to understand the molecular mechanisms underlying the devolvement of multi drug resistance in clinical Mtb strains. They demonstrate the association of isoniazid resistant isolates by rifampicin treatment supporting the idea that selection of MDR is a microenvironment phenomenon and involves a group of isolates.

Strengths:

The methods used in this study are robust and the results support the authors claims to a major extent.<br /> The language has now been corrected and is better to comprehend.

Reviewer #3 (Public Review):

Summary:

In this paper Hajra et al have attempted to identify the role of Sirt1 and Sirt3 in regulating metabolic reprogramming and macrophage host defense. They have performed gene knock down experiments in RAW macrophage cell line to show that depletion of Sirt1 or Sirt3 enhances the ability of macrophages to eliminate Salmonella Typhimurium. However, in mice inhibition of Sirt1 resulted in dissemination of the bacteria but the bacterial burden was still reduced in macrophages. They suggest that the effect they have observed is due to increased inflammation and ROS production by macrophages. They also try to establish a weak link with metabolism. They present data to show that the switch in metabolism from glycolysis to fatty acid oxidation is regulated by acetylation of Hif1a, and PDHA1.

Strengths:

The strength of the manuscript is that the role of Sirtuins in host-pathogen interactions has not been previously explored in-depth making the study interesting. It is also interesting to see that depletion of either Sirt1 or Sirt3 results in a similar outcome.

Weaknesses:

The major weakness of the paper is the low quality of data, making it harder to substantiate the claims. Also, there are too many pathways and mechanisms being investigated. It would have been better if the authors had focussed on either Sirt1 or Sirt3 and elucidated how it reprograms metabolism to eventually modulate host response against Salmonella Typhimurium. Experimental evidence is also lacking to prove the proposed mechanisms. For instance they show correlative data that knock down of Sirt1 mediated shift in metabolism is due to HIF1a acetylation but this needs to be proven with further experiments.

eLife assessment

This study presents valuable findings on the role of the sirtuins SIRT1 and SIRT3 during Salmonella Typhimurium infection. Although the work increases our understanding of the mechanisms used by this pathogen to interact with its host and may have implications for other intracellular pathogens, the reviewers found that the evidence to support the claims is incomplete. In particular, the discrepancy between results obtained using cultured cell lines and the animal model of infection, as well as potential indirect effects through the microbiome stand out.

Reviewer #2 (Public Review):

Dipasree Hajra et al demonstrated that Salmonella was able to modulate the expression of Sirtuins (Sirt1 and Sirt3) and regulate the metabolic switch in both host and Salmonella, promoting its pathogenesis. The authors found Salmonella infection induced high levels of Sirt1 and Sirt3 in macrophages, which were skewed toward the M2 phenotype allowing Salmonella to hyper-proliferate. Mechanistically, Sirt1 and Sirt3 regulated the acetylation of HIF-1alpha and PDHA1, therefore mediating Salmonella-induced host metabolic shift in the infected macrophages. Interestingly, Sirt1 and Sirt3-driven host metabolic switch also had an effect on the metabolic profile of Salmonella. Counterintuitively, inhibition of Sirt1/3 led to increased pathogen burdens in an in vivo mouse model. Overall, this is a well-designed study.

Comments on revised version:

The authors have performed additional experiments to address the discrepancy between in vitro and in vivo data. While this offers some potential insights into the in vivo role of Sirt1/3 in different cell types and how this affects bacterial growth/dissemination, I still believe that Sirt1/3 inhibitors could have some effect on the gut microbiota contributing to increased pathogen counts. This possibility can be discussed briefly to give a better scenario of how Sirt1/3 inhibitors work in vivo. Additionally, the manuscript would improve significantly if some of the flow cytometry analysis and WB data could be better analyzed.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The current manuscript by Hajra et al deals with the role of the prominent Sirtuins SIRT1 and -3 during infection of macrophages with Salmonella Typhimurium (ST). Apparently, ST infection induces upregulation of host cell SRTs to aid its own metabolism during the intracellular lifestyle and to help reprogramming macrophage polarization. The manuscript has two parts, namely one part that deals with Salmonella infection in cells, where RAW 264.7 murine macrophage-like cells, sharing some features with primary macrophages, were employed. Infected RAW cells displayed a tendency to polarize towards wound-healing M2 and not inflammatory M1 macrophages, which was dependent on SRT. Consequently, the inflammatory response in RAW was more robust in the absence of SRT. Moreover, loss of SRTs leads to impaired bacterial proliferation in these cells, which was attributed to defects in metabolic adaption of the bacteria in the absence of SRT-activity and to the increased M1 inflammatory response.

Unfortunately, the line of argumentation remains incomplete because corresponding assays in mice showed the opposite result as compared to the experiments using RAW 264.7 cells. i.e. loss of SRTs leads to increased bacterial load in animals (versus impaired proliferation in RAW 264.7 cells). The authors cannot explain this discrepancy.

Strengths:

Extensive analysis of Salmonella infection in RAW macrophage-like cells and mice in the context of SRT1/3 function.

Weaknesses:

Lack of connection between the cell-based and organismic data, which are not supportive of each other.

We are highly grateful for your valuable and insightful comments. Thank you for appreciating the merit of our manuscript. We agree with the opposing phenotypes among the RAW264.7 cell line (Fig. 2A), primary peritoneal macrophages (ex vivo) (Fig.2B), and in vivo mouse model (Fig.8) findings. Both RAW264.7 macrophage and peritoneal macrophage infection show attenuated intracellular bacterial proliferation owing to the heightened proinflammatory burst. This is in sharp contrast to our in vivo mouse model of infection which shows increased organ burden and bacterial dissemination. The higher bacterial load in the organs including the spleen (Fig.8B) is attributed to increased pro-inflammatory cytokine burst and ROS production (Fig.8F-H, Fig.S9) triggering bacterial dissemination. The pro-inflammatory arsenals like IL-6, IL-1β and ROS that limit bacterial proliferation within the macrophages (F4/80+ macrophages within the spleen or in RAW264.7 macrophages or primary peritoneal macrophages) are facilitating bacterial dissemination in blood and to the other organs (Fig. 8I-L, Fig.S3F-G). This is in line with the following previous findings-

Klebsiella pneumoniae infection triggers an inflammatory response via secretion of IL-6 upon HIF-1α activation that induces bacterial dissemination (Holden VI, Breen P, Houle S, Dozois CM, Bachman MA. Klebsiella pneumoniae Siderophores Induce Inflammation, Bacterial Dissemination, and HIF-1α Stabilization during Pneumonia. mBio. 2016 Sep 13;7(5):e01397-16. doi: 10.1128/mBio.01397-16. PMID: 27624128; PMCID: PMC5021805.).

Correlation analysis of immune responses to Salmonella infection revealed that increased innate immune “cassette” opposes the adaptive immune arm leading to increased bacterial load in mice (Hotson AN, Gopinath S, Nicolau M, Khasanova A, Finck R, Monack D, et al. Coordinate actions of innate immune responses oppose those of the adaptive immune system during Salmonella infection of mice. Science signaling. 2016;9(410):ra4). 

In our revised manuscript, we have assessed additional splenic populations including CD45+, Ly6C+, and CD11c+ populations. Our results show that the CD45+ splenic population depicts increased bacterial loads like that of the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, within the CD11c+ population, CD45+ granulocytes or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator-treated mice group (Fig. M-S, Fig.S8). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

Reviewer #2 (Public Review):

Dipasree Hajra et al demonstrated that Salmonella was able to modulate the expression of Sirtuins (Sirt1 and Sirt3) and regulate the metabolic switch in both host and Salmonella, promoting its pathogenesis. The authors found Salmonella infection induced high levels of Sirt1 and Sirt3 in macrophages, which were skewed toward the M2 phenotype allowing Salmonella to hyper-proliferate. Mechanistically, Sirt1 and Sirt3 regulated the acetylation of HIF-1alpha and PDHA1, therefore mediating Salmonella-induced host metabolic shift in the infected macrophages. Interestingly, Sirt1 and Sirt3-driven host metabolic switch also had an effect on the metabolic profile of Salmonella. Counterintuitively, inhibition of Sirt1/3 led to increased pathogen burdens in an in vivo mouse model. Overall, this is a well-designed study. There are a few comments below that would further strengthen the current study.

Major comments:

In the in vivo study (lines 436-446) - the authors noticed increased pathogen burden in the EX-527 or the 3TYP-treated mice cohorts but decreased pathogen burden within the F4/80+ macrophage population. What are the other cell types that have increased pathogen burden in splenocytes from EX-527 or the 3TYP treated? Can this be further explored and explained?

While the authors indicated that IL-6 cytokine storm and elevated ROS production could result in bacterial dissemination in vivo, one could also argue that Sirt1/3 inhibitors might have an impact on gut function and/or gut microbiota (PMID: 22115311). Did Sirt1/3 inhibitors also lead to increased pathogen burdens in the gut? If so, the potential effect of these in vivo treatments on gut microbiota/colonization resistance should be discussed.

Minor comment:

Sirt1 has been shown to be degraded during Salmonella infection (PMID: 28192515), which is different from the current study. An explanation should be provided for this.

We thank you for your encouraging and gracious comments. We deeply appreciate your time and efforts in providing constructive feedback for the betterment of our work. As per your precious suggestions, we have assessed additional splenic populations including CD45+, Ly6C+, and CD11c+ populations apart from F4/80+ macrophage populations. Our analysis suggests that the CD45+ splenic population show increased bacterial loads similar to the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, CD11c+ population, CD45+ granulocytes or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator treated mice group (Fig. 8M-S). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

We immensely appreciate the reviewer for this insightful question about the effect of SIRT1/3 on the gut per se. To answer your question, we observed increased pathogen loads within the mesenteric lymph nodes of the gut in the SIRT1/3 inhibitor-treated mice groups (Fig.8B). In our revised manuscript, we evaluated gut inflammation via IL1-β estimation in the mice's ileal tissues and have observed heightened IL-1β production in the inhibitor-treated mice cohorts in comparison to the vehicle control (Fig. S3G). We have also examined gut epithelial pathology via Haematoxylin-Eosin (H&E) staining of the ileal sections to address the effect of in vivo treatment on gut microbiota and colonization resistance which is appended here. However, the gut microbiota crosstalk and their effect on colonization resistance is a part of another current study and it is being examined in detail there. Therefore, this appended H&E has not been incorporated in the revised manuscript.

Author response image 1.

In line with the reference PMID: 28192515, where Sirt1 has been shown to be degraded during Salmonella infection at later time points of infection, our study also has shown that both SIRT1 mRNA (Fig. 1A) and protein levels (Fig. S1A) show an elevated expression at 2h and 6h post-infection and show a downregulation at 16h in comparison to the 6h time point.  However, SIRT3 expression levels remain elevated even at later time points of infection. Therefore, we speculate that there is a shared role between SIRT1 and SIRT3 that facilitates the phenotypes reported in our study.

Reviewer #3 (Public Review):

Summary:

In this paper, Hajra et al have attempted to identify the role of Sirt1 and Sirt3 in regulating metabolic reprogramming and macrophage host defense. They have performed gene knockdown experiments in RAW macrophage cell lines to show that depletion of Sirt1 or Sirt3 enhances the ability of macrophages to eliminate Salmonella Typhimurium. However, in mice, inhibition of Sirt1 resulted in dissemination of the bacteria but the bacterial burden was still reduced in macrophages. They suggest that the effect they have observed is due to increased inflammation and ROS production by macrophages. They also try to establish a weak link with metabolism. They present data to show that the switch in metabolism from glycolysis to fatty acid oxidation is regulated by acetylation of Hif1a, and PDHA1.

Strengths:

The strength of the manuscript is that the role of Sirtuins in host-pathogen interactions has not been previously explored in-depth making the study interesting. It is also interesting to see that depletion of either Sirt1 or Sirt3 results in a similar outcome.

Weaknesses:

The major weakness of the paper is the low quality of data, making it harder to substantiate the claims. Also, there are too many pathways and mechanisms being investigated. It would have been better if the authors had focussed on either Sirt1 or Sirt3 and elucidated how it reprograms metabolism to eventually modulate host response against Salmonella Typhimurium. Experimental evidence is also lacking to prove the proposed mechanisms. For instance, they show correlative data that the knockdown of Sirt1-mediated shift in metabolism is due to HIF1a acetylation but this needs to be proven with further experiments.

We appreciate the reviewer’s critical analysis of our work. In the revised manuscript, we aimed to eliminate the low-quality data sets and have tried to substantiate them with better and conclusive ones, as directed in the recommendations for the author section. We agree with the reviewer that the inclusion of both Sirtuins 1 and 3 has resulted in too many pathways and mechanisms and focusing on one SIRT and its mechanism of metabolic reprogramming and immune modulation would have been a less complicated alternative approach. However, as rightly pointed out, our work demonstrated the shared and few overlapping roles of the two sirtuins, SIRT1 and SIRT3, together mediating the immune-metabolic switch upon Salmonella infection. As per the reviewer’s suggestion, we have performed additional experiments with HIF-1α inhibitor treatment in our revised manuscript to substantiate our correlative findings on SIRT1-mediated regulation of host glycolysis (Fig.7G).

Reviewer #1 (Recommendations For The Authors):

The authors state "SIRT1 and SIRT3 inhibition resulted in increased pathogen loads in organs and triggered enhanced bacterial dissemination, together leading to increased susceptibility of the mice to S. Typhimurium infection owing to increased ROS and IL-6 production." How can this be reconciled? To the reviewer, this is not a convincing explanation. The reviewer is not a mouse pathologist, so maybe did not understand the argument in full.

However, in order to clarify whether these phenomena can be brought into context and explained by for instance cell-autonomous (in (RAW) macrophages) versus non-autonomous (in mice) mechanisms, it would be required to bring in context the organismic phenotype with a cellular phenotype, using more physiologic primary macrophages.

(1) The authors show in Figure 8 that in general SRT inhibition leads to increased infection whereas SRT activation results in decreased infection. This is even true for e the spleen (e.g. Figure 8B), which should be full of macrophages upon infection.

(2) Only Figure 8L implies that endogenous primary, splenic macrophages show a higher infection rate upon pharmacologic SRT activation, which would potentially mirror the RAW results. This is however not supportive of their own explanation: Who would now produce more ROS and IL6 if these macrophages are more supportive of intracellular ST? Is there a difference in the roles or SRTs between different types of macrophages and/or neutrophils? And between macrophages and somatic cells concerning ST infection? The reviewer tends to believe that RAW cells display a defective killing response (such as ROS production) as they are highly transformed cells. Therefore, the authors should use cultured peritoneal macrophages or BMDMs in addition to RAW264.7 cells.

The literature cited by the authors also implies that the inflammatory response in mice is higher in the absence of SRTs. This is in line with a role for SRTs in (negatively) regulating M1 inflammatory polarization but probably not with increased bacterial burden in mice. If it was, then increased dissemination could be explained by increased tissue damage. However, the flow cytometry experiments from infected organs then do not confirm that, as the infection of individual cells is higher upon SRT inhibition. Thus there seems a broad gap between the role of SRTs in ST infection in RAW264.7 cells versus non-transformed cells.

I would not discard the RAW results, as I am convinced that they contain valuable data. However, it needs to be clarified what aspect of the host response RAW 264.7 cells represent. Primary macrophages might likely be more aggressive towards the bacteria. Finally, the question arises: what is the role of the metabolic switch in the in vivo setting?

The reviewer recommends repeating some key experiments by in-vitro-infecting BMDMs or isolated peritoneal macrophages (after some days of culturing) to bridge between the present RAW-derived data and the mouse data. How is the bacterial load with and without SRT inhibitor/activator in primary macrophages, when infected outside of the body? Can ex-vivo infection also affect polarization of e.g. peritoneal macrophages or the metabolic switch? If it is possible to find a conclusive explanation for their data, then this story might really add to our understanding of another aspect of how ST manipulates the host to survive.

In case the reviewer understands the mouse experiments correctly, all assays on peritoneal cells were performed after in-vivo-infection and/or treatment.

Together, RAW 264.7 murine macrophage-like cells might not be the right model to understand the phenotypes in full. As far as the reviewer knows, these cells are not capable of killing bacteria as effectively as activated primary macrophages or neutrophils.

A few of the key findings of RAW264.7 macrophages have been replicated in primary peritoneal macrophages (Fig. 2B, S3E-F, S6B, S7B-D). We wanted to clarify that the peritoneal macrophage experiments were performed ex vivo, wherein peritoneal macrophages were isolated from mice were then subjected to SIRT1/3 inhibitor treatments and Salmonella infection and not after in vivo treatment or infection. In ex vivo setting, we have examined the effect of SIRTs on the metabolic switch during Salmonella infection (Fig. S7B-D) which resembled our RAW264.7 macrophage data. Additionally, in in vivo setting, we have analyzed the transcript level expression of host metabolic genes and corresponding bacterial metabolic genes in infected mice liver and spleen tissue under SIRT1/3 inhibitor treatment (Fig.S7E-F, Fig.6C-D). Our primary peritoneal macrophage data exactly mirrors the RAW264.7 macrophage findings showing attenuated intracellular bacterial proliferation owing to the heightened proinflammatory burst upon SIRT1/3 knockdown or inhibition (Fig.2A-B). This is opposite to our in vivo mouse model of infection which shows increased organ burden and bacterial dissemination (Fig.8A-H). The pro-inflammatory arsenals that limit bacterial proliferation within the macrophages (F4/80+ macrophages within the spleen or in RAW264.7 macrophages or primary peritoneal macrophages) are facilitating bacterial dissemination in blood and to the other organs owing to tissue damage (Fig.8E-L). This is in line with the following previous findings-

Klebsiella pneumoniae infection triggers an inflammatory response via secretion of IL-6 upon HIF-1α activation that induces bacterial dissemination (Holden VI, Breen P, Houle S, Dozois CM, Bachman MA. Klebsiella pneumoniae Siderophores Induce Inflammation, Bacterial Dissemination, and HIF-1α Stabilization during Pneumonia. mBio. 2016 Sep 13;7(5):e01397-16. doi: 10.1128/mBio.01397-16. PMID: 27624128; PMCID: PMC5021805.).

Correlation analysis of immune responses to Salmonella infection revealed that increased innate immune “cassette” opposes the adaptive immune arm leading to increased bacterial load in mice (Hotson AN, Gopinath S, Nicolau M, Khasanova A, Finck R, Monack D, et al. Coordinate actions of innate immune responses oppose those of the adaptive immune system during Salmonella infection of mice. Science Signaling. 2016;9(410):ra4). 

As per the reviewer’s suggestions, we have analyzed other populations apart from F4/80+ macrophages and have observed that the CD45+ splenic population depicts increased bacterial loads like that of the total splenic population within the SIRT1/3 inhibited cohorts. However, CD45+ monocytes and Ly6C positive splenic population exhibit compromised burden within the SIRT1/3 inhibited cohorts. Moreover, the CD1c+ population, CD45+ granulocytes, or lymphocytes show comparable organ loads to that of the vehicle control or SIRT1 activator-treated mice group (Fig.8M-S, Fig.S8). Overall, our data suggest heterogeneous bacterial burden in diverse splenic populations.

Reviewer #3 (Recommendations For The Authors):

Abstract

The authors state that perturbing Sirt1 and Sirt3 results in a shift in Salmonella's metabolism. On the contrary, the data reflects the metabolism in the host cell and not the bacteria. This statement is wrong. They only show increased expression of some of the glycolytic genes in Salmonella, which is not sufficient to make the claim that the switch to fatty acid oxidation in macrophages is due to utilisation of glucose by the bacteria.

We value the reviewer’s response and have accordingly reframed our sentence in the abstract (Line 24-25).

Fig 1: Expression of Sirt1 - The data needs to be supported with a western blot for Sirt1 and Sirt3 but the Western blots shown in the supplementary figure are of very poor quality and do not support the authors' claim.

We have repeated the western blot and have supplemented the previous blot with an alternate blot in Fig. S1A as per your precious input.

Why haven't the authors shown any representative blots for Sirt1 and Sirt3 upon infection with Salmonella mutants? They need to italicize the genes when they describe mRNA expression.

Previously we had only performed transcript-level expression of Sirt1 and Sirt3 upon infection with Salmonella mutants and therefore representative blot image was absent. The gene names have been duly italicized while describing mRNA expression (Line 126-154). We regret the inconvenience caused. We have performed the western blotting to assess the protein expression profile upon infection with Salmonella mutants as per the reviewer’s suggestion and the representative blot image has been duly appended in the revised manuscript (Fig. S1B).

What is the rationale for examining Sirt1 and Sirt3 mRNA in M1 and M2 macrophages? Salmonella infection on its own will polarise the macrophages towards M1. How long were these macrophages infected? The time points are missing.

The rationale behind the examination of Sirt1 and Sirt3 mRNA in M1 and M2 polarized was to ascertain whether indeed M1 polarized macrophages exhibit decreased expression of Sirt1 or Sirt3 and polarization of macrophages toward M2 state show upregulation of Sirt1 and Sirt3 upon Salmonella infection. After confirming these above-mentioned findings through this preliminary experiment, we then hypothesized whether Salmonella infection on its own will polarise the macrophages toward an immunosuppressive M2 state at a later time course of infection as infection drives the induction of SIRT expression and whether this is mediated by Sirt1 and Sirt3 (Fig. 3). We are extremely apologetic for not mentioning the 16h time-point in the figure and the missing time point has been duly documented in the revised manuscript (Line 155).

Fig S2 knockdown of Sirt1 and Sirt3 are not convincing.

We are extremely sorry for the inconclusive knockdown blot. An alternative blot has been substantiated in the revised manuscript (Fig. S2,C-D).

Fig 2A and 2B the time point post infection has not been mentioned. Although it is stated that 2h and 16h post-infection samples were analysed. Only one time point has been shown.

We are sorry for the confusion. We wanted to clarify that Fig.2A and Fig. 2B show the fold proliferation where fold proliferation was calculated as CFU at 16hr divided by CFU at 2hr as mentioned in the materials and methods section under the heading of Intracellular proliferation or gentamicin protection assay.

Fold Proliferation= [CFU at 16h]/[CFU at 2h]

The cytokines data are intriguing in that the increase in IL-6 relative to control is seen only at 2h and 20h but not at 6h. Il-6 at 20h in untransfected cells is comparable to uninfected cells. Did the authors investigate cell death? Salmonella induces various forms of cell death which could account for the decreased cytokine production at later time points.

We have investigated the cell death upon Salmonella infection via MTT assay. At later time points of infection, we indeed observed around 16 percent decrease in cell survival compared to the initial time point of 2h. The results have been appended here and it supports our eminent reviewer’s reasoning for the decreased cytokine production at later time points.

Author response image 2.

Additional cytokines such as IL-1b would be helpful. Also, not sure how uninfected macrophages produce nearly 200pg of IL-10.

As per the author’s critical suggestion, we have assessed the IL-1b cytokine production at 16h post-infection in RAW264.7 macrophages and peritoneal macrophages and mice serum samples at 5th day post-infection (Fig.S3C, S3E-F). Our results indicate increased production of IL-b in the infected SIRT1/3 knockdown RAW264.7 macrophages, SIRT1/3 inhibitor-treated peritoneal macrophages and in mice serum samples under SIRT1/3 inhibitor treatment in comparison to the vehicle control. Additionally, we have quantified IL-1b in mice ileal tissues under SIRT1/3 inhibitor treatment (Fig.S3G) and have obtained heightened intestinal IL-1b production in the inhibitor-treated cohorts. We thank the reviewer for raising the concern for 200pg of IL-10 in the uninfected macrophages. We have repeated the experiment and have provided an alternative representative graph for the experiment wherein the IL-10 levels in the uninfected cohorts range between 20-40pg/ml (Fig. S3B).

It is surprising that the authors have found increased Sirt1 binding to NFkB, however there is no change in acetylated NFkB upon infection (Fig 4B). Acetylated p65 is equally high in uninfected Scrambled siRNA, UI shSirt1, STM Scr, and STM shSirt1. Furthermore, increased binding of Sirt1 with NFkb would mean decreased acetylation hence decreased inflammation. However, Salmonella induces profound inflammation.

We thank the reviewers for their insightful and critical questioning. We truly acknowledge that due to oversaturation there was no apparent change in the acetylated p65 among the different sample sets. Therefore, in the revised manuscript we have provided an image at lower exposure where the changes in the acetylation of the p65 subunit are apparent. Salmonella induces inflammation upon challenge similar to any other pathogens and induces acute inflammatory responses. This heightened acute inflammation at the initial phases of infection subsides at a later phase of infection. Here, we have performed the Sirt1 interaction with NFκB at 16hr post-infection where increased binding of Sirt1 with NFκB facilitates the resolution of the Salmonella-_induced acute inflammation. This is in line with previous reports that suggest SIRT1 suppresses acute inflammation through the promotion of p65 acetylation and inhibition of NFκB activity. (Yang H, Zhang W, Pan H, et al. SIRT1 activators suppress inflammatory responses through promotion of p65 deacetylation and inhibition of NF-κB activity. _PLoS One. 2012;7(9):e46364. doi:10.1371/journal.pone.0046364, Liu TF, Yoza BK, El Gazzar M, Vachharajani VT, McCall CE. NAD+-dependent SIRT1 deacetylase participates in epigenetic reprogramming during endotoxin tolerance. J Biol Chem. 2011;286(11):9856–64., Liu TF, Vachharajani V, Millet P, Bharadwaj MS, Molina AJ, McCall CE. Sequential actions of SIRT1-RELB-SIRT3 coordinate nuclear-mitochondrial communication during immunometabolic adaptation to acute inflammation and sepsis. J Biol Chem. 2015;290(1):396–408.)

Please explain how the acetylated p65 was analysed.

Total endogenous p65 subunit was immunoprecipitated using Anti-NFκB p65 antibody and the immunoprecipitated fraction was probed with Anti-Acetylated Lysine antibody to assess acetylated p65.

An increase in ROS production is seen in a relatively small percentage of cells- not more than 4% of cells. How does this contribute to such a significant difference in intracellular bacterial burden? Also, it is not clear how the authors calculated the fold change in proliferation. It is better to show the actual bacterial burden logarithmically.

We strongly agree with the reviewer’s concerns, and we have reanalyzed the flow cytometric data set. The revised data have been presented in Fig. S5 which shows a considerable increase in DCFDA positive population. For instance, the infected scrambled control shows around 2.44% of ROS-producing cells, however knockdown of SIRT1 and SIRT3 increases the ROS-producing cells to 27.34% and 28.64% respectively.

Fold proliferation was calculated as CFU at 16hr divided by CFU at 2hr as mentioned in the materials and methods section under the heading of Intracellular proliferation or gentamicin protection assay. Fold proliferation has been calculated as opposed to absolute CFU values to nullify the differential phagocytosis of bacteria to the macrophages among the samples.

Fold Proliferation= [CFU at 16h]/[CFU at 2h]

An increase in metabolic genes is not sufficient to show that the macrophages are metabolically reprogrammed.

We thank the reviewer for the valuable comment. We agree that an increase in metabolic gene profile is not sufficient to claim metabolic reprogramming. Therefore, in addition to the metabolic gene profile, we have estimated lactate production (end-product of glycolysis) as an indicator of glycolysis (Fig. 5 C-E) and have performed the fatty acid β oxidation activity (Fig. 5G-H) to support our claims.

Figure 5F the band intensities do not visually match the bands shown for PFK. For instance, shSIRT1 STM (1.00) and shSIRT3 STM (0.81).

We are extremely sorry for the erroneous band intensity for shSIRT3. Upon reanalysis of the band intensities, we have corrected the band intensity for shSIRT3 to 2.28 (Fig.5F).

It is surprising that HADHA is not expressed in uninfected samples.

We are extremely apologetic for the inappropriate representative blot. We feel that the discrepancy might have arisen due to the usage of old antibodies. We have provided an alternate blot for the HADHA gene where fresh antibody staining solution was used for probing which shows expression even in the uninfected samples (Fig.5F).

Figure 6A - What is the significance of PFA fixed samples (PI) compared to SI samples? This has not been discussed.

PFA-fixed samples are paraformaldehyde-treated bacterial samples that harbor the immune signals or Pattern Associated Molecular Patterns (PAMPs). The rationale for using PI in addition to SI samples was to show whether the phenomena is driven by live metabolically active pathogens or is mediated by PAMPs.

I understand that the hypothesis is that during the later phase of infection, there is an increase in fatty acid oxidation which correlates with a decrease in inflammation. However, at 6h there is no increase in genes regulating fatty acid oxidation. Why did the authors choose 6h when the previous experiments have been done at 16h?

We indeed agree with the reviewer’s understanding of our hypothesis that there is an increase in fatty acid oxidation along the progression of infection which correlates with a decrease in inflammation. The Salmonella intracellular replication has been reported to commence at 6h post-internalization when SPI-2 effector expression is fully established (Helaine S, Thompson JA, Watson KG, Liu M, Boyle C, Holden DW. Dynamics of intracellular bacterial replication at the single cell level. Proc Natl Acad Sci U S A. 2010;107(8):3746-3751. doi:10.1073/pnas.1000041107). Therefore, we have assessed the 6h timepoint post-infection in addition to the initial and later timepoints of 2h and 16h respectively. Additionally, the nanostring gene profiling data of both host and bacterial genes indicate the onset of both metabolic (Fig. 5A, 6A) and immune genes (Fig. 3A) modulation at 6h post-infection. We have validated these results via qPCR studies and have observed an upregulation in the transcript level of fatty acid oxidation genes as depicted in Fig. S7A in RAW264.7 macrophages.

