eLife assessment

Serotonin is an important neurotransmitter and its synaptic concentration is controlled by re-uptake by the sodium-coupled serotonin transporter SERT. The manuscript by Chan et al reports results from a systematic deep mutagenesis approach to study the surface expression and APP+ (5HT analogue) transport mechanism of the human serotonin transporter. The authors complement this experimental evidence with large-scale molecular simulations of the transporter in the presence of APP+. The use of deep mutagenesis and large-scale adaptive sampling simulations is impressive, and could contribute to understanding the structural requirements for folding and how transporters evolve to recognize different substrates.

Reviewer #1 (Public Review):

Sertonin is an important neurotransmitter and it synaptic concentration is controlled by re-uptake by the sodium-coupled serotonin transporter SERT. In this paper, some 6000 mutations of SERT were made and tested for surface expression and uptake of a serotonin analogue APP+. The SERT mutants were analysed and compared to the SERT structure and dynamics based on MD simulations. The authors have concluded that mutations located on surface exposed regions are tolerated whilst those involved in packing and structural integrity are not. Gain-of-function mutations map onto regions that in most cases favour opening of a solvent-exposed intracellular vestibule. Closure of the intracellular gate is thought to be rate-limiting to the transport cycle, and thus the evolutionary-based screen is consistent with the clustering of gain-of-function mutations.

Strengths:<br /> This paper using a large unbiased data-set to probe the evolution of the serotonin transporter SERT for the substrate APP+. They have been able to compare both localisation and transport data, which is an interesting data-set. Using MD simulations they are further able to provide some rationale basis for the gain-of-function mutants.

Weaknesses:<br /> They can only detect surface expression of myc-tagged SERT based on conjugation with a fluorescent anti-myc antibody. As such, they cannot distinguish between SERT mutants that abolish expression vs. those that are no longer trafficking to the plasma membrane. This is a downside, as it would have been interesting to know the fraction of SERT mutations disrupt trafficking. Indeed, the relationship between misfolding and targeting is poorly understood beyond the calnexin- calreticulin cycle. Furthermore, there seems to be a gap between the large-scale mutagenesis data and the MD simulations in which the main mechanistic conclusions seem to be based on (carried out in a separate publication). Thus, overall while the mutation data-set is impressive its not clear how this aids to our mechanistic understanding of SERT.

Reviewer #2 (Public Review):

The manuscript by Chan et al reports results of a systematic mutagenesis approach to study the surface expression and APP+ transport mechanism of serotonin transporter. They complement this experimental evidence with large-scale molecular simulations of the transporter in the presence of APP+. The use of deep mutagenesis and large-scale adaptive sampling simulations is impressive and could be very exciting contributions to the field.

On the whole, the results appear to provide a fascinating insight into the effects of mutations on transport mechanisms, and how those interrelate with the structural fold and biophysical properties of a dynamic protein and its substrate pathways. A weakness of the conclusions based on the molecular simulation is that it relies on comparison with previously-published work involving non-identical simulation systems (i.e. different protonation states).

Conclusions in this work about the origins of the sodium:serotonin 1:1 stoichiometry should also be considered in the context of the fact that there are two sodium ions bound in the structures of SERT, and more work is needed to explain why this ion is not also released/co-transported.

Some of the methods require additional information to be provided to be reproducible, for example, for the Transition Path Theory results, and so it is not possible to assess these conclusions with the manuscript in its current form.

Reviewer #3 (Public Review):

The results of the deep mutagenesis screen represent a wealth of information on the expression and function of SERT that everyone studying this protein will appreciate. However, as the authors explain, the screen identified mutations that increased APP+ transport but inhibited transport of the cognate substrate, 5-HT. Because of the methods used, 5-HT could not be used as a substrate, somewhat limiting the usefulness of the screen.

However, the authors have taken advantage of this limitation to address the mechanistic features of SERT that discriminate between 5-HT and APP+. From the position of mutations that augment APP+ transport, they have identified the aqueous pathway created in inward facing SERT conformations as a region of importance. Based on the MD simulations, transition to inward facing conformations is facilitated by 5-HT but less so by APP+. The authors conclude, quite reasonably, that mutations interfering with the stability of inward-closed SERT states could overcome the reduced ability of APP+ to open the pathway.

Another reasonable conclusion based on the mutant screen, is that mutations detrimental to surface expression were found in packed hydrophobic regions of the protein, but similar mutations in the permeation pathways were less likely to decrease expression. The authors postulate that this provides an evolutionary advantage by maintaining the structural fold while allowing modification of ion and substrate binding and coupling sites, a reasonable but speculative conclusion.

Not all gain-of-function mutations have to be specific to APP+. The authors point out that Ala173Gly converts SERT to the residue found in NET and DAT at this position. It would have been interesting to know how this mutation and others affect 5-HT transport. Indeed, the lack of any 5-HT transport measurements with the mutants is a glaring weakness of the manuscript.

eLife assessment

The authors provide a high quality genome of the xenacoelomorph worm Xenoturbella bocki and discuss its structure and evolution. Understanding the genomic structure of this group provides important insights into bilaterian evolution. The authors make a solid case that the data they present can support the placement of Xenacoelomorpha within the deuterostomes rather than as a sister group to all other bilaterians, but do not unequivocally reject the competing scenario.

Reviewer #1 (Public Review):

The authors report a high-quality genome assembly for a member of Xenacoelomorpha, a taxon that is at the center of the last remaining great controversies in animal evolution. The taxon and the species in question have "jumped around" the animal tree of life over the past 25 years, and seemed to have found their place as a sister-group to all remaining bilaterians. This hypothesis posits that the earliest split within Bilateria includes Xenacoelomorpha on the one hand and a clade known as Nephrozoa (Protostomia + Deuterostomia) on the other, and is thus referred to as the Nephrozoa hypothesis. Nephrozoa is supported by phylogenomic evidence, by a number of synapomorphic morphological characters in the Nephrozoa (namely, the presence of nephridia) and lack of some key bilaterian characters in Xenacoelomorpha, and by the presence of unique miRNAs in Nephrozoa.

The Nephrozoa hypothesis has been challenged several times by the authors' groups who alternatively suggest placing Xenacoelomorpha within Deuterostomia as a sister group to a clade known as Ambulacraria. This hypothesis (the Xenambulacraria hypothesis) is supported by alternative phylogenomic datasets and by the shared presence of a number of unique molecular signatures. In this contribution, the authors aim to strengthen their case by providing full genome data for Xenoturbella bocki.<br /> The actual sequencing and analysis are technically and methodologically excellent. Some of the analyses were done several years ago using approaches that may now seem obsolete, but there is no reason not to include them. As a detailed report of a newly sequenced genome, the manuscript meets the highest standards.

The authors emphasize a number of key findings. One is the fact that the genome is not as simple as one might expect from a "basal" taxon, and is on par with other bilaterian genomes and even more complex than the genome of secondarily simplified bilaterians. There is an implicit expectation here that the sister group to all Bilateria would represent the primitive state. This is of course not true, and the authors are aware of this, but it sometimes feels as though they are using this implicit assumption as a straw dog argument to say that since the genome is not as simple as expected, X. bocki must be nested within Bilateria. The authors get around this by acknowledging that their finding is consistent with a "weak version of the Nephrozoa hypothesis", which is essentially the Nephrozoa phylogenetic hypothesis without implicit assumptions of simplicity.

Another finding is a refutation of the miRNA data supporting Nephrozoa. This is an important finding although it is somewhat flogging a dead horse, since there is already a fair amount of skepticism about the validity of the miRNA data (now over 20 years old) for higher-level phylogenetics.

The finding that the authors feel is most important is gene presence-absence data that recovers a topology in which X. bocki is sister to Abulacraria. The problem is that the same tree does not support the monophyly of Xenacoelomorpha. This may be an artifact of fast evolving acoel genomes, as the authors suggest, but it still raises questions about the robustness of the data.

In sum, the authors' results and analyses leave an open window for the Xenambulacraria hypothesis, but do not refute the Nephrozoa hypothesis. The manuscript is a valuable contribution to the debate but does not go a significant way towards its resolution.<br /> The manuscript has gone through several rounds of review and revision on a preprint server and is thus fairly clear of typos, inconsistencies and lack of clarity. The authors are honest and open in their interpretation of the results and their strengths.

Reviewer #2 (Public Review):

The manuscript describes the genome assembly and analysis of Xenoturbella bocki, a worm that bears many morphological features ascribed to basal bilateria. The authors aim to analyse this genome in an attempt to determine the phylogenetic position of X. bocki as a representative of Xenacoelomorpha and its associated acoelomorphs. In doing so, they want to inform the debate as to whether xenacoelomorph belong among, or is in fact paraphyletic to all bilaterians.

This paper presents a high-quality assembly of the X. bocki genome. By virtue of the phylogenetic position of this species, this genome has considerable scientific interest. This assembly appears to be highly complete and is a strength of the paper. The further characterisation of the genome is well executed and presented. Solid results from this paper include a comprehensive description of the Hox genes, miRNA and neruopeptide repertoire, as well as a description of the linkage group and how they relate to the ancestral linkage groups.

Where this paper is weaker is that for the central claims and questions of this paper, i.e,. the question of the phylogenetic position of xenacoelomorph and whether X. bocki is a slowly evolving, but otherwise representative member of this clade, remains insufficiently resolved.

The authors have achieved the goal of describing the X. bocki genome very well. By contrast, it is unclear, based on the presented evidence, whether xenacoelomorph is truly a monophyletic group. The balance of the evidence seems to suggest that the X. bocki genome belongs within the bilateria group. However, it is unclear as to what is driving the position of the other acoels. Assumign that X. bocki and the other two species in that group are monophyletic, then the evidence will favour the authors' conclusion (but without clearly rejecting the alternatives).

This paper will likely further animate the debate regarding this basal species, and also questions related to the ancestral characters of bilateria as a whole. In particular the results from the HOX and paraHOX clusters, may provide an interesting counterpoint to the previous results based on the acoels.

eLife assessment

The study presents a valuable finding on quantifying the orientation and organization of chondrocyte columns in the prenatal and postnatal growth plate cartilage using advanced 3D imaging and a sophisticated image analysis pipeline. The evidence supporting the authors' conclusions regarding the lack of columns in the fetal growth plate is considered inadequate due to technical caveats, inconsistencies in the data and corresponding model, and failure to correctly put the findings in context.

Reviewer #1 (Public Review):

Rubin et al. study chondrocyte columns in the prenatal and postnatal growth plate in 3D for the first time, using a novel analysis pipeline in which Confetti clones in the murine growth plate are analysed morphometrically. Prenatal chondrocytes were found not to be organised in columns parallel to the main orientation of the long bone, but rather, prenatal chondrocytes were commonly organised perpendicular to the main direction of growth. In the postnatal (P40) growth plate there was a diverse arrangement of columns, but more of the columns were vertically aligned

I enjoyed reading the work and the analysis is rigorous. However, I think that it is not valid to state that columns do not form in the embryo. The data only supports the finding that strictly vertical columns do not form in the embryo, as the cells are still organised into columns, albeit with a range of orientations. I do not like the term "typically" aligned, as how can we know what is "typical" when orientation has never before been assessed in 3D... And the authors' data demonstrates that it is certainly not "typical" for chondrocyte to organise into vertical columns prenatally.

It would be very interesting to delve deeper into the reason for the change in orientation of columns between pre- and post-natal. For example, does more circumferential growth happen prenatally as compared to postnatally? Is the rate of circumferential vs longitudinal growth different between prenatal and postnatal, and could the change in column orientation be responsible for a (possible) shift in the balance between longitudinal vs circumferential growth before vs after birth? The first sentence of the Discussion refers to the role of chondrocyte columns in driving bone elongation, but aren't they also involved in driving bone morphology?

I feel describing the activity of the cells as "mis-rotations" which implies the orientations are not intentional. It is likely not accidental or mistaken that the chondrocytes align in the ways they do- the diaphysis is largely for longitudinal growth while the epiphyses, and lateral expansion of the joint is also important. I find the data in Figure 4 fascinating, especially the variation in orientations between the regions of the growth plate (from proximal to distal), with the most lateral orientation at the most proximal and distal ends- it would be nice to see more discussion of these variations and what they may be contributing to.

The abstract focuses solely on the analysis of columns prenatally and would benefit from the inclusion of the data from the postnatal growth plate and from the chondrocyte rotations.

Reviewer #2 (Public Review):

The origin and function of proliferative chondrocyte columns in the growth plate that are generally aligned with predicted longitudinal growth vectors have been robustly debated since the implementation of clonal analysis and live cell imaging techniques more than a decade ago. In particular, live cell imaging demonstrated that in the proliferative zone, most daughter pairs rotate fully or partially after division to form columns of stacked cells and a minority of pairs fail to rotate. These observations and others led to a mechanistic model of column formation, but limitations in the live cell imaging methods that only visualize a single round of division and rotation left open an important question - what is the effect of different rotation profiles on column formation, bone growth, and morphology?

This manuscript describes the use of an inducible lineage tracing system in the mouse combined with a novel image analysis pipeline to analyze column formation over multiple cell divisions. The main conclusion is that many clones generate single columns in postnatal mice (as expected), but clones in embryonic growth plate cartilage form clusters distributed laterally, not aligned with longitudinal growth. These findings are interpreted to suggest that column formation is not required for long bone growth in the embryo and that lateral expansion of proliferative chondrocyte clusters may drive an increase in bone width.

Although these findings are intriguing and potentially impactful, there are important caveats to the approach that generate significant uncertainty in both the measurements and the conclusions. (1) The claim that embryonic growth plate chondrocytes do not form columns conflicts with the observation of columnar stacks in the clusters. (2) Interpretation of nuclear elevation data is based on the unproven assumption that nuclei should be stacked in cell columns. (3) Clonal analysis of proliferative chondrocyte cell division and stacking behaviors is only valid if clone labeling is initiated in a proliferative chondrocyte, not when the founder cell is a resting chondrocyte. The data are insufficient to validate this absolute requirement.

Reviewer #3 (Public Review):

The manuscript by Rubin and Agrawal et al presents a very nice imaging analysis of clonal cell organization in the fetal and late juvenile mouse growth cartilages. The authors have performed a thorough quantification of the orientations of clusters and of clones of cells with respect to the growth axis. They conclude that growth cartilage is not as strictly 'columnar' as has been commonly described, especially at the fetal stage. There is value to having such quantifications in the literature as a reminder that interpretations of phenotypes need to be rooted in the cell biology of the stage at hand, as emphasized by the authors. However, although the approach is comprehensive, aspects of the quantification methods are not described adequately to determine if they are correct for the questions. There are also some inequivalent comparisons to prior literature and an oversight of important published observations showing that some of these conclusions have been known for decades, though not as thoroughly quantitative. There have long been observations that some growth cartilages do not have proliferative columns oriented in the axis of growth and that not all columns of a growth cartilage are perfectly organized; these facts do not negate the observations that columnar organization does exist, as re-confirmed here, and that it correlates with and contributes to rapid growth rates. Each of these points is further elaborated below.

eLife assessment

This fundamental work advances our understanding of the central coding and control mechanisms regulating sympathetic nervous system efferent signals to bone. The evidence supporting the conclusion is mostly convincing, although the inclusion of higher resolution images for certain data and further discussions would strengthen the study. This paper holds potential interest for skeletal biologists and neuroscientists who study the brain-bone sympathetic neural circuits.

Reviewer #1 (Public Review):

This manuscript presents, for the first time, the utilization of PRV viral transneuronal tracing to elucidate the central coding and control mechanisms governing sympathetic nervous system (SNS) efferent signals to bone. This groundbreaking work not only holds promising research prospects but also establishes a robust foundation for understanding the neural regulation of bone metabolism.

Reviewer #2 (Public Review):

Summary:<br /> In this study, the authors have used virtual transneuronal tracing technology to identify for the first time the central sympathetic nervous system outflow sites that innervate bone.

Strengths:<br /> The study provides a comprehensive atlas of the brain regions that potentially play a role in coding and decoding sympathetic nervous system signals to bone.

Weaknesses:<br /> While the study provides compelling evidence for the brain-bone sympathetic nervous system neuroaxis, it is unclear if diseases that affect bone (e.g. diabetes, osteoporosis, kidney failure) disrupt brain-bone sympathetic neural circuits.

Reviewer #3 (Public Review):

It has been reported that the sympathetic nervous system (SNS) mediates bone metabolism and nociceptive functions. However, the exact localization and organization of the central SNS circuitry innervating bone and the brain sites have not been mapped and efferent SNS outflow to bone has not yet been characterized yet. Authors used pseudorabies (PRV) viral transneuronal tracing approach to identify central SNS outflow sites that innervate bone. The authors found that the central SNS outflow to bone originates from brain nuclei, sub-nuclei and regions of six brain divisions (midbrain and pons, hypothalamus, hindbrain medulla, forebrain, cerebral cortex, and thalamus). The authors provided compelling evidence for a brain-bone SNS neuroaxis that may regulate bone metabolism and nociceptive functions, which provided a greater understanding of the neural regulation of bone metabolism and would stimulate further research into bone pain and the neural regulation of bone metabolism. Authors may discuss and summarize their results in detail for a better understanding of their findings and enhancing the manuscript's utility for readers.

eLife assessment

This paper is valuable in that it provides a critical missing link between measures of structural connectivity and rhythmic tapping abilities, pointing to some interesting possibilities for how tapping synchronization is carried out. The methodology and findings are solid, and of interest to those studying the neural mechanisms of timing.

Reviewer #1 (Public Review):

Garcia-Saldivar and colleagues present a manuscript investigating connections between diffusion-weighted imaging (DWI) parameters and paced finger tapping measures. A cohort of human participants (n=32) performed a paced finger tapping task with a synchronization-continuation paradigm, in which they were required to listen to a paced metronome, begin tapping in synchrony with it, and then continue tapping at the same rate without it. Both auditory and visual metronomes were used, at a range of intervals. All subjects received structural scans measuring DWI, with an emphasis on superficial and deep white matter structures. This latter analysis was the most innovative, as it allowed the authors to examine microstructural effects in short-range cortical connections.

Behaviorally, the authors replicated some well-known effects in paced finger tapping, with better performance for auditory over visual rhythms, negative lag-1 autocorrelations, and best performance at a range of ~1.5Hz. For the DWI analyses, a large number of correlations were observed across a wide variety of connections with various brain regions. The most salient effects observed were a connection between asynchrony, only for the auditory condition, and connections between the right auditory and motor systems, around the duration of peak performance, as well as a "chronotopic" organization across parts of the corpus callosum, most notably in areas linking motor regions between hemispheres.

Overall, this paper provides a critical missing link between measures of structural connectivity and rhythmic tapping abilities, pointing to some interesting possibilities for how tapping synchronization (at least for auditory intervals) is carried out. Negative aspects of the paper come from the largely exploratory aspects of the analysis, as well as potential biases from the low sample size.

Reviewer #2 (Public Review):

This is a valuable study of the relationships between aspects of white matter structure in the brain and the accuracy of tapping performance on auditory and visual versions of a synchronization-continuation task. The authors find brain-behaviour relationships between absolute asynchrony (precision of phase alignment between taps and stimulus events), but only for certain temporal rates (650 and 750 ms ISI, not 550, 850, or 950 ms ISI). Other behavioural metrics do not significantly correlate with white matter measures, and no visual condition behavioural metrics correlate either. The methodology and findings are solid, and of interest to those studying the neural mechanisms of timing.

The question is interesting, as the neural mechanisms of timing, and the nature of how modality differences in timing arise, are important, given that certain modality differences in timing accuracy (e.g., auditory benefits relative to visual) are less striking in our closest evolutionary relatives. Overall, the methods are well-presented and both behavioural and neural measures are appropriate.

The results are generally well-reported, although there is a lack of clarity about multiple comparison corrections for the number of separate behavioural metrics, different interval lengths examined, and the two sensory modalities.

Some weaknesses:<br /> The use of absolute (unsigned) asynchrony as a measure of 'predictive' ability is not fully justified. Signed asynchrony may be a more informative measure of predictive ability, as (small) negative asynchronies (taps prior to event onset) are often interpreted as indicating prediction, whereas positive asynchronies (taps after the event onset) are not.<br /> The work may benefit from considering the 'phase' and 'period' nature of the different behavioural measures, as they may tap different aspects of timing. Separating the behavioural metrics into those reflecting phase synchrony versus period matching may be a useful distinction, as the period-related metrics are the ones that do not have evidence of correlation with brain metrics.<br /> The manuscript does not present a very clear framework for why certain measures might be predicted to correlate with white matter structure and others not, and the pattern of results is also not easily interpretable. This may just be the nature of the data, but it would help clarify if more justification for the selection of task and stimulus rates was presented, along with an idea of the predictions made by different theoretical approaches for what relationships between this particular set of behavioural and brain data might exist. Similarly, a more nuanced discussion might further explore the potential reasons for the lack of evidence for a relationship at shorter and longer auditory interval lengths, as well as for any of the visual condition measures.

Overall, the authors find white-matter structure relationships with absolute asynchrony measures during auditory (but not visual) synchronization-continuation at certain rates. These findings appear reasonably justified.

Author Response:

eLife assessment

We thank the Editors for identifying qualified reviewers. We agree that the “evidence supporting this claim (that ‘many breast cancer mutations are mildly deleterious’) is incomplete”. Much more detail is needed to state this decisively and we do not claim completeness here. As far as validation, we carried out synthetic testing of the models as suggested by Reviewer #1 and the results seem good.

Reviewer #1:

We thank the Reviewer for a very thorough examination of not only the current paper but also our previous paper. We agree that the illustration material can be overwhelming and we plan to use the Reviewer’s advice in that matter. In addition, we originally put some textbook material in the Appendix, and arguably some of it may be considered superfluous.

Most of the references the Reviewer provides are known to us, although it is likely we should cite and discuss more. All of the above will be included in the revision we are planning.

The Reviewer is certainly correct that population growth and spatial effects play a major role in cancer. However, the effects of constraining environment are quite strong and the reality lies somewhere between the Moran and branching process models; exactly what we attempt to clarify. As for spatial effects, most tumors extracted in clinic are dissected in bulk and sub-sampling is rare, so the spatial information is rarely accessible.

The subsequent point of importance concerns the weak specificity of the site frequency spectra (SFS) with respect to the underlying genetic and demographic forces. This cannot be denied. However, we just meant to state that our SFS are consistent with a model involving slightly deleterious passengers.

Regarding the validation of the estimation procedures which is a point well-taken, we carried out synthetic testing of the models as suggested by Reviewer #1 and the results seem good. This will be discussed in full in the revision.

In our view, the most important remark is the one concerning scaling of the models. The Reviewer is certainly correct that 100 stem cells are insufficient to drive a realistic tumor. However, what we had in mind but not explained sufficiently, is that a sample of 100 cells corresponds to average-depth coverage in bulk sequencing. Therefore, the strict interpretation is that the model mirrors what is observed in the sample. A more accurate approach would be to up-scale the model and then sample 100 cells from it. The Moran-type model can be up-scaled using diffusion approximation, and we hope to include these computations in the revision. The associated criticism concerning tumor growth seems less relevant, since we experimented with less or more stringent constraints in our models.

Reviewer #2:

We thank Reviewer #2 for studying our paper and some very positive comments. Among others, the Reviewer underscores the fact that the Moran-type model generates SFS concordant with the data (with all necessary reservations). The Reviewer concurs with us that conditioning on non-extinction is not very common in the literature, while it should be.

Similarly as the Reviewer, we are somewhat puzzled by the differences in behavior between models A and B. Model B seems more parsimonious, but Model A looks more similar to the critical or slightly supercritical branching process. We will work to clarify these observations.

eLife assessment

This study uses numerical simulations to characterize and compare variants of two widely used mathematical models and then applies those models to inferring evolutionary parameters from breast cancer data. The copious numerical results will be of some interest to mathematical biologists working with similar models. The finding that many breast cancer mutations are mildly deleterious is valuable but the evidence supporting this claim is incomplete because the mathematical modelling and statistical methods are insufficiently justified and inadequately validated.

Reviewer #1 (Public Review):

This paper can be seen as an extension of a recent study by two of the same authors [1]. In the previous paper, the authors considered two variants of the Moran process, labelled Model A and Model B, and examined differences between the evolutionary dynamics of these two models. They further described the site frequency spectra, expected allele counts, and expected singleton counts of these models, building on analytical results from prior studies, and used numerical simulations to investigate the models' evolutionary dynamics. Finally, they compared the site frequency spectra of the two models (using numerical simulations) to spectra derived from a small breast cancer data set (two sets of three samples).

In the new paper, the authors consider the same two Moran process variants (Model A and Model B) and some related branching processes. As before, they compare the site frequency spectra and various summary statistics of these models, but here they present only numerical simulations (except that some prior analytical results are summarized in Appendix A, which are never referred to in the main text and seem unconnected to the study). They then compare the site frequency spectra of these models (again using numerical simulations) to those derived from the same breast cancer samples as before and thus infer some evolutionary parameters.

The first main conclusion is that the critical branching process and the Moran process models behave similarly and generate similar site frequency spectra. This finding is unsurprising (indeed, the authors acknowledge that the result "has been expected"). For a reasonably large population size, the population size in the critical branching process has been shown to vary relatively little over time and the model is thus essentially a continuous time Moran process (see, for example, Equation 8.55 in ref 2). Nor is it surprising that the authors see stronger similarities when they select only the subset of branching process replicates in which the final population size is particularly close to the initial population size (this is because, in these replicates, the population size likely varies even less than usual).

The second main conclusion is that, although "the mutational SFS alone is not adequate" to quantify the strength of selection, "All fitted values for the selective disadvantage of passenger mutations are nonzero, supporting the view that they exert deleterious selection during tumorigenesis". Although the question of whether mildly deleterious mutations play an important role in cancer evolution is of considerable interest, it's debatable whether the results presented here help resolve the issue.

Many prominent researchers have called into question whether cancer evolutionary parameters can be reliably inferred from site frequency spectra (e.g., [3-7]), even using sophisticated statistical methods. The statistical approach used here (though not named as such in the paper) is a crude kind of approximate Bayesian computation. To improve the accuracy of the results, it would have been better to have set reasonably vague priors for the uncertain mutation rates, rather than fixing them arbitrarily. It would also have been better to have chosen a likelihood function explicitly based on an analysis of the sampling and error distributions, rather than just summing the absolute logged deviations. It is well known that "Checking the model is crucial to statistical analysis" and "A good Bayesian analysis, therefore, should include at least some check of the adequacy of the fit of the model to the data and the plausibility of the model for the purposes for which the model will be used" [8]. The authors' failure to describe any attempt to validate or check their model, using simulated data or otherwise, casts doubt on the reliability of their inferences.

Putting aside the potential biassing effects of sampling error, measurement error, and the limitations of the authors' statistical method, it is well established that both population growth and spatial structure profoundly alter the shape of site frequency spectra in ways that can mimic the effects of selection (e.g. [9-11]). Indeed, Figures 3, 4 and 5 show that the critical and super-critical branching processes generate markedly different site frequency spectra. It follows that if the population dynamics and spatial structure of the mathematical model used for inference don't match those of the biological process that produced the data then any inferred evolutionary parameter values will be unreliable. Breast cancer has two indisputable ecological features that shape its evolutionary dynamics: the cell population expands by many orders of magnitude from a single cell, and the population is spatially structured. In the authors' mathematical model, the population size is initially 100 cells and either remains constant or varies little, and there is no spatial structure. These profound mismatches between model and data cast further doubt on what is supposed to be the paper's most important biological finding.

In this paper the authors offer no justification for their decision to model breast cancer as a non-growing, non-spatial cell population. Nor do they engage with the extensive recent literature on the challenges of inferring evolutionary parameters from cancer site frequency spectra (they cite none of the many relevant papers listed at https://www.sottorivalab.org/neutral-evolution.html). Their 2022 paper [1] claims that, "it sometimes makes sense to consider cancer growth in the framework of constant-population models. Our models correspond to the situation in which a constant population of N "healthy" stem cells is gradually replaced by a growing clone of transformed cells with increasing fitness." No evidence was presented to support this hypothesis regarding breast cancer progression. On the other hand, a wealth of evidence supports the consensus view that, in breast cancer and other human solid tumours, the number of cells with unlimited proliferative potential is several orders of magnitude greater than 100 and grows over time (e.g. [12]).

Analytic expressions for the site frequency spectra with neutral mutations are already known. It is well known that the site frequency spectrum of an exponentially growing population has a tail following a power law S_k ~ k^(-2) [13, 14]. Similarly, it is known that for the critical branching process or the Moran process, the site frequency spectrum at equilibrium is S_k ~ k^(-1) [13, 15]. Especially noteworthy yet uncited studies that use those results about site frequency spectra to make inferences based on sequencing data include ref 16, in which selection is inferred, and ref 17, in which evolutionary parameters of constant populations (healthy cell populations) are inferred.

Although the paper is well written, the figures are ineffective in communicating the results. As others have put it, "A figure is meant to express an idea or introduce some facts or a result that would be too long (or nearly impossible) to explain only with words" and "If your figure is able to convey a striking message at first glance, chances are increased that your article will draw more attention from the community" [18]. On the contrary, Figures 3, 4, 5 and 6 are bewilderingly complicated, crowded, and repetitive. These figures comprise no fewer than fifty-six plots, each containing numerous curves or histograms, spread across four pages. To compare the results of different scenarios, the reader is presumably expected to put these figures side by side and try to spot the differences, hampered by inconsistent axis ranges, absence of axis labels, absence of titles, absence of legends, and unreliable captions ("cyan" seems to refer to pale blue, and "orange" to something closer to red). For example, the only notable difference between Figures 3 and 4 is in the shape of a single green curve in panel I. In the main text of a published paper, one would expect fewer, more carefully curated figures drawing attention to salient features, so that the reader can infer the main results with minimal effort. The rest can be put in supplementary figures.

In summary, this paper adds somewhat to our understanding of some standard mathematical models; whether it tells us anything new about cancer is open to debate.

References<br /> (1) Kurpas, Monika K., and Marek Kimmel. "Modes of selection in tumors as reflected by two mathematical models and site frequency spectra." Frontiers in Ecology and Evolution 10 (2022): 889438.<br /> (2) Bailey, Norman TJ. The elements of stochastic processes with applications to the natural sciences. John Wiley & Sons, 1964.<br /> (3) Tarabichi, Maxime, et al. "Neutral tumor evolution?." Nature Genetics 50.12 (2018): 1630-1633.<br /> (4) McDonald, Thomas O., Shaon Chakrabarti, and Franziska Michor. "Currently available bulk sequencing data do not necessarily support a model of neutral tumor evolution." Nature Genetics 50.12 (2018): 1620-1623.<br /> (5) Balaparya, Abdul, and Subhajyoti De. "Revisiting signatures of neutral tumor evolution in the light of complexity of cancer genomic data." Nature Genetics 50.12 (2018): 1626-1628.<br /> (6) Noorbakhsh, Javad, and Jeffrey H. Chuang. "Uncertainties in tumor allele frequencies limit power to infer evolutionary pressures." Nature Genetics 49.9 (2017): 1288-1289.<br /> (7) Bozic, Ivana, Chay Paterson, and Bartlomiej Waclaw. "On measuring selection in cancer from subclonal mutation frequencies." PLoS Computational Biology 15.9 (2019): e1007368.<br /> (8) Neher, Richard A., and Oskar Hallatschek. "Genealogies of rapidly adapting populations." Proceedings of the National Academy of Sciences 110.2 (2013): 437-442.<br /> (9) Gelman, Andrew, et al. Bayesian data analysis (Third Edition). Chapman and Hall/CRC, 2014.<br /> (10) Fusco, Diana, et al. "Excess of mutational jackpot events in expanding populations revealed by spatial Luria-Delbrück experiments." Nature Communications 7.1 (2016): 12760.<br /> (11) Noble, Robert, et al. "Spatial structure governs the mode of tumour evolution." Nature Ecology & Evolution 6.2 (2022): 207-217.<br /> (12) Lawson, Devon A., et al. "Single-cell analysis reveals a stem-cell program in human metastatic breast cancer cells." Nature 526.7571 (2015): 131-135.<br /> (13) Gunnarsson, Einar B., Leder, Kevin, and Foo Jasmine. "Exact site frequency spectra of neutrally evolving tumors: A transition between power laws reveals a signature of cell viability" Theoretical Population Biology 142 (2021) 67-90<br /> (14) Durrett, Richard "Branching Process Models of Cancer" Springer (2015)<br /> (15) Durrett, Richard "Probability Models for DNA Sequence Evolution" Springer Science & Business media (2008)<br /> (16) Williams, Mark J. et al. "Quantification of subclonal selection in cancer from bulk sequencing data." Nature Genetics 50 (6). 895-903 (2018)<br /> (17) Moeller, Marius E. et al. "Measures of genetic diversification in somatic tissues at bulk and single-cell resolution" eLife (2024) 12:RP89780<br /> (18) Rougier, Nicolas P., Michael Droettboom, and Philip E. Bourne. "Ten simple rules for better figures." PLoS Computational Biology 10.9 (2014): e1003833.

Reviewer #2 (Public Review):

Summary:

In this manuscript, the authors present a comparison of two models of cancer evolution with advantageous drivers and deleterious passengers: a fixed-population "Moran" model, and a "Branching Process" (BP) model with dynamic population size. The Moran model is more mathematically-tractable, but since cancer is a disease of uncontrolled growth, it is unclear to me how clinically-relevant it is to consider a model with constant population size. Intriguingly, both models can explain observed Site Frequency Spectrums (SFSs) in three breast cancers, which suggests that the Moran model may have some value. This distinction between the two models is addressed well.

Strengths:

The comparisons of the various BP models (extinction/non-extinction, and balanced/supercritical) are very interesting. The survivability of rare, fitness-disadvantaged clones has huge implications for treatment resistance in general - drug resistant clones are very often disadvantaged in the absence of drug. Clinical sequencing is, most decidedly, investigating population dynamics conditioned on non-extinction, however most published models do not condition on non-extinction - an unfortunate community oversight that this publication rectifies.

Site Frequency Spectrums in three breast cancers are measured with unprecedented resolution to my knowledge (allele abundances below one in a thousand).

Detailed description of the behavior of the various models.

Weaknesses:

I do not believe Moran B is a useful theoretical distinction between Moran A. Incorporating fitness effects into the birth process, instead of the death process, is generally mathematically equivalent when time is measured in generations (or cell divisions). Visible differences in the two models in Figures 2-6 by all accounts seem to be due to the fact that Moran B experiences more evolution in the balanced/driver-dominated case, and less evolution in the passenger dominated case. We generally do not use arbitrary time steps for this reason - we quantify time in 'generations'.

eLife assessment

This investigation marks an important advancement in our understanding of motor thalamus connectivity, illustrating a complex integration of inputs that reshapes previous models. The study utilizes compelling methodologies that expose a dynamic synaptic network, although the evidence of triple-input convergence on individual neurons and for multiple driver type inputs onto motor thalamic neurons remains incomplete. Despite this, the findings provide a persuasive rationale for revisiting our perceptions of the thalamic role in motor control, with a call for further studies to substantiate the breadth of these functional interactions.

Reviewer #1 (Public Review):

The manuscript demonstrates an analysis of the synaptic organization within the motor thalamus, emphasizing the interplay between the ventrolateral (VL) and ventroanterior (VA) nuclei and their respective inputs. The primary aim is to unravel the complexities of synaptic interactions among the motor cortex's layer 5 (M1L5), the cerebellum (Cb), and the basal ganglia output nuclei (GPi and SNr), which converge upon the VA/VL nuclei of the motor thalamus. This examination is executed using a combination of anatomical tracing, optogenetics, and electrophysiological recordings in mouse brain slices, which together yield novel insights into the motor control circuitry.

The study uncovers that contrary to traditional models that presumed segregation, some motor thalamic neurons simultaneously integrate inputs from the cerebellum and basal ganglia. Furthermore, a subset of these neurons also receive convergent inputs from M1L5 and basal ganglia, underscoring the complexity of these synaptic networks. Notably, the study reveals that both M1L5 and Cb inputs exhibit driver-type synaptic properties, suggesting a significant impact on thalamic relay neurons.

The functional implications of this synaptic convergence suggest a complex gating mechanism by the inhibitory outputs of the basal ganglia, which could modulate information flow within the motor thalamus. This modulation is significant not only for transthalamic information processing but also for the integration of cerebellar inputs to the motor cortex. The study also highlights direct projections from M1L5 to the motor thalamus, indicating a potential direct influence on thalamic activity, in addition to the known indirect influence through the cortico-basal ganglia-thalamo-cortical loop.

The manuscript suggests that the traditional understanding of motor thalamic connectivity requires reconsideration, and it emphasizes the necessity of further investigation to understand fully the functional implications of this synaptic convergence. Future research may focus on more direct demonstrations of triple-input convergence and its behavioral consequences, as well as cross-species comparative studies to enhance the findings' applicability.

While the study provides valuable contributions to our knowledge of the motor thalamus, illuminating the intricate synaptic architecture of the motor thalamus and setting the stage for future explorations that will deepen our comprehension of motor control and thalamic function.

Reviewer #2 (Public Review):

This study assesses how inputs from primary motor cortex layer 5 (M1L5), basal ganglia output nuclei (GPi and SNr), and cerebellum (Cb) converge onto motor thalamus nuclei (VA/VL).

Methodology includes anatomical tracing, optogenetics and electrophysiological recordings in mouse brain slices.

The major findings are:<br /> - Some motor thalamic neurons receive input from both cerebellar and basal ganglia. This is contrary to the common belief that assumes these two inputs are segregated in the motor thalamus.

- Some motor thalamus neurons receive converging input from both motor cortex (M1L5) and basal ganglia.

- Both M1L5 and Cb inputs to the motor thalamus have driver-type synaptic properties, indicating a strong influence on thalamic relay neurons.

Functional implications are:<br /> - Given the inhibitory nature of basal ganglia output neurons, the converging inputs can allow for basal ganglia to gate information flow through the motor thalamus. This applies to transthalamic information, ie information conveyed through the thalamus across cortical regions, as well as cerebellar information flow to motor cortex.

- The direct projection from M1L5 to motor thalamus suggests that motor cortex can affect motor thalamic activity not only indirectly, through the traditional cortico-basal ganglia-thalamo-cortical loop, but also through direct projections.

The study is convincing and has important implications for the field. Methodology involves elegant viral techniques.

The main weakness is that there is no direct functional demonstration of all the 3 inputs from motor cortex, cerebellum, and basal ganglia, converging onto the same cells in motor thalamus. All the recordings concern dual area stimulations, and the anatomical studies show a very small overlap of all the 3 inputs onto motor thalamus.

eLife assessment

This paper presents a new method for separating organelles in an unbiased way. The method is applied to the separation of distinct subpopulations of insulin vesicles. There are concerns around whether the vesicles measured are in fact insulin vesicles and whether the observed changes in vesicle populations upon glucose stimulation are biologically meaningful, and thus it is difficult to assess at this stage how well the technique performs. This paper is likely to be of wide interest to cell biologists studying a variety of compartments, as well as to researchers in the beta cell field.

Reviewer #1 (Public Review):

This manuscript presents an exciting new method for separating insulin secretory granules using insulator-based dielectrophoresis (iDEP) of immunolabeled vesicles. The method has the advantage of being able to separate vesicles by subtle biophysical differences that do not need to be known by the experimenter, and hence could in principle be used to separate any type of organelle in an unbiased way. Any individual organelle ("particle") will have a characteristic ratio of electrokinetic to dielectrophoretic mobilities (EKMr) that will determine where it migrates in the presence of an electric field. Particles with different EKMr will migrate differently and thus can be separated. The present manuscript is primarily a methods paper to show the feasibility of the iDEP technique applied to insulin vesicles. Experiments are performed on cultured cells in low or high glucose, with the conclusion that there are several distinct subpopulations of insulin vesicles in both conditions, but that the distributions in the two conditions are different. As it is already known that glucose induces release of mature insulin vesicles and stimulates new vesicle biosynthesis and maturation, this finding is not necessarily new, but is intended as a proof of principle experiment to show that the technique works. This is a promising new technology based on solid theory that has the possibility to transform the study of insulin vesicle subpopulations, itself an emerging field. The technique development is a major strength of the paper. Also, cellular fractionation and iDEP experiments are performed well, and it is clear that the distribution of vesicle populations is different in the low and high glucose conditions. However, more work is needed to characterize the vesicle populations being separated, leaving open the possibility that the separated populations are not only insulin vesicles, but might consist of other compartments as well. It is also unclear whether the populations might represent immature and mature vesicles, distinct pools of mature vesicles such as the readily releasable pool and the reserve pool, or vesicles of different age. Without a better characterization of these populations, it is not possible to assess how well the iDEP technique is doing what is claimed.

Major comments:

(1) There is no attempt to relate the separated populations of vesicles to known subpopulations of insulin vesicles such as immature and mature vesicles, or the more recently characterized Syt9 and Syt7 vesicle subpopulations that differ in protein and lipid composition (Kreutzberger et al. 2020). Given that it is unclear exactly what populations of vesicles will be immunolabeled (see point #2 below), it is also possible that some of the "subpopulations" are other compartments being separated in addition to insulin vesicles. It will be important to examine other markers on these separated populations or to perform EM to show that they look like insulin vesicles.

(2) An antibody to synaptotagmin V is used to immunolabel vesicles, but there has been confusion between synaptotagmins V and IX in the literature and it isn't clear what exactly is being recognized by this antibody (this reviewer actually thinks it is Syt 9). If it is indeed recognizing Syt 9, it might already be labeling a restricted population of insulin vesicles (Kreutzberger et al. 2020). The specificity of this antibody should be clarified. Furthermore, Figure 2 is not convincing at showing that this synaptotagmin antibody specifically labels insulin vesicles nor is there convincing colocalization of this synaptotagmin antibody with insulin vesicles. In the image shown, several cells show very weak or no staining of both insulin and the synaptotagmin. The highlighted cell appears to show insulin mainly in a perinuclear structure (probably the Golgi) rather than in mature vesicles (which should be punctate), and insulin is not particularly well-colocalized with the synaptotagmin. Other cells in the image appear to have even less colocalization of insulin and synaptotagmin, and there is no quantification of colocalization. It seems possible that this antibody is recognizing other compartments in the cell, which would change the interpretation of the populations measured in the iDEP experiments. It would also be good to perform synaptotagmin staining under glucose-stimulating conditions, in case this alters the localization.

(3) The EKMr values of the vesicle populations between the low and high glucose conditions don't seem to precisely match. It is unclear if this just a technical limitation in comparing between experiments or instead suggests that glucose stimulation does not just change the proportion of vesicles in the subpopulations (i.e. the relative fluorescent intensities measured), but rather the nature of the subpopulations (i.e. they have distinct biophysical characteristics). This again gets to the issue of what these vesicle subpopulations represent. If glucose stimulation is simply converting immature to mature vesicles, one might expect it to change the proportion of vesicles, but not the biophysical properties of each subpopulation.

(4) The title of the paper promises "isolation" of insulin vesicles, but the manuscript only presents separation and no isolation of the separated populations. Isolation of the separated populations is important to be able to better define what these populations are (see point #1 above). Isolation is also critical if this is to be a valuable technique in the future. Yet the paper is unclear on whether it is actually technically feasible to isolate the populations separated by iDEP. In line 367, it states "this method provides a mechanism for the isolation and concentration of fractions which show the largest difference between the two population patterns for further bioanalysis (imaging, proteomics, lipidomics, etc.)." However, in line 361 it says "developing the capability to port the collected individual boluses will enable downstream analyses such as mass spectrometry or electron microscopy," suggesting that true isolation of these populations is not yet feasible. This should be clarified.

Reviewer #2 (Public Review):

This manuscript used DC-iDEP, a technology previously used on other organelle preparations to isolate insulin secretory granules from INS1 cells based on differences in dielectrophoretic and electrokinetic properties of synaptotagmin V positive insulin granules.

The major motivation presented for this work is to provide a methodology to allow for more sensitive isolation of subpopulations of granules allowing better understanding of the biochemical composition of these populations. This manuscript clearly demonstrates the ability of this technology to separate these subpopulations which will allow for future biochemical characterizations of insulin granules in future studies.

After proving these subpopulations can be observed, this method was then utilized to show there are shifts in these subpopulations when granules are isolated from glucose stimulated cells. Overall the method of isolation is novel and could provide a tool for further characterization of purified secretory granules.

The observation of glucose stimulation causing shifts in subpopulations is unsurprising. Glucose stimulation could cause a depletion of insulin and other secretory content from a subset of granules. It would be expected that this loss of content would cause a shift in electrochemical properties of the granules, but this is a nice confirmation that the isolation method has the sensitivity to delineate these changes.

Major comments:

(1) It is unclear what Synaptotagmin isoform is being looked at. Synaptotagmin V and IX have been repetitively interchanged in the literature. See note in syt IX section of "Moghadam and Jackson 2013 Front. Endocrinology" or read "Fukuda and Sagi-Eisenberg Calcium Bind Proteins 2008".

The 386 aa. isoform that is abundant in PC12 cells has been robustly observed in INS1 cells in multiple studies and has been frequently referred to as syt IX. The sequence the antibody was raised against should be determined from the company where this was purchased and then this should be mapped to to which isoform of Synaptotagmin by sequence and clarified in the text.

(2) Immunofluorescence of insulin and syt V is confusing. The example images do not appear to show robust punctate structures that are characteristic of secretory granules (in both the insulin and syt V stain).

(3) In the discussion it says, "Finally, this method provides a mechanism for the isolation and concentration of fractions which show the largest difference between the two population patterns for further bioanalysis (imaging, proteomics, lipidomics, etc.) that otherwise would not be possible given the low-abundance components of these subpopulations."

It would help to elaborate more on the yield and concentrations of isolated granules. This would give a better sense of what level of biochemical characterization could be performed on sub-populations of granules.

Reviewer #3 (Public Review):

The manuscript from Barekatain et al. is investigating heterogeneity within the population of insulin vesicles from an insulinoma cell line (INS-1E) in response to glucose stimulation. Prevailing dogma in the beta-cell field suggests that there are distinct pools of mature insulin granules, such as ready-releasable and a reserve pool, which contribute to distinct phases of insulin release in response to glucose stimulation. Whether these pools (and others) are distinct in protein/lipid composition or other aspects is not known, but has been suggested. In this manuscript, the authors use density gradient sedimentation to enrich for insulin vesicles, noting the existence of a number of co-purifying contaminants (ER and mitochondrial markers). Following immunolabeling with synaptotagmin V and fluorescent-conjugated secondary antibodies, insulin vesicles were applied to a microfluidic device and separated by dielectrophoretic and electrokinetic forces following an applied voltage. The equilibrium between these opposing forces was used to physically separate insulin granules. Here some differences were observed in the insulin (Syt V positive) granule populations, when isolated from cells that were either non-stimulated or stimulated with glucose, which has been suggested previously by other studies as noted by the authors; however in the current manuscript, the inclusion of a number of control experiments may provide a better context for what the data reveal about these changes.

The major strength of the paper is in the use of the novel, highly sophisticated methodology to examine physical attributes of insulin granules and thus begin to provide some insight into the existence of distinct insulin granule populations within a beta-cell -these include insulin granules that are maturing, membrane-docked (i.e. readily releasable), in reserve, newly-synthesized, aged, etc. Whether physical differences exist between these various granule pools is not known. In this capacity, the technical abilities of the current manuscript may begin to offer some insight into whether these perceived distinctions are physical.

The major weakness of the manuscript is that the study falls short in terms of linking the biology to the sophisticated changes observed and primarily focuses on differences in response to glucose. Without knowing what the various populations of granules are, it is challenging to understand what the changes in response to glucose mean.

Specific concerns are as follows:

(1) There is confusion on what the DC-iDEP separation between stimulated and stimulated cells reveals. Do these changes reflect maturation state of granules, nascent vs. old granules? Ready-releasable vs. reserve pool? The comments in the text seem to offer all possibilities.

(2) It is unclear what we can infer regarding the physical changes of granules between the stimulated states of the cells. Without an understanding of the magnitude of the effect, it is unclear how biologically significant these changes are. For example, what degree of lipid or protein remodeling would be necessary to give a similar change?

(3) The reliance on a single vesicle marker, Syt V, is concerning given that granule remodeling is the focus.

(4) Additional confirmation that the isolated vesicles are in fact insulin granules would be helpful. As noted, granules were gradient enriched, but did carry contaminants. Note that the microscopy image provided does not provide any real validation for this marker.

Further confirmation that the immune-isolated vesicles are in fact insulin granules should be included. EM with immunogold labeling post-SytV enrichment would be a potential methodology to confirm.

(5) It would be useful to understand if the observed effects are specific to the INS-1E cell line or are a more universal effect of glucose on beta-cells.

eLife assessment

Using continuum theory of elastic solids the authors present evidence that periodic muscle contraction leads to elongation of C. elegans embryos by storing elastic energy that is subsequently released by extending the embryo's long axis. This important finding could apply to other developmental processes and be exploited in soft robotics. The presented evidence is convincing on the phenomenological level adopted in the work. How bending energy is converted into elongation on a more microscopic level remains to be worked out.

eLife assessment

This is an important computational study that applies the machine learning method of bilinear modeling to the problem of relating gene expression to connectivity. Specifically, the author attempts to use transcriptomic data from mouse retinal neurons to predict their known connectivity with promising results. On revision, the approach was tested against a second data set from C. elegans. A limited number of genes studied in this second dataset may have resulted in performance that matched but did not exceed prior models. However, taken together, the results were felt to provide solid evidence for the value of the approach.

eLife assessment

In this important study, the authors report a novel measurement of the Escherichia coli chemotactic response and demonstrate that these bacteria display an attractant response to potassium, which is connected to intracellular pH level. The experimental evidence provided is convincing and the work will be of interest to microbiologists studying chemotaxis.

Author response:

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

eLife assessment

In this important study, the authors report a novel measurement of the Escherichia coli chemotactic response and demonstrate that these bacteria display an attractant response to potassium, which is connected to intracellular pH level. Whilst the experiments are mostly convincing, there are some confounders regards pH changes and fluorescent proteins that remain to be addressed.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This paper shows that E. coli exhibits a chemotactic response to potassium by measuring both the motor response (using a bead assay) and the intracellular signaling response (CheY phosporylation level via FRET) to step changes in potassium concentration. They find increase in potassium concentration induces a considerable attractant response, with amplitude comparable to aspartate, and cells can quickly adapt (and generally over-adapt). The authors propose that the mechanism for potassium response is through modifying intracellular pH; they find both that potassium modifies pH and other pH modifiers induce similar attractant responses. It is also shown, using Tar- and Tsr-only mutants, that these two chemoreceptors respond to potassium differently. Tsr has a standard attractant response, while Tar has a biphasic response (repellent-like then attractant-like). Finally, the authors use computer simulations to study the swimming response of cells to a periodic potassium signal secreted from a biofilm and find a phase delay that depends on the period of oscillation.

Strengths:

The finding that E. coli can sense and adapt to potassium signals and the connection to intracellular pH is quite interesting and this work should stimulate future experimental and theoretical studies regarding the microscopic mechanisms governing this response. The evidence (from both the bead assay and FRET) that potassium induces an attractant response is convincing, as is the proposed mechanism involving modification of intracellular pH. The updated manuscript controls for the impact of pH on the fluorescent protein brightness that can bias the measured FRET signal. After correction the response amplitude and sharpness (hill coefficient) are comparable to conventional chemoattractants (e.g. aspartate), indicating the general mechanisms underlying the response may be similar. The authors suggest that the biphasic response of Tar mutants may be due to pH influencing the activity of other enzymes (CheA, CheR or CheB), which will be an interesting direction for future study.

Weaknesses:

The measured response may be biased by adaptation, especially for weak potassium signals. For other attractant stimuli, the response typically shows a low plateau before it recovers (adapts). In the case of potassium, the FRET signal does not have an obvious plateau following the stimuli of small potassium concentrations, perhaps due to the faster adaptation compared to other chemoattractants. It is possible cells have already partially adapted when the response reaches its minimum, so the measured response may be a slight underestimate of the true response. Mutants without adaptation enzymes appear to be sensitive to potassium only at much larger concentrations, where the pH significantly disrupts the FRET signal; more accurate measurements would require development of new mutants and/or measurement techniques.

We acknowledge and appreciate the reviewer's concerns regarding the potential impact of adaptation on the measured response magnitude. We have estimated the effect of adaptation on the measured response magnitude. The half-time of adaptation at 30 mM KCl was measured to be approximately 80 s, corresponding to a time constant of t = 80/ln(2) = 115.4 s, which is significantly longer than the time required for medium exchange in the flow chamber (less than 10 s). Consequently, the relative effect of adaptation on the measured response magnitude should be less than 1-exp(-10/t) = 8.3%. Even for the fastest adaptation (at the lowest KCl concentration) we measured, the effect should be less than 20%, which is within experimental uncertainties. Nevertheless, we agree that developing new techniques to measure the dose-response curve more precisely would be beneficial.

Reviewer #2 (Public Review):

Zhang et al investigated the biophysical mechanism of potassium-mediated chemotactic behavior in E coli. Previously, it was reported by Humphries et al that the potassium waves from oscillating B subtilis biofilm attract P aeruginosa through chemotactic behavior of motile P aeruginosa cells. It was proposed that K+ waves alter PMF of P aeruginosa. However, the mechanism was this behaviour was not elusive. In this study, Zhang et al demonstrated that motile E coli cells accumulate in regions of high potassium levels. They found that this behavior is likely resulting from the chemotaxis signalling pathway, mediated by an elevation of intracellular pH. Overall, a solid body of evidence is provided to support the claims. However, the impacts of pH on the fluorescence proteins need to be better evaluated. In its current form, the evidence is insufficient to say that the fluoresce intensity ratio results from FRET. It may well be an artefact of pH change.

The authors now carefully evaluated the impact of pH on their FRET sensor by examining the YFP and CFP fluorescence with no-receptor mutant. The authors used this data to correct the impact of pH on their FRET sensor. This is an improvement, but the mathematical operation of this correction needs clarification. This is particularly important because, looking at the data, it is not fully convincing if the correction was done properly. For instance, 3mM KCl gives 0.98 FRET signal both in Fig3 and FigS4, but there is almost no difference between blue and red lines in Fig 3. FigS4 is very informative, but it does not address the concern raised by both reviewers that FRET reporter may not be a reliable tool here due to pH change.

We apologize for not making the correction process clear. We corrected the impact of pH on the original signals for both CFP and YFP channels by

where and represent the pH-corrected and original PMT signal (CFP or YFP channel) from the moment of addition of L mM KCl to the moment of its removal, respectively, and  is the correction factor, which is the ratio of PMT signal post- to pre-KCl addition for the no-receptor mutant at L mM KCl, for CFP or YFP channel as shown Fig. S5. The pH-corrected FRET response is then calculated as the ratio of the pH-corrected YFP to the pH-corrected CFP signals, normalized by the pre-stimulus ratio.