Line 355 it is mentioned that Sirt1 and Sirt3 abrogate metabolic shift by reducing glycolytic flux. This is incorrect as experiments such as carbon chase assays have not been performed to investigate glycolytic flux.

As per the reviewer’s valuable suggestion, we have removed the word ‘flux’ from the above-mentioned statement(Line 351, Line 353).

Lines 392-393: "We immunoprecipitated PDHA1 and checked for its interaction with SIRT3 or SIRT1 under knockdown condition of SIRT3 or upon SIRT3 inhibitor treatment (Fig.7 G-H)"

What is the rationale for checking PDHA1 interaction with Sirt under Sirt knockdown conditions?

We are thankful to the reviewer for the critical comments. The rationale for checking PDHA1 interaction with Sirt was to ascertain that indeed Sirt interacted with PDHA1 under S. Typhimurium infection and abrogation of either protein expression (knockdown) or their enzymatic activity (inhibitor treatment) diminished the interaction.

Moreover, the blots are very confusing and do not represent the authors' claims.

(1) In the input blot I do not see Sirt3 depletion in shSirt3 knockdown sample.

The knockdown has been quantified in the input blot as per your suggestion. A knockdown of 40% has been obtained in the uninfected dataset whereas a knockdown of 47.1% has been obtained in the infected data set at 16h post-infection (Fig.7H).

(2) Why does Sirt1 interact with PDHA1 similar to Sirt3. Do both the proteins bind to PDHA1 at the same time/ competitively? If so do they both deacetylate?

In literature, Sirt3 has been shown to interact with PDHA1 and deacetylate PDHA1. However, the interaction of Sirt1 with PDHA1 has not been reported previously and therefore we are unable to comment on the exact dynamics of the interaction. Future studies need to be performed to explore these phenomena in depth. However, SIRT1 agonist SRT1720 has been shown to impact PDH phosphorylation and its activity (Han Y, Sun W, Ren D, Zhang J, He Z, Fedorova J, Sun X, Han F, Li J. SIRT1 agonism modulates cardiac NLRP3 inflammasome through pyruvate dehydrogenase during ischemia and reperfusion. Redox Biol. 2020 Jul;34:101538).

(3) Figure 7I in the IP: IgG samples Sirt3 seem to bind to IgG non-specifically, which questions the specificity of Sirt3 binding to PDHA1.

We appreciate the reviewer for pointing out this concern. The immunoprecipitation experiment has been repeated and the same has been appended in the revised manuscript and we observe no non-specific binding of Sirt3 antibody to IgG.

(4) In Figure 7I all the bands Ac PDHA1, PDHA1, and Sirt3 look similar with double bands, which has not been seen in other blots. How is this possible?

This cannot explain the increase in beta-oxidation observed.

We thank the reviewer for raising this concern. We have repeated the experiment and provided the alternative blot as per the reviewer’s suggestion.

The rationale for performing this experiment was to show that SIRT plays an important role in the activation of downstream TCA cycle pathways via PDHA1 deacetylation during Salmonella infection. The deacetylation of PDHA1 has been previously reported to cause transcriptional activation of the downstream TCA cycle and oxidative phosphorylation (Zhang Y, Wen P, Luo J, et al., Cell Death Dis.,2021). Additionally, PDHA1 hyperacetylation has been reported to cause lactate overproduction (An, S., Yao, Y., Hu, H. et al. PDHA1 hyperacetylation-mediated lactate overproduction promotes sepsis-induced acute kidney injury via Fis1 lactylation. Cell Death Dis 14, 457 (2023)). In our study, increased lactate production and PDHA1 hyperacetylation have been observed during SIRT3 inhibition conditions upon Salmonella infection.

eLife assessment

The work by Han and collaborators describes valuable findings on the role of Akkermansia muciniphila during ETEC infection. If confirmed, these findings will add to a growing list of beneficial properties of this organism. However, as it stands, the strength of the evidence used to justify the conclusions in the manuscript is incomplete.

Reviewer #2 (Public Review):

The authors indicated that the adherence of ETEC is to intestinal epithelial cells. However, it is also possible that the majority of ETEC may reside in the intestinal mucus, particularly under in vivo infection condition. The colonization of ETEC in the jejunum and colon of piglets (Fig 2C) and in the intestines of mice (Fig S2A) does not necessarily reflect the adherence of ETEC to epithelial cells. Please verify these observations with other methods, such as immunostaining. Also, while Salmonella enterica serovar Typhimurium or Listeria monocytogenes can invade organoids within 1 hour, it is unknown if ETEC invade into organoids in this study. Clarifying this will help resolve if A. muciniphila block the adherence and/or invasion of ETEC. Please also address if A. muciniphila metabolites could prevent ETEC infection in the organoid models.

Reviewer #3 (Public Review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

After revision, the bioinformatics section of the methods is still jumbled and may indicate issues in the pipeline. Important parameters are not included to replicate analyses. Merging the forward and reverse reads may represent a problem for denoising. Chimera detection was performed prior to denoising.

Potential denoising issues for NovaSeq data was not addressed in the response. The authors did not clarify if multiple testing correction was applied; however, it may be assumed not as written. The raw sequencing data made available through the SRA accession (if for the correct project) indicates it was a MiSeq platform; however, the sample names do not appear to link up to this experimental design and metadata not sufficient to replicate analyses.

Author response:

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

Reviewer #1 (Public Review):

Summary:

In this paper, the authors investigate the impact of fecal microbiota transfer (FMT) on intestinal recovery from enterotoxigenic E. coli infection following antibiotic treatment. Using a piglet model of intestinal infection, the authors demonstrate that FMT reduces weight loss and diarrhea and enhances the expression of tight junction proteins. Sequencing analysis of the intestinal microbiota following FMT showed significant increases in Akkermansia muciniphila and Bacteroides fragilis. Using additional mouse and organoid models, the authors examine the impact of these microbes on intestinal recovery and modulation of the Wnt signaling pathway. Overall, the data support the notion that FMT following ETEC infection is beneficial, however, additional investigation is required to fully elucidate the mechanisms involved.

Strengths:

Initial experiments used a piglet model of infection to test the value of FMT on recovery from E. coli. The FMT treatment was beneficial and the authors provide solid evidence that the treatment increased the diversity of the microbiota and enhanced the recovery of the intestinal epithelium. Sequencing data highlighted an increase in Akkermansia muciniphila and Bacteroides fragilis after FMT.

The mouse data are consistent with the observations in pigs, and reveal that daily gavage with A. muciniphila or B. fragilis enhances intestinal recovery based on histological analysis, expression of tight junction proteins, and analysis of intestinal barrier function.

The authors demonstrate the benefit of probiotic treatment following infection using a range of model systems.

Weaknesses:

Without sequencing the pre-infection pig microbiota or the FMT input material itself, it's challenging to firmly say that the observed bloom in Akkermansia muciniphila and Bacteroides fragilis stemmed from the FMT.

Response: We have determined the relative abundance of each bacterium in fecal bacterial suspension, referring to Hu et al. (2018). The absolute abundances of Akkermansia muciniphila and Bacteroides fragilis in the FMT were 1.3 × 103 ± 2.6 × 103 and 4.5 × 103 ± 6.1 × 103 respectively.

Reference:

Hu LS, Geng SJ, Li Y, et al. Exogenous Fecal Microbiota Transplantation from Local Adult Pigs to Crossbred Newborn Piglets. Front. Microbiol. 2018, 8.

The lack of details for the murine infection model, such as weight loss and quantification of bacterial loads over time, make it challenging for a reader to fully appreciate how treatment with Akkermansia muciniphila and Bacteroides fragilis is altering the course of infection. Bacterial loads of E. coli were only quantified at one time point, and the mice that received A. muciniphila and B. fragilis had very low levels of E. coli. Therefore, it is not clear if all mice were subjected to the same level of infection in the first place. The reduced translocation of E. coli to the organs and enhanced barrier function may just reflect the low level of infection in these mice. Further, the authors' conclusion that the effect is specific to A. muciniphila or B. fragilis would be more convincing if the experiments included an inert control bacterium, to demonstrate that gavage with any commensal microbe would not elicit a similar effect.

The weight loss was added in Figure S2A. All mice were subjected to the same level of infection in the first place.

Many of the conclusions in the study are drawn from the microscopy results. However, the methods describing both light microscopy and electron microscopy lack sufficient detail. For example, it is not clear how many sections and fields of view were imaged or how the SEM samples were prepared and dehydrated. The mucus layer does not appear to be well preserved, which would make it challenging to accurately measure the thickness of the mucus layer.

For light microscopy, 3-4 fields were selected from each mouse to count about 30 crypts. The method of electron microscopy was complemented on line 263-270. We have removed data of the mucus layer.

Gene expression data appears to vary across the different models, for example, Wnt3 expression in mice versus organoids. Additional experiments may be required to clarify the mechanisms involved. Considering that both of the bacteria tested elicited similar changes in Wnt signaling, this pathway might be broadly modulated by the microbiota.

The reason why the Wnt3 expression pattern is different in mice and in porcine intestinal organoids may be caused by the different infection periods of ETEC in vivo and in vitro. Furthermore, in vivo, the stem cell niche of intestinal stem cells is not only regulated by intestinal epithelial cells, but also affected by mesenchymal cells in connective tissues (Luo et al., 2022). However, in vitro models, stem cell niche is only regulated by epithelial secretory factors, which may also account for the differences in in vitro and in vivo results.

It has been reported that B. fragilis pretreatment significantly increased the relative abundance of A. muciniphila in the intestine of CDI mice, and the growth and maintenance of A. muciniphila were involved in the restoration of intestinal barrier integrity after CDI infection, indicating that there might exist a bacterial metabolic symbiosis between A. muciniphila and B. fragilis (Deng et al., 2018).

References:

Luo HM, Li MX, Wang F, et al. The role of intestinal stem cell within gut homeostasis: Focusing on its interplay with gut microbiota and the regulating pathways. Int. J. Biol. Sci. 2022, 18(13): 5185-5206.

Deng HM, Yang SQ, Zhang YC, et al. Bacteroides fragilis Prevents Clostridium difficile Infection in a Mouse Model by Restoring Gut Barrier and Microbiome Regulation. Front. Microbiol. 2018, 9.

The unconventional choice to not include references in the results section makes it challenging for the reader to put the results in context with what is known in the field. Similarly, there is a lack of discussion acknowledging that B. fragilis is a potential pathogen, associated with intestinal inflammation and cancer (Haghi et al. BMC Cancer 19, 879 (2019) ), and how this would impact its utility as a potential probiotic.

Bacteroides fragilis is one of the symbiotic anaerobes within the mammalian gut and is also an opportunistic pathogen which often isolated from clinical specimens. Bacteroides fragilis was first isolated from the pathogenic site and considered to be pathogenic bacteria. However, with the deepening of research, it is gradually realized that in the long-term evolution process, Bacteroides fragilis colonized in the gut has established a friendly relationship with the host, which is an essential component for maintaining the health of the host, especially for obesity, diabetes and immune deficiency diseases. We have supplemented the discussion on line 598-603.

Reviewer #2 (Public Review):

Ma X. et al proposed that A. muciniphila was a key strain that promotes the proliferation and differentiation of intestinal stem cells by acting on the Wnt/β-catenin signaling pathway. They used various models, such as the piglet model, mouse model, and intestinal organoids to address how A. muciniphila and B. fragilis offer protection against ETEC infection. They showed that FMT with fecal samples, A. muciniphila or B. fragilis protected piglets and/or mice from ETEC infection, and this protection is manifested as reduced intestinal inflammation/bacterial colonization, increased tight junction/Muc2 proteins, as well as proper Treg/Th17 cells. Additionally, they demonstrated that A. muciniphila protected basal-out and/or apical-out intestinal organoids against ETEC infection via Wnt signaling. While a large body of work has been performed in this study, there are quite a few questions to be addressed.

Major comments:

- The similar protective effect of FMT with fecal samples, A. muciniphila or B. fragilis is perhaps not that surprising, considering that FMT likely restores microbiota-mediated colonization resistance against ETEC infection. While FMT with fecal samples increases SCFAs, it is unclear whether/how FMT with A. muciniphila or B. fragilis alter the microbiota composition/abundance as well as metabolites in the current models in a way that offers protection.

We examined changes in the gut microbiota of mice treated with A. muciniphila and B. fragilis through 16s rRNA, and results showed that both A. muciniphila and B. fragilis improved the alpha and beta diversities of the microbiota, while these results were not included in this manuscript.

- Does ETEC infection in piglets/mice cause histological damage in the intestines? These data should be shown.

The results of scanning electron microscopy (Figure 3A) showed the intestinal damage of piglets after ETEC infection. H&E staining and transmission electron microscopy (Figure 5A and 5B) showed the intestinal damage of mice after ETEC infection.

- Line 447, "ETEC adheres to intestinal epithelial cells". However, there is no data showing the adherence (or invasion) of ETEC to intestinal epithelial cells, irrespective of piglets/mouse/organoids.

The scanning electron microscope (Figure 3A bottom) showed that ETEC K88 infected piglets existed obvious rod-shaped bacterial adhesion on the surface of microvilli. Figure 2C showed the colonization of ETEC K88 in the jejunum and colon of piglets. Figure S2A showed the E. coli colonization in intestines and other tissues of mice.

- In both basal-out and apical-out intestinal organoid models, A. muciniphila protects organoids against ETEC infection. Did ETEC enter into intestinal epithelial cells at all after only one hour of infection? Is the protection through certain A. muciniphila metabolites?

It has been reported that the duration of the co-culture for studying the host-microbiota cross-talk by apical-out organoids model is 1 hour (Poletti et al., 2021). In addition, Co et al. (2019) used apical-out organoids model to study host-pathogen interactions, with Salmonella enterica serovar Typhimurium or Listeria monocytogenes invading organoids for an hour.

References:

Poletti M, Arnauts K, Ferrante M, et al. Organoid-based Models to Study the Role of Host-microbiota Interactions in IBD. J. Crohns Colitis. 2021, 15(7): 1222-1235.

Co JY, Margalef-Catala M, Li XN, et al. Controlling Epithelial Polarity: A Human Enteroid Model for Host-Pathogen Interactions. Cell Reports. 2019, 26(9): 2509-2520.

Reviewer #3 (Public Review):

Summary:

The manuscript by Ma et al. describes a multi-model (pig, mouse, organoid) investigation into how fecal transplants protect against E. coli infection. The authors identify A. muciniphila and B. fragilis as two important strains and characterize how these organisms impact the epithelium by modulating host signaling pathways, namely the Wnt pathway in lgr5 intestinal stem cells.

Strengths:

The strengths of this manuscript include the use of multiple model systems and follow-up mechanistic investigations to understand how A. muciniphila and B. fragilis interacted with the host to impact epithelial physiology.

Weaknesses:

The major weakness is that, as presented, the manuscript is quite difficult to follow, even for someone familiar with the field. The lack of detail in figure legends, organization of the text, and frequent use of non-intuitive abbreviated group names without a clear key (ex. EP/EF, or C E A B) make comprehension challenging. The results section is perhaps too succinct and does not provide sufficient information to understand experimental design and interpretation without reading the methods section first or skipping to the discussion (as an example: WNT-c59 treatment). Extensive revisions could be encouraged to aid in communicating the potentially exciting findings.

The abbreviations of experimental groups are firstly defined in the Methods and Materials, and we have supplemented the experimental design in the results section on line 397-399, 439-442 and 516-520.

The bioinformatics section of the methods requires revision and may indicate issues in the pipeline. Merging the forward and reverse reads may represent a problem for denoising. Also since these were sequenced on a NovaSeq, the error learning would have to be modified or the diversity estimates would be inappropriately multiplied. "Alpha diversity and beta diversity were calculated by normalized to the same sequence randomly." Not sure what this means, does this mean subsampled? "Blast was used for sequence alignment", does this mean the taxonomic alignment? This would need to be elaborated on and database versions should be included. The methods, including if any form of multiple testing was included, for LEFSE was also not included.

Denoising was conducted using UNOISE3 to correct for sequencing errors. Subsequent analysis of alpha diversity and beta diversity were all performed based on the output normalized data. Multiple sequence alignment was performed using MUSCLE (v3.8.31) software to obtain the phylogenetic relationships of all OTUs sequences. We have supplemented the method of multiple testing on line 323-328.

Reviewer #1 (Recommendations For The Authors):

At some points, the rationale for using both porcine and murine models was unclear, and it would be helpful for the reader to elaborate on the benefits of these models and why they were used in the introduction. Similarly, it would be helpful to describe the benefits of basal-in organoids versus injecting standard organoids with bacteria.

The main subject of this study was piglets, supplemented by a mouse model for validation. Interpretation of measurements from organoid microinjection experiments must account for multiple confounding variables such as heterogeneous exposure concentrations and durations, as well as impacts of disrupting the organoid wall. We have added the description in the introduction on line 88-90.

Line 165 -- The number of piglets used seems high, is it correct approximately 100 pigs were used?

Nine litters were selected for processing, while only 18 piglets were finally slaughtered.

There is very little discussion of the preliminary experiment that the authors used to determine how much bacteria to use. I recommend either discussing the data and how the doses were chosen or omitting it. It was not clear if the authors used pasteurized or live bacteria in the experiments. It would also be interesting to include a discussion of the observation that relatively low levels of Akkermansia (10^6 CFU) appeared more beneficial than the higher doses, typically used in these types of experiments.

We removed these results. The experiments used live bacteria.

Microscopy methods for both light microscopy and EM would be stronger with added details including how many sections and fields of view were imaged and how the numbers of goblet cells normalized across samples. Without having a clear cross-section of a crypt, it is not clear to me how the images can be used to accurately quantify the number of cells per crypt. Additional details in the methods on how many total crypts were counted should also be included.

For light microscopy, 3-4 fields were selected from each mouse to count about 30 crypts. We have removed the data of the mucus layer and goblet cells.

Line 236 -- missing which gene was used.

The Genbank Accession was added on line 232-233.

Line 310 -- OTU nomenclature.

We have supplemented the OTU nomenclature on line 314.

Line 413 -- This line seems inconsistent with the data analysis described in the methods section. The authors may need to expand their description of the 16S data analysis to be clear and reproducible.

We have redescribed the 16S data analysis on line 312-328.

Line 413 -- it is not surprising that 16s analysis did not capture species, it will have limited resolution beyond the genus level.

We deleted this sentence.

Methods are missing some details on the data analysis, eg. methods/programs and statistical analysis of PCoA and NMDS, LefSe.

The methods and statistical analysis of PCoA, NMDS and LEfSe were supplemented on line 323-328.

Fig 4C -- The images do not clearly capture the mucus layer or how it was analyzed. The sections appear to be cut at a slight angle, with multiple partial sections of crypts. I think this might make it challenging to count goblet cells, especially if the counts are normalized over the number of crypts or villi. The mucus layer does not appear well preserved. For example, I would expect to see an intact mucus layer lining the colon in the PBS control group. Re-cutting sections with a clean cross-section through the tissue will make data analysis easier.

We have removed data of the mucus layer.

Fig 4D -- The images appear to be of the mouse proximal colon, whereas the mucus layer and most muc2 will be in the distal colon. If the authors have tissue sections of the distal colon, this may give a clearer image of the mucus layer and might be more consistent with the TEM images in Fig. 4B.

We apologize for the absence of the distal colon sections.

To fully preserve the mucus layer, in addition to fixing in Carnoy's solution, the embedding process must be run without the standard washes in 70% ethanol (see: Johansson and Hansson. Methods Mol Biol. (2012) 229; doi: 10.1007/978-1-61779-513-8_13). The mucus will wash away during standard paraffin embedding if the tissue is washed with 70% ethanol, and I wonder if that has occurred in these samples.

The tissue wasn’t washed with 70% ethanol.

Fig 6A and 6B -- Although the legend indicates that the data is representative of two independent experiments, it is not clear how many fields of view or cells were imaged. In the bar graphs, it is not clear how many crypts were analyzed and from how many fields of view.

3-4 fields were selected from each mouse to count about 30 crypts.

**For all of the bar graphs, this could be addressed by displaying all of the data points, rather than just the mean, to give the reader a sense of how many cells were counted. (as was done in Fig 7B).

We have changed the bar graphs with data points.

498-501 -- The text says that the gene expression patterns in the organoids are consistent with the in vivo data, but the data patterns of gene expression appear to be different. For example, patterns for Wnt3 and B-catenin expression in mice, appear to be the opposite of what was observed in the organoid?

Lines 509-512 mean that the expression patterns of mice in organoids and in vivo is consistent. Figure 7C was incorrectly written as Figure 8C, we have changed it.

Since Akkermansia does not grow under aerobic conditions, it should be made clear that the organoid co-culture treatment does not involve actively growing bacterial cultures.

Reunanen et al. found that Akkermansia can tolerate oxygen, more than 90% Akkermansia can keep for 1 h under oxic, 5% CO2 conditions.

Reference:

Reunanen J, Kainulainen V, Huuskonen L, et al. Akkermansia muciniphila Adheres to Enterocytes and Strengthens the Integrity of the Epithelial Cell Layer. Appl. Environ. Microbiol. 2015, 81(11): 3655-3662.

Minor points

Line 50 -"evidence".

We have changed to “evidence” on line 49.

Line 64, 422 - italicize, check italics throughout.

We have checked italics throughout the manuscript.

Line 64 - may need to be reworded.

We have changed to “Clostridioides difficile” on line 66.

Line 77 - pathogen.

We have changed to “pathogen” on line 77.

Line 161 - the.

We have removed “the” on line 161.

Line 178 - mouse.

We have changed to “mouse” on line 179.

Line 313 -- wording is confusing.

We have changed the description on line 319-320.

Line 318 -- Silva version #.

The version is Silva 132. We have added it on line 316.

Line 334 - Manufacturer for Live/Dead cell stain?

The Live/Dead cell stain was used BD Biosciences FVS510. We have added it on line 345.

Line 433 -- FD4 not defined until here.

We have refined the FD4 on line 218-219.

Line 512 -- but did not promote.

We have changed to “but did not promote” on line 526.

Line 517 -- Looks like this should be "basal-in organoids" instead of basal-out?

We have changed the "basal-out" to "apical-to" on line 531.

Line 546 -- induced neonatal should be protected?

They are in separate pens.

Jumps from Fig 7B to Fig 8C in the text.

We apologize for the wrong writing, and we have change it.

Reviewer #2 (Recommendations for The Authors):

The title itself is a bit misleading. Please consider changing it. The authors meant that A. muciniphila prevents pathogen invasion, but does not function in pathogen invasion.

We have changed the title.

Major comments:

- Figures 4A, 4D, and 6B should include presentation of cross-section pictures.

We provided cross-section pictures to the journal.

- Figures 7, 8, and 9 should indicate clearly whether mouse or piglet organoids are used. For instance, in the main text, line 490, it indicates piglet organoids, but in Figure 7A legend, it indicates mouse tissue.

We apologize for the misspelling, and have changed to “mice” on line 501-502.

- In Figure 7A, the 3rd row, 2nd panel, crypts formed into spherical organoids; whereas in Figure 8, ETEC infection of basal-out organoids formed budding organoids. This needs to be better explained.

Mouse intestinal organoids were cultured ex vivo from crypts isolated from mice infected with ETEC, while porcine intestinal organoids were co-cultured with ETEC in vitro.

Minor comments:

- In the result section, the numbering of Figures or supplementary Figures is problematic, i.e it should start with Figure 1..., Figure S1, but not directly go to Figure S2A etc.

The Figure 1 was in Materials and Methods.

- Line 458, please add the gating strategy used in the flow cytometry study.

The gating strategy was added on line 351-356.

- The effect of A. muciniphila on the proliferation of intestinal epithelium through the Wnt/β-catenin signaling pathway is well known (such as PMID: 32138776). The authors should discuss this in detail.

We have supplemented the discussion on line 637-639.

Reviewer #3 (Recommendations For The Authors):

It is somewhat unusual that the results from the piglets are in the supplement as this is a major strength of the manuscript (Fig S2).

We have put these results into Figure 2 of the manuscript.

"Collectively, our results may provide theoretical basis that FMT is a promising mitigation method for pathogenic bacteria infection and a new strategy for precise application of FMT in clinical and livestock production"- This is somewhat of an odd statement as the introduction of the manuscript completely skips over most of what is known about FMTs in the context of C. difficile. Also if anything, does the authors' own data not point mostly at using A. muciniphila on its own? Clinical trials are well underway in humans.

We have changed the sentences to “Collectively, our results may provide theoretical basis that A. muciniphila is a promising method to repair intestinal barrier damage and a new strategy for the precise application of A. muciniphila in livestock production.” on line 98-100.

Line 26: I am not sure probiotic is the right word here given its strict scientific definition. Perhaps beneficial or protective would be more appropriate.

We have changed “probiotic” to “beneficial” on line 25.

Line 27: I believe AIMD is antibiotic-induced microbiome-depletion in most usages which may be more accurate and informative than dysregulated.

The type, dosing, and time of antibiotic we used were applied to induce microbiota disorder.

It would appear that there are issues in the reference formatting where a number of journal names are missing.

We have re-edited the reference formatting.

Line 64- I believe eLife requires the standard practice of italicizing genus and species names. Also Clostridium difficile should now be referred to as Clostridioides difficile.

We have changed to “Clostridioides difficile” and italicized it on line 66 and 569. The italicizing genus and species names were checked throughout the manuscript.

Figure S2C: is it not clear why the melt curve was included here, but the legend should make it more clear what is being shown. I assume this is to provide evidence of specificity?

The melting curve was used to demonstrate that only the ETEC K88 could be amplified by the primers we used. We have added an illustration in the figure legend.

Figure 2D: there should be a quantitative analysis done on the staining of Muc2.

We have quantified the staining of MUC2 in Figure 3D.

Figure 3: The legends are not sufficient. For example: it is not clear what Figure 3A actually shows as the y-axis is not labelled and it is not clear what the relationship is between this and the anosim which is a function for permanova.

Anosim analysis was performed using the R software with anosim package function based on the rank order of Bray-Curtis distance values to test the significance of differences between groups. The y-axis is the rank of the distance between samples.

Line 416- OTU not OUT.

We have changed to “OTU” on line 428.

Figure 4- the naming key needs to be included in the figure legend. C, E, A, and B are immediately obvious.

The naming key was included in the figure legend.

Methods: additional information on the flow cytometry gating strategy/controls should be included.

The gating strategy was added on line 351-356.

Author response:

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

Reviewer #1 (Public Review):

I'll begin by summarizing what I understand from the results presented, and where relevant how my understanding seems to differ from the authors' claims. I'll then make specific comments with respect to points raised in my previous review (below), using the same numbering. Because this is a revision I'll try to restrict comments here to the changes made, which provide some clarification, but leave many issues incompletely addressed.

As I understand it the main new result here is that certain recurrent network architectures promote emergence of coordinated grid firing patterns in a model previously introduced by Kropff and Treves (Hippocampus, 2008). The previous work very nicely showed that single neurons that receive stable spatial input could 'learn' to generate grid representations by combining a plasticity rule with firing rate adaptation. The previous study also showed that when multiple neurons were synaptically connected their grid representations could develop a shared orientation, although with the recurrent connectivity previously used this substantially reduced the grid scores of many of the neurons. The advance here is to show that if the initial recurrent connectivity is consistent with that of a line attractor then the network does a much better job of establishing grid firing patterns with shared orientation.

Beyond this point, things become potentially confusing. As I understand it now, the important influence of the recurrent dynamics is in establishing the shared orientation and not in its online generation. This is clear from Figure S3, but not from an initial read of the abstract or main text. This result is consistent with Kropff and Treves' initial suggestion that 'a strong collateral connection... from neuron A to neuron B... favors the two neurons to have close-by fields... Summing all possible contributions would result in a field for neuron B that is a ring around the field of neuron A.' This should be the case for the recurrent connections now considered, but the evidence provided doesn't convincingly show that attractor dynamics of the circuit are a necessary condition for this to arise. My general suggestion for the authors is to remove these kind of claims and to keep their interpretations more closely aligned with what the results show.

We would like to clarify that the simple (flexible) attractor is a weaker condition than the ones previously used to align grid cells. However, by no means we claim that it is a necessary condition for grid maps to align. Other architectures, certainly more complex ones but perhaps even simpler ones, can align grid maps in our model.

Major (numbered according to previous review)

(1) Does the network maintain attractor dynamics after training? Results now show that 'in a trained network without feedforward Hebbian learning the removal of recurrent collaterals results in a slight increase in gridness and spacing'. This clearly implies that the recurrent collaterals are not required for online generation of the grid patterns. This point needs to be abundantly clear in the abstract and main text so the reader can appreciate that the recurrent dynamics are important specifically during learning.

We respectfully disagree with the interpretation of this result. In this model cells self-organize to produce aligned grid maps. In such systems it makes sense to characterize the equilibrium states of the system. We turned learning off in Figure S3 to show that the recurrent connections have a contractive effect on grid spacing. But artificially turning off learning means that one can no longer make claims about the equilibrium states of the system, since it can no longer evolve freely. In a functional network, if the recurrent attractor is removed, the system will evolve towards poor gridness and no alignment no matter what the starting point is, as also shown in Figure S3. Several experimental results invite us to think of grid cells as the equilibrium solution of a series of constraints that is ready to change at any time: Barry et al, 2012; Yoon et al, 2013; Carpenter et al, 2015; Krupic et al, 2015; Krupic et al, 2018; Jayakumar et al, 2019.

One point in which we perhaps agree with the reviewer is that information about the hexagonal maps is kept in the feedforward weights, while behavior and the recurrent collaterals act as constraints of which these feedforward weights are the equilibrium solution.

(2) Additional controls for Figure 2 to test that it is connectivity rather than attractor dynamics (e.g. drawing weights from Gaussian or exponential distributions). The authors provide one additional control based on shuffling weights. However, this is far from exhaustive and it seems difficult on this basis to conclude that it is specifically the attractor dynamics that drive the emergence of coordinated grid firing.

Again, we do not claim that this is the only way in which grid maps can be aligned, but it is the simplest one proposed so far. We were asked if it was the specific combination of input weights to a cell rather than the organization provided by the attractor which resulted in aligned maps. By shuffling the inputs to a cell we keep the combination of inputs invariant but lose the attractor architecture. Since grid maps in this new situation are not aligned, we can safely conclude that it is not the combination of inputs per se, but the specific organization of these inputs that allows grid alignment. It is not fully clear to us what ‘exhaustive’ means in this context.

(3) What happens if recurrent connections are turned off? The new data clearly show that the recurrent connections are not required for online grid firing, but this is not clear from the abstract and is hard to appreciate from the main text.

This point is related to (1). Absent this constraint, Figure S3 shows that the system evolves toward larger spacing, with poorer gridness and no alignment.

(4) This is addressed, although the legend to Fig. S2D could provide an explanation / definition for the y-axis values.

We have now added: Mean input fields are the sum of all inputs of a given kind entering a neuron at a given moment in time, averaged across cells and time.

(5) Given the 2D structure of the network input it perhaps isn't surprising that the network generates 2D representations and this may have little to do with its 1D connectivity. The finding that the networks maintain coordinated grids when recurrent connections are switched off supports my initial concern and the authors explanation, to me at least, remain confusing. I think it would be helpful to consider that the connectivity is specifically important for establishing the coordinated grid firing, but that the online network does not require attractor dynamics to generate coordinated grid firing.

This point is related to (1) and (3). We agree with the reviewer that the input lies within a 2D manifold, but this is not something that the network has to find out because it receives one datapoint of information at a time. This alone is not enough to form aligned grid cells, since each grid cell can find a roughly equivalent equilibrium in a different direction. It is only the constraint imposed by the recurrent collaterals that aligns grid maps, and, as we show, this constraint does not need to be constructed ad hoc to work on 2D, as previously thought. When recurrent connections are switched off, the system evolves toward unaligned grid maps, with larger spacing and lower gridness. Regarding the results obtained after modifying the network and turning off learning, we think they have a very limited scope (in this case showing the contractive effect of recurrent collaterals on grid spacing), given that the system is artificially being kept out of its natural equilibrium.

(6) Clarity of the introduction. This is somewhat clearer, but I wonder if it would be hard for someone not familiar with the literature to accurately appreciate the key points.

We have made our best effort to improve the clarity of the introduction.

(7) Remapping. I'm not sure why this is ill posed. It seems the proposed model can not account for remapping results (e.g. Fyhn et al. 2007). Perhaps the authors could just clearly state this as a limitation of the model (or show that it can do this).

We view our model as perfectly consistent with Fyhn et al, 2007. Remapping is not triggered by the network itself, though, but rather by a re-arrangement of the inputs requiring the network to learn new associations. Different simulations of the same model with identical parameters can be interpreted as remapping experiments.

Reviewer #3 (Public Review):

Summary:

The paper proposes an alternative to the attractor hypothesis, as an explanation for the fact that grid cell population activity patterns (within a module) span a toroidal manifold. The proposal is based on a class of models that were extensively studied in the past, in which grid cells are driven by synaptic inputs from place cells in the hippocampus. The synapses are updated according to a Hebbian plasticity rule. Combined with an adaptation mechanism, this leads to patterning of the inputs from place cells to grid cells such that the spatial activity patterns are organized as an array of localized firing fields with hexagonal order. I refer to these models below as feedforward models.

It has already been shown by Si, Kropff, and Treves in 2012 that recurrent connections between grid cells can lead to alignment of their spatial response patterns. This idea was revisited by Urdapilleta, Si, and Treves in 2017. Thus, it should already be clear that in such models, the population activity pattern spans a manifold with toroidal topology. The main new contributions in the present paper are (i) in considering a form of recurrent connectivity that was not directly addressed before. (ii) in applying topological analysis to simulations of the model. (iii) in interpreting the results as a potential explanation for the observations of Gardner et al.

We wanted to note that we do not see this paper as proposing an alternative to the attractor hypothesis, given that we use attractor networks, but rather as an exploration of possibilities not yet visited by this hypothesis.

Strengths:

The exploration of learning in a feedforward model, when recurrent connectivity in the grid cell layer is structured in a ring topology, is interesting. The insight that this not only align the grid cells in a common direction but also creates a correspondence between their intrinsic coordinate (in terms of the ring-like recurrent connectivity) and their tuning on the torus is interesting as well, and the paper as a whole may influence future theoretical thinking on the mechanisms giving rise to the properties of grid cells.

Weaknesses:

(1) In Si, Kropff and Treves (2012) recurrent connectivity was dependent on the head direction tuning, in addition to the location on a 2d plane, and therefore involved a ring structure. Urdapilleta, Si, and Treves considered connectivity that depends on the distance on a 2d plane. The novelty here is that the initial connectivity is structured uniquely according to latent coordinates residing on a ring.

The recurrent architectures in the cited works are complex and require arranging cells in a 2D manifold to calculate connectivity based on their relative 2D position. In other words, the 2D structure is imprinted in the architecture, as in our 2D condition. In this work the network is much simpler and only requires neighboring relations in 1D. Such relationships have been shown to spontaneously emerge in the hippocampal formation (Pastalkova et al, 2008; Gonzalo Cogno et al, 2024).

(2) The paper refers to the initial connectivity within the grid cell layer as one that produces an attractor. However, it is not shown that this connectivity, on its own, indeed sustains persistent attractor states. Furthermore, it is not clear whether this is even necessary to obtain the results of the model. It seems possible that (possibly weaker) connections with ring topology, that do not produce attractor dynamics but induce correlations between neurons with similar locations on the ring would be sufficient to align the spatial response patterns during the learning of feedforward weights.

Regarding the first part of the comment, the recurrent collaterals create one or at times multiple bumps of activity in the network so that neighboring (interconnected) cells activate together. An initial random state of activity rapidly falls into this dynamic, constrained by the attractor. To us this is not surprising given that this connectivity is the classical means of creating a continuous attractor. Perhaps there is some deeper meaning in this comment that we are not fully grasping.

Regarding the second part of the comment, we fully agree with the reviewer. We are presenting what so far is the simplest connectivity that can align grid maps, but by no means we claim that it is the simplest possible one. Regarding weaker connections with ring topology, we show in Figure S2 that a ring attractor with too weak or too strong connections is incapable of aligning grids, since a balance between feedforward and feedback inputs is required.

(3) Given that all the grid cells are driven by an input from place cells that span a 2d manifold, and that the activity in the grid cell network settles on a steady state which is uniquely determined by the inputs, it is expected that the manifold of activity states in the grid cell layer, corresponding to inputs that locally span a 2d surface, would also locally span a 2d plane. The result is not surprising. My understanding is that this result is derived as a prerequisite for the topological analysis, and it is therefore quite technical.

We understand that the reviewer is referring to the motivation behind studying local dimensionality. We agree that the topological analysis approach is quite technical, but it provides unique insights. The theorem of closed surfaces, which allows us to deduce a toroidal topology from Betti numbers (1,2,1), only applies to closed surfaces. One thus needs to show that the point cloud is a surface (local dimensionality of 2) and is closed (no borders or singularities). If borders or singularities were present, a toroidal topology could not be claimed from these Betti numbers. Thus, it is a crucial step of the analysis.

(4) The modeling is all done in planar 2d environments, where the feedforward learning mechanism promotes the emergence of a hexagonal pattern in the single neuron tuning curve. Under the scenario in which grid cell responses are aligned (i.e. all neurons develop spatial patterns with the same spacing and orientation) it is already quite clear, even without any topological analysis that the emerging topology of the population activity is a torus.

However, the toroidal topology of grid cells in reality has been observed by Gardner et al also in the wagon wheel environment, in sleep, and close to boundaries (whereas here the analysis is restricted to the a sub-region of the environment, far away from the walls). There is substantial evidence based on pairwise correlations that it persists also in various other situations, in which the spatial response pattern is not a hexagonal firing pattern. It is not clear that the mechanism proposed in the present paper would generate toroidal topology of the population activity in more complex environments. In fact, it seems likely that it will not do so, and this is not explored in the manuscript.

We agree that our work was constrained to exploration in 2D and that the situations posed by the reviewer are challenging, but we do not see them as unsurmountable. The wagon wheel shows a preservation of toroidal topology locally, where the behavior of the animal is rather 2-dimensional. Globally, hexagonal maps are lost, which is compatible with some flexibility in the way grid maps are formed. If sleep meant that all inputs are turned off, our model would predict a dynamic dictated by the architecture (1D for the ring attractor, for example), but we do not really know that this is the case. In the future, we intend to explore predictive activity along the linear attractor, which could both result in path integration and in some level of preservation of the activity when inputs are completely turned off.

Regarding boundaries, as we have argued before, the cited work chooses to filter away what looks like more than half of the overall explained variance through PCA, and this is only before applying a non-linear dimensionality reduction algorithm. It is specifically shown that the analyzed components are the ones with global periodicity throughout the environment. Thus, it is conceivable that through this approach, local irregularities found only at the borders are disregarded in favor of a clearer global picture. While using a different methodology, our approach follows a similar spirit, albeit with far less noisy data.

(5) Moreover, the recent work of Gardner et al. demonstrated much more than the preservation of the topology in the different environments and in sleep: the toroidal tuning curves of individual neurons remained the same in different environments. Previous works, that analyzed pairwise correlations under hippocampal inactivation and various other manipulations, also pointed towards the same conclusion. Thus, the same population activity patterns are expressed in many different conditions. In the present model, this preservation across environments is not expected. Moreover, the results of Figure 6 suggest that even across distinct rectangular environments, toroidal tuning curves will not be preserved, because there are multiple possible arrangements of the phases on the torus which emerge in different simulations.

We agree with this observation. A symmetry in our implementation results in the fact that only ~50% of times the system falls in the preferred solution, and the rest of the times it falls into other local minima. Whether this result is at odds with current observations can be debated on the basis of probabilities. However, we believe that the symmetry we found is purely circumstantial, and that it can be broken by elements such as head direction modulation or other ingredients used to achieve path integration. In other words, we acknowledge that symmetry is an issue of the implementation we show here (which has been kept as simple as possible to serve as a proof-of-principle) but we do not think that it is a defining feature of flexible attractors in general. We expect that future implementations that incorporate path integration capabilities will not present this kind of symmetry in the space of solutions.

Regarding the rigid phase translation across modalities, while this effect is very clear in Gardner et al, it is less so in other datasets. The analyses shown in Hermansen et al (2024) can rather be interpreted as somewhere in the way between perfect rigid translation and fully randomized phases across navigation modalities.

(6) In real grid cells, there is a dense and fairly uniform representation of all phases (see the toroidal tuning of grid cells measured by Gardner et al). Thus, the highly clustered phases obtained in the model (Fig. S1) seem incompatible with the experimental reality. I suspect that this may be related to the difficulty in identifying the topology of a torus in persistent homology analysis based on the transpose of the matrix M.

We partly agree with this observation and note that a pattern of ordered phases is an issue not only for the 1D attractor but also for the 2D one, which appears much more uniform than in experimental data. The low number of neurons we used for computational economy and the full connectivity could be key ingredients to generate these phase patterns. To show that this is not a defining feature of flexible attractors, apart from the fact that these patterns appear also with non-flexible 2D architectures, we included in Figure S1 simulations with ‘fragmented 1D’ architectures. In this case the architecture is a superposition of 20 random 1D stripe-like attractors. While the alignment of maps achieved with this architecture is almost at the same level as the one obtained with 1D and 2D attractors, the phases are much more similar to what has been observed experimentally, and less uniform than what is obtained with 2D attractors.

(7) The motivations stated in the introduction came across to me as weak. As now acknolwledged in the manuscript, attractor models can be fully compatible with distortions of the hexagonal spatial response patterns - they become incompatible with this spatial distortions only if one adopts a highly naive and implausible hypothesis that the attractor state is updated only by path integration. While attractor models are compatible with distortions of the spatial response pattern, it is very difficult to explain why the population activity patterns are tightly preserved across multiple conditions without a rigid two-dimentional attractor structure. This strong prediction of attractor models withstood many experimental tests - in fact, I am not aware of any data set where substantial distortions of the toroidal activity manifold were observed, despite many attempts to challenge the model. This is the main motivation for attractor models. The present model does not explain these features, yet it also does not directly offer an explanation for distortions in the spatial response pattern.

Some interesting examples are experiments in 3D, where grid cells presumably communicate with each other through the same recurrent collaterals, but global periodicity is lost and only some local order is preserved even away from boundaries (Ginosar et al, 2021; Grieves et al, 2021). While these datasets have not been explored using topological analysis, they serve as strong motivators to understanding 2D grid cells as one equilibrium solution that arises under some set of constraints, but belongs to a wider space of possible solutions that may arise as well under more flexible constraints. Even (and especially) if one adheres to the hypothesis that grid cells are pre-wired into a 2D torus, a concept like flexible attractors might become useful to understand how their activity is rendered in 3D. Another strong motivation is our lack of understanding of how a perfectly balanced 2D structure is formed and maintained. Simpler architectures could be thought of as alternatives, but also as an intermediate step towards it.

Regarding the rigid phase translation across modalities, while this effect is very clear in Gardner et al, it is less so in other datasets. The analyses shown in Hermansen et al (2024) can rather be interpreted as somewhere in the way between perfect rigid translation and fully randomized phases.

In a separate point, although it might not be strictly related to the comment, we do not fully share the idea that persistent activity patterns during sleep are necessary or sufficient conditions for attractor dynamics, although we do agree that attractors could be the mechanism behind them and any alternative is at least as complex as attractors. On the necessity side, attractors in the hippocampus are not constantly engaged (Wills et al, 2005). For sufficiency, one should prove that no other network is capable of reproducing the phenomenon, and to our best knowledge we are still far from that point.

(8) There is also some weakness in the mathematical description of the dynamics. Mathematical equations are formulated in discrete time steps, without a clear interpretation in terms of biophysically relevant time scales. It appears that there are no terms in the dynamics associated with an intrinsic time scale of the neurons or the synapses (a leak time constant and/or synaptic time constants). I generally favor simple models without lots of complexity, yet within this style of modelling, the formulation adopted in this manuscript is unconventional, introducing a difficulty in interpreting synaptic weights as being weak or strong, and a difficulty in interpreting the model in the context of other studies.

We chose to keep the model as simple as possible and in the line of previous publications developing it. However, we see the usefulness of putting it in what in the meantime has become a canonical framework. Fortunately this has been done by D’Albis and Kempter (2017). In our simplified version of the model there is no leak term and adaptation on its own brings down activity in the absence of input, but we agree that such a term could be added, albeit not without modifying all other network parameters.

In my view, the weaknesses discussed above limit the ability of the model, as it stands, to offer a compelling explanation for the toroidal topology of grid cell population activity patterns, and especially the rigidity of the manifold across environments and behavioral states. Still, the work offers an interesting way of thinking on how the toroidal topology might emerge.

Reviewer 1:

Reviewer #1 (Recommendations For The Authors):

See comments above. In addition:

(1) Abstract: '...interconnected by a two-dimensional attractor guided by path integration'. This is unclear. I think the intended meaning might be along the lines of '...their being computed by a 2D continous attractor that performs path integration'?

'path integration allowing for no deviations from the hexagonal pattern' This is incorrect. Local modulation of the gain of the speed input to a standard CAN would distort the grid pattern.

'Using topological data analysis, we show that the resulting population activity is a sample of a torus' Activity in the model?

'More generally, our results represent a proof of principle against the intuition that the architecture and the representation manifold of an attractor are topological objects of the same dimensionality, with implications to the study of attractor networks across the brain' I guess one might hold this intuition, but it strikes me as obvious that if you impose an sufficiently strong n-dimensional input on a network then it it's activity could have the same dimensionality. I don't really see this as being a point worth highlighting. Perhaps the more interesting point, it that during learning the recurrent connectivity aligns the grid fields of neurons in the network, and this may be a specific function of the 1D attractor dynamcis, although I don't think the authors have made this point convincing.

'The flexibility of this low dimensional attractor allows it to negotiate the geometry of the representation manifold with the feedforward inputs'. See above for comments on the use of 'negotiate'.

'while the ensemble of maps preserves features of the network architecture'. I don't understand this. What is the 'ensemble of maps' and what are the features referred to.

We have reviewed the abstract considering these points. Regarding the ‘strong n-dimensional input’, we want to point out that it is not the input itself that generates a torus (the no attractor condition does not lead to a torus) but rather the interplay between the input and the attractor.

‘Perhaps the more interesting point …’, we do not fully understand how this sentence deviates from our own conclusions. We here show that a strong n-dimensional input is not enough to align grid cells (produce a n-torus), it is the interplay between inputs and attractor dynamics that does so, even if the attractor is not n-dimensional in terms of architecture.

The ensemble of maps refers to the transpose of the population activity matrix, where each point in the cloud is a map, and the features refer to the persistent homology.

(2) The manuscript still fails to clarify the difference between a model that path integrates in two dimensions and a model that simply represents information with a given dimensionality. The argument that it's surprising that a network with 1D architecture represents a higher dimensional input strikes me as incorrect and an unnecessary attempt to argue for conceptual importance. At least to me this isn't surprising. It would be surprising if the 1D network could path integrate but this doesn't seem to be the case.

In response to the reviewer’s concerns, we have made clear in the introduction and discussion that this model has no path integration capabilities, although we aim to develop a model capable of path integration using the kind of simple architecture presented here. We want to highlight here that equating attractor dynamics with path integration would be a conceptual mistake.

(3) Other wording also seems to make unnecessary conceptual claims. E.g. The repeated use of 'negotiate' implies some degree of intelligence, or at least an exchange of information, that isn't shown to exist. I wonder if more precise language could be used? As I understand it the dimensionality is bounded by the inputs on the one hand, and the network connectivity on the other, with the actual dimensionality being a function of the recurrent and feedforward synaptic weights. There's clearly some role for the relative weights and the properties of plasticity rules, but I don't see any evidence for a negotiation.

An interesting observation in Figure S2 is that grid maps are aligned only if the relative strength of feedforward and recurrent inputs is similar. If one of them can impose over the other, grid maps do not align. This equilibrium can metaphorically be thought of as a negotiation instance, where the negotiation is an emergent property of the system rather than something happening at an individual synapse.

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

Reviewer #1:

Reviewer #1 (Recommendations For The Authors):

Major

(1) What is the evidence that, after training, the 1D network maintains its attractor dynamics when feedforward inputs are active? If the claim is that it does then it's important to provide evidence, e.g. responses to perturbations, or other tests. The alternative is that after training the recurrent inputs are drowned out by the feed forward spatial inputs.

We agree with the reviewer on the importance of this point. In our model, networks are always learning, and the population activity represented by aligned grid maps in a trained network is a dynamic equilibrium that emerges from the interplay between feedforward and collateral constraints. If Hebbian learning is turned off, one gets a snapshot of the network at that moment. We now show in Fig. S3 that in a trained network without feedforward Hebbian learning the removal of recurrent collaterals results in a slight increase in gridness and spacing. The expansion is due to the fact that, as we argue in the Results section, the attractor has a contractive effect on grid maps, which could relate to observations in novel environments (Barry et al, 2007). If Hebbian learning is turned on in the same situation, the maps, no longer constrained by the attractor, drift toward the equilibrium solution of the ‘No attractor’ condition, with significantly larger spacing, no alignment and lower individual gridness. Thus, the attractor is the force preventing them to do so when feedforward Hebbian learning is on.

These observations point to the key role played by the attractor not only in forming but also in sustaining grid activity. The dynamic equilibrium framework fits well known properties of the system, such as its capacity to recalibrate very fast (Jayakumar et al, 2019), although this particular feature cannot be modeled with the current version of our model, that lacks path integration capabilities.

(2) It would be useful to include additional control conditions for Figure 2 to test the hypothesis that it is simply connectivity, rather than attractor dynamics, that drives alignment.

This could be achieved by randomly assigning strengths to the recurrent connections, e.g. drawing from exponential or Gaussian distributions.

We agree and have included Fig. S2b-d, showing that the same distribution of collateral input weights entering each neuron, but lacking the 1D structure provided by the attractor, does not align grid maps. This is achieved by shuffling rows in the connectivity matrix, while avoiding self connections to make the comparison fair (self connections substantially alter the dynamic of the network, making it much more rigid). We observed that individual grid maps have very low gridness levels, even lower than in the no-attractor condition. In contrast, they have levels of population gridness slightly higher than in the no-attractor condition, but closer to 0 than to levels achieved with attractors. Our interpretation of these results is that irregular connectivity achieves some alignment in a few arbitrary directions and/or locations, which improves the coordination between maps at the expense of impairing rather than improving hexagonal responses of individual cells. Such observations stand in clear context to what is observed with continuous attractors with an orderly architecture.

These results suggest that it is the structure of the attractor that allows grid cells to be aligned rather than the mere presence of recurrent collateral connections.

(3) It seems conceivable that once trained the recurrent connections would no longer be required for alignment. Can this be evaluated by considering what happens if the recurrent connections are turned off after training (or slowly turned off during training)? Does the network continue to generate aligned grid fields?

This point has elements in common with point 1. As we argued in that response, the attractor has two main effects on grid maps: it aligns them and it contracts them. If the attractor is turned off, feedforward Hebbian learning progressively drives maps toward the solution obtained for the ‘no attractor’ condition, characterized by maps with larger spacing, poorer gridness and lack of alignment.

(4) After training what is the relative strength of the recurrent and feedforward inputs to each neuron?

Both recurrent and feedforward synaptic-strength matrices are normalized throughout training, so that the overall incoming synaptic strength to each neuron is invariant. Because of this, although individual feed-forward and recurrent input fields vary dynamically, their average is constant, with the exception of the very first instances of the simulation, before a stable regime is reached in grid-cell activity levels. We have included Fig. S2d, showing the dynamics of feedforward and recurrent mean fields throughout learning as well as their ratio. In addition, Fig. S2a shows that the strength of recurrent relative to feedforward inputs is an important parameter, since alignment is only obtained in an intermediate range of ratios.

(5) It would be helpful to also evaluate the low dimensional structure of the input to the network. Assuming it has a 2D structure, as it represents 2D space, can an explanation be provided for why it is surprising that the trained network also encodes activity with a 2D manifold? It strikes me that the more interesting finding might relate to alignment of the grids rather than claims about a 1D attractor encoding a 2D representation. Either way, stronger evidence and clearer discussion would be helpful.

The reviewer is correct in assuming that the input has a 2D structure, that can be represented by a sheet embedded in a high dimensional space and thus has the Betti numbers [1,0,0]. The surprising element in our results is that we are showing for the first time that the population activity of an attractor network is constrained to a manifold that results from the negotiation between the architecture of the attractor and the inputs, and does not merely reflect the former as previously assumed. In this sense, the alignment of grid cells by a 1D attractor is an instance of the more general case that 1D attractors can encode 2D representations.

It is certainly the case that the 2D input is a strong constraint pushing population activity toward a 2D manifold. However, the final form of the 2D manifold is strongly constrained by the attractor, as shown by the contrast with the no-attractor condition (a 2D sheet, as in the input, vs a torus when the attractor is present). The 1D attractor is able to flexibly adapt to the constraint posed by the inputs while doing its job (as demonstrated in previous points), which results in 2D grid maps aligned by a 1D attractor. Generally speaking, this work provides a proof of principle demonstrating that the topology of the attractor architecture and the manifold of the population activity space need not be identical, as previously widely assumed by the attractor community, and need not even have the same dimensionality. Instead, a single architecture can potentially be applied to many purposes. Hence, our work provides a valuable new perspective that applies to the study of attractors throughout the brain.

(6) The introduction should be clearer about the different types of grid model and the computations they implement. E.g. The authors' previous model generates grid fields from spatial inputs, but if my understanding is correct it isn't able to path integrate. By contrast, while the many 2D models with continuous attractor dynamics also generate grid representations, they do so by path integration mechanisms that are computationally distinct from the spatial transformation implemented by feedforward models (see also general comments above).

We agree with the reviewer and have made this point explicit in the introduction.

(7) A prediction from continuous attractor models is that when place cells remap the low dimensional manifold of the grid activity is unaffected, except that the location of the activity bump is moved. It strikes me as important to test whether this is the case for the model presented here (my intuition is that it won't be, but it would be important to establish either way).

We want to emphasize that our model is a continuous attractor model, so the question regarding the difference between what our model and continuous attractor network models predict is an ill-posed one. One of our main conclusions is precisely that attractors can work in a wider spectrum of ways than previously thought.

In lack of a better definition, our multiple simulations could be thought of as training in different arenas. It is true that in our model maps take time to form, but this is also the case in novel environments (Barry et al, 2007 ), and continuous attractor models exclusively or strongly guided by self motion cues struggle to replicate this phenomenon. We show that the current version of our model accepts multiple solutions (in practice four but conceptually infinite countable), all of them resulting in a torus for the population activity (i.e. the same topology or low dimensional manifold). It is not clear to us how easy it would be to differentiate between most of these solutions in experimental data, with only incomplete information. This said, incorporating a symmetry-breaking ingredient to the model, for example related to head direction modulation, could perhaps lead to the prevalence of a single type of solution. We intend to explore this possibility in the future in order to add path-integration capabilities to the system, as described in the discussion.

(8) The Discussion implies that 1D networks could perform path integration in a manner similar to 2D networks. This is a strong claim but isn't supported by evidence in the study. I suggest either providing evidence that this is the case for models of this kind or replacing it with a more careful discussion of the issue.

The current version of our model has no path integration capabilities, as is now made explicit in the Introduction and Discussion. In addition, we have now made clear that the idea that path integration could perhaps be implemented using 1D networks is, although reasonable, purely speculative.

Minor

(1) Introduction. 'direct excitatory communication between them'. Suggest rewording to 'local synaptic interactions', as communication can also be purely inhibitory (e.g. Burak and Fiete, 2009) or indirect by excitation of local interneurons (e.g. Pastoll et al., Neuron, 2013).

We agree and have adopted this phrasing.

(2) The decision to focus the topology analysis on the 60 cm wide central square appears somewhat arbitrary. Are the irregularities referred to a property of the trained networks or would they also emerge with analysis of simulated ideal data? Can more justification be expanded and supplementary analyses be shown when the whole arena is used?

In practical terms, a subsampling of the data to around half was needed because the persistent homology packages struggle to handle large amounts of data, especially in the calculation of H2. We decided to cut a portion of contiguous pixels in the open field at least larger than the hexagonal tile representing the whole grid population period (as represented in Figure 6). Leaving the borders aside was a logical choice since it is known that the solution at the borders is particularly influenced by the speed anisotropy of the virtual rat (see Si, Kropff & Treves, 2012), in a way that mimics how borders locally influence grid maps in actual rats (Krupic et al, 2015). The specific way in which our virtual rat handles borders is arbitrary and might not generalize. A second issue around borders is that maps are differently affected by incomplete smoothing, although this issue does not apply to our data because we did not smooth across neighboring pixels. In sum, considering the central 60 cm wide square was sufficient to contain the whole torus and a reasonable compromise that would allow us to perform all analyses in the part of the environment less influenced by boundaries.

(3) It could help the general reader to briefly explain what a persistence diagram is.

This is developed in the Appendix, but we have now added a reference to it and a brief description in the main text.

(4) For the analyses in Figure 3-4, and separately for Figure 5, it might help the reader to provide visualizations of the low dimensional point cloud.

All these calculations take place in the original high-dimensional point cloud. Doing them in a reduced space would be incorrect because there is no dimensionality reduction technique that guarantees the preservation of topology. In Figure 7 we reduce the dimensionality of data but emphasize that it is only done for visualization purposes, not to characterize topology. We also point out in this Figure that the same non-linear dimensionality reduction technique applied to objects with identical topology yields a wide variety of visualizations, some of them clear and some less clear. This observation further exemplifies why one cannot assume that a dimensionality-reduction technique preserves topology, even for a low-dimensional object embedded in a high-dimensional space.