As shown in Author response image1, which represents the same data as Fig. 3A and Fig. S5A, the original normalized FRET responses to 3 mM KCl are 0.967 for the wild-type strain (Fig. 3) and 0.981 for the no-receptor strain (Fig. S5). The standard deviation of the FRET values under steady-state conditions is 0.003. Thus, the difference in responses between the wild-type and no-receptor strains is significant and clearly exceeds the standard deviation. The pH correction factors CpH at 3 mM KCl are 1.004 for the YFP signal and 1.016 for the CFP signal. Consequently, the pH-corrected FRET responses are 0.967´1.016/1.004=0.979 for the wild-type and 0.981´1.016/1.004=0.993 for the no-receptor strain. The reason the pH-corrected FRET response for the no-receptor strain is 0.993 instead of the expected 1.000 is that this value represents the lowest observed response rather than the average value for the FRET response.

The detailed mathematical operation for correcting the pH impact has now been included in the “FRET assay” section of Materials and Methods.

Author response image 1.

Chemotactic response of the wild-type strain (A, HCB1288-pVS88) and the no-receptor strain (B, HCB1414-pVS88) to stepwise addition and removal of KCl. The blue solid line denotes the original normalized signal. Downward and upward arrows indicate the time points of addition and removal of 3 mM KCl, respectively. The horizontal red dashed line denotes the original normalized FRET response value to 3 mM KCl.

The authors show the FRET data with both KCl and K2SO4, concluding that the chemotactic response mainly resulted from potassium ions. However, this was only measured by FRET. It would be more convincing if the motility assay in Fig1 is also performed with K2SO4. The authors did not address this point. In light of complications associated with the use of the FRET sensor, this experiment is more important.

We thank the reviewer for the suggestion. We agree that additional confirmation with a motility assay is important. To address this, we have now measured the response of the motor rotational signal to 15 mM K2SO4 using the bead assay and compared it with the response to 30 mM KCl. The results are shown in Fig. S2. The response of motor CW bias to 15 mM K2SO4 exhibited an attractant response, characterized by a decreased CW bias upon the addition of K2SO4, followed by an over-adaptation that is qualitatively similar to the response to 30 mM KCl. However, there were notable differences in the adaptation time and the presence of an overshoot. Specifically, the adaptation time to K2SO4 was shorter compared to that for KCl, and there was a notable overshoot in the CW bias during the adaptation phase. These differences may have resulted from the weaker response to K2SO4 (Fig. S1B) and additional modifications due to CysZ-mediated cellular uptake of sulfate (Zhang et al., Biochimica et Biophysica Acta 1838,1809–1816 (2014)). The faster adaptation and overshoot complicated the chemotactic drift in the microfluidic assay as in Fig. 1, such that we were unable to observe a noticeable drift in a K2SO4 gradient under the same experimental conditions used for the KCl gradient.

The response of motor rotational signal to 15 mM K2SO4 has been added to Fig. S2.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

(1) The response curve and adaptation level/time in the main text (Fig. 4) should be replaced by the corrected counterparts (currently in Fig. S5). The current version is especially confusing because Fig. 6 shows the corrected response, but the difference from Fig. 4 is not mentioned.

We thank the reviewer for the suggestion. We have now merged the results of the original Fig. S5 into Fig. 4.

a. The discussion of the uncorrected response with small hill coefficient and potentially negative cooperativity was left in the text (lines 223-234), but the new measurements show this is not true for the actual response. This should be removed or significantly rephrased.

We thank the reviewer for the suggestion. We have now removed the statement about potentially negative cooperativity and added the corrected results for the actual response.

(2) It may be helpful to restate the definition of f_m in the methods (near Eq. 3-4).

Thank you for the suggestion. We have now restated the definition of fm and fL below Eq. 3-4: “In the denominator on the right-hand side of Eq. 3, the two terms within the parentheses of exponential expression represent the methylation-dependent (fm) and ligand-dependent (fL) free energy, respectively.”

Reviewer #1 (Public Review):

Summary:

This paper shows that E. coli exhibits a chemotactic response to potassium by measuring both the motor response (using a bead assay) and the intracellular signaling response (CheY phosporylation level via FRET) to step changes in potassium concentration. They find increase in potassium concentration induces a considerable attractant response, with amplitude comparable to aspartate, and cells can quickly adapt (and generally over-adapt). The authors propose that the mechanism for potassium response is through modifying intracellular pH; they find both that potassium modifies pH and other pH modifiers induce similar attractant responses. It is also shown, using Tar- and Tsr-only mutants, that these two chemoreceptors respond to potassium differently. Tsr has a standard attractant response, while Tar has a biphasic response (repellent-like then attractant-like). Finally, the authors use computer simulations to study the swimming response of cells to a periodic potassium signal secreted from a biofilm and find a phase delay that depends on the period of oscillation.

Strengths:

The finding that E. coli can sense and adapt to potassium signals and the connection to intracellular pH is quite interesting and this work should stimulate future experimental and theoretical studies regarding the microscopic mechanisms governing this response. The evidence (from both the bead assay and FRET) that potassium induces an attractant response is convincing, as is the proposed mechanism involving modification of intracellular pH. The updated manuscript controls for the impact of pH on the fluorescent protein brightness that can bias the measured FRET signal. After correction the response amplitude and sharpness (hill coefficient) are comparable to conventional chemoattractants (e.g. aspartate), indicating the general mechanisms underlying the response may be similar. The authors suggest that the biphasic response of Tar mutants may be due to pH influencing the activity of other enzymes (CheA, CheR or CheB), which will be an interesting direction for future study.

Weaknesses:

The measured response may be biased by adaptation, especially for weak potassium signals. For other attractant stimuli, the response typically shows a low plateau before it recovers (adapts). In the case of potassium, the FRET signal does not have an obvious plateau following the stimuli of small potassium concentrations, perhaps due to the faster adaptation compared to other chemoattractants. It is possible cells have already partially adapted when the response reaches its minimum, so the measured response may be a slight underestimate of the true response. Mutants without adaptation enzymes appear to be sensitive to potassium only at much larger concentrations, where the pH significantly disrupts the FRET signal; more accurate measurements would require the development of new mutants and/or measurement techniques.

Note added after the second revision: The authors made a reasonable argument regarding the effects of adaptation, which were estimated to be small.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The manuscript "Self-inhibiting percolation and viral spreading in epithelial tissue" describes a model based on 5-state cellular automata of development of an infection. The model is motivated and qualitatively justified by time-resolved measurements of expression levels of viral, interferon-producing, and antiviral genes. The model is set up in such a way that the crucial difference in outcomes (infection spreading vs. confinement) depends on the initial fraction of special virus-sensing cells. Those cells (denoted as 'type a') cannot be infected and do not support the propagation of infection, but rather inhibit it in a somewhat autocatalytic way. Presumably, such feedback makes the transition between two outcomes very sharp: a minor variation in concentration of ``a' cells results in qualitative change from one outcome to another. As in any percolation-like system, the transition between propagation and inhibition of infection goes through a critical state with all its attributes. A power-law distribution of the cluster size (corresponding to the fraction of infected cells) with a fairly universal exponent and a cutoff at the upper limit of this distribution.

Strengths:

The proposed model suggests an explanation for the apparent diversity of outcomes of viral infections such as COVID.

Author response: We thank the referee for the concise and accurate summary of our work.

Weaknesses:

Those are not real points of weakness, though I think addressing them would substantially improve the manuscript.

Author response: Below we will address these point by point.

The key point in the manuscript is the reduction of actual biochemical processes to the NOVAa rules. I think more could be said about it, be it referring to a set of well-known connections between expression states of cells and their reaction to infection or justifying it as an educated guess.

Author response: We have now improved this part in the model section. We have added a few sentences explaining how the cell state transitions are motivated by the UMAP results:

“The cell state transitions triggered by IFN signaling or viral replication are known in viral infection, but how exactly the transitions are orchestrated for specific infections is poorly understood. The UMAP cell state distribution hints at possible preferred transitions between states. The closer two cell states are on the UMAP, the more likely transitions between them are, all else being equal. For instance, the antiviral state (𝐴) is easily established from a susceptible cell (𝑂), but not from the fully virus-hijacked cell (𝑉 ). The IFN-secreting cell state (𝑁) requires the co-presence of the viral and antiviral genes and thus the cell cluster is located between the antiviral state (𝐴) and virus-infected state (𝑉 ) but distant from the susceptible cells (𝑂).

Inspired by the UMAP data visualization (Fig. 1a), we propose the following transitions between five main discrete cell states”

Another aspect where the manuscript could be improved would be to look a little beyond the strange and 'not-so-relevant for a biomedical audience' focus on the percolation critical state. While the presented calculation of the precise percolation threshold and the critical exponent confirm the numerical skills of the authors, the probability that an actual infected tissue is right at the threshold is negligible. So in addition to the critical properties, it would be interesting to learn about the system not exactly at the threshold: For example, how the speed of propagation of infection depends on subcritical p_a and what is the cluster size distribution for supercritical p_a.

Author response: We agree that further exploring the model away from the critical threshold is worthwhile. While our main focus has been on explaining the large degree of heterogeneity in outcomes – readily explained as a consequence of the sharp threshold-like behavior – we now include plots of the time-evolution of the infection (as well as the remaining states) over time for subcritical values of pa. The plots can be found in Figure S4 of the supplement.

Reviewer #2 (Public Review):

Xu et al. introduce a cellular automaton model to investigate the spatiotemporal spreading of viral infection. In this study, the author first analyzes the single-cell RNA sequencing data from experiments and identifies four clusters of cells at 48 hours post-viral infection, including susceptible cells (O), infected cells (V), IFN-secreting cells (N), and antiviral cells (A). Next, a cellular automaton model (NOVAa model) is introduced by assuming the existence of a transient pre-antiviral state (a). The model consists of an LxL lattice; each site represents one cell. The cells change their state following the rules depending on the interaction of neighboring cells. The model introduces a key parameter, p_a, representing the fraction of pre-antiviral state cells. Cell apoptosis is omitted in the model. Model simulations show a threshold-like behavior of the final attack rate of the virus when p_a changes continuously. There is a critical value p_c, so that when p_a < p_c, infections typically spread to the entire system, while at a higher p_a > p_c, the propagation of the infected state is inhibited. Moreover, the radius R that quantifies the diffusion range of N cells may affect the critical value p_c; a larger R yields a smaller value of the critical value p_c. The structure of clusters is different for different values of R; greater R leads to a different microscopic structure with fewer A and N cells in the final state. Compared with the single-cell RNA seq data, which implies a low fraction of IFN-positive cells - around 1.7% - the model simulation suggests R=5. The authors also explored a simplified version of the model, the OVA model, with only three states. The OVA model also has an outbreak size. The OVA model shows dynamics similar to the NOVAa model. However, the change in microstructure as a function of the IFN range R observed in the NOVAa model is not observed in the OVA model.

Author response: We thank the referee for the comprehensive summary of our work.

Data and model simulation mainly support the conclusions of this paper, but some weaknesses should be considered or clarified.

Author response: Thank you - we will address these point by point below.

(1) In the automaton model, the authors introduce a parameter p_a, representing the fraction of pre-antiviral state cells. The authors wrote: ``The parameter p_a can also be understood as the probability that an O cell will switch to the N or A state when exposed to the virus of IFNs, respectively.' Nevertheless, biologically, the fraction of pre-antiviral state cells does not mean the same value as the probability that an O cell switches to the N or A state. Moreover, in the numerical scheme, the cell state changes according to the deterministic role N(O)=a and N(a)=A. Hence, the probability p_a did not apply to the model simulation. It may need to clarify the exact meaning of the parameter p_a.

Author response: We acknowledge that this was an imprecise formulation, and have now changed it.

What we tried to convey with that comment was that, alternatively to having a certain fraction of cells be in the a state initially, one could instead have devised a model in which We should note that even the current model has a level of stochasticity, since we choose the cells to be updated with a constant probability rate - we choose N cells to update in each timestep, with replacement.

However, based on your suggestion, we simulated a version of the dynamics which included stochastic conversion, i.e. each action of a cell on a nearby cell happens only with a probability p_conv (and the original model is recovered as the p_conv=1 scenario). Of course, this slows down the dynamics (or effectively rescales time by a factor p_conv), but crucially we find that it does not appreciably affect the location of the threshold p_c. Below we include a parameter scan across p_a values for R=1 and p_conv=0.5, which shows that the threshold continues to appear at around p_a=27%. each O-state cell simply had a probability to act as an a-state cell upon exposure to the virus or to interferons, i.e. to switch to an N state (if exposed to virus) or to the A state (if exposed to interferons). In this simplified model, there would be no functional difference, since it would simply amount to whether each cell had a probability to be designated an a-cell initially (as in our model), or upon exposure. So our remark mainly served to explain that the role of the p_a parameter is simply to encode that a certain fraction of virus-naive cells behave this way (whether predetermined or not).

(2) The current model is deterministic. However, biologically, considering the probabilistic model may be more realistic. Are the results valid when the probability update strategy is considered? By the probability model, the cells change their state randomly to the state of the neighbor cells. The probability of cell state changes may be relevant for the threshold of p_a. It is interesting to know how the random response of cells may affect the main results and the critical value of p_a.

Author response: This is a good point - we are firm believers in the importance of stochasticity. We should note that even the current model has a level of stochasticity, since we choose the cells to be updated with a constant probability rate - we choose N cells to update in each timestep, with replacement.

However, based on your suggestion, we simulated a version of the dynamics which included stochastic conversion, i.e. each action of a cell on a nearby cell happens only with a probability p_conv (and the original model is recovered as the p_conv=1 scenario). Of course, this slows down the dynamics (or effectively rescales time by a factor p_conv), but crucially we find that it does not appreciably affect the location of the threshold p_c. Below we include a parameter scan across p_a values for R=1 and p_conv=0.5, which shows that the threshold continues to appear at around p_a=27%.

We now discuss these findings in the supplement and include the figure below as Fig. S5.

Author response image 1.

(3) Figure 2 shows a critical value p_c = 27.8% following a simulation on a lattice with dimension L = 1000. However, it is unclear if dimension changes may affect the critical value.

Author response: Re-running the simulations on a lattice 4x as large (i.e. L=2000) yields a similar critical value of 27-28% for R=1, so we are confident that finite size effects do not play a major role at L=1000 and beyond. For R=5, however, we find that a minimum lattice size greater than L=1000 is necessary to determine the critical threshold. Concretely, we find that the threshold value pc for R=5 changes somewhat when the lattice size is increased from 1000 to 2000, but is invariant under a change from 2000 to 3000, so we conclude that L=2000 is sufficient for R=5. The pc value for R=5 cited in the manuscript (~0.4%) was determined from simulations at L=2000.

Reviewer #3 (Public Review):

Summary:

This study considers how to model distinct host cell states that correspond to different stages of a viral infection: from naïve and susceptible cells to infected cells and a minority of important interferon-secreting cells that are the first line of defense against viral spread. The study first considers the distinct host cell states by analyzing previously published single-cell RNAseq data. Then an agent-based model on a square lattice is used to probe the dependence of the system on various parameters. Finally, a simplified version of the model is explored, and shown to have some similarity with the more complex model, yet lacks the dependence on the interferon range. By exploring these models one gains an intuitive understanding of the system, and the model may be used to generate hypotheses that could be tested experimentally, telling us "when to be surprised" if the biological system deviates from the model predictions.

Author response: Thank you for the summary! We agree with the role that you describe for a model such as this one.

Strengths:

-  Clear presentation of the experimental findings and a clear logical progression from these experimental findings to the modeling.

-  The modeling results are easy to understand, revealing interesting behavior and percolation-like features.

-  The scaling results presented span several decades and are therefore compelling. - The results presented suggest several interesting directions for theoretical follow-up work, as well as possible experiments to probe the system (e.g. by stimulating or blocking IFN secretion).

Weaknesses:

-  Since the "range" of IFN is an important parameter, it makes sense to consider lattice geometries other than the square lattice, which is somewhat pathological. Perhaps a hexagonal lattice would generalize better.

-  Tissues are typically three-dimensional, not two-dimensional. (Epithelium is an exception). It would be interesting to see how the modeling translates to the three-dimensional case. Percolation transitions are known to be very sensitive to the dimensionality of the system.

Author response: We agree that probing different lattice geometries (2- and 3-dimensional alike) would be interesting and worthwhile. However, for this manuscript, we prefer to confine the analysis to the current, simple case. We do agree, however, that an extensive exploration of the role of geometry is an interesting future possibility.

-  The fixed time-step of the agent-based modeling may introduce biases. I would consider simulating the system with Gillespie dynamics where the reaction rates depend on the ambient system parameters.

-  Single-cell RNAseq data typically involves data imputation due to the high sparsity of the measured gene expression. More information could be provided on this crucial data processing step since it may significantly alter the experimental findings.

Justification of claims and conclusions:

The claims and conclusions are well justified.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

It is necessary to explain what UMAP does. Is clustering done in the space of twenty-something original dimensions or 2D? How UMAP1 and UMAP2 are selected and are those the same in all plots?

Author response: We have now added a few sentences to clarify the point raised above - the second snippet explains how clustering is performed:

“As a dimension reduction algorithm, UMAP is a manifold learning technique that favors the preservation of local distances over global distances (McInnes et al., 2018; Becht et al., 2019). It constructs a weighted graph from the data points and optimizes the graph layout in the low-dimensional space.”

“We cluster the cells with the principal components analysis (PCA) results from their gene expression. With the first 16 principal components, we calculate k-nearest neighbors and construct the shared nearest neighbor graph of the cells then optimize the modularity function to determine clusters. We present the cluster information on the UMAP plane and use the same UMAP coordinates for all the plots in this paper hereafter.”

Figure 1, what do bars in the upper right corners of panels d,e,f, and g indicate? ``Averaged' refers to time average? Something is missing in ``Cell proportions are labeled with corresponding colors in a)' .

Author response: Thank you - we have now modified the figure caption. The bars in the upper right corners of panels d, e, f are color keys for gene expression, the brighter the color is, the higher the gene expression is.

“Averaged” gene expression refers to the mean expression of that particular gene across the cells within each indicated cluster.

The lines in c) correspond to cell proportions in different states at different time points. The same state in 1) and c) is shown in the same color.

Line 46, ``However' does not sound right in this context. Would ``Also' be better?

Author response: We agree and have corrected it in the revised manuscript.

Line 96``The viral genes are also partially expressed in these cells, but different from the 𝑁 cluster, the antiviral genes are fully expressed (Fig. S1 and S2).' The sentence needs to be rephrased.

Author response: We have rephrased the sentence: “As in the N cluster, the viral gene E is barely detected in these cells, indicating incomplete viral replication. However, in contrast to the N cluster, the antiviral genes are expressed to their full extent (Fig. S1 and S2).”

Line 126, missing "be", ``large' -> ``larger'.

Author response: Thank you, we have now corrected these typos.

Line 139-140 The logical link between ignoring apoptosis and the diffusion of IFN is unclear.

Author response: We modified the sentence as “Here, we assume that the secretion of IFNs by the 𝑁 cells is a faster process than possible apoptosis (Wen et al., 1997; Tesfaigzi, 2006) of these cells and that the diffusion of IFNs to the neighborhood is not significantly affected by apoptosis.”

Fig. 2a Do the yellow arrows show the effect of IFN and the purple arrows the propagation of viral infection?

Author response: That is correct. We have added this information to the figure caption: “The straight black arrows indicate transitions between cell states. The curved yellow arrows indicate the effects of IFNs on activating antiviral states. The curved purple arrows indicate viral spread to cells with 𝑂 and 𝑎 states.”

Fig. 3, n(s) as the axis label vs P(s) in the text? How do the curves in panel a) look when the p_a is well above or below p_c?

Author response: Thank you. We have edited the labels in the figure to reflect the symbols used in the text.

Boundary conditions? From Fig. 4, apparently periodic?

Author response: Yes, we use periodic boundary conditions in the model. We clarify it in the model section now (last sentence).

It will be good to see a plot with time dependences of all cell types for a couple of values of p_a, illustrating propagation and cessation of the infection.

Author response: We agree, and have added a Figure S4 in the supplement which explores exactly that. Thank you for the suggestion.

A verbal qualitative description of why p_a has such importance and how the infection is terminated for large p_a would help.

Reviewer #2 (Recommendations For The Authors):

Below are two minor comments:

(1) In the single-cell RNA sequencing data analysis, the authors describe the cell clusters O, V, A, and N. However, showing how the clusters are identified from the data might be more straightforward.

Author response: Technically, we cluster the cells using principal components analysis (PCA) results of their gene expression. With the first 16 principal components, we calculate k-nearest neighbors and construct the shared nearest neighbor graph of the cells and then optimize the modularity function to determine clusters. We manually annotate the clusters with O, V, A, and N based on the detected abundance of viral genes, antiviral genes, and IFNs.

(2) In Figure 3, what does n(s) mean in Figure 3a? And what is the meaning of the distribution P(s) of infection clusters? It may be stated clearly.

Author response: The use of n(s) was inconsistent, and we have now edited the figure to instead say P(s), to harmonize it with the text. P(s) is the distribution of cluster sizes, s, expressed as a fraction of the whole system. In other words, once a cluster has reached its final size, we record s=(N+V)/L^2 where N and V are the number of N and V state cells in the cluster (note that, by design, each simulation leads to a single cluster, since we seed the infection in one lattice point). We now indicate more clearly in the caption and the main text what exactly P(s) and s refer to.

Reviewer #3 (Recommendations For The Authors):

- Would the authors kindly share the simulation code with the community? Also, the data analysis code should be shared to follow current best practices. This needs to be standard practice in all publications. I would go as far as to say that in 2024 publishing a data analysis / simulation study without sharing the relevant code should be ostracized by the community.

Author response: We absolutely agree and have created a GitHub repository in which we share the C++ source code for the simulations and a Python notebook for plotting. The public repository can be found at https://github.com/BjarkeFN/ViralPercolation. We add this information in supplement under section “Code availability”.

­

- I would avoid the use of the wording "critical" threshold since this is almost guaranteed to infuriate a certain type of reader.

­

- Line 265 has a curious use of " ... " which should be replaced with something more appropriate.

Author response: Thank you for pointing it out! We have checked the typos.

Reviewer #1 (Public Review):

Summary:

The manuscript describes a model based on 5-state cellular automata of development of an infection. The model is motivated and qualitatively justified by time-resolved measurements of expression levels of viral, interferon-producing, and antiviral genes. The model is set up in such a way that the crucial difference in outcomes (infection spreading vs. confinement) depends on the initial fraction of special virus-sensing cells. Those cells (denoted as 'type a') cannot be infected and do not support the propagation of infection, but rather inhibit it in a somewhat autocatalytic way. Presumably, such feedback makes the transition between two outcomes very sharp: a minor variation in concentration of 'a' cells results in qualitative change from one outcome to another. As in any percolation-like system, the transition between propagation and inhibition of infection goes through a critical state with all its attributes, including a power-law distribution of the cluster size (corresponding to the fraction of infected cells) with a fairly universal exponent and a cutoff at the upper limit of this distribution.

Strengths:

The proposed model suggests a well-justified explanation for the frequently observed yet puzzling diversity of outcomes of viral infections such as COVID.

Weaknesses:

None.

eLife assessment

This study presents a cellular automaton model to study the dynamics of virus-induced signalling and innate host defense against viruses such as SARS-CoV-2 in epithelial tissue. The simulations and data analysis are convincing and represent a valuable contribution that would be of interest to researchers studying the dynamics of viral propagation.

Reviewer #2 (Public Review):

Xu et al. introduce a cellular automaton model to investigate the spatiotemporal spreading of viral infection. In this study, the author first analyzes the single-cell RNA sequencing data from experiments and identifies four clusters of cells at 48 hours post-viral infection, including susceptible cells (O), infected cells (V), IFN-secreting cells (N), and antiviral cells (A). Next, a cellular automaton model (NOVAa model) is introduced by assuming the existence of a transient pre-antiviral state (a). The model consists of an LxL lattice; each site represents one cell. The cells change their state following the rules depending on the interaction of neighboring cells. The model introduces a key parameter, p_a, representing the fraction of pre-antiviral state cells. Cell apoptosis is omitted in the model. Model simulations show a threshold-like behavior of the final attack rate of the virus when p_a changes continuously. There is a critical value p_c, so that when p_a < p_c, infections typically spread to the entire system, while at a higher p_a > p_c, the propagation of the infected state is inhibited. Moreover, the radius R that quantifies the diffusion range of N cells may affect the critical value p_c; a larger R yields a smaller value of the critical value p_c. The authors further examine the result with stochastic version dynamics, and the main findings are unchanged upon stochastic dynamics. The structure of clusters is different for different values of R; greater R leads to a different microscopic structure with fewer A and N cells in the final state. Compared with the single-cell RNA seq data, which implies a low fraction of IFN-positive cells of around 1.7%, the model simulation suggests R=5. The authors also explored a simplified version of the model, the OVA model, with only three states. The OVA model also has an outbreak size. The OVA model shows dynamics similar to the NOVAa model. However, the change in microstructure as a function of the IFN range R observed in the NOVAa model is not observed in the OVA model.

Reviewer #3 (Public Review):

Summary:

This study considers how to model distinct host cell states that correspond to different stages of a viral infection: from naïve and susceptible cells to infected cells and a minority of important interferon-secreting cells that are the first line of defense against viral spread. The study first considers the distinct host cell states by analyzing previously published single-cell RNAseq data. Then an agent-based model on a square lattice is used to probe the dependence of the system on various parameters. Finally, a simplified version of the model is explored, and shown to have some similarity with the more complex model, yet lacks the dependence on the interferon range. By exploring these models one gains an intuitive understanding of the system, and the model may be used to generate hypotheses that could be tested experimentally, telling us "when to be surprised" if the biological system deviates from the model predictions.

Strengths:

- Clear presentation of the experimental findings and a clear logical progression from these experimental findings to the modeling.<br /> - The modeling results are easy to understand, revealing interesting behavior and percolation-like features.<br /> - The scaling results presented span several decades and are therefore compelling.<br /> - The results presented suggest several interesting directions for theoretical follow-up work, as well as possible experiments to probe the system (e.g. by stimulating or blocking IFN secretion).

Weaknesses:

- The fixed time-step of the agent-based modeling may introduce biases. I would consider simulating the system with Gillespie dynamics where the reaction rates depend on the ambient system parameters.<br /> - Single-cell RNAseq data requires careful handling or it may generate false leads. The strength of the RNAseq evidence presented is not clear.

Two places where the manuscript could be extended:

- Since the "range" of IFN is an important parameter, it makes sense to consider other lattice geometries other than the square lattice, which is somewhat pathological. Perhaps a hexagonal lattice would generalize better.<br /> - Tissues are typically three-dimensional, not two-dimensional. (Epithelium is an exception). It would be interesting to see how the modeling translates to the three-dimensional case. Percolations transitions are known to be very sensitive to the dimensionality of the system.

Justification of claims and conclusions:

The claims and conclusions are well justified.

eLife assessment

This valuable study reports that actin-related proteins may be involved in transcriptional regulation during spermatogenesis. The supporting data remain incomplete, and more extensive disentanglement from the canonical role of these actin-related proteins and the experimental validation of in silico predictions are required. This work will be of interest to reproductive biologists and other researchers working on non-canonical roles of actin and actin-related proteins.