(5) The detailed comparison of the dynamics of each model is limited by the number of data points. Why not address this by new simulations with more neurons?

We are not sure we understand this comment. In Figure 2, the dynamics for each model are markedly different. These are averages over 100 simulations. We are not sure what benefit would be obtained from adding more neurons. Before starting this work we searched for the minimal number of neurons that would result in convergence to an aligned solution in 2D networks, which we found to be around 100. Optimizing this parameter in advance was important to reduce computational costs throughout our work.

(6) Could the variability in Figure 7 also be addressed by increasing the number of data points?

As we argued in a previous point, there is no reason to expect preservation of topology after applying Isomap. We believe this lack of topology preservation to be the main driver of variability.

(7) Page/line numbers would be useful.

We agree. However, the text is curated by biorxiv which, to our best knowledge, does not include them.

Reviewer 2:

Reviewer #2 (Recommendations For The Authors):

(1) I highly suggest that the author rewrite some parts of the Results. There are lots of details which should be put into the Methods part, for example, the implementation details of the network, the analysis details of the toroidal topology, etc. It will be better to focus on the results part first in each section, and then introduce some of the key details of achieving these results, to improve the readability of the work.

This suggestion contrasts with that of Reviewer #1. As a compromise, we decided to include in the Results section only methodological details that are key to understanding the conclusions, and describe everything else in the Methods section.

(2) 'Progressive increase in gridness and decrease in spacing across days have been observed in animals familiarizing with a novel environment...' From Fig.2c I didn't see much decrease. The authors may need to carry out some statistical test to prove this. Moreover, even the changes are significant, this might be not the consequence of the excitatory collateral constraint. To prove this, the authors may need to offer some direct evidence.

We agree that the decrease is not evident in this figure due to the scale, so we are adding the correlation in the figure caption as proof. In addition, several arguments, some related to new analyses, demonstrate that the attractor contracts grid maps. First, the ‘no attractor’ condition has a markedly larger spacing compared to all other conditions (Fig. 2a). We also now show that spacing monotonically decreases with the strength of recurrent relative to feedforward weights, in a way that is rather independent of gridness (Fig. S2a). Second, as we now show in Fig. S2b-d, simulations with a shuffled 1D attractor, such that the sum of input synapses to each neuron are the same as in the 1D condition but no structure is present, lead to a spacing that is mid-way between the ‘no attractor’ condition and the conditions with attractors. Third, as we now show in Fig. S3a, turning off both recurrent connections and feedforward learning in a trained network results in a small increase in spacing. Fourth, as we now show in Fig. S3b, turning off recurrent connections while feedforward learning is kept on increases grid spacing to levels comparable to those of the ‘no attractor’ condition. All these elements support a role of the attractor in contracting grid spacing.

(3) Some of the items need to be introduced first before going into details in the paper, for instance, the stipe-like attractor network, the Betti number, etc.

We have added in the Results section a brief description and references to full developments in the Appendix.

Reviewer 3 (Public Review):

(1) It is not clear to me that the proposal here is fundamentally new. In Si, Kropff and Treves (2012) recurrent connectivity was dependent on the head direction tuning and thus had a ring structure. Urdapilleta, Si, and Treves considered connectivity that depends on the distance on a 2d plane.

In the work of Si et al connectivity is constructed ad-hoc for conjunctive cells to represent a torus, it depends on head-directionality but also on the distance in a 2D plane. The topology of this architecture has not been assessed, but it is close to the typical 2D ‘rigid’ constraint. In the work of Urdapilleta et al, the network is a simple 2D one. The difference with our work is that we focus on the topology of the recurrent network and do not use head-direction modulation. In this context, we prove that a 1D network is enough to align grid cells and, more generally, we provide a proof of principle that the topology of the architecture and the representation space of an attractor network do not need to be identical, as previously assumed by the attractor community. These two important points were neither argued, speculated nor self-evident from the cited works.

(2) The paper refers to the connectivity within the grid cell layer as an attractor. However, would this connectivity, on its own, indeed sustain persistent attractor states? This is not examined in the paper. Furthermore, is this even necessary to obtain the results in the model? Perhaps weak connections that do not produce an attractor would be sufficient to align the spatial response patterns during the learning of feedforward weights, and reproduce the results? In general, there is no exploration of how the strength of collateral interactions affects the outcome.

The reviewer makes several important points. Local excitation combined with global inhibition is the archetypical architecture for continuous attractors (see for example Knierim and Zhang, Annual review of neuroscience, 2012). Thus, in the absence of feedforward input, we observe a bump of activity. As in all continuous attractors, this bump is not necessarily ‘persistent’ and instead is free to move along the attractor.

We cannot prove that there is not a simpler architecture that has the same effect as our 1D or 1DL conditions, and we think that there are some interesting candidates to investigate in the future. What we now prove in new Fig. S2b-d is that it is not the strength of recurrent connections themselves, but instead the continuous attractor structure that aligns grid cells in our model. To demonstrate this, we shuffle incoming recurrent connections to each neuron in the 1D condition (while avoiding self-connections for fairness), and show that training does not lead to grid alignment. We also show in Fig. S1 that an architecture represented by 20 overlapping 1DL attractors, each formed by concatenating 10 random cells, aligns grid cells to levels slightly lower but similar to the 1D or 1DL attractors. This architecture can perhaps be considered as simpler to build in biological terms than all the others, but it is still constituted by continuous attractors.

The strength of recurrent collaterals, or more precisely the recurrent to feedforward ratio, is crucial in our model to achieve a negotiated outcome from constraints imposed by the attractor and the inputs. We now show explicit measures of this ratio in Fig. S2, as well as examples showing that an imbalance in this ratio impairs grid alignment. When the ratio is too high or too low, both individual and population gridness are low. Interestingly, grid spacing behaves differently, decreasing monotonically with the relative strength of recurrent connections.

(3) I did not understand what is learned from the local topology analysis. Given that all the grid cells are driven by an input from place cells that spans a 2d manifold, and that the activity in the grid cell network settles on a steady state that depends only on the inputs, isn't it quite obvious that the manifold of activity in the grid cell layer would have, locally, a 2d structure?

The dimensionality of the input is important, although not the only determinant of the topology of the activity. The recurrent collaterals are the other determinant, and their architecture is a crucial feature. For example, as we now show in Figure S2b-d, shuffled recurrent synaptic weights fail to align grid cells. In the 1D condition, if feedforward inputs were absent, the dynamics of the activity would be confined to a ring. The opposite condition is our ‘no attractor’ condition, in which activity in the grid cell layer mimics the topology of inputs, a 2D sheet (and not a torus). It is in the intermediate range, when both feedforward and recurrent inputs are important, that a negotiated solution (a torus) is achieved.

The analyses of local dimensionality and local homology of Figure 3 are crucial steps to demonstrate toroidal topology. According to the theorem of classification of closed surfaces, global homology is not enough to univocally define the topology of a point cloud, and thus this step cannot be skipped. The step is aimed to prove that the point cloud is indeed a closed surface.

(4) The modeling is all done in planar 2d environments, where the feedforward learning mechanism promotes the emergence of a hexagonal pattern in the single neuron tuning curve. This, combined with the fact that all neurons develop spatial patterns with the same spacing and orientation, implies even without any topological analysis that the emerging topology of the population activity is a torus.

We cannot agree with this intuition. In the ‘no attractor’ condition, individual maps have hexagonal symmetry with standardized spacing, but given the lack of alignment the population activity is not a closed surface and thus not a torus. It can rather be described as a 2D sheet embedded in a high dimensional space, a description that also applies to the input space.

While it is rather evident that an ad hoc toroidal architecture folds this 2D population activity into a torus, it is less evident and rather surprising that 1D architectures have the same capability. This is the main novelty in our work.

(5) Moreover, the recent work of Gardner et al. demonstrated much more than the preservation of the topology in the different environments and in sleep: the toroidal tuning curves of individual neurons remained the same in different environments. Previous works, that analyzed pairwise correlations under hippocampal inactivation and various other manipulations, also pointed towards the same conclusion. Thus, the same population activity patterns are expressed in many different conditions. In the present model, the results of Figure 6 suggest that even across distinct rectangular environments, toroidal tuning curves will not be preserved, because there are multiple possible arrangements of the phases on the torus which emerge in different simulations.

We agree with the reviewer in the main point, although the recently found ring activity in the absence of sensory feedback (Gonzalo Cogno et al, 2023) suggests that what is happening in the EC is more nuanced than a pre-wired torus. Solutions in Figure 6 are different ways of folding a 1D strip into a torus, with or without the condition of periodicity in the 1D strip. Whether or not these different solutions would be discernible from one another in a practical setup is not clear to us. For example, global homology, as addressed in the Gardner paper, is the same for all these solutions. Furthermore, while our solutions of up to order 3 are highly discernable, higher order solutions, potentially achievable with other network parameters, would be impossible to discern by eye in representations similar to the ones in Figure 6. In addition, while we chose to keep our model in the simplest possible form as a clear proof of principle, new elements introduced to the model such as head directionality could break the symmetry and lead to the prevalence of one preferred solution for all simulation replicates. We plan to investigate this possibility in the future when attempting to incorporate path-integration capabilities to the model.

(6) In real grid cells, there is a dense and fairly uniform representation of all phases (see the toroidal tuning of grid cells measured by Gardner et al). Here the distribution of phases is not shown, but Figure 7 suggests that phases are non uniformly represented, with significant clustering around a few discrete phases. This, I believe, is also the origin for the difficulty in identifying the toroidal topology based on the transpose of the matrix M: vectors representing the spatial response patterns of individual neurons are localized near the clusters, and there are only a few of them that represent other phases. Therefore, there is no dense coverage of the toroidal manifold that would exist if all phases were represented equally. This is not just a technical issue, however: there appears to be a mismatch between the results of the model and the experimental reality, in terms of the phase coverage.

As mentioned in the results section, Figure 7 is meant for visualization purposes only, and serves more as cautionary tale regarding the imprevisible risks of non-linear dimensionality reduction than as a proof of the organization of activity in the network. Isomap is a non-linear transformation that deforms each of our solutions in a unique way so that, while all have the topology of a torus embedded in a high dimensional space, only a few of them exhibited one of two possible toroidal visualizations in a 3D Isomap reduction. Isomap, as well as all other popular dimensionality reduction techniques, provide no guarantee of topology invariance. A better argument to judge the homogenous distribution of phases is persistent homology, which identifies relatively large holes (compared to the sampling spacing) in the original manifold embedded in a high dimensional space. In our case, persistent homology identified only two holes significantly larger than noise (the two cycles of a torus) and one cavity in all conditions that included attractors. Regarding the specific distribution of phases in different conditions, however, see our reply below.

(7) The manuscript makes several strong claims that incorrectly represent the relation between experimental data and attractor models, on one hand, and the present model on the other hand. For the latter, see the comments above. For the former, I provide a detailed list in the recommendations to the authors, but in short: the paper claims that attractor models induce rigidness in the neural activity which is incompatible with distortions seen in the spatial response patterns of grid cells. However, this claim seems to confuse distortions in the spatial response pattern, which are fully compatible with the attractor model, with distortions in the population activity patterns, which would be incompatible with the attractor model. The attractor model has withstood numerous tests showing that the population activity manifold is rigidly preserved across conditions - a strong prediction (which is not made, as far as I can see, by feedforward models). I am not aware of any data set where distortions of the population activity manifold have been identified, and the preservation has been demonstrated in many examples where the spatial response pattern is disrupted. This is the main point of two papers cited in the present manuscript: by Yoon et al, and Gardner et al.

First of all, we would like to note that our model is a continuous attractor model. Different attractor models have different outcomes, and one of the main conclusions of our manuscript is that attractors can do a wider range of operations than previously thought.

We agree with the reviewer that distortions in spatial activity (which speak against a purely path-integration guided attractor) should not be confused with distortions in the topology of the population activity (which would instead speak against the attractor dynamics itself). We have rephrased these observations in the manuscript. In fact, we believe that the capacity of grid cells to present distorted maps without a distortion of the population activity topology, as shown for example by Gardner and colleagues, could result from a tension between feedforward and recurrent inputs, the potential equilibriums of which our manuscript aims to characterize.

(8) There is also some weakness in the mathematical description of the dynamics. Mathematical equations are formulated in discrete time steps, without a clear interpretation in terms of biophysically relevant time scales. It appears that there are no terms in the dynamics associated with an intrinsic time scale of the neurons or the synapses, and this introduces a difficulty in interpreting synaptic weights as being weak or strong. As mentioned above, the nature of the recurrent dynamics within the grid cell network (whether it exhibits continuous attractor behavior) is not sufficiently clear.

We agree with the reviewer that our model is rather simple, and we value the extent to which this simplicity allows for a deep characterization. All models are simplifications and the best model in any given setup is the one with the minimum amount of complexity necessary to describe the phenomenon under study. We believe that to understand whether or not a 1D continuous attractor architecture can result in a toroidal population activity, a biophysically detailed model, with prohibitive computational costs, would have been unnecessarily complex. This argument does not intend to demerit biophysically detailed models, which are capable of addressing a wider range of questions regarding, for example, the spiking dynamics of grid cells, which cannot be addressed by our simple model.

Reviewer #3 (Recommendations For The Authors):

The work points to an interesting scenario for the emergence of toroidal topology, but the interpretation of this idea should be more nuanced. I recommend reconsidering the claims about limitations of the attractor theory, and acknowledging the limitations of the present theory.

I don't see the limitations mentioned above as a reason to reject the ideas proposed in this manuscript, for two main reasons: first, additional research might reveal a regime of parameters where some issues can be resolved (e.g. the clustering of phases). In addition, the mechanism described here might act at an early stage in development to set up initial dynamics along a toroidal manifold, while other mechanisms might be responsible for the rigidity of the toroidal manifold in an adult animal. But all this implies that the novelty in the present manuscript is weaker than implied, the ability to explain experimental observations is more limited than implied, and these limitations should be acknowledged and discussed.

I recommend reporting on the distribution of grid cell phases and, if indeed clustered, this should be discussed. It will be helpful to explore whether this is the reason for the difficulty in identifying the toroidal topology based on the collection of spatial response patterns (using the transpose of the matrix M).

Ideally, a more complete work would also explore in a more systematic and parametric way the influence of the recurrent connectivity's strength on the learning, and whether a toroidal manifold emerges also in non-planar, such as the wagon-wheel environment studied in Gardner et al.

Part of these recommendations have been addressed in the previous points (public review). Regarding the reason why the transpose of M does not fully recapitulate architecture with our conservative classification criteria, we believe that there is no reason why it should in the first place. We view the fact that the transpose of M recapitulates some features of the architecture as a purely phenomenological observation, and we think it is important as a proof that M is not exactly the same for the different conditions. We imagined that if M matrices were exactly the same this could be due to poor spatial sampling by our bins. Knowing that they are intrinsically different is important even if the reason why they have these specific features is not fully clear to us.

Although we do not think that the distribution of phases is related to the absence of a cavity in the transpose of M or to the four clusters found in Isomap projections, it remains an interesting question that we did not explore initially. We are now showing examples of the distribution of phases in Figure S1. We observed that in both 2D and 1D conditions phases are distributed following rather regular patterns. Whether or not these patterns are compatible with experimental observations of phase distribution is to our view debatable, given that so far state-of-the-art techniques have only allowed to simultaneously record a small fraction of the neurons belonging to a given module. This said, we think that it is important to note that ordered phase patterns are an anecdotal outcome of our simulations rather than a necessary outcome of flexible attractors or attractors in general. To prove this point, we simulated a condition with a new architecture represented by the overlap of 20 short 1DL attractors, each recruiting 10 random neurons from the pool of 100 available ones.

The rest of the parameters of the simulations were identical to those in the other conditions.

By definition, the topology of this architecture has Betti numbers [20,0,0]. We show in Figure S1 that this architecture aligns grid cells, with individual and population gridness reaching slightly lower levels compared to the 1D condition. However, the distribution of phases of these grid cells has no discernible pattern. This result is an arbitrary example that serves as a proof-of-principle to show that flexible attractors can align grid cells without exhibiting ordered phases, not a full characterization of the outcome of this type of architecture, which we leave for future work. For the rest of our work, we stick to the simplest versions of 1D architectures, which allow for a more in-depth characterization.

The wagon-wheel is an interesting case in which maps loose hexagonal symmetry although the population activity lies in a torus, perhaps evidencing the tension between feedforward and recurrent inputs and suggesting that grid cell response does not obey the single master of path integration. If we modeled it with a 1D attractor, we believe the outcome would strongly depend on virtual rat trajectory. If the trajectory was strictly linear, the population activity would be locally one-dimensional and potentially represented by a ring. Instead, if the trajectory allowed for turns, i.e. a 2D trajectory within a corridor-like maze, the population activity would be toroidal as in our open field simulations, while maps would not have perfect hexagonal symmetry, mimicking experimental results.

More minor comments:

Recurrent dynamics are modeled as if there is no intrinsic synaptic or membrane time constant. This may be acceptable for addressing the goals of this paper, but it is a bit unusual and it will be helpful to explain and justify this choice.

As mentioned above, we believe that the best model in a given setup is the one with the lowest number of complexities that can still address the phenomenon under study. One does not use general relativity to build a bridge, although it provides a ‘more accurate’ description of the physics involved. All models are simplifications, and the more complex a model, the more it has to be taken as a black box.

The Introduction mentions that in most models interaction between co-modular neurons occurs through direct excitatory communication, but in quite a few models the interaction is inhibitory. The crucial feature is that the interaction is strongly inhibitory between neurons that differ in their tuning, and either less inhibitory or excitatory between neurons with similar phases.

We agree that directed inhibition has been shown to be as efficient as directed excitation, and we have modified the introduction to reflect this.

The Discussion claims that the present work is the first one in which the topology of the recurrent architecture differs from the topology of the emergent state space. However, early works on attractor models of grid cells showed how neural connectivity which is arranged on a 2d plane, without any periodic boundary conditions, leads to a state space that exhibits the toroidal topology. Therefore, this claim should be revised.

We agree, although the 2D sheet in this case acts as a piece of the torus, and locally the input space and architecture are identical objects. It could be argued that architectures that represent a 2D local slice of the torus, the whole torus, or several cycles around the torus form a continuous family parametrized by the extension of recurrent connections, and as a consequence it is not surprising that these works have not made claims about the incongruence between architecture and representation topologies. The 2D sheet connectivity is still constructed ad hoc to organize activity in a 2D bump, and there is no negotiation between disparate constraints because locally the constraints imposed by input and architecture are the same. We believe this situation is conceptually different from our flexible 1D attractors. We have adapted our claim to include this technical nuance.

Why are neural responses in the perimeter of the environment excluded from the topological analysis? The whole point of the toroidal manifold analysis on real experimental data is that the toroidal manifold is preserved regardless of the animal's location and behavioral condition.

We agree, although experimental data needs to go through extensive pre-processing such as dimensionality reduction before showing a toroidal topology. Such manipulations might smooth away the specific effects of boundaries on maps, together with other sources of noise. In our case, the original reason to downsample the dataset is related to the explosion in computational time that we experience with the ripser package when using more than ~1000 data points. For a proof-of-principle characterization we were much more interested in what happened in the center of the arena, where a 1D attractor could fold itself to confine population activity into a torus. The area we chose was sufficiently large to contain the whole torus. Borders do affect the way the attractor folds (they also affect grid maps in real rats). We feel that these imperfections could be interesting to study in relation to the parameters controlling how our virtual rat behaves at the borders, but not at this proof-of-principle stage.

The periodic activity observed in Ref. 29 could in principle provide the basis for the ring arrangement of neurons. However, it is not yet clear whether grid cells participate in this periodic activity.

We agree. So far it seems that entorhinal cells in general participate in the ring, which would imply that all kinds of cells are involved. However, it could well be that only some functional types participate in the ring and grid cells specifically do not, as future experiments will tell.

dsdsd

законодательство

что бы ни было

мягкий; снисходительный

заявлять вновь

молить

вне зависимости от

подвергание

подойти

погружать

глобальные перемены;

вписываться в

расплываться;

зазорный

исторически сложившийся

избитое выражение; банальность

обычай

глубоко укоренившийся

автор статьи

This is \|, by the way.

Test aussi :)

Test

Como se hablo en clase, creo que la creatividad es el inicio de una escritura que emociona al lector, sin creatividad no podriamos cautivar la atencion ni el interes de unlector, si creamos un texto a base de imaginacion, podemos hacer verdaderas obras de arte, cuentos magicos, historias envolventes, podemos plasmar nuestras emociones y cada persina que lee, se personifica en cada palabra escrita y en la forma tan creativa que se daria

La escritura creativa nos invita a explorar mundos imaginarios, conectar con nuestras emociones y reflexionar sobre el mundo que nos rodea es una herramienta poderosa para el desarrollo personal, la conexión con los demás y el cambio social me encantó mucho está tesis ya que nos comparte que mejora la comunicación, la expresión escrita, la creatividad y la inteligencia emocional fortalece la autoestima, aumenta confianza en uno mismo. Fomenta el pensamiento crítico, analizar y reflexionar sobre diferentes temas.

En resumen a todo este texto es que la escritura creativa es una herramienta valiosa para el desarrollo del alumno, tanto a nivel personal como académico. La escuela tiene la responsabilidad de fomentar la escritura creativa en el aula, proporcionando a los alumnos las herramientas y el apoyo necesarios para expresarse libremente a través de la escritura. Y así el niño pueda seguir progresando en todo el transcurso de su vida estudiantil.

Este texto me ha parecido muy interesaste pero lo que mas me gusto y me llamo la atención fue sobre la escritura creativa en el aula de Educación Primaria. Este trabajo no solo resalta la importancia de la creatividad en la enseñanza de la escritura, sino que también proporciona metodologías y propuestas didácticas concretas para implementar en el aula. Me ha gustado mucho la forma en que el trabajo combina la teoría con la práctica, proponiendo el uso de nuevas tecnologías, como blogs, para motivar a los estudiantes y desarrollar sus competencias lingüísticas. Además, la revisión del currículo y los métodos históricos de enseñanza de la escritura nos ha permitido conocer y comprender de una mejor manera la evolución de las prácticas educativas en este campo. En definitiva, considero que este trabajo es una valiosa herramienta para cualquier docente que busque fomentar la creatividad y mejorar las habilidades de escritura de sus alumnos ya que la escritura es vital para la expresión y comunicación de docentes a estudiantes y de estudiantes así los docentes.

Es tan cierto como escritores como Vargas Llosa (2001 nos hacen caer en cuenta en como nos hemos acostumbrado solo a escribir mensajes de texto en vez de volver ha escribir cartas en las cuales podemos expresarnos emociones y pensamiento los cuales son muy dificles de expresar o talvez en adentrarnos en leer llibros que nos puede hacer imaginar otras mundos y realidades, mandamos de un 1000000 de mensajes al dia pero no somos capaces de leer un libro escribir algo a mano es por eso que hoy en dia prefieren darle a un niño un celular en vez de un cuaderno o un libro esto causa muchos problemos debido a que genera falencias en el aparendizaje y concentración por que los niños se enfocan en lo artifical a pensar en otras cosas.

Se refiere a que debemos tener la disposición y la motivación para iniciar nuevas ideas, proyectos o acciones sin esperar a que otros te digan qué hacer. Implica ser proactivo y estar dispuesto a tomar la iniciativa para explorar nuevas posibilidades, experimentar con diferentes enfoques y buscar soluciones innovadoras.

Este texto nos menciona el que debemos incentivar la práctica de hablar, ya que así se mejora la claridad y precisión en la pronunciación de palabras y al igual a eso se aprende a apreciar y disfrutar los sonidos y ritmos de cada de las palabras que estamos aprendiendo.

So basically this some guidance on what companies who aren't covered by the CSRD ought to cover to demonstrate following the intent of the law

Why Quality Matters: Choosing the Best Face Paints for Your Art

eLife assessment

This valuable work explores death coding data to understand the impact of COVID-19 on cancer mortality. The work provides solid evidence that deaths with cancer as a contributing cause were not above what would be expected during pandemic waves, suggesting that cancer did not strongly increase the risk of dying of COVID-19. These results are an interesting exploration into the coding of causes of death that can be used to make sense of how deaths are coded during a pandemic in the presence of other underlying diseases, such as cancer.

Reviewer #1 (Public Review):

Summary:

In the paper, the authors study whether the number of deaths in cancer patients in the USA went up or down during the first year (2020) of the COVID-19 pandemic. They found that the number of deaths with cancer mentioned on the death certificate went up, but only moderately. In fact, the excess with-cancer mortality was smaller than expected if cancer had no influence on the COVID mortality rate and all cancer patients got COVID with the same frequency as in the general population. The authors conclude that the data are consistent with cancer not being a risk factor for COVID and that cancer patients were likely actively shielding themselves from COVID infections.

Strengths:

The paper studies an important topic and uses sound statistical and modeling methodology. It analyzes both, deaths with cancer listed as the primary cause of death, as well as deaths with cancer listed as one of the contributing causes. The authors argue, correctly, that the latter is a more important and reliable indicator to study relationships between cancer and COVID. The authors supplement their US-wide analysis with analysing three states separately.

For comparison, the authors study excess mortality from diabetes and from Alzheimer's disease. They show that Covid-related excess mortality in these two groups of patients was expected to be much higher (than in cancer patients), and indeed that is what the data showed.

Author response:

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

eLife assessment

This valuable work explores death coding data to understand the impact of COVID-19 on cancer mortality. The work provides solid evidence that deaths with cancer as a contributing cause were not above what would be expected during pandemic waves, suggesting that cancer did not strongly increase the risk of dying of COVID-19. These results are an interesting exploration into the coding of causes of death that can be used to make sense of how deaths are coded during a pandemic in the presence of other underlying diseases, such as cancer.

We thank the editor and reviewers for the time they took to review our manuscript and for the thoughtful suggestions they provided. We have completed several revisions based on their feedback and we feel our paper is stronger as a result. However, none of these revisions change the overall conclusions of our study.

Reviewer #1 (Public Review):

Summary:

In the paper "Disentangling the relationship between cancer mortality and COVID-19", the authors study whether the number of deaths in cancer patients in the USA went up or down during the first year (2020) of the COVID-19 pandemic. They found that the number of deaths with cancer mentioned on the death certificate went up, but only moderately. In fact, the excess with-cancer mortality was smaller than expected if cancer had no influence on the COVID mortality rate and all cancer patients got COVID with the same frequency as in the general population. The authors conclude that the data show no evidence of cancer being a risk factor for COVID and that the cancer patients were likely actively shielding themselves from COVID infections.

Strengths:

The paper studies an important topic and uses sound statistical and modeling methodology. It analyzes both, deaths with cancer listed as the primary cause of death, as well as deaths with cancer listed as one of the contributing causes. The authors argue, correctly, that the latter is a more important and reliable indicator to study relationships between cancer and COVID. The authors supplement their US-wide analysis by analysing three states separately.

Weaknesses:

The main findings of the paper can be summarized as six numbers. Nationally, in 2022, multiple-cause cancer deaths went up by 2%, Alzheimer's deaths by 31%, and diabetes deaths by 39%. At the same time, assuming no relationship between these diseases and either Covid infection risk or Covid mortality risk, the deaths should have gone up by 7%, 46%, and 28%. The authors focus on cancer deaths and as 2% < 7%, conclude that cancer is not a risk factor for COVID and that cancer patients must have "shielded" themselves against Covid infections.

However, I did not find any discussion of the other two diseases. For diabetes, the observed excess was 39% instead of "predicted by the null model" 28%. I assume this should be interpreted as diabetes being a risk factor for Covid deaths. I think this should be spelled out, and also compared to existing estimates of increased Covid IFR associated with diabetes.

And what about Alzheimer's? Why was the observed excess 31% vs the predicted 46%? Is this also a shielding effect? Does the spring wave in NY provide some evidence here? Why/how would Alzheimer's patients be shielded? In any case, this needs to be discussed and currently, it is not.

We thank the reviewer for their positive feedback on the paper and for these suggestions. It is true that we have emphasized the impact on cancer deaths, as this was the primary aim of the paper. In the revised version, we have expanded the results and discussion sections to more fully describe the other chronic conditions we used as comparators (lines 267-284;346 – 386).

Note that we are somewhat reluctant to designate any of these conditions as risk factors based solely on comparing the time series model with the demographic model of our expectations. As we mention in the discussion, there is considerable uncertainty around estimates from the demographic model in terms of the size of the population-at-risk, the mean age of the population-at-risk, and the COVID-19 infection rates and infection fatality ratios. Our demographic model is primarily used to demonstrate the effects of competing risks across types of cancers and chronic conditions, since these findings are robust to model assumptions. In contrast, the demographic model should be used with caution if the goal is to titrate the level of these risk factors (as the level of imputed risk is dependent on model assumptions). In the updated version of the manuscript, we have included uncertainty intervals in Table 3, using the upper and lower bounds of the estimated infection rates and IFRs, to better represent this uncertainty. We have also discussed this uncertainty more explicitly in the text and ran sensitivity analyses with different infection rate assumptions in the discussion (lines 354-362; 367 -370).

We would like to note that rather than interpreting the absolute results, we used this demographic model as a tool to understand the relative differences between these conditions. From the demographic model we determined that we would expect to see much higher mortality in diabetes and Alzheimer’s deaths compared to cancer deaths due to three factors (1. Size of population-at-risk, 2. Mean age of the population-at-risk, 3. Baseline risk of mortality from the condition), that are separate from the COVID-19 associated IFR. And in general, this is what we observed.

In comparing the results from the demographic model to the observed excess, diabetes does standout as an outlier from cancer and Alzheimer’s disease in that the observed excess is consistently above the null hypothesis which does lend support to the conclusion that diabetes is in fact a risk factor for COVID-19. A conclusion which is also supported by many other studies. Our findings for hematological cancers are also similar, in that we find consistent support for this condition being a risk factor. We have commented on this in the discussion and added a few references (lines 346-354; 395-403).

Our hypothesis regarding non-hematological cancer deaths (lower than anticipated mortality due to shielding) could also apply to Alzheimer’s deaths. Furthermore, we used the COVID-19 attack rate for individuals >65 years (based on the data that is available), but we estimate that the mean age of Alzheimer’s patients is actually 80-81 years, so this attack rate may in fact be a bit too high, which would increase our expected excess. We have commented on this in the discussion (lines 363-377).

Reviewer #2 (Public Review):

The article is very well written, and the approach is quite novel. I have two major methodological comments, that if addressed will add to the robustness of the results.

(1) Model for estimating expected mortality. There is a large literature using a different model to predict expected mortality during the pandemic. Different models come with different caveats, see the example of the WHO estimates in Germany and the performance of splines (Msemburi et al Nature 2023 and Ferenci BMC Medical Research Methodology 2023). In addition, it is a common practice to include covariates to help the predictions (e.g., temperature and national holidays, see Kontis et al Nature Medicine 2020). Last, fitting the model-independent for each region, neglects potential correlation patterns in the neighbouring regions, see Blangiardo et al 2020 PlosONE.

Thank you for these comments and suggestions. We agree there are a range of methods that can be used for this type of analysis, and they all come with their strengths, weaknesses, and caveats. Broadly, the approach we chose was to fit the data before the pandemic (2014-2019), and project forward into 2020. To our knowledge it is not a best practice to use an interpolating spline function to extrapolate to future years. This is demonstrated by the WHO estimates in Germany in the paper you mention. This was our motivation for using polynomial and harmonic terms.

Based on the above:

a. I believe that the authors need to run a cross-validation to justify model performance. I would suggest training the data leaving out the last year for which they have mortality and assessing how the model predicts forward. Important metrics for the prediction performance include mean square error and coverage probability, see Konstantinoudis et al Nature Communications 2023. The authors need to provide metrics for all regions and health outcomes.

Thank you for this suggestion. We agree that our paper could be strengthened by including cross validation metrics to justify model performance. Based on this suggestion, and your observations regarding Alzheimer’s disease, we have done two things. First, for the full pre-pandemic period (2014-2019) for each chronic condition and location we tested three different models with different degree polynomials (1. linear only, 2. linear + second degree polynomial, 3. linear + second degree polynomial + third degree polynomial) and used AIC to select the best model for each condition and location. Next, also in response to your suggestion, we estimated coverage statistics. Using the best fit model from the previous step, we then fit the model to data from 2014-2018 only and used the model to predict the 2019 data. We calculated the coverage probability as the proportion of weekly observed data points that fell within the 95% prediction interval. For all causes of death and locations the coverage probability was 100% (with the exception of multiple cause kidney disease in California, which is only shown in the appendix). The methods and results have been updated to reflect this change and we have added a figure to the appendix showing the selected model and coverage probability for each cause of death and location (lines 504 – 519; 847-859; Appendix 1- Figure 11).

b. In the context of validating the estimates, I think the authors need to carefully address the Alzheimer case, see Figure 2. It seems that the long-term trends pick an inverse U-shape relationship which could be an overfit. In general, polynomials tend to overfit (in this case the authors use a polynomial of second degree).It would be interesting to see how the results change if they also include a cubic term in a sensitivity analysis.

Thank you for this observation. Based on the changes described above, the model for Alzheimer’s disease now includes a cubic term in the national data and in Texas and California. The model with the second-degree polynomial remained the best fit for New York (Appendix 1 – Figure 11).

c. The authors can help with the predictions using temperature and national holidays, but if they show in the cross-validation that the model performs adequately, this would be fine.

At the scale of the US, adding temperature or environmental covariates is difficult and few US-wide models do so (see Goldstein 2012 and Quandelacy 2014 for examples from influenza). Furthermore, because we are looking at chronic disease outcomes, it is unclear that viral covariates or national holidays would drive these outcomes in the same way as they would if we were looking at mortality outcomes more directly related to transmissible diseases (such as respiratory mortality). Our cross validation also indicates that our models fit well without these additional covariates.

d. It would be nice to see a model across the US, accounting for geography and spatial correlation. If the authors don't want to fit conditional autoregressive models in the Bayesian framework, they could just use a random intercept per region.

We think the reviewer is mistaken here about the scale of our national analysis. Our national analysis did not fit independent models for each state or region. Rather, we fit a single model to the weekly-level national mortality data where counts for the whole of the US have been aggregated. We have clarified in the text (lines 156, 464). As such, we do not feel a model accounting for spatial correlation would be appropriate nor would we be able to include a random intercept for each region. We did fit three states independently (NY, TX, CA), but these states are very geographically distant from each other and unlikely to be correlated. These states were chosen in part because of their large population sizes, yet even in these states, confidence intervals were very wide for certain causes of death. Fitting models to each of the 50 US states, most of which are smaller than those chosen here, would exacerbate this issue.

(2) I think the demographic model needs further elaboration. It would be nice to show more details, the mathematical formula of this model in the supplement, and explain the assumptions

Thank you for this comment. We have added additional details on the demographic model to the methods. We have also extended this analysis to each state to further strengthen our conclusions (lines 548-590).

Reviewing Editor Recommendations:

I think that perhaps something that is missing is that the authors never make their underlying assumption explicit: they are assuming that if cancer increases the risk of dying of COVID-19, this would be reflected in the data on multiple causes of death where cancer would be listed as one of the multiple causes rather than as the underlying cause, and that their conclusions are predicated on this assumption. I would suggest explicitly stating this assumption, as opposed to other reasons why cancer mortality would increase (ex. if cancer care worsened during pandemic waves leading to poorer cancer survival).

Response: Thank you for this suggestion. We have added a few sentences to the introduction to make this assumption clear (lines 106-112).

Reviewer #1 (Recommendations For The Authors):

- It could make sense to add "in the United States" into the title, as the paper only analyses US data.

- It may make sense to reformulate the title from "disentangling the relationship..." into something that conveys the actual findings, e.g. "Lack of excess cancer mortality during Covid-19 pandemic" or something similar. Currently, the title tells nothing about the findings.

Thank you for these suggestions. We have added “in the US” to the title. However, we feel that our findings are a bit more subtle than the suggested reformulation would imply, and we prefer to leave it in its current form.

- Abstract, lines 42--45: This is the main finding of the paper, but I feel it is simplified too strongly in the abstract. Your simulations do *not* "largely explain" excess mortality with cancer; they give higher numbers! Which you interpret as "shielding" etc., but this is completely absent from the abstract. This sentence makes the impression that you got a good fit between simulated excess and real excess, which I would say is not the case.

Thank you for this comment. We have rephrased the sentence in the abstract to better reflect our intentions for using the demographic model (lines 46-49). As stated above, the purpose of the demographic model was not to give a good fit with the observed excess mortality. Rather, we used the demographic model as a tool to understand the relative differences between these conditions in terms of expected excess mortality given the size, age-distribution, and underlying risk of death from the condition itself, assuming similar IFR and attack rates. And based on this, we conclude that it is not necessarily surprising that we see higher excess mortality for diabetes and Alzheimer’s compared to cancer.

- Results line 237: you write that it's "more consistent with the null hypothesis", however clearly it is *not* consistent with the null hypothesis either (because 2% < 7%). You discuss in the Discussion that it may be due to shielding, but it would be good to have at least one sentence about it already here in the Results, and refer to the Discussion.

We have mentioned this in the results and refer to the discussion (lines 277-278).

- Results line 239: why was it closer to the assumption of relative risk 2? If I understand correctly, your model prediction for risk=1 was 7% and for risk=2 it was 13%. In NY you observed 8% (line 187). How is this closer to risk=2?

Thank you for this observation. We have updated the demographic model with new data, extended the model to state-level data, and included confidence intervals on these estimates. We have also added additional discussion around the differences between our observations and expectations (lines 249-284).

- Discussion line 275: "we did not expect to see large increases" -- why exactly? Please spell it out here. Was it due to the age distribution of the cancer patients? Was it due to the high cancer death risk?

We demonstrate that it is the higher baseline risk of death for cancer that seems to be driving our low expectations for cancer excess mortality (lines 304-320). We have added this to the sentence to clarify our conclusions on this point and have added a figure to better illustrate this concept of competing risks (Figure 6).

- Methods, line 405: perhaps it makes sense to cite some other notable papers on Covid excess mortality such as Msemburi et al Nature 2023, Karlinsky & Kobak eLife 2021, Islam et al BMJ 2021, etc.

Thank you for mentioning this oversight. We certainly should have cited these papers and have included them in the updated version.

- Methods line 410: why did you use a 5-week moving average? Why not fit raw weekly death counts? NB regression should be able to deal with it.

Smoothing time series data with a moving average prior to running regression models is a very common practice. We did a sensitivity analysis using the raw data. This produced excess estimates with slightly larger confidence intervals, but does not change the overall conclusions of the paper.

- Methods line 416: please indicate the software/library/package you used for fitting NB regression.

We fit the NB regression using the MASS package in R version 4.3. We have added this to the methods (line 519).

- Line 489: ORCHID -> ORCID

This is a test

( ~ 10:20)

Kathleen recommends as part of an inspectional reading to find out who the author is. This is valuable and I believe not something Adler & van Doren mentioned in their book.

Knowing who the author is gives more context to the book and potentially some information about credibility.

Will implement this.

(~5:40)

It appears she put some sticky notes at important points/structure references while reading inspectionally...

Does this not take too much effort/time for an inspectional read a la Adler?

( ~ 10:45)

This is basically layered learning and making use of the creation of prior knowledge.

(~10:00)

It's not just about your domain knowledge on a subject, it's also about your reading skill in general and how difficult a book is written.

This video tells me I need to spend more time actually reflecting on the table of contents and title. As well as with the pigeonholing; classify in the mind in what categories this book falls.

( ~1:55)

Interesting sentiment. Library Lin supposes that most people who do not like reading don't like it because of bad reading habits and that when they improve on their reading habits, they will start liking it.

Off-topic, I love this woman's accent.

位运算不需要额外空间，但是left和right不能相等。 while(left < right){ s[left] ^= s[right]; s[right] ^= s[left]; s[left] ^= s[right]; ++left; --right; }

Yes this is non-examinable but useful to know!

测试标注

consider adding mood stabilisers

maybe drop prochlorperazine

likely to be coincidental rather than cause of prescription

Discover the natural power of essential oils in skincare! From moisturisation to soothing properties, these plant-derived wonders offer a sensorial treat for your skin. Elevate your bathing experience with luxurious soaps enriched with the goodness of nature. Read more: https://techlics.com/how-do-essential-oils-enhance-our-soaps/

I think this is why I've enjoyed building keyboards in recent years. Steep learning curve for me (no background in electronics or C), but it's enough to know there's a solution for pretty much every problem I might encounter. It's an on-ramp to engaging with something like writing. It fosters the notion that there is a "solution", I just need to keep digging. Keyboard building as training for a resilient creative practice. Yes.

galima matyti su vizualizatoriumi priešais prizmutę

neišimt galimai geriau ...ir išimt prizmutę

eLife assessment

This work provides a valuable characterization of neural activity in the anterior insular cortex during fear. Using behavior, single unit recording, and optogenetic control of neural activity, the paper provides convincing data on the role of anterior insular circuits in bidirectionally controlling fear. The study is a great starting point on the path to testing hypotheses about bidirectional control of behavior via neural activity in anatomically defined output populations.

Reviewer #1 (Public Review):

The authors tested whether anterior insular cortex neurons that increase or decrease firing during fear behavior, freezing, bidirectionally control fear via separate, anatomically defined outputs. Using a fairly simple behavior where mice were exposed to tone-shock pairings, they found roughly equal populations that increased or decreased firing during freezing. They next tested whether these distinct populations also had distinct outputs. Using retrograde tracers they found that the anterior insular cortex contains non-overlapping neurons which project to the mediodorsal thalamus or amygdala. Mediodorsal thalamus-projecting neurons tended to cluster in deep cortical layers, while amygdala-projecting neurons were primarily in more superficial layers. Stimulation of insula-thalamus projection decreased freezing behavior, and stimulation of insula-amygdala projections increased fear behavior. Given that the neurons which increased firing were located in deep layers, that thalamus projections occurred in deep layers, and that stimulation of insula-thalamus neurons decreased freezing, the authors concluded that the increased-firing neurons were likely thalamic projections. Similarly, given that decreased-firing neurons tended to occur in more superficial layers, that insula-amygdala projections were primarily superficial, and that insula-amygdala stimulation increased freezing behavior, authors concluded that the decreased firing cells were likely amygdala projections. The study has several strengths though also some caveats. Overall, the authors provide a valuable contribution to the field by demonstrating bidirectional control of behavior, linking the underlying anatomy and physiology.

Strengths:

The potential link between physiological activity, anatomy, and behavior is well laid out and is an interesting question. The activity contrast between the units that increase/decrease firing during freezing is clear.

It is nice to see the recording of extracellular spiking activity, which provides a clear measure of neural output, whereas similar studies often use bulk calcium imaging, a signal which rarely matches real neural activity even when anatomy suggests it might.

Weaknesses:

The link between spiking, anatomy, and behavior requires assumptions/inferences: the anatomically/genetically defined neurons which had distinct outputs and opposite behavioral effects can only be assumed the increased/decreased spiking neurons, based on the rough area of cortical layer they were recorded. This is, of course, discussed as a future experiment.

Reviewer #2 (Public Review):

In this study, the authors aim to understand how neurons in the anterior insular cortex (insula) modulate fear behaviors. They report that the activity of a subpopulation of insula neurons is positively correlated with freezing behaviors, while the activity of another subpopulation of neurons is negatively correlated to the same freezing episodes. They then used optogenetics and showed that activation of anterior insula excitatory neurons during tones predicting a footshock increases the amount of freezing outside the tone presentation, while optogenetic inhibition had no effect. Finally, they found that two neuronal projections of the anterior insula, one to the amygdala and another to the medial thalamus, are increasing and decreasing freezing behaviors respectively.

Author response:

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

Reviewer #1 (Public Review):

The authors sought to test whether anterior insular cortex neurons increase or decrease firing during fear behavior and freezing, bi-directionally control fear via separate, anatomically defined outputs. Using a fairly simple behavior where mice were exposed to tone-shock pairings, they found roughly equal populations that do indeed either increase or decrease firing during freezing. Next, they sought to test whether these distinct populations may also have distinct outputs. Using retrograde tracers they found that the anterior insular cortex contains non-overlapping neurons which project to the mediodorsal thalamus or amygdala. Mediodorsal thalamus-projecting neurons tended to cluster in deep cortical layers while amygdala-projecting neurons were primarily in more superficial layers. Stimulation of insula-thalamus projection decreased freezing behavior, and stimulation of insula-amygdala projections increased fear behavior. Given that the neurons that increased firing were located in deep layers, that thalamus projections occurred in deep layers, and that stimulation of insula-thalamus neurons decreased freezing, the authors concluded that the increased firing neurons may be thalamus projections. Similarly, given that decreased-firing neurons tended to occur in more superficial layers, that insula-amygdala projections were primarily superficial, and that insula-amygdala stimulation increased freezing behavior, authors concluded that the decreased firing cells may be amygdala projections. The study has several strengths though also some caveats.

Strengths:

The potential link between physiological activity, anatomy, and behavior is well laid out and is an interesting question. The activity contrast between the units that increase/decrease firing during freezing is clear.

It is nice to see the recording of extracellular spiking activity, which provides a clear measure of neural output, whereas similar studies often use bulk calcium imaging, a signal that rarely matches real neural activity even when anatomy suggests it might (see London et al 2018 J Neuro - there are increased/decreased spiking striatal populations, but both D1 and D2 striatal neurons increase bulk calcium).

Weaknesses:

The link between spiking, anatomy, and behavior requires assumptions/inferences: the anatomically/genetically defined neurons which had distinct outputs and opposite behavioral effects can only be assumed the increased/decreased spiking neurons, based on the rough area of the cortical layer they were recorded.

Yes, we are aware that we could not provide a direct link between spiking, anatomy and behavior. We have specifically noted this in the discussion section and added a possible experiment that could be carried out to provide a more direct link in a future study.

[Lines 371-375] We would like to provide a more direct evidence between the neuronal response types and projection patterns in future studies by electrophysiologically identifying freezing-excited and freezing-inhibited aIC neurons and testing whether those neurons activates to optogenetic activation of amygdala or medial thalamus projecting aIC neurons.

The behavior would require more control to fully support claims about the associative nature of the fear response (see Trott et al 2022 eLife) - freezing, in this case, could just as well be nonassociative. In a similar vein, fixed intertrial intervals, though common practice in the fear literature, pose a problem for neurophysiological studies. The first is that animals learn the timing of events, and the second is that neural activity is dynamic and changes over time. Thus it is very difficult to determine whether changes in neural activity are due to learning about the tone-shock contingency, timing of the task, simply occur because of time and independently of external events, or some combination of the above.

Trott et al. (2022) stated that "...freezing was the purest reflection of associative learning." The nonassociative processes mentioned in the study were related to running and darting behaviors, which the authors argue are suppressed by associative learning. Moreover, considerable evidence from immediate postshock freezing and immediate postshock context shift studies all indicate that the freezing response is an associative (and not nonassociative) response (Fanselow, 1980 and 1986; and Landeira-Fernandez et al., 2006). Thus, our animals' freezing response to the tone CS presentation in a novel context, following three tone CS-footshock US pairings, most likely reflects associative learning. 

Concerning the issue of fixed inter-trial intervals (ITIs), which are standard in fear conditioning studies, particularly those with few CS-US paired trials, we acknowledge the challenge in interpreting the neural correlates of behavior. However, the ITIs in our extinction study was variable and we still found neural activities that had significant correlation with freezing. The results of our extinction study, carried out with variable it is, suggest that the aIC neural activity changes measured in this study is likely due to freezing behavior associated with fear learning, not due to learning the contingencies of fixed ITIs.

Reviewer #2 (Public Review):

In this study, the authors aim to understand how neurons in the anterior insular cortex (insula) modulate fear behaviors. They report that the activity of a subpopulation of insula neurons is positively correlated with freezing behaviors, while the activity of another subpopulation of neurons is negatively correlated to the same freezing episodes. They then used optogenetics and showed that activation of anterior insula excitatory neurons during tones predicting a footshock increases the amount of freezing outside the tone presentation, while optogenetic inhibition had no effect. Finally, they found that two neuronal projections of the anterior insula, one to the amygdala and another to the medial thalamus, are increasing and decreasing freezing behaviors respectively. While the study contains interesting and timely findings for our understanding of the mechanisms underlying fear, some points remain to be addressed.

We are thankful for the detailed and constructive comments by the reviewer and addressed the points. Specifically, we included possible limitations of using only male mice in the study, included two more studies about the insula as references, specified the L-ratio and isolated distance used in our study, added the ratio of putative-excitatory and putative-inhibitory neurons obtained from our study, changed the terms used to describe neuronal activity changes (freezing-excited and freezing-inhibited cells), added new analysis (Figure 2H), rearranged Figure 2 for clarity, added new histology images, and added atlas maps with viral expressions (three figure supplements).

Reviewer #1 (Recommendations For The Authors):

- I would suggest keeping the same y-axis for all figures that display the same data type - Figure 5D, for example.

Thank you for the detailed suggestion. We corrected the y-axis that display the same data type to be the same for all figures.

- In the methods, it says 30s bins were used for neural analysis (line 435). I cannot imagine doing this, and looking at the other figures, it does not look like this is the case so could you please clarify what bins, averages, etc were used for neural and behavioral analysis?

Bin size for neural analysis varied; 30s, 5s, 1s bins were used depending on the analysis. We corrected this and specified what time bin was used for which figure in the methods.

Bin size for neural and freezing behavior was 30s and we also added this to the methods.

- I would not make any claims about the fear response here being associative/conditional. This would require a control group that received an equal number of tone and shock exposures, whether explicitly unpaired or random.

The unpaired fear conditioning paradigm, unpaired tone and shock, suggested by the reviewer is well characterized not to induce fear behavior by CS (Moita et al., 2003 and Kochli et al., 2015). In addition, considerable evidence from immediate post-shock freezing and immediate post-shock context shift studies all indicate that the freezing response is an associative (and not nonassociative) response (Fanselow, 1980 and 1986; and Landeira-Fernandez et al., 2006). Thus, our animals' freezing response to the tone CS presentation in a novel context, following three tone CS-footshock US pairings, most likely reflects associative learning.

- I appreciate the discussion about requiring some inference to conclude that anatomically defined neurons are the physiologically defined ones. This is a caveat that is fully disclosed, however, I might suggest adding to the discussion that future experiments could address this by tagging insula-thalamus or insula-amygdala neurons with antidromic (opto or even plain old electric!) stimulation. These experiments are tricky to perform, of course, but this would be required to fully close all the links between behavior, physiology, and anatomy.

As suggested, we have included that, in a future study, we would like to elucidate a more direct link between physiology, anatomy and behaviors by optogenetically tagging the insula-thalamus/insula-amygdala neurons and identifying whether it may be a positive or a negative cell (now named the freezing-excited and freezing-inhibited cells, respectively) in the discussion.

[Lines 371-375] We would like to provide a more direct evidence between the neuronal response types and projection patterns in future studies by electrophysiologically identifying freezing-excited and freezing-inhibited aIC neurons and testing whether those neurons activates to optogenetic activation of amygdala or medial thalamus projecting aIC neurons.

Reviewer #2 (Recommendations For The Authors):

Major comments:

(1) As all experiments have been performed only in male mice, the authors need to clearly state this limit in the introduction, abstract, and title of the manuscript.

With increasing number of readers becoming interested in the biological sex used in preclinical studies, we also feel that it should be mentioned in the beginning of the manuscript. As suggested, we explicitly wrote that we only used male mice in the title, abstract, and introduction. In addition, we discussed possible limitations of only using male mice in the discussion section as follows:

[Lines 381-386] Another factor to consider is that we have only used male mice in this study. Although many studies report that there is no biological sex difference in cued fear conditioning (42), the main experimental paradigm used in this study, it does not mean that the underlying brain circuit mechanism would also be similar. The bidirectional fear modulation by aIC→medial thalamus or the aIC→amygdala projections may be different in female mice, as some studies report reduced cued fear extinction in females (42).

(2) The authors are missing important publications reporting findings on the insular cortex in fear and anxiety. For example, the authors should cite studies showing that anterior insula VIP+ interneurons inhibition reduces fear memory retrieval (Ramos-Prats et al., 2022) and that posterior insula neurons are a state-dependent regulator of fear (Klein et al., 2021). Also, regarding the anterior insula to basolateral amygdala projection (aIC-BLA), the author should include recent work showing that this population encodes both negative valence and anxiogenic spaces (Nicolas et al., 2023). 

We appreciate the detailed suggestions and we added appropriate publications in the discussion section. The anterior insula VIP+ interneuron study (Ramos-Prats et al., 2022) is interesting, but based on the evidence provided in the paper, we felt that the role of aIC VIP+ interneuron in fear conditioning is low. VIP+ interneurons in the aIC seem to be important in coding sensory stimuli, however, it’s relevance to conditioned stimuli seems to be low; overall VIP intracellular calcium activity to CS was low and did not differ between acquisition and retrieval. Also, inhibition of VIP did not influence fear acquisition. VIP inhibition during fear acquisition did reduce fear retrieval (CS only, no light stimulation), but this does not necessarily mean that VIP activity will be involved in fear memory storage or retrieval, especially because intracellular calcium activity of VIP+ neurons was low during fear conditioning and retrieval.

Studies by Klein et al. (2021) and Nicolas et al. (2023) are integrated in the discussion section as follows.

[Lines 297-301] Group activity of neurons in the pIC measured with fiberphotometry, interestingly, exhibited fear state dependent activity changes—decreased activity with high fear behavior and increased activity with lower fear behavior (29)—suggesting that group activity of the pIC may be involves in maintain appropriate level of fear behavior.

[Lines 316-319] Another distinction between the aIC and pIC may be related with anxiety, as a recent study showed that group activity of aIC neurons, but not that of the pIC, increased when mice explored anxiogenic space (open arms in an elevated plus maze, center of an open field box) (32).

(3) The authors should specify how many neurons they excluded after controlling the L-ratio and isolation distance. It is also important to specify the percentage of putative excitatory and inhibitory interneurons recorded among the 11 mice based on their classification (the number of putative inhibitory interneurons in Figure 1D seems too low to be accurate).

We use manual cluster cutting and only cut clusters that are visually well isolated. So we hardly have any neurons that are excluded after controlling for L-ratio and isolation distance. The criterion we used was L-ratio<0.3 and isolation distance>15, and we specified this in the methods as follows.

[Lines 454-458] We only used well-isolated units (L-ratio<0.3, isolation distance>15) that were confirmed to be recorded in the aIC (conditioned group: n = 116 neurons, 11 mice; control group: n = 14 neurons, 3 mice) for the analysis (46). The mean of units used in our analysis are as follows: L-ratio = 0.09 ± 0.012, isolation distance = 44.97 ± 5.26 (expressed as mean ± standard deviation).

As suggested, we also specified the percentage of putative excitatory and inhibitory interneurons recorded from our study in the results and methods section. The relative percentage of putative excitatory and inhibitory interneurons were similar for both the conditioned and the control groups (conditioned putative-excitatory: 93.1%, putative-inhibitory: 6.9%; control putative-excitatory: 92.9%, putative-inhibitory: 7.1%). Although the number of putative-interneurons isolated from our recordings is low that is what we obtained. Putative inhibitory neurons, probably because of their relatively smaller size, has a tendency to be underrepresented than the putative excitatory cells.

[Lines 83-87] Of the recorded neurons, we analyzed the activity of 108 putative pyramidal neurons (93% of total isolated neurons) from 11 mice, which were distinguished from putative interneurons (n = 8 cells, 7% of total isolated neurons) based on the characteristics of their recorded action potentials (Figure 1D; see methods for details).

[Lines 464-467] The percentage of putative excitatory neurons and putative inhibitory interneurons obtained from both groups were similar (conditioned putative-excitatory: 93.1%, putative-inhibitory: 6.9%; control putative-excitatory: 92.9%, putative-inhibitory: 7.1%).

(4) While the use of correlation of single-unit firing frequency with freezing is interesting, classically, studies analyze the firing in comparison to the auditory cues. If the authors want to keep the correlation analysis with freezing, rather than correlations to the cues, they should rename the cells as "freezing excited" and "freezing inhibited" cells instead of positive and negative cells.

As suggested, we used the terms “freezing-excited” and “freezing-inhibited” cells instead of positive and negative cells.

(5) To improve clarity, Figure 2 should be reorganized to start with the representative examples before including the average of population data. Thus Panel D should be the first one. The authors should also consider including the trace of the firing rate of these representative units over time, on top of the freezing trace, as well as Pearson's r and p values for both of them. Then, the next panels should be ordered as follows: F, G, H, C, A, B, I, and finally E.

We have rearranged Figure 2 based on the suggestions.

(6) It is unclear why the freezing response in Figure 2 is different in current panels F, G, and H. Please clarify this point.

It was because the freezing behaviors of slightly different population of animals were averaged. Some animals did not have positive/negative (or both) cells and only the behavior of animals with the specified cell-type were used for calculating the mean freezing response. With rearrangement of Figure 2, now we do not have plots with juxtaposed mean neuronal response-types and behavior.

(7) Even though the peak of tone-induced firing rate change between negative and positive cells is 10s later for positive cells, the conclusion that this 'difference suggests differential circuits may regulate the activities of different neuron types in response to fear' is overstating the observation. This statement should be rephrased. Indeed, it could be the same circuits that are regulated by different inputs (glutamatergic, GABA, or neuromodulatory inputs).

We agree and delete the statement from the manuscript.

(8) The authors mention they did not find tone onset nor tone offset-induced responses of anterior insula neurons. It would be helpful to represent this finding in a Figure, especially, which were the criteria for a cell to be tone onset or tone offset responding.

We added how tone-onset and tone-offset were analyzed in the methods section and added a plot of the analysis in Figure 2H.

(9) Based on the spread of the viral expression shown in Figure 3B, it appears that the authors are activating/inhibiting insula neurons in the GI layer, whereas single-unit recordings report the electrodes were located in DI, AID, and AIV layers. The authors should provide histology maps of the viral spread for ChR2, NpHR3, and eYFP expression.