Reviewer #1 (Public Review):

Summary:

This study offers a new perspective. ACTL7A and ACTL7B play roles in epigenetic regulation in spermiogenesis. Actin-like 7 A (ACTL7A) is essential for acrosome formation, fertilization, and early embryo development. ACTL7A variants cause acrosome detachment responsible for male infertility and early embryonic arrest. It has been reported that ACTL7A is localized on the acrosome in mouse sperms (Boëda et al., 2011). Previous studies have identified ACTL7A mutations (c.1118G>A:p.R373H; c.1204G>A:p.G402S, c.1117C>T:p.R373C), All these variants were located in the actin domain and were predicted to be pathogenic, affecting the number of hydrogen bonds or the arrangement of nearby protein structures (Wang et al., 2023; Xin et al., 2020; Zhao et al., 2023; Zhou et al., 2023). This work used AI to model the role of ACTL7A/B in the nucleosome remodeling complex and proposed a testis-specific conformation of SCRAP complex. This is different from previous studies.

Strengths:

This study provides a new perspective to reveal the additional roles of these proteins.

Weaknesses:

The results section contains a substantial background description. However, the results and discussion sections require streamlining. There is a lack of mutual support for data between the sections, and direct data to support the authors' conclusions are missing.

Reviewer #2 (Public Review):

Summary:

How dynamics of gene expression accompany cell fate and cellular morphological changes is important for our understanding of molecular mechanisms that govern development and diseases. The phenomenon is particularly prominent during spermatogenesis, the process which spermatogonia stem cells develop into sperm through a series of steps of cell division, differentiation, meiosis, and cellular morphogenesis. The intricacy of various aspects of cellular processes and gene expression during spermatogenesis remains to be fully understood. In this study, the authors found that testis-specific actin-related proteins (which usually participate in modifying cells' cytoskeletal systems) ACTL7A and ACTL7B were expressed and localized in the nuclei of mouse spermatocytes and spermatids. Based on this observation, the authors analyzed protein sequence conservations of ACTL7B across dozens of species and identified a putative nuclear localization sequence (NLS) that is often responsible for the nuclear import of proteins that carry them. Using molecular biology experiments in a heterologous cell system, the authors verified the potential role of this internal NLS and found it indeed could facilitate the nuclear localization of marker proteins when expressed in cells. Using gene-deleted mouse models they generated previously, the authors showed that deletion of Actl7b caused changes in gene expression and mis-localization of nucleosomal histone H3 and chromatin regulator histone deacetylase HDAC1 and 2, supporting their proposed roles of ACTL7B in regulating gene expression. The authors further used alpha-Fold 2 to model the potential protein complexes that could be formed between the ARPs (ACTL7A and ACTL7B) and known chromatin modifiers, such as INO80 and SWI/SNF complexes and found that consistent with previous findings, it is likely that ACTL7A and ACTL7B interact with the chromatin-modifying complexes through binding to their alpha-helical HSA domain cooperatively. These results suggest that ACTL7B possesses novel functions in regulating chromatin structure and thus gene expression beyond conventional roles of cytoskeleton regulation, providing alternative pathways for understanding how gene expression is regulated during spermatogenesis and the etiology of relevant infertility diseases.

Strengths:

The authors provided sufficient background to the study and discussions of the results. Based on their previous research, this study utilized numerous methods, including protein complex structural modeling method alpha-fold 2 Multimers, to further investigate the functional roles of ACTL7B. The results presented here are in general of good quality. The identification of a potential internal NLS in ACTL7B is mostly convincing, in line with the phenotypes presented in the gene deletion model.

Weaknesses:

While the study offered an interesting new look at the functions of ARP proteins during spermatogenesis, some of the study is mainly theoretical speculations, including the protein complex formation. Some of the results may need further experimental verifications, for example, differentially expressed genes that were found in potentially spermatogenic cells at different developmental stages, in order to support the conclusions and avoid undermining the significance of the study.

Reviewer #3 (Public Review):

In this manuscript, Pierre Ferrer and colleagues explore the exciting possibility that, in the male germ line, the composition and function of deeply conserved chromatin remodeling complexes is fine-tuned by the addition of testis-specific actin-related proteins (ARPs). In this regard, the Authors aim to extend previously reported non-canonical (transcriptional) roles of ARPs in somatic cells to the unique developmental context of the germ line. The manuscript is focused on the potential regulatory role in post-meiotic transcription of two ARPs: ACTL7A and ACTL7B (particularly the latter). The canonical function of both testis-specific ARPs in spermatogenesis is well established, as they have been previously shown to be required for the extensive cellular morphogenesis program driving post-meiotic development (spermiogenesis). Disentangling the actual functions of ACTL7A and ACTL7B as transcriptional regulators from their canonical role in the profound morphological reshaping of post-meiotic cells (a process that also deeply impacts nuclear architecture and regulation) represents a key challenge in terms of interpreting the reported findings (see below).

The authors begin by documenting, via fluorescence microscopy, the intranuclear localization of ACTL7B. This ARP is convincingly shown to accumulate in the nucleus of spermatocytes and spermatids. Using a series of elegant reporter-based experiments in a somatic cell line, the authors map the driver of this nuclear accumulation to a potential NLS sequence in the ACTL7B actin-like body domain. Ferrer and colleagues then performed a testicular RNA-seq analysis in ACTL7B KO mice to define the putative role of ACTL7B in male germ cell transcription. They report substantial changes to the testicular transcriptome - particularly the upregulation of several classes of genes - in ACTL7B KO mice. However, wild-type testes were used as controls for this experiment, thus introducing a clear confounding effect to the analysis (ACTL7B KO testes have extensive post-meiotic defects due to the canonical role of ACTL7B in spermatid development). Then, the authors employ cutting-edge AI-driven approaches to predict that both ACTL7A and ACTL7B are likely to bind to four key chromatin remodeling complexes. Although these predictions are based on a robust methodology, they would certainly benefit from experimental validation. Finally, the authors associate the loss of ACTL7B with decreased lysine acetylation and lower levels of the HDAC1 and HDAC3 chromatin remodelers in the nucleus of developing spermatids.

Globally, these data may provide important insight into the unique processes male germ cells employ to sustain their extraordinarily complex transcriptional program. Furthermore, the concept that (comparably younger) testis-specific proteins can be incorporated into ancient chromatin remodeling complexes to modulate their function in the germ line is timely and exciting.

It is my opinion that the manuscript would benefit from additional experimental validation to better support the authors' conclusions. In particular, I believe that addressing two critical points would substantially strengthen the message of the manuscript:

(1) The proposed role of ACTL7B in post-meiotic transcriptional regulation temporally overlaps with the protein's previously reported canonical functions in spermiogenesis (PMID: 36617158 and 37800308). Indeed, the canonical functions of ACTL7B have been shown to have a profound effect at the level of spermatid morphology and to impact nuclear organization. This potentially renders the observed transcriptional deregulation in ACTL7B KO testes an indirect consequence of spermatid morphology defects. I acknowledge that it is experimentally difficult to disentangle the proposed intranuclear roles of ACTL7B from the protein's well-documented cytoplasmic function. Perhaps the generation of a NLS-scrambled ACTL7B variant could offer some insight. In light of the substantial investment this approach would represent, I would suggest, as an alternative, that instead of using wild-type testes as controls for the transcriptome and chromatin localization assays, the authors consider the possibility of using testicular tissue from a mutant with similarly abnormal spermiogenesis but due to transcription-independent defects. This would, in my opinion, offer a more suitable baseline to compare ACTL7B KO testes with.

(2) The manuscript would greatly benefit if experimental validation of the AI-driven predictions were to be provided (in terms of the binding capacity of ACTL7A and ACTL7B to key chromatin remodeling complexes). More so it seems that the authors have the technical expertise / available mass spectrometry data required for this purpose (lines 664-665). Still on this topic, given the predicted interactions of ACTL7A and ACTL7B with the SRCAP, EP400, SMARCA2 and SMARCA4 complexes (Figure 7), it is rather counter-intuitive that the Authors chose for their immunofluorescence assays, in ACTL7B KO testes, to determine the chromatin localization of HDAC1 and HDAC3, rather than that of any of above four complexes.

eLife assessment

The authors develop a novel genetic strategy for specific and comprehensive labeling of axo-axonic cells, also referred to as chandelier cells, in the mouse brain. The approach and analysis are rigorous such that the data convincingly support the key conclusions, including the expanded distribution of axo-axonic cells throughout the brain. This study provides important new information about the distribution of a significant neuronal cell type, as well as new tools for future studies. This work will be of broad interest to neuroscientists who work on the anatomical and functional organization of neural circuits.

Reviewer #2 (Public Review):

Summary:

The goals of this study were to develop a genetic approach that would specifically and comprehensively target axo-axonic cells (AACs) throughout the brain and then to describe the patterns and characteristics of the targeted AACs in multiple, selected brain regions. The investigators have been successful in providing the most complete description of the regional distribution of putative (pAACs) throughout the brain to date. The supporting evidence is convincing, and the findings should serve as a guide for more detailed studies of AACs within each brain region and lead to new insights into their connectivity and functional organization of this important group of GABAergic interneurons.

Strengths:

The study has numerous strengths. A major strength is the development of a unique intersectional genetic strategy that uses cell lineage (Nkx2.1) and molecular (Unc5b or Pthlh) markers to identify AACs specifically and, apparently, nearly completely throughout the mouse brain. While AACs have been described previously in the cerebral cortex, hippocampus and amygdala, there has been no specific genetic marker that selectively identifies all AACs in these regions.

Importantly, the current genetic strategy labels pAACs in additional brain regions, including the claustrum-insular complex, extended amygdala, and several olfactory centers in which AACs have not been previously recognized. In general, the findings provide support for the specificity of the methods for targeting AACs and include several examples of labeling near markers of axon initial segments, providing validation of their AAC identity.

The descriptions and numerous low magnification images of the brain provide a roadmap for subsequent, detailed studies of AACs in numerous brain regions. The overview and summaries of the findings in the Abstract, Introduction and Discussion are particularly clear and helpful in placing the extensive regional descriptions of AACs in context.

Weaknesses:

Considering the unique and striking characteristics of AACs, it would have been ideal to include a clear, high resolution confocal image of an AAC from the Unc5b;Nkx2.1 mouse that would display the beauty of these cells with their numerous cartridges of axon terminals, emanating from a single AAC. While several cells are illustrated, the processes are often obscured by other labeling or the background created by the blue Dapi labeling. A high-resolution image of an isolated cell would not only support the identity of the cells as AACs but also demonstrate the potential advantages of their labeling for more detailed anatomical and neurophysiological studies. High magnification views of the axon terminals adjacent to AnkG-labeled axon initial segments are included and provide strong support for the identity of the cells. However, they cannot convey the extensiveness and patterns of the axonal arborizations of these cells.

The intersectional genetic methods included use of the lineage marker Nkx2.1 with either Unc5b or Pthlh as the molecular marker. As described, the mice with intersectional targeting of Nkx2.1 and Unc5b appear to show the most specific brain-wide labeling for AACs, and the majority of the descriptions are from these mice. The targeting with Nkx2.1 and Pthlh is less convincing and there appears to be a disconnect between the descriptions and the images. While the descriptions emphasize that the labeling is very similar in the two types of mice, the images suggest distinct differences, including labeling of non AACs in striatum and layer 4 of the cortex in the Pthlh;Nkx2.1 mouse, as described in the manuscript. In addition, the Pthlh;Nkx2.1 mouse has higher cell targeting in some regions and fewer labeled cells in others. Perhaps it would be more accurate to present the Pthlh;Nkx2.1 mouse as differing from the Unc5b;Nkx2.1 mouse, but useful for AAC labeling in select regions and under some conditions, such as following tamoxifen administration at specific ages. As currently presented, the inclusion of the Pthlh;Nkx2.1 detracts from the otherwise convincing argument that the Unc5b;Nkx2.1 mouse provides a specific and comprehensive way to identify AACs.

Reviewer #3 (Public Review):

Summary:

Raudales et al. aimed at providing an insight into the brain-wide distribution and synaptic connectivity of bona fide GABAergic inhibitory interneuron subtypes focusing on the axo-axonic cell (AAC), one of the most distinctive interneuron subtypes, which innervates the axon initial segments of glutamatergic projection neurons. They establish intersectional genetic strategies that enable them to specifically and comprehensively capture AACs based on their lineage (Nkx2.1) and marker expression (Unc5b, Pthlh). They find that AACs are deployed across essentially all the pallium-derived brain structures as well as anterior olfactory nucleus, taenia tecta, and lateral septum. They show that AACs in distinct areas and layers of the neocortex as well as different subregions of the hippocampal formation display unique soma and synaptic density and morphological variations. Rabies virus-based retrograde monosynaptic input tracing reveals that AACs in the neocortex, the hippocampus, and the basolateral amygdala receive synaptic inputs from common as well as specific brain regions and supports the utility of this novel genetic approach. This study elucidates brain-wide neuroanatomical features and morphological variations of AACs with solid techniques and analysis. Their novel AAC-targeting strategies will facilitate the study of their development and function in different brain regions. The conclusions in this paper are well supported by the data. However, there are a few minor comments.

(1) The authors added a description about validation of ChCs in the method section: "Validation was conducted with high-magnification confocal microscopy and defined by a cell exhibiting at least two RFP-labelled axons colocalized with AIS labelled by AnkryinG or Phospho-IκBα". However, this does not clearly define pAACs themselves. If they follow this criteria, an RFP-labeled cell exhibiting only one synaptic cartridge that is colocalized with an AIS should be a pAAC. Is this what the authors are triying to say?

On the other hand, in the response to reviewers, the authors apparently define pAACs in a different way, in which they more focus on the number of cells exhibiting cartridges that are associated with AISs in a certain anatomical region rather than the number of cartridges per cell.

"For BNST we did not positively identify more than a few exhibiting overlap with AnkryinG/IκBα, so we currently leave them as pAACs"<br /> "Putative AAC (pAACs) refers to populations in which relatively few single cell examples of AACs exhibiting co-localized cartridges were found"

The authors need to directly define pAACs.

(2) In the response to reviewers, the authors claimed that both Pthlh and Unc5b mice are useful for studying developing AACs. It would be nice if they include this content in the text (e.g. Discussion).

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

In this manuscript, the authors set out to develop genetic tools that can specifically and comprehensively label Axo-Axonic Cells (AACs), also known as Chandelier cells. These AACs possess unique morphological and connectivity features, making them an ideal subject for studying various aspects of cell types across different experimental methods. To achieve both specificity and comprehensiveness in AAC labeling, the authors employ an intersectional strategy that combines lineage origin and molecular markers. This approach successfully targets AACs across the mouse brain and reveals their widespread distribution in various brain structures beyond the previously known regions. Additionally, the authors utilize rabies transneuronal labeling to provide a comprehensive overview of AACs, their variations, and input sources throughout the brain. This experimental approach offers a powerful model system for investigating the role of AACs in circuit development and function across diverse brain regions.

Strengths:

Genetic Tools and Specificity: The authors' genetic tools show qualitative evidence of specificity for AACs, opening new avenues for targeted research on these cells. The use of intersectional strategies enhances the precision of AAC labeling.

Widespread Distribution: The study significantly broadens our understanding of AAC distribution, revealing their presence in brain regions beyond what was previously documented. This expanded knowledge is a valuable contribution to the field.

Transneuronal Labeling: The inclusion of rabies transneuronal labeling provides a comprehensive view of AACs, their variations, and input sources, allowing for a more holistic understanding of their role in neural circuits.

Weaknesses:

Quantitative Analysis: While the claim of specificity appears qualitatively convincing, the manuscript could be improved with more quantitative analysis.

We are glad that the reviewers appreciated our multimodal and brain-wide characterizations of the AAC population. We include many qualitative AAC examples and would like to highlight the quantitative nature of our whole brain cell body and cartridge analyses, made possible by transgenic targeting and our serial two-photon tomography imaging platform (STP). In addition to providing this brain wide AAC atlas, we also propose AACs as perhaps one of the best case examples for a bona fide cell type, which may inspire further in-depth anatomical and functional studies of AACs, and efforts to capture other ground truth cell types.

Comprehensiveness Claim: The assertion of comprehensiveness, implying labeling "almost all" AACs in all brain regions, is challenging to substantiate conclusively. Acknowledging the limitations of proving complete comprehensiveness and discussing them in the discussion section would be more appropriate than asserting it in the results section.

We thank the reviewer for this suggestion and have revised the results and discussion sections accordingly. The issue of how to access comprehensiveness in AAC labeling is a fair and important point, as dense brain-wide AAC labeling has not been achieved and assessed before. Previous studies had used less efficient and specific methods for capturing AACs, primarily in select areas of cortex, hippocampus, and amygdala. These AAC populations are recapitulated by our genetic strategies with higher density and specificity. It does not seem that we have missed any previously-reported AAC populations; in fact, we discovered multiple previously unreported populations. Another evidence supporting our “comprehensive” labeling of AACs is that two independent Unc5b and Pthlh transgenic strategies showed very similar AAC distribution patterns (Fig. 1 Suppl. 3). However, we recognize that probably the only way to fully assess “completeness” of labeling may be to compare with anatomical ground truth, such as by dense EM reconstruction of all AACs across the brain volume. This is currently not technically possible but may become feasible in the future. 

Local Inputs: While the manuscript focuses on inter-areal inputs to AACs, it would benefit from exploring local inputs as well. Identifying the local neurons that target AACs and analyzing their patterns could provide valuable insights into AAC function within specific brain regions.

This is a good suggestion. However, our serial two-photon tomography imaging platform does not have the capability for reliably preserving tissue sections for immunohistochemical processing afterward. Additionally, though our starter AAV injections were limited to 100-150nL, there were far too many input cells labelled at the injection side to resolve individual input cells and correlate with their synaptic partners (e.g. a rabies-labelled pyramidal cell within the injection site may still project to starter cell few hundred microns away). Thus, our rabies input mapping was best suited for characterizing long-range inputs and was the focus here. For studying local inputs to AACs, future studies could combine very dilute starter AAV injections with multi-marker characterization of cell types by immunohistochemistry or FISH.  

Discussion Focus: The discussion section should delve deeper into the biological implications of the findings, moving beyond technical significance. Exploring similarities and differences in input patterns between AACs and other cell types, and linking them to the locations of starter cells or specific connectivity patterns in the brain, would enrich the discussion. For instance, investigating whether input patterns can be predicted based on the locations of starter cells or connectivity specificity could provide valuable insights.

We thank the reviewer for this suggestion. We have expanded the discussion to include more on the relevance and implications of our input mapping results to different starter populations of AACs.

Reviewer #2 (Public Review):

Summary:

The goals of this study were to develop a genetic approach that would specifically and comprehensively target axo-axonic cells (AACs) throughout the brain and then to describe the patterns and characteristics of the targeted AACs in multiple, selected brain regions. The investigators have been successful in providing the most complete description of the regional distribution of putative (pAACs) throughout the brain to date. The supporting evidence is convincing, even though incomplete in some brain regions. The findings should serve as a guide for more detailed studies of AACs within each brain region and lead to new insights into the connectivity and functional organization of this important group of GABAergic interneurons.

Strengths:

The study has numerous strengths. A major strength is the development of a unique intersectional genetic strategy that uses cell lineage (Nkx2.1) and molecular (Unc5b or Pthlh) markers to identify axo-axonic AACs specifically and, apparently, nearly completely throughout the mouse brain. While AACs have been described previously in the cerebral cortex, hippocampus, and amygdala, there has been no specific genetic marker that selectively identifies all AACs in these regions.

The current genetic strategy has labeled pAACs in a large number of additional brain regions, including the claustrum-insular complex, extended amygdala, and several olfactory centers. In general, the findings provide support for the specificity of the methods for targeting AACs, and include some examples of labeling near markers of axon initial segments. However, the Investigators are careful to refer to labeled neurons as "putative AACs" as they have not been fully characterized and their identity verified.

The descriptions and numerous low-magnification images of the brain provide a roadmap for subsequent, detailed studies of AACs in numerous brain regions. The overview and summaries of the findings in the Abstract, Introduction, and Discussion are particularly clear and helpful in placing the extensive regional descriptions of AACs in context.

Weaknesses:

One weakness of the study is the lack of an illustration of the high-resolution cell labeling that can be achieved with the methods, including labeling of numerous rows of axon terminals in contact with axon initial segments. The initial images of the brain-wide distribution of putative AACs are necessarily presented at low magnification. Although the authors indicate that the cells have "highly characteristic AAC labeling patterns throughout the neocortex, hippocampus and BLA", these morphological details cannot be visualized by the reader at the current magnification, even when the images are enlarged on the computer screen. Some of the details become evident in later Figures, but an initial illustration of single cell labeling with confocal microscopy, or tracing of their characteristic axonal arbors, would support the specificity of the labeling in the low magnification images.

We thank the reviewer for the suggestion. We have now added high-resolution images showing the colocalization of AAC axon boutons (cartridges) along AnkG positive postsynaptic axon initial segments in Fig. 2 Suppl. 1, Figure 1 panels a, d, e, and Fig. 4 panels b, c. These images unequivocally demonstrate AAC identity and specificity.

Table 1 indicates that the AAC identity of the cells has been validated in many brain regions but not in all. The methods used for validation have not been described and should be included for completeness. The authors are careful to acknowledge that labeled cells in some regions have not been validated and refer to such cells as pAACs.

Validation was defined by colocalization of RFP-labelled AAC cartridges and AnkryinG or Phospho-IκBα-labelled axon initial segments, imaged by confocal microscopy. We provide high-magnification examples throughout figures 2-6 and supplements. We have also tried to clarify this better in the methods section entitled “Immunohistochemistry.” Putative AAC (pAACs) refers to populations in which relatively few single cell examples of AACs exhibiting co-localized cartridges were found, largely due to the sparsity of the low tamoxifen dosage used (see response above).

The intersectional genetic methods included the use of the lineage marker Nkx2.1 with either Unc5b or Pthlh as the molecular marker. As described, the mice with intersectional targeting of Nkx2.1 and Unc5b appear to show the most specific brain-wide labeling for AACs, and the majority of the descriptions are from these mice. The targeting with Nkx2.1 and Pthlh is less convincing. The title for Figure 1 Supplemental Figure 3 suggests a similar AAC distribution in the Pthlh;Nkx2.1 mouse compared to the Unc5b;Nkx2.1 mouse. However, the descriptions of the individual panels suggest a number of inconsistencies and non-AAC labeling. The heavy labeling in the caudate and cells in layer 4 is particularly problematic. Based on the data presented, it appears that heavy labeling achieved in these mice could not be relied on for specific labeling of all AACs, although specific labeling could be achieved under some conditions, such as following tamoxifen administration at select ages.

The reviewer is correct about Pthlh being less specific for AACs than Unc5b when crossed to a constitutive Nkx2.1 recombinase driver line. Pthlh/Nkx2.1 intersection labeled a set of layer 4 cells in somatosensory cortex and dense cells in striatum, which are clearly not AACs. But these are the only main difference compared to Unc5b/Nkx2.1 intersection. As the reviewer points out, it is only when Pthlh is crossed to an inducible Nkx2.1-CreER line and induced embryonically with tamoxifen that there is more specific AAC labeling (at least in cortex). We included this data as well as the intersection with VIP-Cre in case either of these are useful to researchers studying fate-mapping of AACs or bipolar cell interneurons. We have also revised the title of Fig. 1 Suppl. 3 to better convey this.

The methods described for dense labeling and single-cell labeling are described briefly in the methods. Some discussion of the development of the methods would be useful, including how it was determined that methods for heavy labeling identified AACs specifically and completely.

We have added a description on the development of these to the methods section entitled “Animals.”

Reviewer #3 (Public Review):

Summary:

Raudales et al. aimed at providing an insight into the brain-wide distribution and synaptic connectivity of bona fide GABAergic inhibitory interneuron subtypes focusing on the axo-axonic cell (AAC), one of the most distinctive interneuron subtypes, which innervates the axon initial segments of glutamatergic projection neurons. They establish intersectional genetic strategies that enable them to specifically and comprehensively capture AACs based on their lineage (Nkx2.1) and marker expression (Unc5b, Pthlh). They find that AACs are deployed across essentially all the pallium-derived brain structures as well as the anterior olfactory nucleus, taenia tecta, and lateral septum. They show that AACs in distinct areas and layers of the neocortex as well as different subregions of the hippocampal formation display unique soma and synaptic density and morphological variations. Rabies virus-based retrograde monosynaptic input tracing reveals that AACs in the neocortex, the hippocampus, and the basolateral amygdala receive synaptic inputs from common as well as specific brain regions and supports the utility of this novel genetic approach. This study elucidates brain-wide neuroanatomical features and morphological variations of AACs with solid techniques and analysis. Their novel AAC-targeting strategies will facilitate the study of their development and function in different brain regions. The conclusions in this paper are well supported by the data. However, there are a few comments to strengthen this study.

(1) The definition of putative AAC (pAAC) is unclear and Table 1 may not be accurate. Although the authors find synaptic cartridges of RFP-labeled cells in the claustro-insular complex and the dorsal endopiriform nuclei, they still consider these cells as pAACs (not validated). The authors claim that without examining the presence of synaptic cartridges, RFP-labeled cells in the hypothalamus and the bed nuclei of the stria terminalis (BNST) are pAACs while those in the L4 of the somatosensory cortex in Pthlh;Nkx2.1;Ai65 mice are non-AACs. In Table 1, the BNST is supposed to contain AACs (validated), but in the text, the authors claim that RFP-labeled cells in the BNST are pAACs. Could the authors clarify how AACs, pAACs, and non-AACs are defined?

We thank the reviewer for their interest and comments on our work. Please see our response to reviewer 2 for clarification on putative pAACs. Additionally, we have clarified in the methods under “Immunohistochemistry” how we defined AACs, pAAC, and non-AACs. For BNST we did not positively identify more than a few exhibiting overlap with AnkryinG/IκBα, so we currently leave them as pAACs—Table 1 has been corrected to reflect this.

(2) The intersectional strategies presented in this study could also specifically capture developing AACs. If so, how early are AACs labeled in the brain? It would also be nice if the authors could add a simple schematic like Fig. 1a showing the time course of Pthlh expression.

We thank the reviewer for suggesting the application of our method in studying AAC development. As the onset of Unc5b is in early postnatal time, tamoxifen induction of Unc5b-CreER in early postnatal days can enable studies of AAC neurite and synapse development, maturation, and plasticity. Similarly, Pthlh expression in the brain is relatively low/absent at P4 and present at P14 and later timepoints. Pthlh-Flp;Nkx2.1-Cre intersection can be used to study postnatal AAC development and plasticity.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

While the claim of specificity appears qualitatively convincing, additional quantitative analysis would make the authors' claim much stronger. For example in Figure 4 (f-h), where the authors show an overlap of AAC axons with AnkG labeling, there also appears to be a region of AAC axon lacking adjacent AnkG labeling. The author could quantify the fraction of cartridges that overlap with AnkG labeling in different brain regions, potentially stringing their claim that pAACs are AACs as well as providing important documentation of the diversity or homogeneity of compartment targeting across the brain.

As mentioned previously, we only performed AnkG co-labeling analysis on low-dose tamoxifen/sparsely labelled samples in which we could readily differentiate individual cells. This was performed on samples with the Ai65 cytoplasmic reporter—for validation purposes we could positively identify co-labelled cartridges, but it would be more difficult to accurately identify any cartridges not co-labeled (since the entire axon was labelled with RFP). For precisely identifying and mapping AAC cartridge locations we found the intersectional synaptophysin-EGFP reporter (Fig. 2k-n) to be a more precise method for specifically labeling the “cartridge” segment of AAC axons. However, we did not try AnkG staining on samples from this reporter line, as they were set aside for STP imaging.