Thank you for the excellent suggestion. Now the histological sample in Figure 3B is a sample with expression in the GI/DI/AID layers and it also has an image taken at higher resolution (x40) to show that viral vectors are expressed inside neurons. We also added histological maps with overlay of viral expression patterns of the ChR2, eYFP, and NpHR3 groups in Figure 3—figure supplement 1.

(10) In Figure 5B, the distribution of terminals expressing ChR2 appears much denser in CM than in MD. This should be quantified across mice and if consistent with the representative image, the authors should refer to aIC-CM rather than aIC-MD terminals.

Overall, we referred to the connection as aIC-medial thalamus, which collectively includes both the CM and the MD. Microscopes we have cannot determine whether terminals end at the CM or MD, but the aIC projections seems to pass through the CM to reach the MD. The Allen Brain Institute’s Mouse brain connectivity map (https://connectivity.brain-map.org/projection/experiment/272737914) of a B6 mouse, the mouse strain we used in our study, with tracers injected in similar location as our study also supports our speculation and shows that aIC neuronal projections terminate more in the MD than in the CM. In addition, the power of light delivered for optogenetic manipulation is greatly reduced over distance, and therefore, the MD projecting terminals which is closer to the optic fiber will be more likely to be activated than the CM projecting terminals. However, since we could not determine whether the aIC terminate at the CM or the MD, we collectively referred to the connection as the aIC-medial thalamus throughout the manuscript.

Author response image 1.

(11) Histological verifications for each in vivo electrophysiology, optogenetic, and tracing experiments need to include a representative image of the implantation/injection site, as well as a 40x zoom-in image focusing on the cell bodies or terminals right below the optic fiber (for optogenetic experiments). Moreover, an atlas map including all injection locations with the spread of the virus and fiber placement should be added in the Supplement Figures for each experiment (see Figure S1 Klein et al., 2021). Similarly, the authors need to add a representation of the spread of the retrograde tracers for each mouse used for this tracing experiment.

As suggested, we added a histology sample showing electrode recording location for in-vivo electrophysiology in Figure 1 and added atlas maps for the optogenetic and tracing experiments in supplementary figures. We also provide a 40x zoom-in image of the expression pattern for the optogenetic experiments (Figure 3B).

(12) To target anterior insula neurons, authors mention coordinates that do not reach the insula on the Paxinos atlas (AP: +1.2 mm, ML: -3.4 mm, DV: -1.8 mm). If the DV was taken from the brain surface, this has to be specified, and if the other coordinates are from Bregma, this also needs to be specified. Finally, the authors cite a review from Maren & Fanselow (1996), for the anterior insula coordinates, but it remains unclear why.

AP and ML coordinates are measurement made in reference to the bregma. DV was calculated from the brain surface. We specified these in the Methods. We did not cite a review from Maren & Fenselow for the aIC coordinates.

Minor comments:

(1) A schematic of the microdrive and tetrodes, including the distance of each tetrode would also be helpful.

We used a handcrafted Microdrives with four tetrodes. Since they were handcrafted, the relative orientation of the tetrodes varies and tetrode recording locations has to be verified histologically. We, however, made sure that the distance between tetrodes to be more than 200 μm apart so that distinct single-units will be obtained from different tetrodes. We added this to the methods as follows.

[Lines 430-431] The distance between the tetrodes were greater than 200 μm to ensure that distinct single-units will be obtained from different tetrodes.

(2) Figure 2E: representation of the baseline firing (3-min period before the tone presentation) is missing.

Figure 2E is the 3 min period before tone presentation

(3) Figure 2: Averages Pearson's correlation r and p values should be stated on panels F, G, and H (positive cell r = 0.81, P < 0.05; negative cell r = -0.68, P < 0.05).

They were all originally stated in the figures. But with reorganization of Figure 2, we now have a plot of the Pearson’s Correlation with r and p values in Figure 2F.

(4) Figure 2I: Representation of the absolute value of the normalized firing is highly confusing. Indeed, as the 'negative cells' are inhibited to freezing, firing should be represented as normalized, and negative for the inhibited cells.

To avoid confusion, we did not take an absolute value of the “negative cells”, which are now called the “freezing-inhibited cells”.

(5) Figure 4E (retrograde tracing): representation of individual values is missing.

Figure 4E now has individual values.

References:

London, T. D., Licholai, J. A., Szczot, I., Ali, M. A., LeBlanc, K. H., Fobbs, W. C., & Kravitz, A. V. (2018). Coordinated ramping of dorsal striatal pathways preceding food approach and consumption. Journal of Neuroscience, 38(14), 3547-3558.

Trott, J. M., Hoffman, A. N., Zhuravka, I., & Fanselow, M. S. (2022). Conditional and unconditional components of aversively motivated freezing, flight and darting in mice. Elife, 11, e75663.

Fanselow, M. S. (1980). Conditional and unconditional components of post-shock freezing. The Pavlovian journal of biological science: Official Journal of the Pavlovian, 15(4), 177-182.

Fanselow, M. S. (1986). Associative vs topographical accounts of the immediate shock-freezing deficit in rats: implications for the response selection rules governing species-specific defensive reactions. Learning and Motivation, 17(1), 16-39.

Landeira-Fernandez, J., DeCola, J. P., Kim, J. J., & Fanselow, M. S. (2006). Immediate shock deficit in fear conditioning: effects of shock manipulations. Behavioral neuroscience, 120(4), 873.

Moita, M. A., Rosis, S., Zhou, Y., LeDoux, J. E., & Blair, H. T. (2003). Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conditioning. Neuron, 37(3), 485-497.

Kochli, D. E., Thompson, E. C., Fricke, E. A., Postle, A. F., & Quinn, J. J. (2015). The amygdala is critical for trace, delay, and contextual fear conditioning. Learning & memory, 22(2), 92-100.

Ramos-Prats, A., Paradiso, E., Castaldi, F., Sadeghi, M., Mir, M. Y., Hörtnagl, H., ... & Ferraguti, F. (2022). VIP-expressing interneurons in the anterior insular cortex contribute to sensory processing to regulate adaptive behavior. Cell Reports, 39(9).

Klein, A. S., Dolensek, N., Weiand, C., & Gogolla, N. (2021). Fear balance is maintained by bodily feedback to the insular cortex in mice. Science, 374(6570), 1010-1015.

Nicolas, C., Ju, A., Wu, Y., Eldirdiri, H., Delcasso, S., Couderc, Y., ... & Beyeler, A. (2023). Linking emotional valence and anxiety in a mouse insula-amygdala circuit. Nature Communications, 14(1), 5073.

Maren, S., & Fanselow, M. S. (1996). The amygdala and fear conditioning : Has the nut been cracked? Neuron, 16(2), 237‑240. https://doi.org/10.1016/s0896-6273(00)80041-0

eLife assessment

This important study reports that FBXO24 is essential for the normal formation and function of the sperm flagellum, motility, and male fertility in mice. The evidence supporting the direct role of this protein in preventing RNP granule formation in the sperm flagellum is compelling. This work will be of interest to biomedical researchers who work on testicular biology and male fertility.

Reviewer #1 (Public Review):

Summary:

The main goal of the authors was to study the testis-specific role of the protein FBXO24 in the formation and function of the ribonucleoprotein granules (membrane-less electron-dense structures rich in RNAs and proteins).

Strengths:

The wide variety of methods used to support their conclusions (including transgenic models)

Weaknesses:

The complex phenotype observed, in some situations, cannot be fully explained by the experiments presented by the authors (i.e., AR or the tail structure).

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The main goal of the authors was to study the testis-specific role of the protein FBXO24 in the formation and function of the ribonucleoprotein granules (membraneless electron-dense structures rich in RNAs and proteins).

We appreciate the summary comment of reviewer #1.

Strengths:

The wide variety of methods used to support their conclusions (including transgenic models)

We appreciate the positive comment of reviewer #1.

Weaknesses:

The lack of specific antibodies against FBXO24. Some of the experiments showing a specific phenotype are descriptive and lack of logical explanation about the possible mechanism (i.e. AR or the tail structure).

Because we could not obtain specific antibodies against FBXO24, we generated Fbxo24-FLAG transgenic mice, which can be used to show the interaction between FBXO24 and IPO5. For the mechanism of impaired acrosome reaction, we added some results and discussion as written in the response to the question (1) of reviewer #1 (public review). For the mechanism of abnormal flagellar structure, we added new results and fixed the manuscript as written in the response to the major comments of reviewer #3 (recommendations for the authors).

Questions:

The paper is excellent and employs a wide variety of methods to substantiate the conclusions. I have very few questions to ask:

(1) KO mice cannot undergo acrosome reaction (AR) even spontaneously. How do you account for this, given that no visible defects were observed in the acrosome?

One possibility is that Fbxo24 KO spermatozoa cannot undergo capacitation; however, it is difficult to analyze the capacitation status such as tyrosine phosphorylation because most Fbxo24 KO spermatozoa are not alive (Figure S3A). Other possibility is that AR-related proteins are affected in Fbxo24 KO spermatozoa. Therefore, we analyzed the amounts of AR-related proteins with mass spectrometry (Figure S3C). Although previous studies indicate that the assembly of the SNARE complex is a key event prior to AR [Hutt et al., 2005 (PMID: 15774481); Katafuchi et al., 2000 (PMID: 11066067); Schulz et al., 1997 (PMID: 9356173); Tomes et al., 2002 (PMID: 11884041)], no clear differences were detected for SNARE proteins (Figure S3C and D). PLCD4 that is important for AR [Fukami et al., 2001 (PMID: 11340203)) was also detected in Fbxo24 KO spermatozoa (Figure S3C). Although we could not find differences in the amounts of AR-related proteins, it is still possible that FER1L5, another AR-related protein [Morohoshi et al., 2023 (PMID: 36696506)] not detected in the mass spectrometry analyses, or AR-related proteins not yet identified are affected in Fbxo24 KO spermatozoa. We added these results and discussion (line 160-166 and 305-312).

(2) KO sperm are unable to migrate in the female tract, and, more intriguingly, they do not pass through the utero-tubal junction (UTJ). The levels of ADAM3 are normal, suggesting that the phenotype is influenced by other factors. The authors should investigate the levels of Ly6K since mice also exhibit the same phenotype but with normal levels of ADAM3.

We detected LY6K in Fbxo24 KO spermatozoa with immunoblotting, but no difference was found.

We added the results (Figure S3E and line 172–175).

(3) In Figure 4A, the authors assert that "RBGS Tg mice revealed that mitochondria were abnormally segmented in Fbxo24 KO spermatozoa." I am unable to discern this from the picture shown in that panel. Could you please provide a more detailed explanation or display the information more explicitly?

We are sorry for the ambiguous explanation on the morphology of sperm mitochondria sheath. Fbxo24 KO cauda epidydimal spermatozoa shows disorganized mitochondria sheath rather than “segmented”. We fixed the sentence (line 190-192) and added white arrowheads that indicate the disorganized regions (Figure 4A).

Reviewer #2 (Public Review):

Summary:

The manuscript by Kaneda et al "FBXO24 ensures male fertility by preventing abnormal accumulation of membraneless granules in sperm flagella" is a significant paper on the role of FBXO24 in murine male germ cell development and sperm ultrastructure and function. The body of experimental evidence that the authors present is extraordinarily strong in both breadth and depth. The authors investigate the protein's functions in male germ cells and sperm using a wide variety of approaches but focusing predominantly on their novel mouse model featuring deletion of the Fbxo24 gene and its product. Using this mouse, and a cross of it with another model that expresses reporters in the head and midpiece, they logically build from one experiment to the next. Together, their data show that this protein is involved in the regulation of membraneless electron-dense structures; loss of FBXO24 led to an accumulation of these materials and defects in the sperm flagellum and fertilizing ability. Interestingly, the authors found that several of the best-known components of electron-dense ribonucleoprotein granules that are found in the intermitochondrial cement and chromatoid body were not disrupted in the Fbxo24 knockout, suggesting that the electron-dense material and these structures are not all the same, and the biology is more complicated than some might have thought. They found evidence for the most changes in IPO5 and KPNB1, and biochemical evidence that FBXO24 and IPO5 could interact.

We appreciate the summary comment of reviewer #2.

Strengths:

The authors are to be commended for the thoroughness of their experimental approaches and the extent to which they investigated impacts on sperm function and potential biochemical mechanisms. Very briefly, they start by showing that the Fbxo24 message is present in spermatids and that the protein can interact with SKP1, in a way that is dependent on its F-box domain. This points toward a potential function in protein degradation. To test this, they next made the knockout mouse, validated it, and found the males to be sterile, although capable of plugging a female. Looking at the sperm, they identified a number of ultrastructural and morphological abnormalities, which they looked at in high resolution using TEM. They also cross their model with RBGS mice so that they have reporters in both the acrosome and mitochondria. The authors test a variety of sperm functions, including motility parameters, ability to fertilize by IVF, cumulus-free IVF, zona-free-IVF, and ICSI. They found that ICSI could rescue the knockout but not other assisted reproductive technologies. Defects in male fertility likely resulted from motility disruption and failure to get through the utero-tubal junction but defects in acrosome exocytosis also were noted. The authors performed thorough investigations including both targeted and unbiased approaches such as mass spectrometry. These enabled them to show that although the loss of the FBXO24 protein led to more RNA and elevated levels of some proteins, it did not change others that were previously identified in the electron-dense RNP material.

The manuscript will be highly significant in the field because the exact functions of the electron-dense RNP materials have remained somewhat elusive for decades. Much progress has been made in the past 15 years but this work shows that the situation is more complex than previously recognized. The results show critical impacts of protein degradation in the differentiation process that enables sperm to change from non-descript round cells into highly polarized and compartmentalized mature sperm, with an equally highly compartmentalized flagellum. This manuscript also sets a high bar for the field in terms of how thorough it is, which reveals wide-ranging impacts on processes such as mitochondrial compaction and arrangement in the midpiece, the correct building of the major cytoskeletal elements in the flagellum, etc.

We appreciate the positive comment of reviewer #2.

Weaknesses:

There are no real weaknesses in the manuscript that result from anything in the control of the authors. They attempted to rescue the knockout by expressing a FLAG-tagged Fbxo24 transgene, but that did not rescue the phenotype, either because of inappropriate levels/timing/location of expression, or because of interference by the tag. They also could not make anti-FBXO24 that worked for coimmunoprecipitation experiments, so relied on the FLAG epitope, an approach that successfully showed co-IP with IPO5 and SKP1.

We could not rescue the phenotype with Fbxo24-FLAG transgene, but different Fbxo24 mutant mice show the same phenotypes (Figure S6G). Further, another group showed that Fbxo24 KO mice exhibited abnormal mitochondrial coiling [Li et al., 2024 (PMID: 38470475)], confirming that

FBXO24 is involved in the mitochondrial sheath formation.

Reviewer #3 (Public Review):

Summary:

In this manuscript, the authors found that FBXO24, a testis-enriched F-box protein, is indispensable for male fertility. Fbxo24 KO mice exhibited malformed sperm flagellar and compromised sperm motility.

We appreciate the summary comment of reviewer #3.

Strengths:

The phenotype of Fbxo24 KO spermatozoa was well analyzed.

We appreciate the positive comment of reviewer #3.

Weaknesses:

The authors observed numerous membraneless electron-dense granules in the Fbxo24 KO spermatozoa. They also showed abnormal accumulation of two importins, IPO5 and KPNB1, in the Fbxo24 KO spermatozoa. However, the data presented in the manuscript do not support the conclusion that FBXO24 ensures male fertility by preventing the abnormal accumulation of membraneless granules in sperm flagella, as indicated in the manuscript title.

Fbxo24 KO mice showed abnormal accumulation of membraneless granules in sperm flagella and male infertility, suggesting that FBXO24 is involved in these processes, but there are no results that show the direct relationship as reviewer #3 mentioned. Therefore, we fixed the title.

Recommendations For The Authors:

Reviewer #2 (Recommendations For The Authors):

On page 4, lines 152-154, the authors introduce the RBGS mouse model and use it in their experiments.

However, they left out an obvious but helpful sentence that tells the reader that they crossed the Fbxo24-null mouse with the RBGS. As one continues reading it is clear, but best to avoid even slight confusion.

We revised the explanation in the result section (line 150-153).

Reviewer #3 (Recommendations For The Authors):

In this manuscript, the authors found that FBXO24, a testis-enriched F-box protein, is indispensable for male fertility. Fbxo24 KO mice exhibited malformed sperm flagellar and compromised sperm motility. The phenotype of Fbxo24 KO spermatozoa was well analyzed.

The authors observed numerous membraneless electron-dense granules in the Fbxo24 KO spermatozoa. They also showed abnormal accumulation of two importins, IPO5 and KPNB1, in the Fbxo24 KO spermatozoa. However, the data presented in the manuscript do not support the conclusion that FBXO24 ensures male fertility by preventing the abnormal accumulation of membraneless granules in sperm flagella, as indicated in the manuscript title.

Fbxo24 KO mice showed abnormal accumulation of membraneless granules in sperm flagella and male infertility, suggesting that FBXO24 is involved in these processes, but there are no results that show the direct relationship as reviewer #3 mentioned. Therefore, we fixed the title.

Major comments:

In the title, abstract, introduction, and some sections such as lines 275-276, the authors conclude that FBXO24 prevents the accumulation of importins and RNP granules during spermiogenesis. However, the provided data do not substantiate this claim. To provide conclusive evidence to support the current title, the authors need to present evidence supporting: 1) direct degradation of IPO5 and KPNB1 by FBXO24; 2) the direct requirement of IPO5 for the formation of the membraneless granules, and 3) infertility resulting from the presence of membraneless granules, rather than other issues such as abnormal ODF and AX.

(1) direct degradation of IPO5 and KPNB1 by FBXO24.

To examine if IPO5 can be degraded by FBXO24, we performed a ubiquitination assay using HEK293T cells. Ubiquitination of IPO5 was upregulated in the presence of WT FBXO24 but not with the mutant ΔF-box FBXO24, suggesting that IPO5 can be ubiquitinated by FBXO24. We did not examine the ubiquitination of KPNB1 because we failed to construct a plasmid vector expressing mouse KPNB1. We think that KPNB1 is not the substrate because we did not detect the interaction between FBXO24 and KPNB1 (Figure 5E). We added the results of the ubiquitination assay (Figure

5F and line 261-265) and mentioned it in the abstract (line 35).

(2) the direct requirement of IPO5 for the formation of the membraneless granules.

(3) infertility resulting from the presence of membraneless granules, rather than other issues such as abnormal ODF and AX.

We revealed that IPO5 aggregate under stress condition in COS7 cells (Figure 6C and D); however, we did not examine whether IPO5 is required for the formation of the membraneless granules. We consider that protein degradation systems such as PROTAC or Trim-Away to knockdown IPO5 at the protein level in Fbxo24 KO mice could be a good way to see if the membraneless granules are diminished and male fertility is rescued. However, it takes time to apply the degradation systems in vivo. Therefore, we would like to leave this rescue experiment for future studies. We fixed the title and  abstract (line 37-38), and removed the last sentence of the introduction.

Also, the other group reported the analyses of Fbxo24 KO mice [Li et al., 2024 (PMID: 38470475)] right after we submitted our manuscript to the eLife. They reported not only disorganized flagellar structures but also abnormal head morphology, which may lead to male infertility. The differences from our study may be due to different mouse genetic backgrounds. We mentioned it in the discussion section (line 348-353).

Minor comments:

(1) The authors claimed a significant increase in the total amount of RNAs in Fbxo24 KO spermatozoa (lines 259-261), suggesting that the ...contain RNAs. More direct evidence supporting this claim should be provided.

We show that the amounts of IPO5 and KBNB1 increased in Fbxo24 KO spermatozoa (Figure 5A and B), both of which could be incorporated into RNP granules in COS7 cells (Figure 6C and D), supporting the idea that membraneless electron-dense structures may be RNP granules. However, because we did not show direct evidence that electron-dense structures contain RNAs, we removed the sentences (line 259-261 of the 1st submission manuscript). 

(2) The author should provide an explanation for the absence of a FLAG band in the input Tg in Figure 5D and the larger size of the IPO5 band in the FLAG-IP group compared to the input. Similar observations are also noted in Figure 5E.

The FLAG band is weak because the protein amount is low. When we increase the contrast, we can see the FLAG band. We added an image with high contrast (Figure 5D). Sometimes, proteins run differently with SDS-PAGE after immunoprecipitation, likely due to varying protein composition in the sample. We explained it in the figure legend (line 868-869).

(3) In Line 526, clarify the procedure for sperm purification, and determine the potential for contamination from somatic cells.

We did not perform sperm purification, but when we observed spermatozoa obtained from cauda epididymis, we rarely observed either somatic cells or immature spermatogenic cells. We added  pictures in Figure S7. Further, we added detailed explanation about how to collect spermatozoa from the epididymis (line 549-550).

(4) Define the Y-axis in Figure 2E, F, and G.

We have revised the figures.

Author response:

Reviewer #1 (Public Review):

Using the UK Biobank, this study assessed the value of nuclear magnetic resonance measured metabolites as predictors of progression to diabetes. The authors identified a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. In general, this is a well-performed study, and the findings may provide a new approach to identifying those at high risk of developing diabetes. I have some comments that may improve the importance of this study.

We deeply appreciate the reviewer's invaluable time dedicated to the review of this manuscript and the insightful comments to enhance its overall quality.

(1) It is unclear why the authors only considered the top 20 variables in the metabolite selection and why they did not set a wider threshold.

Thank you for the comment. We set the top 20 variables in the metabolite selection balancing the performance of the final diabetes risk prediction model and the clinical applicability due to measurement costs. We have added this explanation in the “Methods” section.

“We chose the intersection set of the top 20 most important variables selected by the three machine learning models, after balancing the performance of the final diabetes risk prediction model and the clinical applicability associated with measurement costs of metabolites.”

(2) The methods section would benefit from a more detailed exposition of how parameter tuning was conducted and the range of parameters explored during the training of the RSF model.

According to the reviewer’s suggestion, we have added a more detailed description of parameters tunning and the range of parameters explored during the training of the RSF model in the “Method S2” section in the Supplementary material.

“The RSF model was fitted using the “randomForestSRC” package and the grid search method was used for hyperparameter tuning. Specifically, the grid search method was used to tune hyperparameters among the RSF model, through minimizing out-of-sample or out-of-bag error1. Each tree in the RSF is constructed from a random sample of the data, typically a bootstrap sample or 63.2% of the sample size (as in the present study). Consequently, not all observations are used to construct each tree. The observations that are not used in the construction of a tree are referred to as out-of-bag observations. In an RSF model, each tree is built from a different sample of the original data, so each observation is “out-of-bag” for some of the trees. The prediction for an observation can then be obtained using only those trees for which the observation was not used for the construction. A classification for each observation is obtained in this way and the error rate can be estimated from these predictions. The resulting error rate is referred to as the out-of-bag error. Through calculating the out-of-bag error in each iteration, the best hyperparameters were finally determined.

The hyperparameters to be tuned and range of grid search in the present study were below: number of trees (50-1000, by 50), number of variables to possibly split at each node (3-6, by 1), and minimum size of terminal node (1-20, by 1)2.”

(3) It is hard to understand the meaning of the decision curve analysis and the clinical implications behind the net benefit, which are required to clarify the application values of models.

Thank you for the comment. We have added more description and discussion about the decision curve analysis in the “Methods” and “Discussion” sections.

“Furthermore, we used decision curve analysis (DCA) to assess the clinical usefulness of prediction model-based guidance for prediabetes management, which calculates a clinical “net benefit” for one or more prediction models in comparison to default strategies of treating all or no patients3.”

“Most importantly, a model with good discrimination does not necessarily have high clinical value. Hence, DCA was used to compare the clinical utility of the model before and after adding the metabolites, and this showed a higher net benefit for the latter than the basic model, suggesting the addition of the metabolites increased the clinical value of prediction, i.e., the potential benefit of guiding management in individuals with prediabetes3,4. These results provided novel evidence supporting the value of metabolic biomarkers in risk prediction and stratification for the progression from prediabetes to diabetes.”

(4) Notably, the NMR platform utilized within the UK Biobank primarily focused on lipid species. This limitation should be discussed in the manuscript to provide context for interpreting the results and acknowledge the potential bias from the measuring platform.

Thank you for the comment. We acknowledged this limitation that NMR platform within the UK Biobank primarily focused on lipid species and the potential bias from the measuring platform and have added this in “Discussion” section.

“Third, the Nightingale metabolomics platform primarily focused on lipids and lipoprotein sub-fractions, and thus the predictive value of other metabolites in the progression from prediabetes to diabetes warranted further research using an untargeted metabolomics approach.”

(5) The manuscript should explain the potential influence of non-fasting status on the findings, particularly concerning lipoprotein particles and composition. There should be a detailed discussion of how non-fasting status may impact the measurement and the findings.

According to the reviewer’s suggestion, we have added more details to explain the potential influence of non-fasting status on our findings in the “Discussion” section.

“Additionally, the use of non-fasting blood samples might increase inter-individual variation in metabolic biomarker concentrations, however, fasting duration has been reported to account for only a small proportion of variation in plasma metabolic biomarker concentrations5. Therefore, we believe the impact of non-fasting samples on our findings would be minor.”

(6) Cross-platform standardization is an issue in metabolism, and further descriptions of quality control are recommended.

Thank you for the comment. We have added more description of quality control in the “Method S1” section in the Supplementary material.

“Metabolic biomarker profiling by Nightingale Health’s NMR platform provides consistent results over time and across spectrometers. Furthermore, the sample preparation is minimal in the Nightingale Health’s metabolic biomarker platform, circumventing all extraction steps. These aspects result in highly repeatable biomarker measurements. Pre-specified quality metrics were agreed between UK Biobank and Nightingale Health to ensure consistent results across the samples, and pilot measurements were conducted. Nightingale Health performed real-time monitoring of the measurement consistency within and between spectrometers throughout the UK Biobank samples. Two control samples provided by Nightingale Health were included in each 96-well plate for tracking the consistency across multiple spectrometers. Furthermore, two blind duplicate samples provided by the UK Biobank were included in each well plate, with the position information unlocked only after results delivery. Coefficient of variation (CV) targets across the metabolic biomarker profile were pre-specified for both Nightingale Health’s internal control samples and UK Biobank’s blind duplicates. The targets were met for each consecutively measured batch of ~25,000 samples. For the majority of the metabolic biomarkers, the CVs were below 5% (https://biobank.ndph.ox.ac.uk/showcase/refer.cgi?id=3000). Further, the distributions of measured biomarkers from 5 sample batches indicated absence of batch effects (https://biobank.ctsu.ox.ac.uk/ukb/ukb/docs/nmrm_app1).”

Reviewer #2 (Public Review):

Deciphering the metabolic alterations characterizing the prediabetes-diabetes spectrum could provide early time windows for targeted preventive measures to extend precision medicine while avoiding disproportionate healthcare costs. The authors identified a panel of 9 circulating metabolites combined with basic clinical variables that significantly improved the prediction from prediabetes to diabetes. These findings provided insights into the integration of these metabolites into clinical and public health practice. However, the interpretation of these findings should take account of the following limitations.

We appreciate the reviewer’s positive comments and encouragement.

(1) First, the causal relationship between identified metabolites and diabetes or prediabetes deserves to be further examined particularly when the prediabetic status was partially defined. Some metabolites might be the results of prediabetes rather than the casual factors for progression to diabetes.

Thank you for your insightful comments. We agree with you that the panel of metabolites in this study might not be the causal factor for progression from prediabetes to diabetes, which needs further validation in experimental studies. We have added this limitation in the “Discussion” section.

“Fifth, we could not draw any conclusion about the causality between the identified metabolites and the risk for progression to diabetes due to the observational nature, which remained to be validated in further experimental studies.”

(2) The blood samples were taken at random (not all in a non-fasting state) and so the findings were subjected to greater variability. This should be discussed in the limitations.

According to the reviewer’s suggestion, we have added more details to explain the potential influence of non-fasting status on our findings in the “Discussion” section.

“Additionally, the use of non-fasting blood samples might increase inter-individual variation in metabolic biomarker concentrations, however, fasting duration has been reported to account for only a small proportion of variation in plasma metabolic biomarker concentrations5. Therefore, we believe the impact of non-fasting samples on our findings would be minor.”

(3) The strength of NMR in metabolic profiling compared to other techniques (i.e., mass spectrometry [MS], another commonly used metabolic profiling method) could be added in the Discussion section.

According to the reviewer’s suggestion, we have added the strength of NMR in metabolic profiling compared to other techniques in the “Discussion” section.

“Circulating metabolites were quantified via NMR-based metabolome profiling within the UK Biobank, which offers metabolite qualification with relatively lower costs and better reproducibility6.”

(4) Fourth, the applied platform focuses mostly on lipid species which may be a limitation as well.

Thank you for the comment. We acknowledged this limitation that NMR platform within the UK Biobank primarily focused on lipid species and the potential bias from the measuring platform and have added this in the “Discussion” section.

“Third, the Nightingale metabolomics platform primarily focused on lipids and lipoprotein sub-fractions, and thus the predictive value of other metabolites in the progression from prediabetes to diabetes warranted further research using an untargeted metabolomics approach.”

(5) it is a very large group with pre-diabetes, but the results only apply to prediabetes and not to the general population. This should be clear, although the authors have also validated the predictive value of these metabolites in the general population.

Thank you for the comment. We agree with you that the results only apply to prediabetes and not to the general population, though they also showed potential predictive value among participants with normoglycemia. We have accordingly modified the relevant expressions in the “Conclusion” section to restrict these findings to participants with prediabetes.

“In this large prospective study among individuals with prediabetes, we detected a panel of circulating metabolites that were associated with an increased risk of progressing to diabetes.”

References

(1) Janitza S, Hornung R. On the overestimation of random forest's out-of-bag error. PLoS One. 2018;13(8):e0201904.

(2) Tian D, Yan HJ, Huang H, et al. Machine Learning-Based Prognostic Model for Patients After Lung Transplantation. JAMA Netw Open. 2023;6(5):e2312022.

(3) Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019;3:18.

(4) Li J, Xi F, Yu W, Sun C, Wang X. Real-Time Prediction of Sepsis in Critical Trauma Patients: Machine Learning-Based Modeling Study. JMIR Form Res. 2023;7:e42452.

(5) Li-Gao R, Hughes DA, le Cessie S, et al. Assessment of reproducibility and biological variability of fasting and postprandial plasma metabolite concentrations using 1H NMR spectroscopy. PLoS One. 2019;14(6):e0218549.

(6) Geng T-T, Chen J-X, Lu Q, et al. Nuclear Magnetic Resonance–Based Metabolomics and Risk of CKD. American Journal of Kidney Diseases. 2023.

eLife assessment

This valuable study combines prospective cohort, metabolomics, and machine learning to identify a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. The findings are solid and the methods, data, and analyses support the claims. However, the interpretation would benefit from a more rigorous description. With revision of these weaknesses, this paper would provide insights into the integration of these metabolites into clinical and public health practice.

Reviewer #1 (Public Review):

Using the UK Biobank, this study assessed the value of nuclear magnetic resonance measured metabolites as predictors of progression to diabetes. The authors identified a panel of 9 circulating metabolites that improved the ability in risk prediction of progression from prediabetes to diabetes. In general, this is a well-performed study, and the findings may provide a new approach to identifying those at high risk of developing diabetes.

I have some comments that may improve the importance of this study.

(1) It is unclear why the authors only considered the top 20 variables in the metabolite selection and why they did not set a wider threshold.

(2) The methods section would benefit from a more detailed exposition of how parameter tuning was conducted and the range of parameters explored during the training of the RSF model.

(3) It is hard to understand the meaning of the decision curve analysis and the clinical implications behind the net benefit, which are required to clarify the application values of models.

(4) Notably, the NMR platform utilized within the UK Biobank primarily focused on lipid species. This limitation should be discussed in the manuscript to provide context for interpreting the results and acknowledge the potential bias from the measuring platform.

(5) The manuscript should explain the potential influence of non-fasting status on the findings, particularly concerning lipoprotein particles and composition. There should be a detailed discussion of how non-fasting status may impact the measurement and the findings.

(6) Cross-platform standardization is an issue in metabolism, and further descriptions of quality control are recommended.

Reviewer #2 (Public Review):

Deciphering the metabolic alterations characterizing the prediabetes-diabetes spectrum could provide early time windows for targeted preventive measures to extend precision medicine while avoiding disproportionate healthcare costs. The authors identified a panel of 9 circulating metabolites combined with basic clinical variables that significantly improved the prediction from prediabetes to diabetes. These findings provided insights into the integration of these metabolites into clinical and public health practice. However, the interpretation of these findings should take account of the following limitations.

First, the causal relationship between identified metabolites and diabetes or prediabetes deserves to be further examined particularly when the prediabetic status was partially defined. Some metabolites might be the results of prediabetes rather than the casual factors for progression to diabetes.

Second, the blood samples were taken at random (not all in a non-fasting state) and so the findings were subjected to greater variability. This should be discussed in the limitations.

Third, the strength of NMR in metabolic profiling compared to other techniques (i.e., mass spectrometry [MS], another commonly used metabolic profiling method) could be added in the Discussion section.

Fourth, the applied platform focuses mostly on lipid species which may be a limitation as well.

Fifth, it is a very large group with pre-diabetes, but the results only apply to prediabetes and not to the general population. This should be clear, although the authors have also validated the predictive value of these metabolites in the general population.

Паттерн наблюдатель кароче: мы как клинет сервера, только ждём новых данные, которые сокет нам скинет.

то есть, мы как сервер подписываемся на сокет, и когда у него появляются новые данные, он нас об этом уведомляет.

构成

封装

跨地址空间消息传递

不相交的

容量

存

明确的

令人印象深刻的

健全的、合理的

...的两倍

只有

gender?

plus innovation (both technological and managerial)

why? Just briefly explain (2-3 sentences). I like how you position the calico workers within the historiographical debates. Will work well in the introduction to the thesis

is that something you also aim to do?

https://web.archive.org/web/20240715081505/https://www.tagesschau.de/wirtschaft/fluege-klimaneutral-100.html

К чему относится? Если к API, то который, наверное, а если к функциям, то которые...

Обострились у меня проблемы со зрением, похоже... На скрине значок параграфа, до и после: §attr§ Заменю в тексте, но стоит проверить.

Yep. Same here. I’ve been working with my calendar as a log for so long I forget it’s not the way most people use calendars. Nice to read someone else who’s in the same camp.

eLife assessment

This study provides important findings regarding the stability over time of the response properties of neurons in the auditory cortex, including their nonlinear sensitivity to sound context. The data obtained from chronic recordings combined with nonlinear stimulus-response estimation provide convincing evidence that auditory cortical representations are stable over a period of days to weeks. While this study should be of widespread interest to sensory neuroscientists, the paper would be strengthened by a more thorough assessment and discussion of the effects of context and of the stability of the responses, as well as by the inclusion of more information about the location and types of neurons that were sampled.

Reviewer #1 (Public Review):

Summary:

Recent studies have used optical or electrophysiological techniques to chronically measure receptive field properties of sensory cortical neurons over long time periods, i.e. days to weeks, to ask whether sensory receptive fields are stable properties. Akritas et al expand on prior studies by investigating whether nonlinear contextual sensitivity, a property not previously investigated in the context of so-called 'representational drift,' remains stable over days or weeks of recording. They performed chronic tetrode recordings of auditory cortical neurons over at least five recording days while also performing daily measurements of both the linear spectro-temporal receptive field (principal receptive field, PRF) and non-linear 'contextual gain field' (CGF), which captures the neuron's sensitivity to acoustic context. They found that spike waveforms could be reliably matched even when recorded weeks apart. In well-matched units, by comparing the correlation between tuning within one day's session to sessions across days, both PRFs and CGFs showed remarkable stability over time. This was the case even when recordings were performed over weeks. Meanwhile, behavioral and brain state, measured with locomotion and pupil diameter, respectively, resulted in small but significant shifts in the ability of the PRF/CGF model to predict fluctuations in the neuronal response over time.

Strengths:

The study addresses a fundamental question, which is whether the neural underpinnings of sensory perception, which encompasses both sensory events and their context, are stable across relevant timescales over which our experiences must be stable, despite biological turnover. Although two-photon calcium imaging is ideal for identifying neurons stably regardless of their activity levels and tuning, it lacks temporal precision and is therefore limited in its ability to capture the complexity of sensory responses. Akritas et al performed painstaking chronic extracellular recordings in the auditory cortex with the temporal resolution to investigate complex receptive field properties, such as neural sensitivities to acoustic context. Prior studies, particularly in the auditory cortex, focused on basic tuning properties or sensory responsivity, but Akritas et al expand on this work by showing that even the nonlinear, contextual elements of sensory neurons' responses can remain stable, providing a mechanism for the stability of our complex perception. This work is both novel and broadly applicable to those investigating cortical stability across sensory modalities.

Weaknesses:

Apart from some aspects such as single-unit versus multi-unit, the study largely treats their dataset as a monolith rather than showing how factors such as firing rate, depth, and cell type could define more or less stable subpopulations. It is likely that their methodology did not enable an even sampling over these qualities, and the authors should discuss these biases to put their findings more in context with related studies.

Reviewer #2 (Public Review):

Summary:

This study explores the fundamental neuroscience question of the stability of neuronal representation. The concept of 'representational-drift' has been put forward after observations made using 2-photon imaging of neuronal activity over many days revealed that neurons contribute in a time-limited manner to population representation of stimuli or experiences. The authors contribute to the still contested concept of 'drifts' by measuring representation across days using electrophysiology and thus with sufficient temporal resolution to characterize the receptive fields of neurons in timescales relevant to the stimuli used. The data obtained from chronic recordings over days combined with nonlinear stimulus-response estimation allows the authors to conclude that both the spectrotemporal receptive fields as well as contextual gain fields dependent on combination sensitivity to complex stimuli were stable over time. This suggests that when a neuron is responsive to experimental parameters across long periods of time (days), its sensory receptive field is stable.

Strengths:

The strength of this study lies in the capacity to draw novel conclusions on auditory cortex representation based on the experimentally difficult combination of stable recordings of neuronal activity, behavior, and pupil over days and state-of-the-art analysis of receptive fields.

Weaknesses:

It would have been desirable, but too ambitious in the current setting, to be able to assess what proportion if any of the neurons drop out or in to draw a closer parallel with the 2-photon studies.

Reviewer #3 (Public Review):

Summary:

In their study on "Nonlinear sensitivity to acoustic context is a stable feature of neuronal responses to complex sounds in auditory cortex of awake mice", Akritas et al. investigate the stability of the response properties of neurons in the auditory cortex of mice. They estimate a model with restricted non-linearities for individual neurons and compare the model properties between recordings on the same day and subsequent days. They find that both the linear and nonlinear components of the model stay rather constant over this period and conclude that on the level of the tuning properties, there is no evidence for representational drift on this time scale.

Strengths:

- The study has a clear analytical approach that goes beyond linear models and investigates this in a rigorous way, in particular comparing across-day variability to within-day variability.<br /> - The use of tetrodes is a rather reliable way in electrophysiological recordings to assess neuron identity over multiple days.<br /> - The comparison with pupil and motion activity was useful and insightful.<br /> - The presentation of the study is very logical and pretty much flawless on the writing level.

Weaknesses:

- The stability results across cells show a good amount of variability, which is only partially addressed.<br /> - In particular, no attempt is made to localize the cells in space, in order to check whether these differences could be layer or area-dependent.<br /> - The full context model also includes the possibility to estimate the input non-linearity, which was not done here, but could have been insightful.

Author response:

Reviewer #1 (Public Review):

Summary:

Recent studies have used optical or electrophysiological techniques to chronically measure receptive field properties of sensory cortical neurons over long time periods, i.e. days to weeks, to ask whether sensory receptive fields are stable properties. Akritas et al expand on prior studies by investigating whether nonlinear contextual sensitivity, a property not previously investigated in the context of so-called 'representational drift,' remains stable over days or weeks of recording. They performed chronic tetrode recordings of auditory cortical neurons over at least five recording days while also performing daily measurements of both the linear spectro-temporal receptive field (principal receptive field, PRF) and non-linear 'contextual gain field' (CGF), which captures the neuron's sensitivity to acoustic context. They found that spike waveforms could be reliably matched even when recorded weeks apart. In well-matched units, by comparing the correlation between tuning within one day's session to sessions across days, both PRFs and CGFs showed remarkable stability over time. This was the case even when recordings were performed over weeks. Meanwhile, behavioral and brain state, measured with locomotion and pupil diameter, respectively, resulted in small but significant shifts in the ability of the PRF/CGF model to predict fluctuations in the neuronal response over time.

Strengths:

The study addresses a fundamental question, which is whether the neural underpinnings of sensory perception, which encompasses both sensory events and their context, are stable across relevant timescales over which our experiences must be stable, despite biological turnover. Although two-photon calcium imaging is ideal for identifying neurons stably regardless of their activity levels and tuning, it lacks temporal precision and is therefore limited in its ability to capture the complexity of sensory responses. Akritas et al performed painstaking chronic extracellular recordings in the auditory cortex with the temporal resolution to investigate complex receptive field properties, such as neural sensitivities to acoustic context. Prior studies, particularly in the auditory cortex, focused on basic tuning properties or sensory responsivity, but Akritas et al expand on this work by showing that even the nonlinear, contextual elements of sensory neurons' responses can remain stable, providing a mechanism for the stability of our complex perception. This work is both novel and broadly applicable to those investigating cortical stability across sensory modalities.

Weaknesses:

Apart from some aspects such as single-unit versus multi-unit, the study largely treats their dataset as a monolith rather than showing how factors such as firing rate, depth, and cell type could define more or less stable subpopulations. It is likely that their methodology did not enable an even sampling over these qualities, and the authors should discuss these biases to put their findings more in context with related studies.

We did, in fact, investigate whether firing rate and other physiological response properties of units might differentiate subpopulations with different stability. This analysis is shown in Figure 7B-D. There was no apparent relationship between stability of nonlinear contextual gain fields and physiological properties such as mean evoked firing rate, signal-to-noise ratio for evoked firing, or predictive power of the context model (a measure of model goodness-of-fit).

The reviewer is correct, however, that we did not address possible differences between units recorded at different cortical depths or of different cell types, due to limitations of our methodology and sampling.

Reviewer #2 (Public Review):

Summary:

This study explores the fundamental neuroscience question of the stability of neuronal representation. The concept of 'representational-drift' has been put forward after observations made using 2-photon imaging of neuronal activity over many days revealed that neurons contribute in a time-limited manner to population representation of stimuli or experiences. The authors contribute to the still contested concept of 'drifts' by measuring representation across days using electrophysiology and thus with sufficient temporal resolution to characterize the receptive fields of neurons in timescales relevant to the stimuli used. The data obtained from chronic recordings over days combined with nonlinear stimulus-response estimation allows the authors to conclude that both the spectrotemporal receptive fields as well as contextual gain fields dependent on combination sensitivity to complex stimuli were stable over time. This suggests that when a neuron is responsive to experimental parameters across long periods of time (days), its sensory receptive field is stable.

Strengths:

The strength of this study lies in the capacity to draw novel conclusions on auditory cortex representation based on the experimentally difficult combination of stable recordings of neuronal activity, behavior, and pupil over days and state-of-the-art analysis of receptive fields.

Weaknesses:

It would have been desirable, but too ambitious in the current setting, to be able to assess what proportion if any of the neurons drop out or in to draw a closer parallel with the 2-photon studies.

We certainly agree that this comparison would have been desirable in principle. In practice, however, it was technically infeasible and would have been likely to produce misleading results. Our criteria for spike waveform matching across days were extremely conservative, to minimise the potential for a false positive match (which could artifactually decrease apparent stability of unit responses). Therefore, we were likely to have missed some neurons that did in fact remain active over days, due to small changes in extracellular waveform or just noise (which could artifactually decrease apparent stability of population representations). Two-photon imaging is more appropriate for analysing population stability, because cell identity is determined by spatial location. However, as we mention in the paper, electrophysiology is more appropriate for analysing receptive-field stability, because the temporal resolution is sufficient to resolve structure at the millisecond timescales relevant to auditory perception.

Reviewer #3 (Public Review):

Summary:

In their study on "Nonlinear sensitivity to acoustic context is a stable feature of neuronal responses to complex sounds in auditory cortex of awake mice", Akritas et al. investigate the stability of the response properties of neurons in the auditory cortex of mice. They estimate a model with restricted non-linearities for individual neurons and compare the model properties between recordings on the same day and subsequent days. They find that both the linear and nonlinear components of the model stay rather constant over this period and conclude that on the level of the tuning properties, there is no evidence for representational drift on this time scale.

Strengths:

- The study has a clear analytical approach that goes beyond linear models and investigates this in a rigorous way, in particular comparing across-day variability to within-day variability.

- The use of tetrodes is a rather reliable way in electrophysiological recordings to assess neuron identity over multiple days.

- The comparison with pupil and motion activity was useful and insightful.

- The presentation of the study is very logical and pretty much flawless on the writing level.

Weaknesses:

- The stability results across cells show a good amount of variability, which is only partially addressed.

- In particular, no attempt is made to localize the cells in space, in order to check whether these differences could be layer or area-dependent.

- The full context model also includes the possibility to estimate the input non-linearity, which was not done here, but could have been insightful.

We agree with these comments and acknowledge these limitations, which arise from technological constraints. In particular, the tangential trajectory of our chronic tetrode implant, used to maximise stability of chronic recordings, limited our ability to sample cells from different cortical layers/areas and to explore how these factors might relate to variability in stability across units. Estimating input nonlinearities would have been valuable but also would have increased the number of parameters in the model and the data required to obtain reliable, predictive model fits.

eLife assessment

This study shows that a peptide called galanin can decrease or increase seizure activity in experimental models of seizures depending on the way seizures are induced (genetic vs. pharmacological). The authors use zebrafish and several methods to address the effects of galanin. The study will be useful to researchers who use zebrafish as experimental animals and who are interested in how the peptides in the brain (neuropeptides) regulate seizures. However, the strength of evidence was considered incomplete at the present time due to several limitations of the results.

Reviewer #1 (Public Review):

Summary:

In this study, the authors explored how galanin affects whole-brain activity in larval zebrafish using wide-field Ca2+ imaging, genetic modifications, and drugs that increase brain activity. The authors conclude that galanin has a sedative effect on the brain under normal conditions and during seizures, mainly through the galanin receptor 1a (galr1a). However, acute "stressors(?)" like pentylenetetrazole (PTZ) reduce galanin's effects, leading to increased brain activity and more seizures. The authors claim that galanin can reduce seizure severity while increasing seizure occurrence, speculated to occur through different receptor subtypes. This study confirms galanin's complex role in brain activity, supporting its potential impact on epilepsy.

Strengths:

The overall strength of the study lies primarily in its methodological approach using whole-brain Calcium imaging facilitated by the transparency of zebrafish larvae. Additionally, the use of transgenic zebrafish models is an advantage, as it enables genetic manipulations to investigate specific aspects of galanin signaling. This combination of advanced imaging and genetic tools allows for addressing galanin's role in regulating brain activity.

Weaknesses:

The weaknesses of the study also stem from the methodological approach, particularly the use of whole-brain Calcium imaging as a measure of brain activity. While epilepsy and seizures involve network interactions, they typically do not originate across the entire brain simultaneously. Seizures often begin in specific regions or even within specific populations of neurons within those regions. Therefore, a whole-brain approach, especially with Calcium imaging with inherited limitations, may not fully capture the localized nature of seizure initiation and propagation, potentially limiting the understanding of Galanin's role in epilepsy.

Furthermore, Galanin's effects may vary across different brain areas, likely influenced by the predominant receptor types expressed in those regions. Additionally, the use of PTZ as a "stressor" is questionable since PTZ induces seizures rather than conventional stress. Referring to seizures induced by PTZ as "stress" might be a misinterpretation intended to fit the proposed model of stress regulation by receptors other than Galanin receptor 1 (GalR1).

The description of the EAAT2 mutants is missing crucial details. EAAT2 plays a significant role in the uptake of glutamate from the synaptic cleft, thereby regulating excitatory neurotransmission and preventing excitotoxicity. Authors suggest that in EAAT2 knockout (KO) mice galanin expression is upregulated 15-fold compared to wild-type (WT) mice, which could be interpreted as galanin playing a role in the hypoactivity observed in these animals.

However, the study does not explore the misregulation of other genes that could be contributing to the observed phenotype. For instance, if AMPA receptors are significantly downregulated, or if there are alterations in other genes critical for brain activity, these changes could be more important than the upregulation of galanin. The lack of wider gene expression analysis leaves open the possibility that the observed hypoactivity could be due to factors other than, or in addition to, galanin upregulation.

Moreover, the observation that in double KO mice for both EAAT2 and galanin, there was little difference in seizure susceptibility compared to EAAT2 KO mice alone further supports the idea that galanin upregulation might not be the reason for the observed phenotype. This indicates that other regulatory mechanisms or gene expressions might be playing a more pivotal role in the manifestation of hypoactivity in EAAT2 mutants.

These methodological shortcomings and conceptual inconsistencies undermine the perceived strengths of the study, and hinders understanding of Galanin's role in epilepsy and stress regulation.

Reviewer #2 (Public Review):

Summary:

This study is an investigation of galanin and galanin receptor signaling on whole-brain activity in the context of recurrent seizure activity or under homeostatic basal conditions. The authors primarily use calcium imaging to observe whole-brain neuronal activity accompanied by galanin qPCR to determine how manipulations of galanin or the galr1a receptor affect the activity of the whole-brain under non-ictal or seizure event conditions. The authors' Eaat2a-/- model (introduced in their Glia 2022 paper, PMID 34716961) that shows recurrent seizure activity alongside suppression of neuronal activity and locomotion in the time periods lacking seizures is used in this paper in comparison to the well-known pentylenetetrazole (PTZ) pharmacological model of epilepsy in zebrafish. Given the literature cited in their Introduction, the authors reasonably hypothesize that galanin will exert a net inhibitory effect on brain activity in models of epilepsy and at homeostatic baseline, but were surprised to find that this hypothesis was only moderately supported in their Eaat2a-/- model. In contrast, under PTZ challenge, fish with galanin overexpression showed increased seizure number and reduced duration while fish with galanin KO showed reduced seizure number and increased duration. These results would have been greatly enriched by the inclusion of behavioral analyses of seizure activity and locomotion (similar to the authors' 2022 Glia paper and/or PMIDs 15730879, 24002024). In addition, the authors have not accounted for sex as a biological variable, though they did note that sex sorting zebrafish larvae precludes sex selection at the younger ages used. It would be helpful to include smaller experiments taken from pilot experiments in older, sex-balanced groups of the relevant zebrafish to increase confidence in the findings' robustness across sexes. A possible major caveat is that all of the various genetic manipulations are non-conditional as performed, meaning that developmental impacts of galanin overexpression or galanin or galr1a knockout on the observed results have not been controlled for and may have had a confounding influence on the authors' findings. Overall, this study is important and solid (yet limited), and carries clear value for understanding the multifaceted functions that neuronal galanin can have under homeostatic and disease conditions.

Strengths:

- The authors convincingly show that galanin is upregulated across multiple contexts that feature seizure activity or hyperexcitability in zebrafish, and appears to reduce neuronal activity overall, with key identified exceptions (PTZ model).

- The authors use both genetic and pharmacological models to answer their question, and through this diverse approach, find serendipitous results that suggest novel underexplored functions of galanin and its receptors in basal and disease conditions. Their question is well-informed by the cited literature, though the authors should cite and consider their findings in the context of Mazarati et al., 1998 (PMID:982276). The authors' Discussion places their findings in context, allowing for multiple interpretations and suggesting some convincing explanations.

- Sample sizes are robust and the methods used are well-characterized, with a few exceptions (as the paper is currently written).

- Use of a glutamatergic signaling-based genetic model of epilepsy (Eaat2a-/-) is likely the most appropriate selection to test how galanin signaling can alter seizure activity, as galanin is known to reduce glutamatergic release as an inhibitory mechanism in rodent hippocampal neurons via GalR1a (alongside GIRK activation effects). Given that PTZ instead acts through GABAergic signaling pathways, it is reasonable and useful to note that their glutamate-based genetic model showed different effects than did their GABAergic-based model of seizure activity.

Weaknesses:

- The authors do not include behavioral assessments of seizure or locomotor activity that would be expected in this paper given their characterizations of their Eaat2a-/- model in the Glia 2022 paper that showed these behavioral data for this zebrafish model. These data would inform the reader of the behavioral phenotypes to expect under the various conditions and would likely further support the authors' findings if obtained and reported.

- No assessment of sex as a biological variable is included, though it is understood that these specific studied ages of the larvae may preclude sex sorting for experimental balancing as stated by the authors.

- The reported results may have been influenced by the loss or overexpression of galanin or loss of galr1a during developmental stages. The authors did attempt to use the hsp70l system to overexpress galanin, but noted that the heat shock induction step led to reduced brain activity on its own (Supplementary Figure 1). Their hsp70l:gal model shows galanin overexpression anyways (8x fold) regardless of heat induction, so this model is still useful as a way to overexpress galanin, but it should be noted that this galanin overexpression is not restricted to post-developmental timepoints and is present during development.

Reviewer #3 (Public Review):

Summary:

The neuropeptide galanin is primarily expressed in the hypothalamus and has been shown to play critical roles in homeostatic functions such as arousal, sleep, stress, and brain disorders such as epilepsy. Previous work in rodents using galanin analogs and receptor-specific knockout has provided convincing evidence for the anti-convulsant effects of galanin.

In the present study, the authors sought to determine the relationship between galanin expression and whole-brain activity. The authors took advantage of the transparent nature of larval zebrafish to perform whole-brain neural activity measurements via widefield calcium imaging. Two models of seizures were used (eaat2a-/- and pentylenetetrazol; PTZ). In the eaat2a-/- model, spontaneous seizures occur and the authors found that galanin transcript levels were significantly increased and associated with a reduced frequency of calcium events. Similarly, two hours after PTZ galanin transcript levels roughly doubled and the frequency and amplitude of calcium events were reduced. The authors also used a heat shock protein line (hsp70I:gal) where galanin transcript levels are induced by activation of heat shock protein, but this line also shows higher basal transcript levels of galanin. Again, the higher level of galanin in hsp70I:gal larval zebrafish resulted in a reduction of calcium events and a reduction in the amplitude of events. In contrast, galanin knockout (gal-/-) increased calcium activity, indicated by an increased number of calcium events, but a reduction in amplitude and duration. Knockout of the galanin receptor subtype galr1a via crispants also increased the frequency of calcium events.

In subsequent experiments in eaat2a-/- mutants were crossed with hsp70I:gal or gal-/- to increase or decrease galanin expression, respectively. These experiments showed modest effects, with eaat2a-/- x gal-/- knockouts showing an increased normalized area under the curve and seizure amplitude.

Lastly, the authors attempted to study the relationship between galanin and brain activity during a PTZ challenge. The hsp70I:gal larva showed an increased number of seizures and reduced seizure duration during PTZ. In contrast, gal-/- mutants showed an increased normalized area under the curve and a stark reduction in the number of detected seizures, a reduction in seizure amplitude, but an increase in seizure duration. The authors then ruled out the role of Galr1a in modulating this effect during PTZ, since the number of seizures was unaffected, whereas the amplitude and duration of seizures were increased.

Strengths:

(1) The gain- and loss-of function galanin manipulations provided convincing evidence that galanin influences brain activity (via calcium imaging) during interictal and/or seizure-free periods. In particular, the relationship between galanin transcript levels and brain activity in Figures 1 & 2 was convincing.

(2) The authors use two models of epilepsy (eaat2a-/- and PTZ).

(3) Focus on the galanin receptor subtype galr1a provided good evidence for the important role of this receptor in controlling brain activity during interictal and/or seizure-free periods.

Weaknesses:

(1) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the manuscript currently lacks mechanistic insight in the role of galanin during seizure-like activity induced by PTZ.

(2) Calcium imaging is the primary data for the paper, but there are no representative time-series images or movies of GCaMP signal in the various mutants used.

(3) For Figure 3, the authors suggest that hsp70I:gal x eaat2a-/-mutants would further increase galanin transcript levels, which were hypothesized to further reduce brain activity. However, the authors failed to measure galanin transcript levels in this cross to show that galanin is actually increased more than the eaat2a-/- mutant or the hsp70I:gal mutant alone.

(4) Similarly, transcript levels of galanin are not provided in Figure 2 for Gal-/- mutants and galr1a KOs. Transcript levels would help validate the knockout and any potential compensatory effects of subtype-specific knockout.

(5) The authors very heavily rely on calcium imaging of different mutant lines. Additional methods could strengthen the data, translational relevance, and interpretation (e.g., acute pharmacology using galanin agonists or antagonists, brain or cell recordings, biochemistry, etc).

Author response:

Reviewer #1 (Public Review):

Summary:

In this study, the authors explored how galanin affects whole-brain activity in larval zebrafish using wide-field Ca2+ imaging, genetic modifications, and drugs that increase brain activity. The authors conclude that galanin has a sedative effect on the brain under normal conditions and during seizures, mainly through the galanin receptor 1a (galr1a). However, acute "stressors(?)" like pentylenetetrazole (PTZ) reduce galanin's effects, leading to increased brain activity and more seizures. The authors claim that galanin can reduce seizure severity while increasing seizure occurrence, speculated to occur through different receptor subtypes. This study confirms galanin's complex role in brain activity, supporting its potential impact on epilepsy.

Strengths:

The overall strength of the study lies primarily in its methodological approach using whole-brain Calcium imaging facilitated by the transparency of zebrafish larvae. Additionally, the use of transgenic zebrafish models is an advantage, as it enables genetic manipulations to investigate specific aspects of galanin signaling. This combination of advanced imaging and genetic tools allows for addressing galanin's role in regulating brain activity.