Regarding the claim of comprehensiveness, labeling "almost all" AACs in all brain regions is a high standard and challenging to demonstrate conclusively. The study already significantly expands our understanding of AAC distribution, and the authors might consider discussing the limitations of proving complete comprehensiveness in the discussion rather than claiming it in the results section.

We again thank the reviewer for this critique. As mentioned above, we have revised the results and discussion sections to better convey this point across.

Furthermore, the manuscript connectivity section primarily focuses on inter-areal inputs to AACs, but it could benefit from exploring local inputs as well. By identifying the local neurons that target AACs, the authors could ask if there is any general property or rule of the local projections to AACs across the brain, or at least within the cortex. Moreover, a clear indication of the injection site would be helpful, particularly in Figure 7, where there seems to be some discrepancy between the histograms and fluorescent images regarding local projections. The histograms of Figure 7, seem to indicate that the local projection to AACs is a small fraction of all the presynaptic neurons, however, the fluorescent image for the SSp seems to suggest otherwise with many fluorescent cells in the injected area.

We thank the reviewer for these comments. Regarding the local inputs in the rabies tracing datasets, it is a limitation (as mentioned above) of our STP platform’s inability to preserve tissue for immunohistochemistry labeling as well as our relatively dense starter cell labeling. Instead, our focus here was on long-range inputs (i.e. outside the ipsilateral ARA area of injection), which was simply not known for these AAC populations. We have revised the Figure 7 legend and added a description in the methods section to more clearly indicate that we only included long-range input projections in the Figure 7 histograms.

In the discussion, the authors should delve more into the biological implications of their findings rather than solely emphasizing the technical significance. They could explore the similarities and differences in input patterns between AACs and other cell types, potentially linking them to the locations of their starter cells or specific connectivity patterns in the brain. For example, the authors could check if the input patterns could be predicted from the projections to the layers where their starter cells are located (either from an Atlas like the Allen Connectivity Atlas, or from retrograde rabies injections in the same locations). Can the differences between the input patterns to PVC and AAC be predicted for their location versus some specificity of connections?

Thank you for the extensive comment. We address this point above, and have revised our discussion accordingly.

Reviewer #2 (Recommendations For The Authors):

The Figure legends vary in completeness and quality.

(1) The legend for Figure 1 is very informative, and section e-g serves as a useful guide, as the legend includes the names of the brain regions related to the abbreviations and also indicates the specific panels that show the identified structures. Because of the large number of structures and the number of panels in each Figure, it would be ideal to follow the same pattern in the remaining figures.

(2) Several edits are needed in the legend for Figure 1 Supplement Figure 1. The descriptions of a-f could be improved by providing general terms to describe the brain regions associated with the latter list of abbreviations (as has been done with the identification of the cerebral cortex, hippocampus, and olfactory centers and their related panels). One suggestion would be to write out insula, claustrum, and endopiriform prior to listing the abbreviations (AI, CLA, EP) (b-c) and adding amygdaloid complex and extended amygdala before the abbreviations (COA, BLA, MeA) (d-f) and (BST) (d).

We thank the reviewer, as the suggestion of further expanding the abbreviations is a good one. As such, we have revised/reorganized the anatomical abbreviations in the figure legends for Figure 1 Supplement Figures 1, 2, and 3.

Descriptions for Panels g-j require editing to link the appropriate panels and the descriptions. Panels for BSTpr appear to be g-h (rather than f-g) and i,j (rather than h-i.

We have fixed this typo in the legend for Figure 1 Supplement Figure 1.

Descriptions for Panels k-n could be edited to include abbreviations for the identified brain regions. For example, include the abbreviation ARHP after arcuate nuclei and indicate panels m-n (rather than j-l); include PVP after paraventricular and indicate panel n (rather than m); include DMPH after dorsomedial nuclei and indicate k-m (rather than j-l).

Thank you for the suggestion. We have expanded the abbreviations in Figure 1 Supplement 1 accordingly.

Reviewer #3 (Recommendations For The Authors):

(1) Please clarify if tdTomato, EGFP (from helper AAVs), and RFP (from rabies virus) are native signals or IHC signals in legends.

We have added the descriptors “native” or “stained” to all figure legends containing fluorescent images.

(2) Fig. 4b and c: Please add insets of high-magnification images showing AAC boutons along AnkG-labeled AISs.

We have added these insets to Fig. 4b and c.

(3) Fig. 7S1: It appears that d and e are reversed. Judging from the positions of starter cells, d is for PV-Cre? Please make sure. It is also better to draw the laminar border in d and e.

The original genotype labels are correct for Fig. 7S1 d and e. We have added the laminar borders as suggested.

(4) Fig. 9b: Just for consistency, please label with the name of the helper AAV.

Added.

(5) Line 617: intragranular>>>infragranular?

Corrected, thank you.

(6) It may be unclear to some readers if the images in the figures are from confocal or STP. The authors may want to clarify that all images in the figures are generated by confocal microscopy in the method section.

We have clarified this better in the methods section, “Microcopy and image analysis.”

(7) The authors should clarify that STP was used to map input cells to the brain in the result section.

We have added this description in the results section.

eLife assessment

This useful study provides a novel method to detect sleep cycles based on variations in the slope of the power spectrum from electroencephalography signals. The method, dispensing with time-consuming and potentially subjective manual identification of sleep cycles, is supported by solid evidence and analyses but some aspects could be better illustrated and the source of the discrepancies between classical and fractal cycles should be identified. This study will be of interest to researchers and clinicians working on sleep and brain dynamics.

Reviewer #1 (Public Review):

Summary:

In this study, Rosenblum et al introduce a novel and automatic way of calculating sleep cycles from human EEG. Previous results have shown that the slope of the non-oscillatory component of the power spectrum (called the aperiodic or fractal component) changes with the sleep stage. Building on this, the authors present an algorithm that extracts the continuous-time fluctuations in the fractal slope and propose that peaks in this variable can be used to identify sleep cycle limits. Cycles defined in this way are termed "fractal cycles". The main focus of the article is a comparison of fractal and classical, manually defined sleep cycles in numerous datasets.

Strengths:

The manuscript amply illustrates through examples the strong overlap between fractal and classical cycle identification. Accordingly, a high percentage (81%) can be matched one-to-one between methods and sleep cycle duration is well correlated (around R = 0.5). Moreover, the methods track certain global changes in sleep structure in different populations: shorter cycles in children and longer cycles in patients medicated with REM-suppressing anti-depressants. Finally, a major strength of the results is that they show similar agreement between fractal and classical sleep cycle length in 5 different data sets, showing that it is robust to changes in recording settings and methods.

These results suggest that the fractal cycle methodology could provide a valuable new method to study sleep architecture and avoid the time-consuming steps of manual cycle identification. Moreover, it has the potential to be applied to animal studies which rarely deal with sleep cycle structure.

Weaknesses:

The match between fractal and classical cycles is not one-to-one. For example, the fractal method identifies a correlation between age and cycle duration in adults that is not apparent with the classical method. This raises the question as to whether differences are due to one method being more reliable than another or whether they are also identifying different underlying biological differences. It is not clear for example whether the agreement between the two methods is better or worse than between two human scorers, which generally serve as a gold standard to validate novel methods. The authors provide some insight into differences between the methods that could account for differences in results. However, given that the fractal method is automatic it would be important to clearly identify criteria for recordings in which it will produce similar results to the classical method.

Reviewer #2 (Public Review):

Summary:

This study focused on using strictly the slope of the power spectral density (PSD) to perform automated sleep scoring and evaluation of the durations of sleep cycles. The method appears to work well because the slope of the PSD is highest during slow-wave sleep, and lowest during waking and REM sleep. Therefore, when smoothed and analyzed across time, there are cyclical variations in the slope of the PSD, fit using an IRASA (Irregularly resampled auto-spectral analysis) algorithm proposed by Wen & Liu (2016).

Strengths:

The main novelty of the study is that the non-fractal (oscillatory) components of the PSD that are more typically used during sleep scoring can be essentially ignored because the key information is already contained within the fractal (slope) component. The authors show that for the most part, results are fairly consistent between this and conventional sleep scoring, but in some cases show disagreements that may be scientifically interesting.

Weaknesses:

One weakness of the study, from my perspective, was that the IRASA fits to the data (e.g. the PSD, such as in Figure 1B), were not illustrated. One cannot get a sense of whether or not the algorithm is based entirely on the fractal component or whether the oscillatory component of the PSD also influences the slope calculations. This should be better illustrated, but I assume the fits are quite good.

The cycles detected using IRASA are called fractal cycles. I appreciate the use of a simple term for this, but I am also concerned whether it could be potentially misleading? The term suggests there is something fractal about the cycle, whereas it's really just that the fractal component of the PSD is used to detect the cycle. A more appropriate term could be "fractal-detected cycles" or "fractal-based cycle" perhaps?

The study performs various comparisons of the durations of sleep cycles evaluated by the IRASA-based algorithm vs. conventional sleep scoring. One concern I had was that it appears cycles were simply identified by their order (first, second, etc.) but were not otherwise matched. This is problematic because, as evident from examples such as Figure 3B, sometimes one cycle conventionally scored is matched onto two fractal-based cycles. In the case of the Figure 3B example, it would be more appropriate to compare the duration of conventional cycle 5 vs. fractal cycle 7, rather than 5 vs. 5, as it appears is currently being performed.

There are a few statements in the discussion that I felt were either not well-supported. L629: about the "little biological foundation" of categorical definitions, e.g. for REM sleep or wake? I cannot agree with this statement as written. Also about "the gradual nature of typical biological processes". Surely the action potential is not gradual and there are many other examples of all-or-none biological events.

The authors appear to acknowledge a key point, which is that their methods do not discriminate between awake and REM periods. Thus their algorithm essentially detected cycles of slow-wave sleep alternating with wake/REM. Judging by the examples provided this appears to account for both the correspondence between fractal-based and conventional cycles, as well as their disagreements during the early part of the sleep cycle. While this point is acknowledged in the discussion section around L686. I am surprised that the authors then argue against this correspondence on L695. I did not find the "not-a-number" controls to be convincing. No examples were provided of such cycles, and it's hard to understand how positive z-values of the slopes are possible without the presence of some wake unless N1 stages are sufficient to provide a detected cycle (in which case, then the argument still holds except that its alterations between slow-wave sleep and N1 that could be what drives the detection).

To me, it seems important to make clear whether the paper is proposing a different definition of cycles that could be easily detected without considering fractals or spectral slopes, but simply adjusting what one calls the onset/offset of a cycle, or whether there is something fundamentally important about measuring the PSD slope. The paper seems to be suggesting the latter but my sense from the results is that it's rather the former.

Author response:

We thank the reviewers and editors for their review and assessment of our manuscript and comprehensive feedback. The manuscript will be revised to address all the reviewers’ comments. Specifically, to address the comment of Reviewer 1 and the editor regarding the lack of quantitative comparison between the classical and fractal cycle approaches and identification of the source of the discrepancies between classical and fractal cycles, we plan to perform and report the following analyses and comparisons:

(1) Intra-method reliability

a) Classical cycles. An additional scorer will independently define onsets and offsets of all classical sleep cycles for all datasets and mark sleep cycles with skipped REM sleep. Likewise, we will perform automatic sleep cycle detection. We will add a new Supplementary table showing the averaged cycle durations obtained by the two scorers and automatic algorithm as well as the inter-scorer rate agreement and update the Supplemental Excel file with corresponding information for each cycle for each participant for each dataset.

b) Fractal cycles. We will correlate the durations of fractal cycles calculated using the parameters defined in the Main text with those calculated using different parameters, namely, the longer and shorter smoothing window lengths, higher and lower minimum peak prominence. Likewise, we will correlate the durations of fractal cycles calculated using frontal vs other available electrodes.

(2) Origin of method differences

In the current version of our Manuscript, we describe a few possible sources of discrepancies between classical and fractal cycle durations and numbers. Following the suggestion of one of the reviewers, in the revised Manuscript, we will quantify the sources of discrepancies between the two methods in order to identify the “criteria for recordings in which fractal cycles will produce similar results to the classical method”. Specifically, we will calculate the correlation between the difference in classical vs fractal sleep cycle durations on one side, and either the amplitudes of fractal descend/ascend, relative durations of cycles with skipped REM sleep and wake after sleep onset, or peak flatness on the other side.    

In addition, we will include a new figure, illustrating the goodness of fit of the data as assessed by the IRASA method. Likewise, we will update Supplementary File 1 (that shows classical and fractal sleep cycles for each participant) with marks that highlight the onsets and offsets of sleep cycles as well as the cycles with skipped REM sleep.

Author response:

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

Reviewer #1 (Public Review):

In this paper, the authors evaluate the utility of brain age derived metrics for predicting cognitive decline by performing a 'commonality' analysis in a downstream regression that enables the different contribution of different predictors to be assessed. The main conclusion is that brain age derived metrics do not explain much additional variation in cognition over and above what is already explained by age. The authors propose to use a regression model trained to predict cognition ('brain cognition') as an alternative suited to applications of cognitive decline. While this is less accurate overall than brain age, it explains more unique variance in the downstream regression.  

Importantly, in this revision, we clarified that we did not intend to use Brain Cognition as an alternative approach. This is because, by design, the variation in fluid cognition explained by Brain Cognition should be higher or equal to that explained by Brain Age. Here we made this point more explicit and further stated that the relationship between Brain Cognition and fluid cognition indicates the upper limit of Brain Age’s capability in capturing fluid cognition. By examining what was captured by Brain Cognition, over and above Brain Age and chronological age via the unique effects of Brain Cognition, we were able to quantify the amount of co-variation between brain MRI and fluid cognition that was missed by Brain Age. 

REVISED VERSION: while the authors have partially addressed my concerns, I do not feel they have addressed them all. I do not feel they have addressed the weight instability and concerns about the stacked regression models satisfactorily.

Please see our responses to Reviewer #1 Public Review #3 below

I also must say that I agree with Reviewer 3 about the limitations of the brain age and brain cognition methods conceptually. In particular that the regression model used to predict fluid cognition will by construction explain more variance in cognition than a brain age model that is trained to predict age. This suffers from the same problem the authors raise with brain age and would indeed disappear if the authors had a separate measure of cognition against which to validate and were then to regress this out as they do for age correction. I am aware that these conceptual problems are more widespread than this paper alone (in fact throughout the brain age literature), so I do not believe the authors should be penalised for that. However, I do think they can make these concerns more explicit and further tone down the comments they make about the utility of brain cognition. I have indicated the main considerations about these points in the recommendations section below. 

Thank you so much for raising this point. We now have the following statement in the introduction and discussion to address this concern (see below). 

Briefly, we made it explicit that, by design, the variation in fluid cognition explained by Brain Cognition should be higher or equal to that explained by Brain Age. That is, the relationship between Brain Cognition and fluid cognition indicates the upper limit of Brain Age’s capability in capturing fluid cognition. More importantly, by examining what was captured by Brain Cognition, over and above Brain Age and chronological age via the unique effects of Brain Cognition, we were able to quantify the amount of co-variation between brain MRI and fluid cognition that was missed by Brain Age. And this is the third goal of this present study. 

From Introduction:

“Third and finally, certain variation in fluid cognition is related to brain MRI, but to what extent does Brain Age not capture this variation? To estimate the variation in fluid cognition that is related to the brain MRI, we could build prediction models that directly predict fluid cognition (i.e., as opposed to chronological age) from brain MRI data. Previous studies found reasonable predictive performances of these cognition-prediction models, built from certain MRI modalities (Dubois et al., 2018; Pat, Wang, Anney, et al., 2022; Rasero et al., 2021; Sripada et al., 2020; Tetereva et al., 2022; for review, see Vieira et al., 2022). Analogous to Brain Age, we called the predicted values from these cognition-prediction models, Brain Cognition. The strength of an out-of-sample relationship between Brain Cognition and fluid cognition reflects variation in fluid cognition that is related to the brain MRI and, therefore, indicates the upper limit of Brain Age’s capability in capturing fluid cognition. This is, by design, the variation in fluid cognition explained by Brain Cognition should be higher or equal to that explained by Brain Age. Consequently, if we included Brain Cognition, Brain Age and chronological age in the same model to explain fluid cognition, we would be able to examine the unique effects of Brain Cognition that explain fluid cognition beyond Brain Age and chronological age. These unique effects of Brain Cognition, in turn, would indicate the amount of co-variation between brain MRI and fluid cognition that is missed by Brain Age.”

From Discussion:

“Third, by introducing Brain Cognition,  we showed the extent to which Brain Age indices were not able to capture the variation in fluid cognition that is related to brain MRI. More specifically, using Brain Cognition allowed us to gauge the variation in fluid cognition that is related to the brain MRI, and thereby, to estimate the upper limit of what Brain Age can do. Moreover, by examining what was captured by Brain Cognition, over and above Brain Age and chronological age via the unique effects of Brain Cognition, we were able to quantify the amount of co-variation between brain MRI and fluid cognition that was missed by Brain Age.

From our results, Brain Cognition, especially from certain cognition-prediction models such as the stacked models, has relatively good predictive performance, consistent with previous studies (Dubois et al., 2018; Pat, Wang, Anney, et al., 2022; Rasero et al., 2021; Sripada et al., 2020; Tetereva et al., 2022; for review, see Vieira et al., 2022). We then examined Brain Cognition using commonality analyses (Nimon et al., 2008) in multiple regression models having a Brain Age index, chronological age and Brain Cognition as regressors to explain fluid cognition. Similar to Brain Age indices, Brain Cognition exhibited large common effects with chronological age. But more importantly, unlike Brain Age indices, Brain Cognition showed large unique effects, up to around 11%. As explained above, the unique effects of Brain Cognition indicated the amount of co-variation between brain MRI and fluid cognition that was missed by a Brain Age index and chronological age. This missing amount was relatively high, considering that Brain Age and chronological age together explained around 32% of the total variation in fluid cognition. Accordingly, if a Brain Age index was used as a biomarker along with chronological age, we would have missed an opportunity to improve the performance of the model by around one-third of the variation explained.” 

This is a reasonably good paper and the use of a commonality analysis is a nice contribution to understanding variance partitioning across different covariates. I have some comments that I believe the authors ought to address, which mostly relate to clarity and interpretation 

Reviewer #1 Public Review #1

First, from a conceptual point of view, the authors focus exclusively on cognition as a downstream outcome. I would suggest the authors nuance their discussion to provide broader considerations of the utility of their method and on the limits of interpretation of brain age models more generally. 

Thank you for your comments on this issue. 

We now discussed the broader consideration in detail:

(1) the consistency between our findings on fluid cognition and other recent works on brain disorders, 

(2) the difference between studies investigating the utility of Brain Age in explaining cognitive functioning, including ours and others (e.g., Butler et al., 2021; Cole, 2020, 2020; Jirsaraie, Kaufmann, et al., 2023) and those explaining neurological/psychological disorders (e.g., Bashyam et al., 2020; Rokicki et al., 2021)

and 

(3) suggested solutions we and others made to optimise the utility of Brain Age for both cognitive functioning and brain disorders.

From Discussion:

“This discrepancy between the predictive performance of age-prediction models and the utility of Brain Age indices as a biomarker is consistent with recent findings (for review, see Jirsaraie, Gorelik, et al., 2023), both in the context of cognitive functioning (Jirsaraie, Kaufmann, et al., 2023) and neurological/psychological disorders (Bashyam et al., 2020; Rokicki et al., 2021). For instance,  combining different MRI modalities into the prediction models, similar to our stacked models, ocen leads to the highest performance of age prediction models, but does not likely explain the highest variance across different phenotypes, including cognitive functioning and beyond (Jirsaraie, Gorelik, et al., 2023).”

“There is a notable difference between studies investigating the utility of Brain Age in explaining cognitive functioning, including ours and others (e.g., Butler et al., 2021; Cole, 2020, 2020; Jirsaraie, Kaufmann, et al., 2023) and those explaining neurological/psychological disorders (e.g., Bashyam et al., 2020; Rokicki et al., 2021). We consider the former as a normative type of study and the lader as a case-control type of study (Insel et al., 2010; Marquand et al., 2016). Those case-control Brain Age studies focusing on neurological/psychological disorders often build age-prediction models from MRI data of largely healthy participants (e.g., controls in a case-control design or large samples in a population-based design), apply the built age-prediction models to participants without vs. with neurological/psychological disorders and compare Brain Age indices between the two groups. On the one hand, this means that case-control studies treat Brain Age as a method to detect anomalies in the neurological/psychological group (Hahn et al., 2021). On the other hand, this also means that case-control studies have to ignore underfided models when applied prediction models built from largely healthy participants to participants with neurological/psychological disorders (i.e., Brain Age may predict chronological age well for the controls, but not for those with a disorder). On the contrary, our study and other normative studies focusing on cognitive functioning often build age prediction models from MRI data of largely healthy participants and apply the built age prediction models to participants who are also largely healthy. Accordingly, the age prediction models for explaining cognitive functioning in normative studies, while not allowing us to detect group-level anomalies, do not suffer from being under-fided. This unfortunately might limit the generalisability of our study into just the normative type of study. Future work is still needed to test the utility of brain age in the case-control case.”

“Next, researchers should not select age-prediction models based solely on age-prediction performance. Instead, researchers could select age-prediction models that explained phenotypes of interest the best. Here we selected age-prediction models based on a set of features (i.e., modalities) of brain MRI. This strategy was found effective not only for fluid cognition as we demonstrated here, but also for neurological and psychological disorders as shown elsewhere (Jirsaraie, Gorelik, et al., 2023; Rokicki et al., 2021). Rokicki and colleagues (2021), for instance, found that, while integrating across MRI modalities led to age prediction models with the highest age-prediction performance, using only T1 structural MRI gave age-prediction models that were better at classifying Alzheimer’s disease. Similarly, using only cerebral blood flow gave age-prediction models that were better at classifying mild/subjective cognitive impairment, schizophrenia and bipolar disorder. 

As opposed to selecting age-prediction models based on a set of features, researchers could also select age-prediction models based on modelling methods. For instance, Jirsaraie and colleagues (2023) compared gradient tree boosting (GTB) and deep-learning brain network (DBN) algorithms in building age-prediction models. They found GTB to have higher age prediction performance but DBN to have better utility in explaining cognitive functioning. In this case, an algorithm with better utility (e.g., DBN) should be used for explaining a phenotype of interest. Similarly, Bashyam and colleagues (2020) built different DBN-based age-prediction models, varying in age-prediction performance. The DBN models with a higher number of epochs corresponded to higher age-prediction performance. However, DBN-based age-prediction models with a moderate (as opposed to higher or lower) number of epochs were better at classifying Alzheimer’s disease, mild cognitive impairment and schizophrenia. In this case, a model from the same algorithm with better utility (e.g., those DBN with a moderate epoch number) should be used for explaining a phenotype of interest.

Accordingly, this calls for a change in research practice, as recently pointed out by Jirasarie and colleagues (2023, p7), “Despite mounting evidence, there is a persisting assumption across several studies that the most accurate brain age models will have the most potential for detecting differences in a given phenotype of interest”. Future neuroimaging research should aim to build age-prediction models that are not necessarily good at predicting age, but at capturing phenotypes of interest.”

Reviewer #1 Public Review #2

Second, from a methods perspective, there is not a sufficient explanation of the methodological procedures in the current manuscript to fully understand how the stacked regression models were constructed. I would request that the authors provide more information to enable the reader to beUer understand the stacked regression models used to ensure that these models are not overfit. 

Thank you for allowing us an opportunity to clarify our stacked model. We made additional clarification to make this clearer (see below). We wanted to confirm that we did not use test sets to build a stacked model in both lower and higher levels of the Elastic Net models. Test sets were there just for testing the performance of the models.  

From Methods:

“We used nested cross-validation (CV) to build these prediction models (see Figure 7). We first split the data into five outer folds, leaving each outer fold with around 100 participants. This number of participants in each fold is to ensure the stability of the test performance across folds. In each outer-fold CV loop, one of the outer folds was treated as an outer-fold test set, and the rest was treated as an outer-fold training set. Ultimately, looping through the nested CV resulted in a) prediction models from each of the 18 sets of features as well as b) prediction models that drew information across different combinations of the 18 separate sets, known as “stacked models.” We specified eight stacked models: “All” (i.e., including all 18 sets of features),  “All excluding Task FC”, “All excluding Task Contrast”, “Non-Task” (i.e., including only Rest FC and sMRI), “Resting and Task FC”, “Task Contrast and FC”, “Task Contrast” and “Task FC”. Accordingly, there were 26 prediction models in total for both Brain Age and Brain Cognition.

To create these 26 prediction models, we applied three steps for each outer-fold loop. The first step aimed at tuning prediction models for each of 18 sets of features. This step only involved the outer-fold training set and did not involve the outer-fold test set. Here, we divided the outer-fold training set into five inner folds and applied inner-fold CV to tune hyperparameters with grid search. Specifically, in each inner-fold CV, one of the inner folds was treated as an inner-fold validation set, and the rest was treated as an inner-fold training set. Within each inner-fold CV loop, we used the inner-fold training set to estimate parameters of the prediction model with a particular set of hyperparameters and applied the estimated model to the inner-fold validation set. Acer looping through the inner-fold CV, we, then, chose the prediction models that led to the highest performance, reflected by coefficient of determination (R2), on average across the inner-fold validation sets. This led to 18 tuned models, one for each of the 18 sets of features, for each outer fold.

The second step aimed at tuning stacked models. Same as the first step, the second step only involved the outer-fold training set and did not involve the outer-fold test set. Here, using the same outer-fold training set as the first step, we applied tuned models, created from the first step, one from each of the 18 sets of features, resulting in 18 predicted values for each participant. We, then, re-divided this outer-fold training set into new five inner folds. In each inner fold, we treated different combinations of the 18 predicted values from separate sets of features as features to predict the targets in separate “stacked” models. Same as the first step, in each inner-fold CV loop, we treated one out of five inner folds as an inner-fold validation set, and the rest as an inner-fold training set. Also as in the first step, we used the inner-fold training set to estimate parameters of the prediction model with a particular set of hyperparameters from our grid. We tuned the hyperparameters of stacked models using grid search by selecting the models with the highest R2 on average across the inner-fold validation sets. This led to eight tuned stacked models.

The third step aimed at testing the predictive performance of the 18 tuned prediction models from each of the set of features, built from the first step, and eight tuned stacked models, built from the second step. Unlike the first two steps, here we applied the already tuned models to the outer-fold test set. We started by applying the 18 tuned prediction models from each of the sets of features to each observation in the outer-fold test set, resulting in 18 predicted values. We then applied the tuned stacked models to these predicted values from separate sets of features, resulting in eight predicted values. 

To demonstrate the predictive performance, we assessed the similarity between the observed values and the predicted values of each model across outer-fold test sets, using Pearson’s r, coefficient of determination (R2) and mean absolute error (MAE). Note that for R2, we used the sum of squares definition (i.e., R2 \= 1 – (sum of squares residuals/total sum of squares)) per a previous recommendation (Poldrack et al., 2020). We considered the predicted values from the outer-fold test sets of models predicting age or fluid cognition, as Brain Age and Brain Cognition, respectively.”

Author response image 1.

Diagram of the nested cross-validation used for creating predictions for models of each set of features as well as predictions for stacked models. 