Weaknesses:

The weaknesses of the study also stem from the methodological approach, particularly the use of whole-brain Calcium imaging as a measure of brain activity. While epilepsy and seizures involve network interactions, they typically do not originate across the entire brain simultaneously. Seizures often begin in specific regions or even within specific populations of neurons within those regions. Therefore, a whole-brain approach, especially with Calcium imaging with inherited limitations, may not fully capture the localized nature of seizure initiation and propagation, potentially limiting the understanding of Galanin's role in epilepsy.

Furthermore, Galanin's effects may vary across different brain areas, likely influenced by the predominant receptor types expressed in those regions. Additionally, the use of PTZ as a "stressor" is questionable since PTZ induces seizures rather than conventional stress. Referring to seizures induced by PTZ as "stress" might be a misinterpretation intended to fit the proposed model of stress regulation by receptors other than Galanin receptor 1 (GalR1).

The description of the EAAT2 mutants is missing crucial details. EAAT2 plays a significant role in the uptake of glutamate from the synaptic cleft, thereby regulating excitatory neurotransmission and preventing excitotoxicity. Authors suggest that in EAAT2 knockout (KO) mice galanin expression is upregulated 15-fold compared to wild-type (WT) mice, which could be interpreted as galanin playing a role in the hypoactivity observed in these animals.

Indeed, our observation of the unexpected hypoactivity in EAAT2a mutants, described in our description of this mutant (Hotz et al., 2022), prompted us to initiate this study formulating the hypothesis that the observed upregulation of galanin is a neuroprotective response to epilepsy.

However, the study does not explore the misregulation of other genes that could be contributing to the observed phenotype. For instance, if AMPA receptors are significantly downregulated, or if there are alterations in other genes critical for brain activity, these changes could be more important than the upregulation of galanin. The lack of wider gene expression analysis leaves open the possibility that the observed hypoactivity could be due to factors other than, or in addition to, galanin upregulation.

We have performed a transcriptome analysis that we are still evaluation. We can already state that AMPA receptor genes are not significantly altered in the mutant.

Moreover, the observation that in double KO mice for both EAAT2 and galanin, there was little difference in seizure susceptibility compared to EAAT2 KO mice alone further supports the idea that galanin upregulation might not be the reason for the observed phenotype. This indicates that other regulatory mechanisms or gene expressions might be playing a more pivotal role in the manifestation of hypoactivity in EAAT2 mutants.

We agree that upregulation of galanin transcripts is at best one of a suite of regulatory mechanisms that lead to hypoactivity in EAAT2 zebrafish mutants.

These methodological shortcomings and conceptual inconsistencies undermine the perceived strengths of the study, and hinders understanding of Galanin's role in epilepsy and stress regulation.

Reviewer #2 (Public Review):

Summary:

This study is an investigation of galanin and galanin receptor signaling on whole-brain activity in the context of recurrent seizure activity or under homeostatic basal conditions. The authors primarily use calcium imaging to observe whole-brain neuronal activity accompanied by galanin qPCR to determine how manipulations of galanin or the galr1a receptor affect the activity of the whole-brain under non-ictal or seizure event conditions. The authors' Eaat2a-/- model (introduced in their Glia 2022 paper, PMID 34716961) that shows recurrent seizure activity alongside suppression of neuronal activity and locomotion in the time periods lacking seizures is used in this paper in comparison to the well-known pentylenetetrazole (PTZ) pharmacological model of epilepsy in zebrafish. Given the literature cited in their Introduction, the authors reasonably hypothesize that galanin will exert a net inhibitory effect on brain activity in models of epilepsy and at homeostatic baseline, but were surprised to find that this hypothesis was only moderately supported in their Eaat2a-/- model. In contrast, under PTZ challenge, fish with galanin overexpression showed increased seizure number and reduced duration while fish with galanin KO showed reduced seizure number and increased duration. These results would have been greatly enriched by the inclusion of behavioral analyses of seizure activity and locomotion (similar to the authors' 2022 Glia paper and/or PMIDs 15730879, 24002024). In addition, the authors have not accounted for sex as a biological variable, though they did note that sex sorting zebrafish larvae precludes sex selection at the younger ages used. It would be helpful to include smaller experiments taken from pilot experiments in older, sex-balanced groups of the relevant zebrafish to increase confidence in the findings' robustness across sexes. A possible major caveat is that all of the various genetic manipulations are non-conditional as performed, meaning that developmental impacts of galanin overexpression or galanin or galr1a knockout on the observed results have not been controlled for and may have had a confounding influence on the authors' findings. Overall, this study is important and solid (yet limited), and carries clear value for understanding the multifaceted functions that neuronal galanin can have under homeostatic and disease conditions.

Strengths:

- The authors convincingly show that galanin is upregulated across multiple contexts that feature seizure activity or hyperexcitability in zebrafish, and appears to reduce neuronal activity overall, with key identified exceptions (PTZ model).

- The authors use both genetic and pharmacological models to answer their question, and through this diverse approach, find serendipitous results that suggest novel underexplored functions of galanin and its receptors in basal and disease conditions. Their question is well-informed by the cited literature, though the authors should cite and consider their findings in the context of Mazarati et al., 1998 (PMID:982276). The authors' Discussion places their findings in context, allowing for multiple interpretations and suggesting some convincing explanations.

- Sample sizes are robust and the methods used are well-characterized, with a few exceptions (as the paper is currently written).

- Use of a glutamatergic signaling-based genetic model of epilepsy (Eaat2a-/-) is likely the most appropriate selection to test how galanin signaling can alter seizure activity, as galanin is known to reduce glutamatergic release as an inhibitory mechanism in rodent hippocampal neurons via GalR1a (alongside GIRK activation effects). Given that PTZ instead acts through GABAergic signaling pathways, it is reasonable and useful to note that their glutamate-based genetic model showed different effects than did their GABAergic-based model of seizure activity.

Weaknesses:

- The authors do not include behavioral assessments of seizure or locomotor activity that would be expected in this paper given their characterizations of their Eaat2a-/- model in the Glia 2022 paper that showed these behavioral data for this zebrafish model. These data would inform the reader of the behavioral phenotypes to expect under the various conditions and would likely further support the authors' findings if obtained and reported.

We agree that a thorough behavioral assessment would have strengthened the study, but we deemed it outside of the scope of this study.

- No assessment of sex as a biological variable is included, though it is understood that these specific studied ages of the larvae may preclude sex sorting for experimental balancing as stated by the authors.

The study was done on larval zebrafish (5 days post fertilization). The first signs of sexual differentiation become apparent at about 17 days post fertilization (reviewed in Ye and Chen, 2020). Hence sex is no biological variable at the stage studied. 

- The reported results may have been influenced by the loss or overexpression of galanin or loss of galr1a during developmental stages. The authors did attempt to use the hsp70l system to overexpress galanin, but noted that the heat shock induction step led to reduced brain activity on its own (Supplementary Figure 1). Their hsp70l:gal model shows galanin overexpression anyways (8x fold) regardless of heat induction, so this model is still useful as a way to overexpress galanin, but it should be noted that this galanin overexpression is not restricted to post-developmental timepoints and is present during development.

The developmental perspective is an important point to consider. Due to the rapid development of the zebrafish it is not trivial to untangle this. In the zebrafish we first observe epileptic seizures as early as 3 days post fertilization (dpf), where the brain is clearly not well developed yet (e.g. behavioral response to light are still minimal). Even the 5 dpf stage, where most of our experiments have been conducted, cannot by far not be considered post-development.  

Reviewer #3 (Public Review):

Summary:

The neuropeptide galanin is primarily expressed in the hypothalamus and has been shown to play critical roles in homeostatic functions such as arousal, sleep, stress, and brain disorders such as epilepsy. Previous work in rodents using galanin analogs and receptor-specific knockout has provided convincing evidence for the anti-convulsant effects of galanin.

In the present study, the authors sought to determine the relationship between galanin expression and whole-brain activity. The authors took advantage of the transparent nature of larval zebrafish to perform whole-brain neural activity measurements via widefield calcium imaging. Two models of seizures were used (eaat2a-/- and pentylenetetrazol; PTZ). In the eaat2a-/- model, spontaneous seizures occur and the authors found that galanin transcript levels were significantly increased and associated with a reduced frequency of calcium events. Similarly, two hours after PTZ galanin transcript levels roughly doubled and the frequency and amplitude of calcium events were reduced. The authors also used a heat shock protein line (hsp70I:gal) where galanin transcript levels are induced by activation of heat shock protein, but this line also shows higher basal transcript levels of galanin. Again, the higher level of galanin in hsp70I:gal larval zebrafish resulted in a reduction of calcium events and a reduction in the amplitude of events. In contrast, galanin knockout (gal-/-) increased calcium activity, indicated by an increased number of calcium events, but a reduction in amplitude and duration. Knockout of the galanin receptor subtype galr1a via crispants also increased the frequency of calcium events.

In subsequent experiments in eaat2a-/- mutants were crossed with hsp70I:gal or gal-/- to increase or decrease galanin expression, respectively. These experiments showed modest effects, with eaat2a-/- x gal-/- knockouts showing an increased normalized area under the curve and seizure amplitude.

Lastly, the authors attempted to study the relationship between galanin and brain activity during a PTZ challenge. The hsp70I:gal larva showed an increased number of seizures and reduced seizure duration during PTZ. In contrast, gal-/- mutants showed an increased normalized area under the curve and a stark reduction in the number of detected seizures, a reduction in seizure amplitude, but an increase in seizure duration. The authors then ruled out the role of Galr1a in modulating this effect during PTZ, since the number of seizures was unaffected, whereas the amplitude and duration of seizures were increased.

Strengths:

(1) The gain- and loss-of function galanin manipulations provided convincing evidence that galanin influences brain activity (via calcium imaging) during interictal and/or seizure-free periods. In particular, the relationship between galanin transcript levels and brain activity in Figures 1 & 2 was convincing.

(2) The authors use two models of epilepsy (eaat2a-/- and PTZ).

(3) Focus on the galanin receptor subtype galr1a provided good evidence for the important role of this receptor in controlling brain activity during interictal and/or seizure-free periods.

Weaknesses:

(1) Although the relationship between galanin and brain activity during interictal or seizure-free periods was clear, the manuscript currently lacks mechanistic insight in the role of galanin during seizure-like activity induced by PTZ.

We completely agree and concede that this study constitutes only a first attempt to understand the (at least for us) perplexing complexity of galanin function on the brain.

(2) Calcium imaging is the primary data for the paper, but there are no representative time-series images or movies of GCaMP signal in the various mutants used.

We are in the process of preparing some time series images and will include them in the next revision.

(3) For Figure 3, the authors suggest that hsp70I:gal x eaat2a-/-mutants would further increase galanin transcript levels, which were hypothesized to further reduce brain activity. However, the authors failed to measure galanin transcript levels in this cross to show that galanin is actually increased more than the eaat2a-/- mutant or the hsp70I:gal mutant alone.

This is an excellent suggestion. We will perform the necessary qPCR experiments and will include the data in the next revision.

(4) Similarly, transcript levels of galanin are not provided in Figure 2 for Gal-/- mutants and galr1a KOs. Transcript levels would help validate the knockout and any potential compensatory effects of subtype-specific knockout.

(5) The authors very heavily rely on calcium imaging of different mutant lines. Additional methods could strengthen the data, translational relevance, and interpretation (e.g., acute pharmacology using galanin agonists or antagonists, brain or cell recordings, biochemistry, etc).

Again, we agree and concede that a number of additional approaches are needed to get more insight into the complex role of galanin in regulation overall brain activity. These include, among others, also behavioral, multiple single cell recordings and pharmacological interventions.

У меня, наверное, вопрос был скорее, память принадлежит другой учетной записи или процесс принадлежит.

Это точно этот файл? Не connection.xml?

Перенос смешной.

Ein neuer Bericht von Oxfam analysiert die unzureichenden Klimaanpassungsmaßnahmen in Frankreich und insbesondere deren soziale Aspekte. Der Bericht kommt zu dem Ergebnis, dass mehr als die Hälfte der rechtlich fixierten Menschenrechte durch mangelnde Vorbereitung und Unterfinanzierung bedroht ist. Interview mit Cécile Duflot, der Leiterin von Oxfam France. https://www.liberation.fr/environment/climat/rechauffement-climatique-plus-de-la-moitie-des-droits-humains-en-france-est-menacee-par-limprovisation-des-pouvoirs-publics-20240715_2FTYYCVZLFANRP6UL3VTXEHI6M/

防止你对字典做的修改被覆盖。

eLife assessment

The work introduces a valuable new method for depleting the ribosomal RNA from bacterial single-cell RNA sequencing libraries and shows that this method is applicable to studying the heterogeneity in microbial biofilms. The evidence for a small subpopulation of cells at the bottom of the biofilm which upregulates PdeI expression is solid. However, more investigation into the unresolved functional relationship between PdeI and c-di-GMP levels with the help of other genes co-expressed in the same cluster would have made the conclusions more significant.

Reviewer #1 (Public Review):

Summary:

In this manuscript, Yan and colleagues introduce a modification to the previously published PETRI-seq bacterial single-cell protocol to include a ribosomal depletion step based on a DNA probe set that selectively hybridizes with ribosome-derived (rRNA) cDNA fragments. They show that their modification of the PETRI-seq protocol increases the fraction of informative non-rRNA reads from ~4-10% to 54-92%. The authors apply their protocol to investigating heterogeneity in a biofilm model of E. coli, and convincingly show how their technology can detect minority subpopulations within a complex community.

Strengths:

The method the authors propose is a straightforward and inexpensive modification of an established split-pool single-cell RNA-seq protocol that greatly increases its utility, and should be of interest to a wide community working in the field of bacterial single-cell RNA-seq.

Weaknesses:

The manuscript is written in a very compressed style and many technical details of the evaluations conducted are unclear and processed data has not been made available for evaluation, limiting the ability of the reader to independently judge the merits of the method.

Reviewer #2 (Public Review):

Summary:

This work introduces a new method of depleting the ribosomal reads from the single-cell RNA sequencing library prepared with one of the prokaryotic scRNA-seq techniques, PETRI-seq. The advance is very useful since it allows broader access to the technology by lowering the cost of sequencing. It also allows more transcript recovery with fewer sequencing reads. The authors demonstrate the utility and performance of the method for three different model species and find a subpopulation of cells in the E.coli biofilm that express a protein, PdeI, which causes elevated c-di-GMP levels. These cells were shown to be in a state that promotes persister formation in response to ampicillin treatment.

Strengths:

The introduced rRNA depletion method is highly efficient, with the depletion for E.coli resulting in over 90% of reads containing mRNA. The method is ready to use with existing PETRI-seq libraries which is a large advantage, given that no other rRNA depletion methods were published for split-pool bacterial scRNA-seq methods. Therefore, the value of the method for the field is high. There is also evidence that a small number of cells at the bottom of a static biofilm express PdeI which is causing the elevated c-di-GMP levels that are associated with persister formation. Given that PdeI is a phosphodiesterase, which is supposed to promote hydrolysis of c-di-GMP, this finding is unexpected.

Weaknesses:

With the descriptions and writing of the manuscript, it is hard to place the findings about the PdeI into existing context (i.e. it is well known that c-di-GMP is involved in biofilm development and is heterogeneously distributed in several species' biofilms; it is also known that E.coli diesterases regulate this second messenger, i.e. https://journals.asm.org/doi/full/10.1128/jb.00604-15).<br /> There is also no explanation for the apparently contradictory upregulation of c-di-GMP in cells expressing higher PdeI levels. Perhaps the examination of the rest of the genes in cluster 2 of the biofilm sample could be useful to explain the observed association.

Energieminister Ed Miliband wird die britische Delegation bei der COP29 leiten. Anders als bei den letzten konservativen Vorgängerregierungen ist damit wieder eine Top-Level Figur federführend. Die neue Labor-Regierung möchte, dass Großbritannien wieder eine führende Rolle in der Klimapolitik übernimmt https://www.theguardian.com/environment/article/2024/jul/15/ed-miliband-lead-uk-negotiations-cop29-climate-summit

eLife assessment

This descriptive study reports the genetic requirements for growth and fitness of multiple clinical strains of a relatively understudied species of mycobacteria, Mycobacterium intracellulare. The findings are valuable however, the study is incomplete as the primary claims related to hypoxia adaptation need additional experimental support and data presentation requires more clarity. The work will be of interest to microbiologists.

Reviewer #1 (Public Review):

Summary:

Tateishi et al. report a Tn-seq-based analysis of genetic requirements for growth and fitness in 8 clinical strains of Mycobacterium intracellulare Mi), and compare the findings with a type strain ATCC13950. The study finds a core set of 131 genes that are essential in all nine strains, and therefore are reasonably argued as potential drug targets. Multiple other genes required for fitness in clinical isolates have been found to be important for hypoxic growth in the type strain.

Strengths:

The study has generated a large volume of Tn-seq datasets of multiple clinical strains of Mi from multiple growth conditions, including from mouse lungs. The dataset can serve as an important resource for future studies on Mi, which despite being clinically significant remains a relatively understudied species of mycobacteria.

Weaknesses:

The paper lacks clarity in data presentation and organization. For example, some of the key data on cfu counts of clinical Mi strains in a mouse model can be presented along with the Tn-seq dataset in Figure 6, the visualization of which can be improved with volcano plots. etc. Improvement in data visualization is perhaps necessary throughout the paper.

The primary claim of the study that the clinical strains are better adapted for hypoxic growth is not well-supported by the data presented in Figure 7.

The title of the paper is misleading as the study doesn't provide any mechanistic aspect of hypoxic adaptation in Mi.

Reviewer #2 (Public Review):

Summary:

In the study titled "Functional genomics reveals the mechanism of hypoxic adaptation in nontuberculous mycobacteria" by Tateishi et al., the authors have used TnSeq to identify the common essential and growth-defect-associated genes that represent the genomic diversity of clinical M. intracellulare strains in comparison to the reference type strain. By estimating the frequency of Tn insertion, the authors speculate that genes involved in gluconeogenesis, the type VII secretion system, and cysteine desulfurase are relatively critical in the clinical MAC-PD strains than in the type strain, both for the extracellular survival and in a mouse lung infection model.

Based on their analysis, the authors proposed to identify the mechanism of hypoxic adaptation in nontuberculous mycobacteria (NTM) which offer promising drug targets in the strains causing clinical Mycobacterium avium-intracellulare complex pulmonary disease (MAC-PD).

Strengths:

A major strength of the manuscript is the performance of the exhaustive set of TnSeq experiments with multiple strains of M. intracellulare during in vitro growth and animal infection.

Weaknesses:

(1) The study suffers from the authors' preconceived bias toward a small subset of genes involved in hypoxic pellicle formation in ATCC13950.

(2) An important set of data with the ATCC13950 reference strain is missing in the mouse infection study. In the absence of this, it is difficult to establish whether the identified genes are critical for infection/intracellular proliferation, specifically in the clinical isolates that are relatively more adapted for hypoxia.

(3) Statistical enrichment analysis of gene sets by GSEA wrongly involves genes required for hypoxic pellicle formation in ATCC13950 together with the gene sets found essential in the clinical MAC-PD strains, to claim that a significant % of genes belong to hypoxia-adaptation pathways. It could be factually incorrect because a majority of these might overlap with those found critical for the in vitro survival of MAC-PD strains (and may not be related to hypoxia).

(4) Validation of mouse infection experiments with individual mutants is missing.

(5) Phenotypes with TnSeq and CRISPRi-based KD exhibit poor correlation with misleading justifications by the authors.

In summary, this study is unable to provide mechanistic insights into why and how different MAC-PD mutant strains exhibit differential survival (in vitro and in animals) and adaptation to hypoxia. It remains to understand why the clinical strains show better adaptation to hypoxia and what is the impact of other stresses on their growth rates.

Reviewer #3 (Public Review):

Summary:

The study by Tateishi et al. utilized TnSeq in nine genetically diverse M. intracellulare strains, identifying 131 common essential and growth-defect-associated genes across those strains, which could serve as potential drug targets. The authors also provided an overview of the differences in gene essentiality required for hypoxic growth between the reference strain and the clinical strains. Furthermore, they validated the universal and accessory/strain-dependent essential genes by knocking down their expression using CRISPRi technique. Overall, this study offers a comprehensive assessment of gene requirements in different clinical strains of M. intracellular.

(1) The rationale for using ATCC13950 versus clinical strains needs to be clarified. The reference strain ATCC13950 was obtained from the abdominal lymph node of a patient around 10 years ago and is therefore considered a clinical strain that has undergone passages in vitro. How many mutations have accumulated during these in vitro passages? Are these mutations significant enough to cause the behavior of ATCC13950 to differ from other recently sampled clinical strains? From the phylogenetic tree, ATCC13950 is located between M018 and M.i.27. Did the authors observe a similarity in gene essentiality between ATCC13950 and its neighbor strains? What is the key feature that separates ATCC13950 from these clinical strains? The authors should provide a strong rationale for how to interpret the results of this comparison in a clinical or biological context.

(2) Regarding the 'nine representative strains of M. intracellulare with diverse genotypes in this study,' how were these nine strains selected? To what extent do they represent the genetic diversity of the M. intracellulare population? A phylogenetic tree illustrating the global genetic diversity of the M. intracellulare population, with these strains marked on it, would be important to demonstrate their genetic representativeness.

(3) The authors observed a considerable amount of differential gene requirements in clinical strains. However, the genetic underpinning underlying the differential requirement of genes in clinical strains was not investigated or discussed. Because M. intracellulare has a huge number of accessory genes, the authors should at least check whether the differential requirement could be explained by the existence of a second copy of functional analogous genes or duplications.

(4) Growth in aerobic and hypoxic conditions: The authors concluded that clinical strains are better adapted to hypoxia, as reflected by their earlier entry into the log phase. They presented the 'Time at midpoint' and 'Growth rate at midpoint.' However, after reviewing the growth curves, I noticed that ATCC13950 had a longer lag phase compared to other strains under hypoxic conditions, and its phylogenetic neighbor M018 also had a longer lag phase. Hence, I do not believe a conclusion can be drawn that clinical strains are better adapted to hypoxia, as this behavior could be specific to a particular clade. It's also possible that the ATCC13950 strain has adapted to aerobic growth. I would suggest that the authors include growth curves in the main figures. The difference in 'Time at midpoint' could be attributed to several factors, and visualizing the growth curves would provide additional context and clarity.

(5) Lack of statistical statement: The authors emphasized the role of pellicle-formation-associated genes in strain-dependent essential and accessory essential genes. Additionally, the authors observed that 10% of the genes required for mouse infection are also required for hypoxic pellicle formation. However, these are merely descriptive statements. There is no enrichment analysis to justify whether pellicle-formation-associated genes are significantly enriched in these groups.

Hi this is impotent

eLife assessment

This important study shows the effect of gut dysbiosis on the colonization of mycobacteria in the lung. The data with comprehensive analysis of gene expression profiles in the lung with dysbiotic mice is compelling and goes beyond the current state of the art. However, the mechanistic insight and the experiments with Mtb infection are incomplete. With those parts strengthened, this paper would be of interest to researchers working on Mtb infection.

Reviewer #1 (Public Review):

Summary:

This work sought to demonstrate that gut microbiota dysbiosis may promote the colonization of mycobacteria, and they tried to prove that Nos2 down-regulation was a key mediator of such gut-lung pathogenesis transition.

Strengths:

They did large-scale analysis of RNAs in lungs to analyze the gene expression of mice upon gut dysbiosis in MS-infected mice. This might help provide an overview of gene pathways and critical genes for lung pathology in gut dysbiosis. This data is somewhat useful and important for the TB field.

Weaknesses:

(1) They did not use wide-type Mtb strain (e.g. H37Rv) to develop mouse TB infection models, and this may lead to the failure of the establishment of TB granuloma and other TB pathology icons.

(2) The usage of in vitro assays based on A542 to examine the regulation function of Nos2 expression on NO and ROS may not be enough. A542 is not the primary Mtb infection target in the lungs.

(3) They did not examine the lung pathology upon gut dysbiosis to examine the true significance of increased colonization of Mtb.

(4) Most of the studies are based on MS-infected mouse models with a lack of clinical significance.

Reviewer #2 (Public Review):

The manuscript entitled "Intestinal microbiome dysbiosis increases Mycobacteria pulmonary colonization in mice by regulating the Nos2-associated pathways" by Han et al reported that using clindamycin, an antibiotic to selectively disorder anaerobic Bacteriodetes, intestinal microbiome dysbiosis resulted in Mycobacterium smegmatis (MS) colonization in the mice lungs. The authors found that clindamycin induced damage of the enterocytes and gut permeability and also enhanced the fermentation of cecum contents, which finally increased MS colonization in the mice's lungs. The study showed that gut microbiota dysbiosis up-regulated the Nos2 gene-associated pathways, leading to increased nitric oxide (NO) levels and decreased reactive oxygen species (ROS) and β-defensin 1 (Defb1) levels. These changes in the host's immune response created an antimicrobial and anti-inflammatory environment that favored MS colonization in the lungs. The findings suggest that gut microbiota dysbiosis can modulate the host's immune response and increase susceptibility to pulmonary infections by altering the expression of key genes and pathways involved in innate immunity. The authors reasonably provided experimental data and subsequent gene profiles to support their conclusion. Although the overall outcomes are convincing, there are several issues that need to be addressed:

(1) In Figure S1, the reviewer suggests checking the image sizes of the pathological sections of intestinal tissue from the control group and the CL-treatment group. When compared to the same intestinal tissue images in Figure S4, they do not appear to be consistently magnified at 40x. The numerical scale bars should be presented instead of just magnification such as "40x".

(2) In Figure 4d, the ratio of Firmicutes in the CL-FMT group decreased compared to the CON-FMT group, whereas the CL-treatment group showed an increase in Firmicutes compared to the Control group in Figure 3b. The author should explain this discrepancy and discuss its potential implications on the study's findings.

(3) In Figure 6, did the authors have a specific reason for selecting Nos2 but not Tnf for further investigation? The expression level of the Tnf gene appears to be the most significant in both RT-qPCR and RNA-sequencing results in Figure 5f. Tnf is an important cytokine involved in immune responses to bacterial infections, so it is also a factor that can influence NO, ROS, and Defb1 levels.

eLife assessment

The manuscript by Carbo et al. reports a novel role for the MltG homolog AgmT in gliding motility in M. xanthus. The authors provide convincing data to demonstrate that AgmT is a cell wall lytic enzyme (likely a lytic transglycosylase), its lytic activity is required for gliding motility, and that its activity is required for proper binding of a component of the motility apparatus to the cell wall. The findings are valuable as they contribute to our understanding of the molecular mechanisms underlying the interaction between gliding motility and the bacterial cell wall.

Reviewer #1 (Public Review):

Summary:

This manuscript nicely outlines a conceptual problem with the bFAC model in A-motility, namely, how is the energy produced by the inner membrane AglRQS motor transduced through the cell wall into mechanical force on the cell surface to drive motility? To address this, the authors make a significant contribution by identifying and characterizing a lytic transglycosylase (LTG) called AgmT. This work thus provides clues and a future framework work for addressing mechanical force transmission between the cytoplasm and the cell surface.

Strengths:

(1) Convincing evidence shows AgmT functions as an LTG and, surprisingly, that mltG from E. coli complements the swarming defect of an agmT mutant.

(2) Authors show agmT mutants develop morphological changes in response to treatment with a -lactam antibiotic, mecillinam.

(3) The use of single-molecule tracking to monitor the assembly and dynamics of bFACs in WT and mutant backgrounds.

(4) The authors understand the limitations of their work and do not overinterpret their data.

Weaknesses:

(1) A clear model of AgmT's role in gliding motility or interactions with other A-motility proteins is not provided. Instead, speculative roles for how AgmT enzymatic activity could facilitate bFAC function in A-motility are discussed.

(2) Although agmT mutants do not swarm, in-depth phenotypic analysis is lacking. In particular, do individual agmT mutant cells move, as found with other swarming defective mutants, or are agmT mutants completely nonmotile, as are motor mutants?

(3) The bioinformatic and comparative genomics analysis of agmT is incomplete. For example, the sequence relationships between AgmT, MltG, and the 13 other LTG proteins in M. xanthus are not clear. Is E. coli MltG the closest homology to AgmT? Their relationships could be addressed with a phylogenetic tree and/or sequence alignments. Furthermore, are there other A-motility genes in proximity to agmT? Similarly, does agmT show specific co-occurrences with the other A-motility genes across genera/species?

(4) Related to iii, what about the functional relationship of the endogenous 13 LTG genes? Although knockout mutants were shown to be motile, presumably because AgmT is present, can overexpression of them, similar to E. coli MltG, complement an agmT mutant? In other words, why does MltG complement and the endogenous LTG proteins appear not to be relevant?

(5) Based on Figure 2B, overexpression of MltG enhances A-motility compared to the parent strain and the agmT-PAmCh complemented strain, is this actually true? Showing expanded swarming colony phenotypes would help address this question.

(6) Cell flexibility is correlated with gliding motility function in M. xanthus. Since AgmT has LTG activity, are agmT mutants less flexible than WT cells and is this the cause of their motility defect?