Note some previous research, including ours (Tetereva et al., 2022), splits the observations in the outer-fold training set into layer 1 and layer 2 and applies the first and second steps to layers 1 and 2, respectively. Here we decided against this approach and used the same outer-fold training set for both first and second steps in order to avoid potential bias toward the stacked models. This is because, when the data are split into two layers, predictive models built for each separate set of features only use the data from layer 1, while the stacked models use the data from both layers 1 and 2. In practice with large enough data, these two approaches might not differ much, as we demonstrated previously (Tetereva et al., 2022).

Reviewer #1 Public Review #3

Please also provide an indication of the different regression strengths that were estimated across the different models and cross-validation splits. Also, how stable were the weights across splits? 

The focus of this article is on the predictions. Still, it is informative for readers to understand how stable the feature importance (i.e., Elastic Net coefficients) is. To demonstrate the stability of feature importance, we now examined the rank stability of feature importance using Spearman’s ρ (see Figure 4). Specifically, we correlated the feature importance between two prediction models of the same features, used in two different outer-fold test sets. Given that there were five outer-fold test sets, we computed 10 Spearman’s ρ for each prediction model of the same features.  We found Spearman’s ρ to be varied dramatically in both age-prediction (range\=.31-.94) and fluid cognition-prediction (range\=.16-.84) models. This means that some prediction models were much more stable in their feature importance than others. This is probably due to various factors such as a) the collinearity of features in the model, b) the number of features (e.g., 71,631 features in functional connectivity, which were further reduced to 75 PCAs, as compared to 19 features in subcortical volume based on the ASEG atlas), c) the penalisation of coefficients either with ‘Ridge’ or ‘Lasso’ methods, which resulted in reduction as a group of features or selection of a feature among correlated features, respectively, and d) the predictive performance of the models. Understanding the stability of feature importance is beyond the scope of the current article. As mentioned by Reviewer 1, “The predictions can be stable when the coefficients are not,” and we chose to focus on the prediction in the current article.   

Author response image 2.

Stability of feature importance (i.e., Elastic Net Coefficients) of prediction models. Each dot represents rank stability (reflected by Spearman’s ρ) in the feature importance between two prediction models of the same features, used in two different outer-fold test sets. Given that there were five outer-fold test sets, there were 10 Spearman’s ρs for each prediction model.  The numbers to the right of the plots indicate the mean of Spearman’s ρ for each prediction model.  

Reviewer #1 Public Review #4

Please provide more details about the task designs, MRI processing procedures that were employed on this sample in addition to the regression methods and bias correction methods used. For example, there are several different parameterisations of the elastic net, please provide equations to describe the method used here so that readers can easily determine how the regularisation parameters should be interpreted.  

Thank you for the opportunity for us to provide more methodical details.

First, for the task design, we included the following statements:

From Methods:

“HCP-A collected fMRI data from three tasks: Face Name (Sperling et al., 2001), Conditioned Approach Response Inhibition Task (CARIT) (Somerville et al., 2018) and VISual MOTOR (VISMOTOR) (Ances et al., 2009). 

First, the Face Name task (Sperling et al., 2001) taps into episodic memory. The task had three blocks. In the encoding block [Encoding], participants were asked to memorise the names of faces shown. These faces were then shown again in the recall block [Recall] when the participants were asked if they could remember the names of the previously shown faces. There was also the distractor block [Distractor] occurring between the encoding and recall blocks. Here participants were distracted by a Go/NoGo task. We computed six contrasts for this Face Name task: [Encode], [Recall], [Distractor], [Encode vs. Distractor], [Recall vs. Distractor] and [Encode vs. Recall].

Second, the CARIT task (Somerville et al., 2018) was adapted from the classic Go/NoGo task and taps into inhibitory control. Participants were asked to press a budon to all [Go] but not to two [NoGo] shapes. We computed three contrasts for the CARIT task: [NoGo], [Go] and [NoGo vs. Go]. 

Third, the VISMOTOR task (Ances et al., 2009) was designed to test simple activation of the motor and visual cortices. Participants saw a checkerboard with a red square either on the lec or right. They needed to press a corresponding key to indicate the location of the red square. We computed just one contrast for the VISMOTOR task: [Vismotor], which indicates the presence of the checkerboard vs. baseline.” 

Second, for MRI processing procedures, we included the following statements.

From Methods:

“HCP-A provides details of parameters for brain MRI elsewhere (Bookheimer et al., 2019; Harms et al., 2018). Here we used MRI data that were pre-processed by the HCP-A with recommended methods, including the MSMALL alignment (Glasser et al., 2016; Robinson et al., 2018) and ICA-FIX (Glasser et al., 2016) for functional MRI. We used multiple brain MRI modalities, covering task functional MRI (task fMRI), resting-state functional MRI (rsfMRI) and structural MRI (sMRI), and organised them into 19 sets of features.”

“Sets of Features 1-10: Task fMRI contrast (Task Contrast)

Task contrasts reflect fMRI activation relevant to events in each task. Bookheimer and colleagues (2019) provided detailed information about the fMRI in HCP-A. Here we focused on the pre-processed task fMRI Connectivity Informatics Technology Initiative (CIFTI) files with a suffix, “_PA_Atlas_MSMAll_hp0_clean.dtseries.nii.” These CIFTI files encompassed both the cortical mesh surface and subcortical volume (Glasser et al., 2013). Collected using the posterior-to-anterior (PA) phase, these files were aligned using MSMALL (Glasser et al., 2016; Robinson et al., 2018), linear detrended (see hdps://groups.google.com/a/humanconnectome.org/g/hcp-users/c/ZLJc092h980/m/GiihzQAUAwAJ) and cleaned from potential artifacts using ICA-FIX (Glasser et al., 2016). 

To extract Task Contrasts, we regressed the fMRI time series on the convolved task events using a double-gamma canonical hemodynamic response function via FMRIB Software Library (FSL)’s FMRI Expert Analysis Tool (FEAT) (Woolrich et al., 2001). We kept FSL’s default high pass cutoff at 200s (i.e., .005 Hz). We then parcellated the contrast ‘cope’ files, using the Glasser atlas (Gordon et al., 2016) for cortical surface regions and the Freesurfer’s automatic segmentation (aseg) (Fischl et al., 2002) for subcortical regions. This resulted in 379 regions, whose number was, in turn, the number of features for each Task Contrast set of features. “ 

“Sets of Features 11-13: Task fMRI functional connectivity (Task FC)

Task FC reflects functional connectivity (FC ) among the brain regions during each task, which is considered an important source of individual differences (Elliod et al., 2019; Fair et al., 2007; Gradon et al., 2018). We used the same CIFTI file “_PA_Atlas_MSMAll_hp0_clean.dtseries.nii.” as the task contrasts. Unlike Task Contrasts, here we treated the double-gamma, convolved task events as regressors of no interest and focused on the residuals of the regression from each task (Fair et al., 2007). We computed these regressors on FSL, and regressed them in nilearn (Abraham et al., 2014). Following previous work on task FC (Elliod et al., 2019), we applied a highpass at .008 Hz. For parcellation, we used the same atlases as Task Contrast (Fischl et al., 2002; Glasser et al., 2016). We computed Pearson’s correlations of each pair of 379 regions, resulting in a table of 71,631 non-overlapping FC indices for each task. We then applied r-to-z transformation and principal component analysis (PCA) of 75 components (Rasero et al., 2021; Sripada et al., 2019, 2020). Note to avoid data leakage, we conducted the PCA on each training set and applied its definition to the corresponding test set. Accordingly, there were three sets of 75 features for Task FC, one for each task. 

Set of Features 14: Resting-state functional MRI functional connectivity (Rest FC) Similar to Task FC, Rest FC reflects functional connectivity (FC ) among the brain regions, except that Rest FC occurred during the resting (as opposed to task-performing) period. HCPA collected Rest FC from four 6.42-min (488 frames) runs across two days, leading to 26-min long data (Harms et al., 2018). On each day, the study scanned two runs of Rest FC, starting with anterior-to-posterior (AP) and then with posterior-to-anterior (PA) phase encoding polarity. We used the “rfMRI_REST_Atlas_MSMAll_hp0_clean.dscalar.nii” file that was preprocessed and concatenated across the four runs.  We applied the same computations (i.e., highpass filter, parcellation, Pearson’s correlations, r-to-z transformation and PCA) with the Task FC. 

Sets of Features 15-18: Structural MRI (sMRI)

sMRI reflects individual differences in brain anatomy. The HCP-A used an established preprocessing pipeline for sMRI (Glasser et al., 2013). We focused on four sets of features: cortical thickness, cortical surface area, subcortical volume and total brain volume. For cortical thickness and cortical surface area, we used Destrieux’s atlas (Destrieux et al., 2010; Fischl, 2012) from FreeSurfer’s “aparc.stats” file, resulting in 148 regions for each set of features. For subcortical volume, we used the aseg atlas (Fischl et al., 2002) from FreeSurfer’s “aseg.stats” file, resulting in 19 regions. For total brain volume, we had five FreeSurfer-based features: “FS_IntraCranial_Vol” or estimated intra-cranial volume, “FS_TotCort_GM_Vol” or total cortical grey mader volume, “FS_Tot_WM_Vol” or total cortical white mader volume, “FS_SubCort_GM_Vol” or total subcortical grey mader volume and “FS_BrainSegVol_eTIV_Ratio” or ratio of brain segmentation volume to estimated total intracranial volume.”

Third, for regression methods and bias correction methods used, we included the following statements:

From Methods:

“For the machine learning algorithm, we used Elastic Net (Zou & Hastie, 2005). Elastic Net is a general form of penalised regressions (including Lasso and Ridge regression), allowing us to simultaneously draw information across different brain indices to predict one target variable. Penalised regressions are commonly used for building age-prediction models (Jirsaraie, Gorelik, et al., 2023). Previously we showed that the performance of Elastic Net in predicting cognitive abilities is on par, if not better than, many non-linear and morecomplicated algorithms (Pat, Wang, Bartonicek, et al., 2022; Tetereva et al., 2022). Moreover, Elastic Net coefficients are readily explainable, allowing us the ability to explain how our age-prediction and cognition-prediction models made the prediction from each brain feature (Molnar, 2019; Pat, Wang, Bartonicek, et al., 2022) (see below). 

Elastic Net simultaneously minimises the weighted sum of the features’ coefficients. The degree of penalty to the sum of the feature’s coefficients is determined by a shrinkage hyperparameter ‘a’: the greater the a, the more the coefficients shrink, and the more regularised the model becomes. Elastic Net also includes another hyperparameter, ‘ℓ! ratio’, which determines the degree to which the sum of either the squared (known as ‘Ridge’; ℓ! ratio=0) or absolute (known as ‘Lasso’; ℓ! ratio=1) coefficients is penalised (Zou & Hastie, 2005). The objective function of Elastic Net as implemented by sklearn (Pedregosa et al., 2011) is defined as:

where X is the features, y is the target, and b is the coefficient. In our grid search, we tuned two Elastic Net hyperparameters: a using 70 numbers in log space, ranging from .1 and 100, and ℓ!-ratio using 25 numbers in linear space, ranging from 0 and 1.

To understand how Elastic Net made a prediction based on different brain features, we examined the coefficients of the tuned model. Elastic Net coefficients can be considered as feature importance, such that more positive Elastic Net coefficients lead to more positive predicted values and, similarly, more negative Elastic Net coefficients lead to more negative predicted values (Molnar, 2019; Pat, Wang, Bartonicek, et al., 2022). While the magnitude of Elastic Net coefficients is regularised (thus making it difficult for us to interpret the magnitude itself directly), we could still indicate that a brain feature with a higher magnitude weights relatively stronger in making a prediction. Another benefit of Elastic Net as a penalised regression is that the coefficients are less susceptible to collinearity among features as they have already been regularised (Dormann et al., 2013; Pat, Wang, Bartonicek, et al., 2022).

Given that we used five-fold nested cross validation, different outer folds may have different degrees of ‘a’ and ‘ℓ! ratio’, making the final coefficients from different folds to be different. For instance, for certain sets of features, penalisation may not play a big part (i.e., higher or lower ‘a’ leads to similar predictive performance), resulting in different ‘a’ for different folds. To remedy this in the visualisation of Elastic Net feature importance, we refitted the Elastic Net model to the full dataset without spli{ng them into five folds and visualised the coefficients on brain images using Brainspace (Vos De Wael et al., 2020) and Nilern (Abraham et al., 2014) packages. Note, unlike other sets of features, Task FC and Rest FC were modelled acer data reduction via PCA. Thus, for Task FC and Rest FC, we, first, multiplied the absolute PCA scores (extracted from the ‘components_’ attribute of ‘sklearn.decomposition.PCA’) with Elastic Net coefficients and, then, summed the multiplied values across the 75 components, leaving 71,631 ROI-pair indices. “

References

Abraham, A., Pedregosa, F., Eickenberg, M., Gervais, P., Mueller, A., Kossaifi, J., Gramfort, A., Thirion, B., & Varoquaux, G. (2014). Machine learning for neuroimaging with scikitlearn. Frontiers in Neuroinformatics, 8, 14. hdps://doi.org/10.3389/fninf.2014.00014

Ances, B. M., Liang, C. L., Leontiev, O., Perthen, J. E., Fleisher, A. S., Lansing, A. E., & Buxton, R. B. (2009). Effects of aging on cerebral blood flow, oxygen metabolism, and blood oxygenation level dependent responses to visual stimulation. Human Brain Mapping, 30(4), 1120–1132. hdps://doi.org/10.1002/hbm.20574

Bashyam, V. M., Erus, G., Doshi, J., Habes, M., Nasrallah, I. M., Truelove-Hill, M., Srinivasan, D., Mamourian, L., Pomponio, R., Fan, Y., Launer, L. J., Masters, C. L., Maruff, P., Zhuo, C., Völzke, H., Johnson, S. C., Fripp, J., Koutsouleris, N., Saderthwaite, T. D., … on behalf of the ISTAGING Consortium,  the P. A. disease C., ADNI, and CARDIA studies. (2020). MRI signatures of brain age and disease over the lifespan based on a deep brain network and 14 468 individuals worldwide. Brain, 143(7), 2312–2324. hdps://doi.org/10.1093/brain/awaa160

Bookheimer, S. Y., Salat, D. H., Terpstra, M., Ances, B. M., Barch, D. M., Buckner, R. L., Burgess, G. C., Curtiss, S. W., Diaz-Santos, M., Elam, J. S., Fischl, B., Greve, D. N., Hagy, H. A., Harms, M. P., Hatch, O. M., Hedden, T., Hodge, C., Japardi, K. C., Kuhn, T. P., … Yacoub, E. (2019). The Lifespan Human Connectome Project in Aging: An overview. NeuroImage, 185, 335–348. hdps://doi.org/10.1016/j.neuroimage.2018.10.009

Butler, E. R., Chen, A., Ramadan, R., Le, T. T., Ruparel, K., Moore, T. M., Saderthwaite, T. D., Zhang, F., Shou, H., Gur, R. C., Nichols, T. E., & Shinohara, R. T. (2021). Pi alls in brain age analyses. Human Brain Mapping, 42(13), 4092–4101. hdps://doi.org/10.1002/hbm.25533

Cole, J. H. (2020). Multimodality neuroimaging brain-age in UK biobank: Relationship to biomedical, lifestyle, and cognitive factors. Neurobiology of Aging, 92, 34–42. hdps://doi.org/10.1016/j.neurobiolaging.2020.03.014

Destrieux, C., Fischl, B., Dale, A., & Halgren, E. (2010). Automatic parcellation of human cortical gyri and sulci using standard anatomical nomenclature. NeuroImage, 53(1), 1–15. hdps://doi.org/10.1016/j.neuroimage.2010.06.010

Dormann, C. F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., Marquéz, J. R. G., Gruber, B., Lafourcade, B., Leitão, P. J., Münkemüller, T., McClean, C., Osborne, P. E., Reineking, B., Schröder, B., Skidmore, A. K., Zurell, D., & Lautenbach, S. (2013). Collinearity: A review of methods to deal with it and a simulation study evaluating their performance. Ecography, 36(1), 27–46. hdps://doi.org/10.1111/j.16000587.2012.07348.x

Dubois, J., Galdi, P., Paul, L. K., & Adolphs, R. (2018). A distributed brain network predicts general intelligence from resting-state human neuroimaging data. Philosophical Transactions of the Royal Society B: Biological Sciences, 373(1756), 20170284. hdps://doi.org/10.1098/rstb.2017.0284

Elliod, M. L., Knodt, A. R., Cooke, M., Kim, M. J., Melzer, T. R., Keenan, R., Ireland, D., Ramrakha, S., Poulton, R., Caspi, A., Moffid, T. E., & Hariri, A. R. (2019). General functional connectivity: Shared features of resting-state and task fMRI drive reliable and heritable individual differences in functional brain networks. NeuroImage, 189, 516–532. hdps://doi.org/10.1016/j.neuroimage.2019.01.068

Fair, D. A., Schlaggar, B. L., Cohen, A. L., Miezin, F. M., Dosenbach, N. U. F., Wenger, K. K., Fox, M. D., Snyder, A. Z., Raichle, M. E., & Petersen, S. E. (2007). A method for using blocked and event-related fMRI data to study “resting state” functional connectivity. NeuroImage, 35(1), 396–405. hdps://doi.org/10.1016/j.neuroimage.2006.11.051

Fischl, B. (2012). FreeSurfer. NeuroImage, 62(2), 774–781. hdps://doi.org/10.1016/j.neuroimage.2012.01.021

Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., van der Kouwe, A., Killiany, R., Kennedy, D., Klaveness, S., Montillo, A., Makris, N., Rosen, B., & Dale, A. M. (2002). Whole Brain Segmentation. Neuron, 33(3), 341–355. hdps://doi.org/10.1016/S0896-6273(02)00569-X

Glasser, M. F., Smith, S. M., Marcus, D. S., Andersson, J. L. R., Auerbach, E. J., Behrens, T. E. J., Coalson, T. S., Harms, M. P., Jenkinson, M., Moeller, S., Robinson, E. C., Sotiropoulos, S. N., Xu, J., Yacoub, E., Ugurbil, K., & Van Essen, D. C. (2016). The Human Connectome Project’s neuroimaging approach. Nature Neuroscience, 19(9), 1175– 1187. hdps://doi.org/10.1038/nn.4361

Glasser, M. F., Sotiropoulos, S. N., Wilson, J. A., Coalson, T. S., Fischl, B., Andersson, J. L., Xu, J., Jbabdi, S., Webster, M., Polimeni, J. R., Van Essen, D. C., & Jenkinson, M. (2013). The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage, 80, 105–124. hdps://doi.org/10.1016/j.neuroimage.2013.04.127

Gordon, E. M., Laumann, T. O., Adeyemo, B., Huckins, J. F., Kelley, W. M., & Petersen, S. E. (2016). Generation and Evaluation of a Cortical Area Parcellation from Resting-State Correlations. Cerebral Cortex, 26(1), 288–303. hdps://doi.org/10.1093/cercor/bhu239

Gradon, C., Laumann, T. O., Nielsen, A. N., Greene, D. J., Gordon, E. M., Gilmore, A. W., Nelson, S. M., Coalson, R. S., Snyder, A. Z., Schlaggar, B. L., Dosenbach, N. U. F., & Petersen, S. E. (2018). Functional Brain Networks Are Dominated by Stable Group and Individual Factors, Not Cognitive or Daily Variation. Neuron, 98(2), 439-452.e5. hdps://doi.org/10.1016/j.neuron.2018.03.035

Hahn, T., Fisch, L., Ernsting, J., Winter, N. R., Leenings, R., Sarink, K., Emden, D., Kircher, T., Berger, K., & Dannlowski, U. (2021). From ‘loose fi{ng’ to high-performance, uncertainty-aware brain-age modelling. Brain, 144(3), e31–e31. hdps://doi.org/10.1093/brain/awaa454

Harms, M. P., Somerville, L. H., Ances, B. M., Andersson, J., Barch, D. M., Bastiani, M., Bookheimer, S. Y., Brown, T. B., Buckner, R. L., Burgess, G. C., Coalson, T. S., Chappell, M. A., Dapredo, M., Douaud, G., Fischl, B., Glasser, M. F., Greve, D. N., Hodge, C., Jamison, K. W., … Yacoub, E. (2018). Extending the Human Connectome Project across ages: Imaging protocols for the Lifespan Development and Aging projects. NeuroImage, 183, 972–984. hdps://doi.org/10.1016/j.neuroimage.2018.09.060

Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D. S., Quinn, K., Sanislow, C., & Wang, P. (2010). Research Domain Criteria (RDoC): Toward a New Classification Framework for Research on Mental Disorders. American Journal of Psychiatry, 167(7), 748–751. hdps://doi.org/10.1176/appi.ajp.2010.09091379

Jirsaraie, R. J., Gorelik, A. J., Gatavins, M. M., Engemann, D. A., Bogdan, R., Barch, D. M., & Sotiras, A. (2023). A systematic review of multimodal brain age studies: Uncovering a divergence between model accuracy and utility. PaUerns, 4(4), 100712. hdps://doi.org/10.1016/j.pader.2023.100712

Jirsaraie, R. J., Kaufmann, T., Bashyam, V., Erus, G., Luby, J. L., Westlye, L. T., Davatzikos, C., Barch, D. M., & Sotiras, A. (2023). Benchmarking the generalizability of brain age models: Challenges posed by scanner variance and prediction bias. Human Brain Mapping, 44(3), 1118–1128. hdps://doi.org/10.1002/hbm.26144

Marquand, A. F., Rezek, I., Buitelaar, J., & Beckmann, C. F. (2016). Understanding Heterogeneity in Clinical Cohorts Using Normative Models: Beyond Case-Control Studies. Biological Psychiatry, 80(7), 552–561. hdps://doi.org/10.1016/j.biopsych.2015.12.023

Molnar, C. (2019). Interpretable Machine Learning. A Guide for Making Black Box Models Explainable. hdps://christophm.github.io/interpretable-ml-book/

Nimon, K., Lewis, M., Kane, R., & Haynes, R. M. (2008). An R package to compute commonality coefficients in the multiple regression case: An introduction to the package and a practical example. Behavior Research Methods, 40(2), 457–466. hdps://doi.org/10.3758/BRM.40.2.457

Pat, N., Wang, Y., Anney, R., Riglin, L., Thapar, A., & Stringaris, A. (2022). Longitudinally stable, brain-based predictive models mediate the relationships between childhood cognition and socio-demographic, psychological and genetic factors. Human Brain Mapping, hbm.26027. hdps://doi.org/10.1002/hbm.26027

Pat, N., Wang, Y., Bartonicek, A., Candia, J., & Stringaris, A. (2022). Explainable machine learning approach to predict and explain the relationship between task-based fMRI and individual differences in cognition. Cerebral Cortex, bhac235. hdps://doi.org/10.1093/cercor/bhac235

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Predenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., & Duchesnay, É. (2011). Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 12(85), 2825–2830.

Poldrack, R. A., Huckins, G., & Varoquaux, G. (2020). Establishment of Best Practices for Evidence for Prediction: A Review. JAMA Psychiatry, 77(5), 534–540. hdps://doi.org/10.1001/jamapsychiatry.2019.3671

Rasero, J., Sentis, A. I., Yeh, F.-C., & Verstynen, T. (2021). Integrating across neuroimaging modalities boosts prediction accuracy of cognitive ability. PLOS Computational Biology, 17(3), e1008347. hdps://doi.org/10.1371/journal.pcbi.1008347

Robinson, E. C., Garcia, K., Glasser, M. F., Chen, Z., Coalson, T. S., Makropoulos, A., Bozek, J., Wright, R., Schuh, A., Webster, M., Huder, J., Price, A., Cordero Grande, L., Hughes, E., Tusor, N., Bayly, P. V., Van Essen, D. C., Smith, S. M., Edwards, A. D., … Rueckert, D. (2018). Multimodal surface matching with higher-order smoothness constraints. NeuroImage, 167, 453–465. hdps://doi.org/10.1016/j.neuroimage.2017.10.037

Rokicki, J., Wolfers, T., Nordhøy, W., Tesli, N., Quintana, D. S., Alnæs, D., Richard, G., de Lange, A.-M. G., Lund, M. J., Norbom, L., Agartz, I., Melle, I., Nærland, T., Selbæk, G., Persson, K., Nordvik, J. E., Schwarz, E., Andreassen, O. A., Kaufmann, T., & Westlye, L. T. (2021). Multimodal imaging improves brain age prediction and reveals distinct abnormalities in patients with psychiatric and neurological disorders. Human Brain Mapping, 42(6), 1714–1726. hdps://doi.org/10.1002/hbm.25323

Somerville, L. H., Bookheimer, S. Y., Buckner, R. L., Burgess, G. C., Curtiss, S. W., Dapredo, M., Elam, J. S., Gaffrey, M. S., Harms, M. P., Hodge, C., Kandala, S., Kastman, E. K., Nichols, T. E., Schlaggar, B. L., Smith, S. M., Thomas, K. M., Yacoub, E., Van Essen, D. C., & Barch, D. M. (2018). The Lifespan Human Connectome Project in Development: A large-scale study of brain connectivity development in 5–21 year olds. NeuroImage, 183, 456–468. hdps://doi.org/10.1016/j.neuroimage.2018.08.050

Sperling, R. A., Bates, J. F., Cocchiarella, A. J., Schacter, D. L., Rosen, B. R., & Albert, M. S. (2001). Encoding novel face-name associations: A functional MRI study. Human Brain Mapping, 14(3), 129–139. hdps://doi.org/10.1002/hbm.1047

Sripada, C., Angstadt, M., Rutherford, S., Kessler, D., Kim, Y., Yee, M., & Levina, E. (2019). Basic Units of Inter-Individual Variation in Resting State Connectomes. Scientific Reports, 9(1), Article 1. hdps://doi.org/10.1038/s41598-018-38406-5

Sripada, C., Angstadt, M., Rutherford, S., Taxali, A., & Shedden, K. (2020). Toward a “treadmill test” for cognition: Improved prediction of general cognitive ability from the task activated brain. Human Brain Mapping, 41(12), 3186–3197. hdps://doi.org/10.1002/hbm.25007

Tetereva, A., Li, J., Deng, J. D., Stringaris, A., & Pat, N. (2022). Capturing brain-cognition relationship: Integrating task-based fMRI across tasks markedly boosts prediction and test-retest reliability. NeuroImage, 263, 119588. hdps://doi.org/10.1016/j.neuroimage.2022.119588

Vieira, B. H., Pamplona, G. S. P., Fachinello, K., Silva, A. K., Foss, M. P., & Salmon, C. E. G. (2022). On the prediction of human intelligence from neuroimaging: A systematic review of methods and reporting. Intelligence, 93, 101654. hdps://doi.org/10.1016/j.intell.2022.101654

Vos De Wael, R., Benkarim, O., Paquola, C., Lariviere, S., Royer, J., Tavakol, S., Xu, T., Hong, S.J., Langs, G., Valk, S., Misic, B., Milham, M., Margulies, D., Smallwood, J., & Bernhardt, B. C. (2020). BrainSpace: A toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets. Communications Biology, 3(1), 103. hdps://doi.org/10.1038/s42003-020-0794-7

Woolrich, M. W., Ripley, B. D., Brady, M., & Smith, S. M. (2001). Temporal Autocorrelation in Univariate Linear Modeling of FMRI Data. NeuroImage, 14(6), 1370–1386. hdps://doi.org/10.1006/nimg.2001.0931

Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 67(2), 301–320. hdps://doi.org/10.1111/j.1467-9868.2005.00503.x

eLife assessment

This useful manuscript challenges the utility of current paradigms for estimating brain-age with magnetic resonance imaging measures. It presents solid evidence to support the suggestion that an alternative approach focused on predicting cognition may be more beneficial. This work will be of interest to researchers working on brain-age and related models.

Reviewer #1 (Public Review):

In this paper, the authors evaluate the utility of brain-age derived metrics for predicting cognitive decline by performing a 'commonality' analysis in a downstream regression that enables the different contribution of different predictors to be assessed. The main conclusion is that brain-age derived metrics do not explain much additional variation in cognition over and above what is already explained by age. The authors propose to use a regression model trained to predict cognition ('brain cognition') as an alternative suited to applications of cognitive decline. While this is less accurate overall than brain age, it explains more unique variance in the downstream regression.

Comments on revised version:

I thank the authors for the revision of the manuscript and for being more explicit about the inherent conceptual limitations of Brain Age / Brain Cognition. I have no further comments.

eLife assessment

This work explores the role of one the most abundant circRNAs, circHIPK3, in bladder cancer cells, showing with convincing data that circHIPK3 depletion affects thousands of genes and that those downregulated (including STAT3) share an 11-mer motif with circHIPK3, corresponding to a binding site for IGF2BP2. The experiments demonstrate that circHIPK3 can compete with the downregulated mRNAs targets for IGF2BP2 binding and that IGF2BP2 depletion antagonizes the effect of circHIPK3 depletion by upregulating the genes containing the 11-mer. These important findings contribute to the growing recognition of the complexity of cancer signaling regulation and highlight the intricate interplay between circRNAs and protein-coding genes in tumorigenesis.

Reviewer #1 (Public Review):

In this work the authors propose a new regulatory role for one the most abundant circRNAs, circHIPK3. They demonstrate that circHIPK3 interacts with an RNA binding protein (IGF2BP2), sequestering it away from its target mRNAs. This interaction is shown to regulates the expression of hundreds of genes that share a specific sequence motif (11-mer motif) in their untranslated regions (3'-UTR), identical to one present in circHIPK3 where IGF2BP2 binds. The study further focuses on the specific case of STAT3 gene, whose mRNA product is found to be downregulated upon circHIPK3 depletion. This suggests that circHIPK3 sequesters IGF2BP2, preventing it from binding to and destabilizing STAT3 mRNA. The study presents evidence supporting this mechanism and discusses its potential role in tumor cell progression. These findings contribute to the growing complexity of understanding cancer regulation and highlight the intricate interplay between circRNAs and protein-coding genes in tumorigenesis.

Strengths:

The authors show mechanistic insight into a proposed novel "sponging" function of circHIPK3 which is not mediated by sequestering miRNAs but rather a specific RNA binding protein (IGF2BP2). They address the stoichiometry of the molecules involved in the interaction, which is a critical aspect that is frequently overlooked in this type of studies. They provide both genome-wide analysis and a specific case (STAT3) which is relevant for cancer progression. Overall, the authors have significantly improved their manuscript in their revised version.

Weaknesses:

There are seemingly contradictory effects of circHIPK3 and STAT3 depletion in cancer progression. However, the authors have addressed these issues in their revised manuscript, incorporating potential reasons that might explain such complexity.

Reviewer #2 (Public Review):

The manuscript by Okholm and colleagues identified an interesting new instance of ceRNA involving a circular RNA. The data are clearly presented and support the conclusions. Quantification of the copy number of circRNA and quantification of the protein were performed, and this is important to support the ceRNA mechanism.

This is the second rebuttal and the authors further improved the manuscript. The data are of interest for the large spectrum of readers of the journal.

Reviewer #3 (Public Review):

Summary:

In Okholm et al., the authors evaluate the functional impact of circHIPK3 in bladder cancer cells. By knocking it down and performing an RNA-seq analysis, the authors found thousand deregulated genes which look unaffected by miRNAs sponging function and that are, instead, enriched for a 11-mer motif. Further investigations showed that the 11-mer motif is shared with the circHIPK3 and able to bind the IGF2BP2 protein. The authors validated the binding of IGF2BP2 and demonstrated that IGF2BP2 KD antagonizes the effect of circHIPK3 KD and leads to the upregulation of genes containing the 11-mer. Among the genes affected by circHIPK3 KD and IGF2BP2 KD, resulting in downregulation and upregulation respectively, the authors found STAT3 gene which also consistently leads to the concomitant upregulation of one of its targets TP53. The authors propose a mechanism of competition between circHIPK3 and IGF2BP2 triggered by IGF2BP2 nucleation, potentially via phase separation.

Strengths:

The number of circRNAs continues to drastically grow however the field lacks detailed molecular investigations. The presented work critically addresses some of the major pitfalls in the field of circRNAs and there has been a careful analysis of aspects frequently poorly investigated. The time-point KD followed by RNA-seq, investigation of miRNAs-sponge function of circHIPK3, identification of 11-mer motif, identification and validation of IGF2BP2, and the analysis of copy number ratio between circHIPK3 and IGF2BP2 in assessing the potential ceRNA mode of action have been extensively explored and, comprehensively convincing.

Weaknesses:

The authors addressed the majority of the weak points raised initially. However the role played by the circHIPK3 in cancer remains elusive and not elucidated in full in this study.

Overall, the presented study surely adds some further knowledge in describing circHIPK3 function, its capability to regulate some downstream genes, and its interaction and competition for IGF2BP2. However, whereas the experimental part sounds technically logical, it remains unclear the overall goal of this study and the achieved final conclusions.

This study is a promising step forward in the comprehension of the functional role of circHIPK3. These data could possibly help to better understand the circHIPK3 role in cancer

Author response:

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

eLife assessment 

This study explores the role of one the most abundant circRNAs, circHIPK3, in bladder cancer cells, providing convincing data that circHIPK3 depletion affects thousands of genes and that those downregulated (including STAT3) share an 11-mer motif with circHIPK3, corresponding to a binding site for IGF2BP2. The experiments demonstrate that circHIPK3 can compete with the downregulated mRNAs targets for IGF2BP2 binding and that IGF2BP2 depletion antagonizes the effect of circHIPK3 depletion by upregulating the genes containing the 11mer motif. These valuable findings contribute to the growing recognition of the complexity of cancer signaling regulation and highlight the intricate interplay between circRNAs and protein-coding genes in tumorigenesis. 

Public Reviews: 

Reviewer #1 (Public Review): 

In this work the authors propose a new regulatory role for one the most abundant circRNAs, circHIPK3. They demonstrate that circHIPK3 interacts with an RNA binding protein (IGF2BP2), sequestering it away from its target mRNAs. This interaction is shown to regulates the expression of hundreds of genes that share a specific sequence motif (11-mer motif) in their untranslated regions (3'-UTR), identical to one present in circHIPK3 where IGF2BP2 binds. The study further focuses on the specific case of STAT3 gene, whose mRNA product is found to be downregulated upon circHIPK3 depletion. This suggests that circHIPK3 sequesters IGF2BP2, preventing it from binding to and destabilizing STAT3 mRNA. The study presents evidence supporting this mechanism and discusses its potential role in tumor cell progression. These findings contribute to the growing complexity of understanding cancer regulation and highlight the intricate interplay between circRNAs and protein-coding genes in tumorigenesis.

Strengths:

The authors show mechanistic insight into a proposed novel "sponging" function of

circHIPK3 which is not mediated by sequestering miRNAs but rather a specific RNA binding protein (IGF2BP2). They address the stoichiometry of the molecules involved in the interaction, which is a critical aspect that is frequently overlooked in this type of studies. They provide both genome-wide analysis and a specific case (STAT3) which is relevant for cancer progression. Overall, the authors have significantly improved their manuscript in their revised version.

Weaknesses:

While the authors have performed northern blots to measure circRNA levels, an estimation of the circRNA overexpression efficiency, namely the circular-to-linear expression ratio, would be desired. The seemingly contradictory effects of circHIPK3 and STAT3 depletion in cancer progression, are now addressed by the authors in their revised manuscript, incorporating potential reasons that might explain such complexity.

We have now included a full version of the northern blot, where no discernible linear precursor can be detected, supporting efficient circHIPK3 WT and circHIPK3 MUT production (please see the detailed description in the specific comments below). We agree that the observations about STAT3 homeostasis and cancer progression, is not a straightforward extrapolation as discussed. 

Reviewer #2 (Public Review):

Summary: 

The authors have diligently addressed most of the points raised during the review process (except the important point of "additional in vitro experiments [...] needed to investigate the implication of circHIPK3 in bladder cancer cell phenotype" for which no additional experiments were performed), resulting in an improvement in the study. The data are now described with clarity and conciseness, enhancing the overall quality of the manuscript. 

Strengths: 

New, well-defined molecular mechanism of circRNAs involvement in bladder cancer. 

Weaknesses: 

Lack of solid translational significance data. 

The focus of this study has been to disclose molecular mechanisms of action by circHIPK3, with implications for cancer. We agree that further studies are needed to fully understand the impact of circHIPK3 in bladder cancer.  

Reviewer #3 (Public Review):

In Okholm et al., the authors evaluate the functional impact of circHIPK3 in bladder cancer cells. By knocking down circHIPK3 and performing an RNA-seq analysis, the authors found thousands of deregulated genes which look unaffected by miRNAs sponging function and that are, instead, enriched for a 11-mer motif. Further investigations showed that the 11mer motif is shared with the circHIPK3 and able to bind the IGF2BP2 protein. The authors validated the binding of IGF2BP2 and demonstrated that IGF2BP2 KD antagonizes the effect of circHIPK3 KD and leads to the upregulation of genes containing the 11-mer. Among the genes affected by circHIPK3 KD and IGF2BP2 KD, resulting in downregulation and upregulation respectively, the authors found the STAT3 gene, which also consistently has concomitant upregulation of one of its targets TP53. The authors propose a mechanism of competition between circHIPK3 and IGF2BP2 triggered by IGF2BP2 nucleation, potentially via phase separation. 

Strengths: 

Although the number of circRNAs continues to grow, this field lacks many instances of detailed molecular investigations. The presented work critically addresses some of the major piaalls in the field of circRNAs, and there has been a careful analysis of aspects frequently poorly investigated. Experiments involving use of time-point knockdown followed by RNAseq, investigation of miRNA-sponge function of circHIPK3, identification of 11-mer motif, identification and validation of IGF2BP2, and the analysis of copy number ratio between circHIPK3 and IGF2BP2 in assessing the potential ceRNA mode of action are thorough and convincing. 

Weaknesses: 

It is unclear why the authors used certain bladder cancer cells versus non-bladder cells in some experiments. The efficacy of certain experiments (specifically rescue experiments) and some control conditions is still questionable. Overall, the presented study adds some further knowledge in describing circHIPK3 function, its capability to regulate some downstream genes, and its interaction and competition for IGF2BP2. 

We have provided a discussion and argumentation of how certain bladder cancer cells (and non-bladder cancer cells) have been used in this study in our previous rebuttal letter and also clarified this further in the materials and methods section in the first revision. Regarding control conditions for experiments, we believe we have included all necessary controls and explanations for these in the revised version (please see the detailed description in the specific comments below). 

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

Major points about revised manuscript

(1) In Supplementary Figure S5H, the membrane may have been trimmed too closely to the circRNA band, potentially resulting in the absence of the linear RNA band. Could the authors provide a full image of the membrane that includes the loading points? Having access to the complete image would allow for a more comprehensive evaluation of the results, including the presence or absence of expected linear and circular RNA bands.

I have taken the liberty to move this “major point” from the public review section, as I believe it would be too detailed for this section. We have included the full section of the northern blot, according to the reviewers recommendations. 

As described in the previous rebuttal letter our northern blots suffered from heavy background signal arising from the rRNA bands, which was the reason for cuttng the northern blot in the previous version of Supplementary figure S5H. We have now shown the entire blot as suggested by the reviewer, so that the reader can more clearly inspect any potential linear precursor band. We previously stated that we could not assess the circular-to-linear ratio due to background signal, since a potential linear HIPK3 precursor RNA could be masked by the rRNA signal. However, the theoretical size of a linear precursor is ~2.9 kb – a region where we do not detect any distinct bands (just above the 18S band), making a rather efficient circularization very likely. In support of this claim, we are using the Laccase2 vector described in Kramer et al, 2015 (Genes dev), which is proven to produce high levels of circHIPK3 compared to negligable amounts of linear precursor (although in a different cell line). We have also included a 5.8S rRNA probe to control for loading and RNase R activity (can also be ascertained by the disappearence of 18S/28S bands). Since we do not have the option to use another probe (limited by the BSJ-specific probe) and it is not practical to deplete for rRNA from 20 µg samples of total RNA, prior to running the northern blot, we find that this data sufficiently proves that our vector constructs produce a decent amount of RNase R-resistant circHIPK3, with no visible/discernible linear precursor.    

Minor points about revised manuscript

(1) In Supplementary Figure S3B, the authors offer no explanation as to why genes that become upregulated upon circHIPK3 knockdown generally contain more circHIPK3-RBP binding sites other than for IGF2BP2. A clarification would be of help.

Again, this issue has been addressed in the previous rebuttal letter. Our response is repeated below:

We do not have any evidence to explain this observation. One possibility is that other RBPs elicit mRNA-stabilizing effects on average, whereas abundant IGF2BP2 (~ 120.000200.000 copies per cell) now able to bind more target mRNAs and elicit destabilization. This remains highly speculative though.

(2) In Supplementary Figure S3D, the authors' claim that the 11-mer motif is found more bound to IGF2BP2 than for other circHIPK3-RBPs should referred to the corresponding dataset/reference.

Again, this issue has been addressed in the previous rebuttal letter. Our response is repeated below:

This information is stated in the figure legend (K562) and we have now included it in the main text as well: “We evaluated how often binding sites of circHIPK3-RBPs overlap the 11-mer motif and found that this is more often the case for IGF2BP2 binding sites than binding sites of the other circHIPK3-RBPs when scrutinizing K562 datasets (Supplementary Figure S3D)”.

(3) In the rescue experiment where both circHIPK3 and IGF2BP2 are downregulated, using the term "normalization" to mean reestablishing normal levels of gene expression can lead to confusion with the concept of normalization as it is commonly understood in the context of data processing (i.e. the mathematical process of adjusting data to account for various factors that might affect measurements). I would recommend the authors to use a term that more specifically describes the biological process they are referring to, such as "restoration of normal expression levels" or simply "return to normal levels".

We agree that this term could be misunderstood. This has now been changed as recommended.

(4) The figure legend of Supplementary Figure 5F is wrongly labeled. The legend for panel F actually corresponds to panel G and vice versa. 

This has now been corrected.  

Reviewer #2 (Recommendations For The Authors): 

The authors have diligently addressed most of the points raised during the review process (except the important point of "additional in vitro experiments [...] needed to investigate the implication of circHIPK3 in bladder cancer cell phenotype" for which no additional experiments were performed), resulting in an improvement in the study. The data are now described with clarity and conciseness, enhancing the overall quality of the manuscript. Therefore, I support the publication of this work. 

We thank the reviewer for the positive comments.

Reviewer #3 (Recommendations For The Authors): 

Please ensure that when the changes are made (especially for major points) by addressing the reviewer's comments, these are all appropriately incorporated in the text (for example the use of Act B as a low affinity positive control (now in Fig 4A), is not explained in the text neither the legends/methods) 

This has now been included.

Please ensure that all the legends correspond to the right figures (eg: Supplementary Figure with rescue experiment is 5F, but the corresponding legend in the manuscript is the S5G). 

This has now been corrected.

Please for future reviewing processes ensure the new parts are properly highlighted or coloured differently in the manuscript

This has now been done more thoroughly.

Author response:

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

Below, we provide a detailed account of the changes we made. For clarity and ease of review:

•        Original reviewers' comments are included and highlighted in grey

•        Our responses to each comment are written in black text

•        Print screens illustrating the specific changes made to the manuscript are enclosed within black squares

eLife assessment

The authors aim to develop a CRISPR system that can be activated upon sensing an RNA. As an initial step to this goal, they describe RNA-sensing guide RNAs for controlled activation of CRISPR modification. Many of the data look convincing and while several steps remain to achieve the stated goal in an in vivo setting and for robust activation by endogenous RNAs, the current work will be important for many in the field.  

The eLife assessment summarises our ambition to create a CRISPR system controlled by RNA sensing. The synopsis provided encapsulates the essence of our research, emphasising both the progress we have made and the challenges that lie ahead. This assessment fully resonates with our views.

Public Reviews:

Reviewer #1 (Public Review):

This paper describes RNA-sensing guide RNAs for controlled activation of CRISPR modification. This works by having an extended guide RNA with a sequence that folds back onto the targeting sequence such that the guide RNA cannot hybridise to its genomic target. The CRISPR is "activated" by the introduction of another RNA, referred to as a trigger, that competes with this "back folding" to make the guide RNA available for genome targeting. The authors first confirm the efficacy of the approach using several RNA triggers and a GFP reporter that is activated by dCas9 fused to transcriptional activators. A major potential application of this technique is the activation of CRISPR in response to endogenous biomarkers. As these will typically be longer than the first generation triggers employed by the authors they test some extended triggers, which also work though not always to the same extent. They then introduce MODesign which may enable the design of bespoke or improved triggers. After that, they determine that the mode of activation by the RNA trigger involves cleavage of the RNA complexes. Finally, they test the potential for their system to work in a developmental setting - specifically zebrafish embryos. There is some encouraging evidence, though the effects appear more subtle than those originally obtained in cell culture. 

Overall, the potential of a CRISPR system that can be activated upon sensing an RNA is high and there are a myriad of opportunities and applications for it. This paper represents a reasonable starting point having developed such a system in principle. 

The weakness of the study is that it does not demonstrate that the system can be used in a completely natural setting. This would require an endogenous transcript as the RNA trigger with a clear readout. Such an experiment would clearly strengthen the paper and provide strong confidence that the method could be employed for one of the major applications discussed by the authors. The zebrafish data relied on exogenous RNA triggers whereas the major applications (as I understood them) would use endogenous triggers. 

Related, most endogenous RNAs are longer than the various triggers tested and may require extensive modification of the system to be detected or utilised effectively. 

While additional data would clearly be beneficial, there should nevertheless be a more detailed discussion of these caveats and/or the strengths and applications of the system as it is presented (i.e. utility with synthetic triggers).  

We agree with the observation regarding the subtler effects in the zebrafish embryos and the reliance on exogenous RNA triggers. Indeed, the utilisation of endogenous transcripts as triggers in a natural setting is a logical next step. We further acknowledge the need to delve deeper into the complexities and challenges of our system, particularly concerning the detection of endogenous RNA, thus offering valuable insights for researchers looking to adapt our system for various applications. In order to clarify these limitations, we made some changes in the final version of our paper. The following paragraphs have been therefore included in the manuscript discussion:

“In their current iteration, iSBH-sgRNAs show considerable promise for mammalian synthetic biology applications. Specifically, their ability to detect synthetic triggers could be pivotal in the development of complex synthetic RNA circuits and logic gates, thereby advancing the field of cellular reprogramming. However, further work is required to achieve better ON/OFF activation ratios in vivo and more homogeneous activity across tissues in the presence of RNA triggers. Additional chemical modifications could improve iSBH-sgRNA properties, and we believe that chemical modification strategies adopted for siRNA drugs or antisense oligos (Khvorova and Watts (2017)) could also be essential for further iSBH-sgRNA technology development. As iSBH-sgRNAs might be targeted by endogenous nucleases, leading to their degradation, a strategy for preventing this could involve additional chemical modifications. When inserted at certain key positions, such modifications could prevent interaction between iSBH-sgRNAs and cellular enzymes by introducing steric clashes or inhibiting RNA hydrolysis.

Once achieving superior dynamic ranges of iSBH-sgRNA activation in vivo, the next steps would involve understanding the classes of endogenous RNAs that could act as triggers. The chances that an iSBH-sgRNA encounters an endogenous RNA trigger inside a cell would depend on the relative concentrations of the two RNA species. Therefore, a first step towards determining potential endogenous RNA triggers will involve identifying RNA species with comparable expression levels as iSBH-sgRNAs. Then, iSBH-sgRNAs could be designed against these RNA species, followed by experimental validation. It is important to note that eukaryotic cells express a wide range of transcripts of varying sizes, expression levels, and subcellular localisations, all of which could greatly affect iSBH-sgRNA activation levels. Based on the data presented here, we speculate that RNA species up to 300nt that are also highly expressed might act as good triggers. Furthermore, as sgRNAs are involved in targeting Cas9 to genomic DNA in the nucleus, attempting to detect transcripts that are sequestered in the nucleus might also provide additional benefit.”

Reviewer #3 (Public Review):

In this work, the authors describe engineering of sgRNAs that render Cas9 DNA binding controllable by a second RNA trigger. The authors introduce several iterations of their engineered sgRNAs, as well as a computational pipeline to identify designs for user-specified RNA triggers which offers a helpful alternative to purely rational design. Also included is an investigation of the fate of the engineered sgRNAs when introduced into cells, and the use of this information to inform installation of modified nucleotides to improve engineered sgRNA stability. Engineered sgRNAs are demonstrated to be activated by trigger RNAs in both cultured mammalian cells and zebrafish. 

The conclusions made by the authors in this work are predominantly supported by the data provided. However, some claims are not consistent with the data shown and some of the figures would benefit from revision or further clarification. 

Strengths: 

- The sgRNA engineering in this paper is performed and presented in a systematic and logical fashion.

- Inclusion of a computational method to predict iSBH-sgRNAs adds to the strength of the engineering. 

- Investigation into the cellular fate of the engineered sgRNAs and the use of this information to guide inclusion of chemically modified nucleotides is also a strength. 

- Demonstration of activity in both cultured mammalian cells and in zebrafish embryos increases the impact and utility of the technology reported in this work. 

Weaknesses: 

- While the methods here represent an important step forward in advancing the technology, they still fall short of the dynamic range and selectivity likely required for robust activation by endogenous RNA.

- While the iSBH-sgRNAs where the RNA trigger overlaps with the spacer appear to function robustly, the modular iSBH-sgRNAs seem to perform quite a bit less well. The authors state that modular iSBHsgRNAs show better activity without increasing background when the SAM system is added, but this is not supported by the data shown in Figure 3D, where in 3 out of 4 cases CRISPR activation in the absence of the RNA trigger is substantially increased.

- There is very little discussion of how the performance of the technology reported in this work compares to previous iterations of RNA-triggered CRISPR systems, of which there are many examples.  

Concerning the methods falling short of the dynamic range and selectivity required for robust activation by endogenous RNA, we acknowledge this limitation and recognise the need for improvement in this area. In the resubmitted version of the manuscript, we provided a detailed discussion on how the selection of appropriate triggers might partially improve dynamic ranges and selectivity. This includes an exploration of various strategies and considerations that may enhance the robustness of our system (print screen above, also used for addressing Reviewer #1 comments). 

Regarding the inconsistent performance of the modular iSBH-sgRNAs, we acknowledge that modular iSBH-sgRNAs seem to perform slightly less well than first- and second-generation designs. In order to illustrate this, we modified corresponding bar graphs to include fold turn-on iSBH-sgRNA activation in addition to significance (Figures 1, 2 and 3 of the manuscript). We also acknowledge this fact in the text, as well as we recognise this discrepancy in the Figure 3.D and provide further clarifications. To help conveying this message even further, we introduced a new figure (Figure 3- figure supplement 2) to accompany the heat map shown in the Figure 3.D. with corresponding bar graphs. These changes are documented below:

“…promoters. We ran 11 MODesign simulations for each trigger, incrementally extending the loop size while keeping the sgRNA 2 spacer input constant. HEK293T validation experiments showed that choosing modular iSBH-sgRNAs that detect the 4 U6-expressed triggers is possible (Figure 3.D, Figure 3- figure supplement 1.C). Despite not performing quite as well as second-generation designs (Figure 2.A.,Figure 3.D),modular iSBH-sgRNA still enable efficient RNA detection, especially for smaller RNAs such as triggers A and D. For highly efficient designs such asmodular iSBH-sgRNA (D), addition of the SAM effector system (Konermann et al. (2015)) boosted ON-state activation with only a negligible increase in the the OFF-state non-specific activation. Orthogonality tests suggested that activation of modular iSBH-sgRNA designs was specifically conditioned by complementary RNA triggers (Figure 3.E, Figure 3 - figure supplement 2), showing the exquisite specificity of the system.”

Author response image 1.

This supplementary figure reinterprets the data presented in Figure 3.E. using bar plots for enhanced clarity and comparison. It depicts the results of cotransfecting HEK293T cells with four modular iSBH-sgRNAs (A, B, C, and D) and examines all combinations of iSBH-sgRNA: RNA trigger pairings. The bar plots provide a visual representation of mean values with error bars indicating the standard deviation, based on three biological replicates.

Regarding the concern about the lack of comparison with previous iterations of RNA-triggered CRISPR systems, we also acknowledged other similar technologies within the discussion. We also point readers to a literature review we recently published (doi/full/10.1089/crispr.2022.0052) where we describe other similar technologies in more detail.

“To date, a variety of RNA-inducible gRNA designs have been developed (Hanewich-Hollatz et al. (2019); Hochrein et al. (2021); Jakimo et al. (2018); Jiao et al. (2021); Jin et al. (2019); Li et al. (2019); Liu et al. (2022); Lin et al. (2020); Siu and Chen (2019); Galizi et al. (2020); Hunt and Chen (2022b,a); Ying et al. (2020); Choi et al. (2023)). Nevertheless, there is a lack of direct, head-to-head comparisons of these designs under standardised experimental conditions. Some designs were evaluated in vitro, others in bacterial systems, and some in mammalian cells. Consequently, it is challenging to conclusively determine which design exhibits superior properties (Pelea et al. (2022)). Notably, to the best of our knowledge, the iSBH-sgRNA systemis the first RNA-inducible gRNA design tested in vivo and characterising the iSBH-sgRNA activation mechanism was essential for implementing iSBH-sgRNA technology in zebrafish embryos. In vivo, chemical modifications in the spacer sequence were vital for iSBH-sgRNA stability and function.”

eLife assessment

The authors aim to develop a CRISPR system that can be activated upon sensing an RNA. As an initial step to this goal, they describe RNA-sensing guide RNAs for controlled activation of CRISPR modification. Many of the data look convincing and while several steps remain to achieve the stated goal in an in vivo setting and for robust activation by endogenous RNAs, the current work will be important for many in the field.

Reviewer #1 (Public Review):

This paper describes RNA-sensing guide RNAs for controlled activation of CRISPR modification. This works by having an extended guide RNA with a sequence that folds back onto the targeting sequence such that the guide RNA cannot hybridise to its genomic target. The CRISPR is "activated" by the introduction of another RNA, referred to as a trigger, that competes with this "back folding" to make the guide RNA available for genome targeting. The authors first confirm the efficacy of the approach using several RNA triggers and a GFP reporter that is activated by dCas9 fused to transcriptional activators. A major potential application of this technique is the activation of CRISPR in response to endogenous biomarkers. As these will typically be longer than the first generation triggers employed by the authors they test some extended triggers, which also work though not always to the same extent. They then introduce MODesign which may enable the design of bespoke or improved triggers. After that, they determine that the mode of activation by the RNA trigger involves cleavage of the RNA complexes. Finally, they test the potential for their system to work in a developmental setting - specifically zebrafish embryos. There is some encouraging evidence, though the effects appear more subtle than those originally obtained in cell culture.

Overall, the potential of a CRISPR system that can be activated upon sensing an RNA is high and there are a myriad of opportunities and applications for it. This paper represents a reasonable starting point having developed such a system in principle.<br /> The weakness of the study is that it does not demonstrate that the system can be used in a completely natural setting. This would require an endogenous transcript as the RNA trigger with a clear readout. The authors now acknowledge this limitation in their revised manuscript. Future studies and experiments should focus on these aspects in order for the system to be employed to its full and intended potential.

eLife assessment

This study presents valuable findings describing how two brain regions, the midbrain periaqueductal gray matter and basolateral amygdala, communicate when a predator threat is detected. Though the periaqueductal gray is usually viewed as a downstream effector, this work contributes to a growing body of literature from this lab showing that the periaqueductal gray produces effects by acting on the basolateral amygdala. The experimental design, data collection and analysis methods provide solid evidence for the main claims. Although anatomical and immediately early gene results suggest the paraventricular nucleus of the thalamus may serve as a mediator of dorsolateral periaqueductal gray to basolateral amygdala neurotransmission, this finding would benefit from a functional assessment. This study will appeal to a broad audience, including basic scientists interested in neural circuits, basic and clinical researchers interested in fear, and behavioral ecologists interested in foraging.

Reviewer #1 (Public Review):

In the presence of predators, animals display attenuated foraging responses and increased defensive behaviors that serve to protect them from potential predatory attacks. Previous studies have shown that the basolateral nucleus of the amygdala (BLA) and the periaqueductal gray matter (PAG) are necessary for the acquisition and expression of conditioned fear responses. However, it remains unclear how BLA and PAG neurons respond to predatory threats when animals are foraging for food. To address this question, Kim and colleagues conducted in vivo electrophysiological recordings from BLA and PAG neurons and assessed approach-avoidance responses while rats search for food in the presence of a robotic predator.

The authors observed that rats exhibited a significant increase in the latency to obtain the food pellets and a reduction in the pellet success rate when the predator robot was activated. A subpopulation of PAG neurons showing increased firing rate in response to the robot activation didn't change their activity in response to food pellet retrieval during the pre- or post-robot sessions. Optogenetic stimulation of PAG neurons increased the latency to procure the food pellet in a frequency- and intensity-dependent manner, similar to what was observed during the robot test. Combining optogenetics with single-unit recordings, the authors demonstrated that photoactivation of PAG neurons increased the firing rate of 10% of BLA cells. A subsequent behavioral test in 3 of these same rats demonstrated that BLA neurons responsive to PAG stimulation displayed higher firing rates to the robot than BLA neurons nonresponsive to PAG stimulation. Next, because the PAG does not project monosynaptically to the BLA, the authors used a combination of retrograde and anterograde neural tracing to identify possible regions that could convey robot-related information from PAG to the BLA. They observed that neurons in specific areas of the paraventricular nucleus of the thalamus (PVT) that are innervated by PAG fibers contained neurons that were retrogradely labeled by the injection of CTB in the BLA. In addition, PVT neurons showed increased expression of the neural activity marker cFos after the robot test, suggesting that PVT may be a mediator of PAG signals to the BLA.

Overall, the idea that the PAG interacts with the BLA via the midline thalamus during a predator vs. foraging test is new and quite interesting. The authors have used appropriated tools to address their questions.

In this revised version of the manuscript, the authors have made important modifications in the text, inserted new data analyses, and incorporated additional references, as recommended by the reviewers. These modifications have significantly improved the quality of the manuscript.

Reviewer #2 (Public Review):

The authors characterized activity of the dorsal periaqueductal gray (dPAG) - basolateral amygdala (BLA) circuit. They show that BLA cells that are activated by dPAG stimulation are also more likely to be activated by a robot predator. These same cells are also more likely to display synchronous firing.

The authors also replicate prior results showing that dPAG stimulation evokes fear and the dPAG is activated by a predator.

Lastly, the report performs anatomical tracing to show that the dPAG may act on the BLA via the paraventricular thalamus (PVT). Indeed, the PVT receives dPAG projections and also projects to the BLA. However, the authors do not show if the PVT mediates dPAG to BLA communication with any functional behavioral assay. Furthermore, the authors also do not thoroughly characterize the activity of BLA cells during the predatory assay.

The major impact in the field would be to add evidence to their prior work, strengthening the view that the BLA can be downstream of the dPAG.

Reviewer #3 (Public Review):

In the present study, the authors examined how dPAG neurons respond to predatory threats and how dPAG and BLA communicate threat signals. The authors employed single-unit recording and optogenetics tools to address these issues in an 'approach food-avoid predator' paradigm. They characterized dPAG and BLA neurons responsive to a looming robot predator and found that dPAG opto-stimulation elicited fleeing and increased BLA activity. Importantly, they found that dPAG stimulation produces activity changes in subpopulations of BLA neurons related to predator detection, thus supporting the idea that dPAG conveys innate fear signals to the amygdala. In addition, injections of anterograde and retrograde tracers into the dPAG and BLA, respectively, along with the examination of c-FOS activity in midline thalamic relay stations, suggest that the paraventricular nucleus of the thalamus (PVT) may serve as a mediator of dPAG to BLA neurotransmission. Of relevance, the study helps to validate an important concept that dPAG mediates primal fear emotion and may engage upstream amygdalar targets to evoke defensive responses. The series of experiments provide a compelling case for supporting their conclusions. The study brings important concepts revealing dynamics of fear-related circuits particularly attractive to a broad audience, from basic scientists interested in neural circuits to psychiatrists.

Author response:

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

Recommendations for the authors:

We sincerely value the insightful and constructive feedback (italicized) provided by the reviewers, which has been instrumental in identifying areas of our manuscript that required further clarification or amendment. In response to these valuable comments, we have significantly revised the manuscript to enhance clarity and accuracy. Specifically, we have corrected an oversight related to the robot’s velocity and secondary antibody ratios, and addressed previously missing values in Figs. 3E and 4E. Importantly, these corrections did not alter the outcomes of our results. Additionally, we have enriched our manuscript with new data analyses, as reflected in Figures 1B, 1F, 2H-J, 4D, 4F-H, S1A, S1C-E, S3H, S5, and Table 1, ensuring a more comprehensive presentation of our findings. Below are our responses detailing each comment and explaining the modifications integrated into the revised manuscript.

Reviewer 1:

(1) To address the question of whether PAG photostimulation biases the cells that respond to the robot, a counterbalanced experiment, in which the BLA activity is initially recorded during the foraging vs. robot test and the PAG stimulation happens at the end of the session, should have been performed.

In our study, we investigated fear behavior and BLA cell responses to intrinsic dPAG photostimulation (320 pulses) in naïve animals, followed by their reactions to an extrinsic predatory robot. We recognize the reviewer's concern regarding the potential  influence of initial dPAG photostimulation on BLA neuron responses to the robot. We address this issue in our discussion (pg. 13) as follows: “However, it is crucial to consider the recent discovery that optogenetic stimulation of CA3 neurons (3000 pulses) leads to gain-of-function changes in CA3-CA3 recurrent (monosynaptic) excitatory synapses (Oishi et al., 2019). Although there is no direct connection between dPAG neurons and the BLA (Vianna and Brandao 2003, McNally, Johansen, and Blair 2011, Cameron et al. 1995), and no studies have yet demonstrated gain-of-function changes in polysynaptic pathways to our knowledge, the potential for our dPAG photostimulation (320 pulses) to induce similar changes in amygdalar neurons, thereby enhancing their sensitivity to predatory threats, cannot be dismissed.”

(2) In Figure 3, it is unclear which criteria (e.g. response latency, minimum Z score, spike fidelity) was used to identify the BLA neurons that were indirectly activated by PAG stimulation. A graphic containing at least the distribution of the response latencies for each BLA neuron after PAG laser activation is needed.

We have specified the criteria for determining the responsiveness of BLA neurons to dPAG stimulation on page 22. This involves analyzing the first 500-ms post-stimulation across five 0.1-s bins. Units were classified as ‘stim cells’ if they showed z-scores greater than 3 (z > 3) in any of the bins during the initial 500-ms period post-stimulation. Neurons activated by both pellet procurement and dPAG stimulation were not included in the 'stim cell' category. Additionally, we have included a graphic in the revised manuscript (Fig. S3C) that presents the distribution of response latencies of BLA neurons to dPAG stimulation.

(3) To strengthen the claim that it is a BLA-PVT-PAG circuit that carries information about predatory threat, a new experiment using CTB and cFos could be used to demonstrate that PAG neurons that project to PVT are recruited during the robot exposure.

Our study primarily aimed to explore the transmission of threat signals between the dPAG and BLA. We acknowledge that our evidence for the PVT’s intermediary role, derived from CTB injections in the BLA and subsequent CTB+cFos co-labeling analysis in the PVT (Fig. 4G and 4H), is limited. Accordingly, we have moderated the emphasis on the PVT’s involvement in both the abstract and introduction. We now present the PVT’s role as a promising direction for future research in the discussion section of our revised manuscript.

(4) In Fig 2, the authors' interpretation is that photostimulation of PAG neurons elicits fleeing responses in the rats. However, there is a vast literature demonstrating that the PAG is also involved in nociception. Although this is recognized by the authors in the first part of the introduction and briefly described in the discussion, the authors should more explicitly explain that PAG stimulation produces analgesia and thus is unlikely to underlie the escaping responses observed. This may not be intuitive for a broader audience.

We appreciate the reviewer's insightful suggestion to elaborate on the PAG involvement in nociception and analgesia, as supported by the literature. While our initial manuscript acknowledged these functions, we have now expanded our discussion to address the PAG’s multifaceted roles (pg. 12): “As mentioned in the introduction, the dPAG is recognized as part of the ascending nociceptive pathway to the BLA (De Oca et al. 1998, Gross and Canteras 2012, Herry and Johansen 2014, Kim, Rison, and Fanselow 1993, Ressler and Maren 2019, Walker and Davis 1997). The dPAG is also implicated in non-opioid analgesia (e.g., Bagley and Ingram 2020, Cannon et al. 1982, Fields 2000). However, it is essential to emphasize that, despite its roles in pain modulation, the primary behavior observed in dPAG-stimulated, naive rats foraging for food in an open arena was goal-directed escape to the safe nest, underscoring the dPAG’s critical function in survival behaviors.” Note that this aligns with human studies on PAG stimulation (e.g., Carrive and Morgan 2012, Magierek et al. 2003), particularly those by Amano et al. (Amano et al. 1982), which reported patients feeling an urge to escape, similar to being chased, upon PAG stimulation.

(5) To truly demonstrate the functional links between the PAG and BLA, more experiments are needed. For example, one could record from BLA neurons during the robot surge while performing optogenetic inhibition of the PAG neurons. There is also no evidence that activity in the indirect pathway that connects the PAG to the BLA is indispensable for the expression of defensive responses towards the robot (e.g., causality tests using chemogenetic or optogenetic inactivation).

We agree that incorporating optogenetic inhibition of PAG neurons while simultaneously recording from BLA neurons during a robot surge would strengthen the evidence for the functional connectivity between the PAG and BLA. Such an experiment would necessitate the transfection and photoinhibition of a wide array of dPAG neurons responsive to predatory threats. This procedure is technically more viable in transgenic mouse models, given their suitability for genetic manipulation. In light of this, and in response to the suggestions in the Joint Public Review, we have revised the abstract, introduction, and discussion to offer a more cautious interpretation of our findings. This revision reflects a careful consideration of both the evidence and the limitations inherent in our study (pg. 13): “While our findings demonstrate that opto-stimulation of the dPAG is sufficient to trigger both fleeing behavior and increased BLA activity, we have not established that the dPAG is necessary for the BLA’s response to predatory threats. To establish causality, it is essential to conduct experiments such as optogenetic inhibition to determine whether the dPAG is indispensable for activating BLA neurons and initiating escape behavior in the face of threats. The complexity of targeting the dPAG, which includes its dorsomedial, dorsolateral, lateral, and ventrolateral subdivisions (e.g., Bandler, Carrive, and Zhang 1991, Bandler and Keay 1996, Carrive 1993), suggests the need for future studies using transgenic mouse models. Should inactivation of the dPAG negate the BLA's response to predatory threats, it would underscore the dPAG's central role in this defensive mechanism. Conversely, if BLA responses remain unaffected by dPAG inactivation, this could indicate the existence of multiple pathways for antipredatory defense mechanisms.”

(6) The manuscript lacks information about the number of rats and trials that were used across the experiments (e.g. Fig 2G-J). In some occasions, the authors start the experiments with a specific number of animals and then reduce the N by half without providing a rationale (e.g. Fig. 3). Equally confusing is the experimental timeline. For example: a) Were the pre-robot, robot, and post-robot sessions always performed within the same day? b) It was described that microdrivable arrays were used, but did the same rats experienced the robot test more than one time? c) How many bins were used for normalization during the Z-score calculation and when were the data binned at 100 ms versus 1 s? d) How many trials were used for each analysis? For example, to identify robot cells, did the authors establish a minimum number of trials per animal to calculate the peristimulus time histograms? Having a significant number of trials is critical to make sure that the observed neuronal responses are replicable across the trials. e) How was the neuronal activity related to "pellet retrieval" aligned during robot sessions? Was the activity aligned with the moment in which the rat touches the pellet or when the animal returns to the nest with the pellet? f) How did the authors control for trials in which the rat consumed the pellets in the same local vs. those in which they returned to the nest to eat it? All these points are extremely important for future replicability.

We apologize for any confusion caused by the initial lack of detail in our experimental procedures. The revised manuscript has been updated with comprehensive methodological details:  

(i) The study involved thirteen rats (ChR2, n = 9; EYFP, n = 4), subjected to dPAG stimulation using fixed light parameters (473 nm, 20 Hz, 10-ms pulse width, 2 s duration) during Long and Short pellet distance trials (refer to Fig. 2E-G). The stimulation intensity was adjusted to each animal's response (fleeing behavior), ranging from 1-3 mW. Additional testing occurred over multiple days, with incremental adjustments to stimulation parameters (intensity, frequency, duration) after confirming normal baseline foraging behavior (Fig. 2H-J, at x = 0). These details are now clearly depicted in the manuscript.

(ii) The primary objective was to investigate BLA neuron responses to dPAG opto-stimulation. Six rats were initially tested, with three later assessed for their reactions to dPAG stimulation in the presence of an actual predator, to gauge behavioral effects.

(iii) Regarding the experimental timeline:

a) Pre-robot, robot, and post-robot sessions were conducted successively on the same day.

b) Sessions with the robot predator were repeated until habituation occurred or when unit recordings were deemed invalid due to microdrive limitations or the absence of unit detection. Throughout these sessions, the success rate for pellet retrieval remained consistently low. Specifically, the mean success rate for the dPAG recordings was 2.803% + 1.311. For the BLA recordings, animals did not succeed in retrieving pellets during any of the robot trials. To provide a more detailed account of the methodology, the manuscript has been updated to include the number of recording days and the units recorded in the "Behavioral Procedures" section.

c) As described in Materials and Methods, unit recording data were binned at 0.1-s intervals and normalized against a 5-s pre-event baseline (50 bins). For statistical analyses in Figure 1F’s rightmost column, 1-s bins were used to simplify post-hoc analysis corrections.

d) Each recording session consisted of 5-15 trials. Trials were excluded if rats attempted to procure the pellet within 10 s post-dPAG stimulation or robot activation, ensuring accurate characterization of unit responsiveness. Consequently, the number of trials varied among subjects.

e) Pellet retrieval was indicated by the animal entering a designated zone 19 cm from the pellet, driven by hunger.

f) Animals were trained to retrieve pellets and return to their nest for consumption prior to robot testing sessions, as elaborated in the “Baseline foraging” section.

(7) In the abstract, the authors mention that predictive cues are ambiguous during naturalistic predatory threats, but it is not clear what do they mean by ambiguous. In addition, in the introduction section, the authors describe that the present study will investigate how the dPAG and BLA communicate threat signals. However, the author should clarify right in the beginning that these two regions are not monosynaptically connected with each other and cite the proper references.

The abstract’s original sentence, “…where predictive cues are ambiguous and do not afford reiterative trial-and-error learning…” has been refined to “…characterized by less explicit cues and the absence of reiterative trial-and-error learning events …” This adjustment more accurately reflects that cues in natural settings often lack the clear and consistent quality of those in controlled experimental settings, which is necessary for the straightforward process of trial-and-error learning.

Regarding the dPAG and BLA connectivity, the revised introduction (pg. 5) now states: “Considering the lack of direct monosynaptic projections between dPAG and BLA neurons (Vianna and Brandao 2003, McNally, Johansen, and Blair 2011, Cameron et al. 1995), we utilized anterograde and retrograde tracers in the dPAG and BLA, respectively. This was complemented by c-Fos expression analysis following exposure to predatory threats. Our anatomical findings suggest that the paraventricular nucleus of the thalamus (PVT) may be part of a network that conveys predatory threat information from the dPAG to the BLA.”

(8) In the introduction section, the authors should clarify that the US information is conveyed from the PAG to BLA via the lateral thalamus (posterior intralaminar nucleus, medial geniculate nucleus) or dorsal midline thalamus (paraventricular nucleus of the thalamus). The statement regarding how "the PAG functions as part of the ascending pain transmission pathway, providing footshock US information to the BLA" is misleading because the PAG does not send monosynaptic projections directly to the BLA.

The revised text (pg. 3) now reads: “…suggest that the dPAG is part of the ascending US pain transmission pathway to the BLA, the presumed site for CS-US association formation (De Oca et al. 1998, Gross and Canteras 2012, Herry and Johansen 2014, Kim, Rison, and Fanselow 1993, Ressler and Maren 2019, Walker and Davis 1997). This pathway is thought to be mediated through the lateral and dorsal-midline thalamus regions, including the posterior intralaminar nucleus and paraventricular nucleus of the thalamus (Krout and Loewy, 2000; McNally, Johansen, and Blair, 2011; Yeh, Ozawa, and Johansen, 2021; but see Brunzell and Kim, 2001).”

(9) The author's assumption that threat information flows from the PAG to the BLA, rather than BLA to PAG, based on electrical stimulation and lesion experiments performed in previous studies is problematic for at least three reasons: a) Electrical stimulation can activate fibers of passage as well as presynaptic neurons antidromically. b) The lesion approach may not have targeted 100% of the neurons in PAG, which extends anatomically along the antero-posterior axis of the midbrain for several millimeters in rats. This observation also disagrees with more recent studies using optogenetics and imaging tools demonstrating that the PAG is the downstream target of the BLA-CeA pathway. c) The authors cited prior reports describing the role of the amygdala-PAG pathway in dampening the US response and providing a negative signal to the PAG. However, a series of previous studies demonstrating that the PAG serves as the downstream target of the central nucleus of the amygdala for the expression of defensive response are completely ignored by the authors. Here are just some examples: Massi et al, 2023, PMID: 36652513; Tovote et al 2016, PMID: 27279213; Penzo et al, 2014 PMID: 24523533).

We recognize the complexities in interpreting findings from electrical stimulation and lesion studies. Our prior work (Kim et al. 2013) supports the conclusion that predatory threat information directionally flows from the dPAG to the BLA, as evidenced by distinct behavioral outcomes from experimental manipulations of dPAG and BLA. Specifically, dPAG stimulation-induced fleeing behavior was blocked by BLA lesions (as well as muscimol inactivation), whereas BLA stimulation-induced fleeing was unaffected by dPAG or combined dPAG+vPAG lesions (refer to Fig. 5A), suggesting a flow from dPAG to BLA. Our manuscript further clarifies that dPAG optostimulation results confirmed that escape behavior in foraging rats, induce by dPAG electrical stimulation (Kim et al. 2013), was activated by intrinsic dPAG neurons rather than by fibers of passage or current spread to other brain regions.  

Furthermore, the PAG’s anatomical and functional diversity, with distinct segments along its longitudinal axis associated with different defensive behaviors, reinforces our conclusions. The dPAG is implicated in flight responses, while the vPAG is associated with freezing behavior (e.g., Bandler and Shipley 1994, Kim, Rison, and Fanselow 1993, Lefler, Campagner, and Branco 2020, Morgan, Whitney, and Gold 1998). The critiques' referenced studies primarily focus on the BLA-CeA-vPAG circuit's role in freezing during Pavlovian fear conditioning, contrasting with our emphasis on the dPAG-PVT-BLA circuit and its mediation in escape behavior in response to naturalistic predatory threats.

We also note that different invasive procedures can yield varying behavioral outcomes. For example, both acute (e.g., optogenetic and muscimol inactivation) and chronic (e.g., surgical ablation) manipulations within the same brain circuit have shown diverse effects across species (Otchy et al. 2015). Moreover, optogenetics comes with its own set of conceptual and technical challenges (Adamantidis et al. 2015), including the difficulty of targeting, quantifying and photo-inhibiting 100% of PAG neurons. Despite the limitations of each technique, our collective evidence from lesions, inactivation, electrical stimulation (Kim et al. 2013), optostimulation, and single-unit recordings (the present study) supports the premise that the dPAG acts upstream of the BLA in processing predatory threat information.

(10) In the discussion, the authors suggest that the PVT may be the interface between the PAG and the BLA for the expression of antipredatory defensive behavior during their foraging vs. robot test, but previous studies looking at the role of PVT in antipredator defensive behavior and/or approach-avoidance conflict tasks are not cited and discussed in the manuscript (Engelke et al, 2021, PMID: 33947849; Choi et al 2019, PMID: 30979815; Choi and McNally 2017, PMID: 28193686).

We thank the reviewer for pointing out these pivotal studies, which we have carefully reviewed and integrated into the revised manuscript (pg. 14): “These results, in conjunction with previous research on the roles of the dPAG, PVT, and BLA in producing flight behaviors in naïve rats (Choi and Kim 2010, Daviu et al. 2020, Deng, Xiao, and Wang 2016, Kim et al. 2013, Kim et al. 2018, Kong et al. 2021, Ma et al. 2021, Reis et al. 2021), the anterior PVT’s involvement in cat odor-induced avoidance behavior (Engelke et al. 2021), and the PVT’s regulation of behaviors motivated by both appetitive and aversive stimuli (Choi and McNally 2017, Choi et al. 2019), suggest the involvement of the dPAGàPVTàBLA pathways in antipredatory defensive mechanisms, particularly as rats leave the safety of the nest to forage in an open arena (Figure 4I) (Reis et al. 2023).”  

(11) The authors use the expression "looming robot predator" in many cases throughout the manuscript. However, it is unclear whether the defensive responses observed in the rats are elicited by the looming stimulus produced by the movement of the robot towards the rats. The authors describe that rats do not respond to a stationary robot, but would the sound produced by the movement of the robot elicit defensive responses? Would non-approaching lateral or dorsoventral movements (not associated with looming) be sufficient to induce defensive behavior in the rats? There is a vast literature in the field about defensive behaviors induced by looming stimuli. The authors should empirically demonstrate that the escaping responses induced by the robot are mediated by looming or refrain to use the looming terminology to avoid confusion.

Our use of "looming robot predator" is based on empirical evidence from a prior parametric study, which identified the forward, or 'looming,' motion of the Robogator as the key stimulus eliciting a flight response in rats (Kim, Choi, and Lee 2016). This reaction significantly decreased when the robot moved backward from the same starting position, producing a similar sound, and was absent when the robot remained stationary. This suggests that neither sound alone nor the mere presence of a novel object provokes goal-directed escape behavior (Kong et al. 2021). This aligns with studies indicating that simulated looming stimuli, like an expanding disk, induce flight or freezing responses in mice (De Franceschi et al. 2016, Yilmaz and Meister 2013).

It should be noted that the 2013 study by Yilmaz & Meister (Yilmaz and Meister 2013) on the looming disk paradigm showed that not all mice responded to the stimuli (e.g., Figs. 2A and 3A), with those that did exhibiting rapid habituation by the second exposure. This contrasts with our predatory robot paradigm (Choi and Kim 2010), where all rats consistently fled from the looming robotic predator across multiple trials, underscoring the critical role of looming motion in simulating predator attacks that trigger flight behavior in rats.

Thus, the term "looming" accurately captures the nature of the robot's movement and its effect on eliciting defensive responses in rats. Nonetheless, should the editors agree with the reviewer's suggestion to minimize potential confusion, we are willing to substitute "looming" with "approaching," although we consider the terms to be synonymous in the context of our study.

(12) If the authors are citing the Rescorla-Wagner model, they should include at least one additional sentence to explain it, as many people in the field are not familiar with this model.

In response to the request for clarification on the Rescorla-Wagner model, we have added an explanatory sentence (pg. 4): “Fundamentally, the negative feedback circuit between the amygdala and the dPAG serves as a biological implementation of the Rescorla–Wagner (1972) model, a foundational theory of associative learning that emphasizes the importance of prediction errors in reinforcement (i.e., US), as applied to FC (Fanselow 1998).”

(13) The authors need to include the normality test used to determine whether a parametric or non-parametric statistical analysis was the most appropriate test for each experiment.

We have included the outcomes of the normality tests, detailed in Table S1.

(14) In Fig. 1F, the authors show a representative PAG neuron with peristimulus-time histogram and rasters reaching frequencies higher than 100 Hz and sustained firing rates of >50 Hz following robot activation. The authors should include a firing rate analysis (e.g., average firing rate and maximum firing rate before and after robot activation) of the 22 robot-responsive PAG neurons recorded during the session to clarify whether this high firing rate, which is atypical in other brain regions, is commonly observed in the PAG. Showing the isolated waveforms of some representative neurons would help to clarify whether the activity is being recorded from a single-isolated unit instead of multiple units within the same channel.

In response to the critique, we have expanded our analysis to include both average and maximum firing rates before and after robot activation for the 22 robot-responsive PAG neurons. This detailed firing rate analysis, illustrating their distribution, has been incorporated into the revised manuscript (refer to Figure S1C and S1D). Furthermore, to alleviate concerns regarding the identification of single-unit activity versus potential multi-unit recordings, we have included peri-event raster plots and waveforms for two additional representative neurons in Figure 1F.

(15) In Figure 2, the authors should indicate when the recordings are performed on anesthetized vs. freely-moving awake animals.

In the original manuscript, we specified that the optrode recordings depicted in Figure 2B were conducted on anesthetized rats. To enhance clarity and directly address the critique, we have now clearly indicated this condition in Figure 2A as well.

(16) The optogenetic stimulation parameters used in Fig 2H indicate that 0.5 mW was sufficient to induce behavioral changes. This is surprising because most optogenetic experiments in the field use much higher intensities (> 5mW). If much lower intensities are sufficient to drive PAG-mediated behaviors, this may be a very important observation that should be conveyed to the field. I recommend the reviewers clarify if they in fact used 0.5 mW and then discuss that the laser intensity used in the experiments was 10X lower than that required for other brain regions

In our study, we indeed observed that 0.5 mW of dPAG stimulation increased the latency to procure the pellet without completely preventing the action. Notably, at 1 mW, more than half of the animals (n = 5/9 rats; Fig. 2H) and at 3 mW, all rats (9/9) failed to procure the pellet and fled from the foraging area to the nest (Fig. 2G). These results indicate that even lower intensities were sufficient to elicit behavioral changes through dPAG stimulation in a large foraging arena, highlighting the dPAG's sensitivity to optogenetic manipulation. This finding is consistent with our earlier research on dPAG electrical stimulation, which required significantly lower intensities to provoke defensive behaviors compared to the BLA. Specifically, the stimulation intensity needed for aversive behavior in the dPAG was substantially lower (dPAG: 65.0 ± 6.85 µA) than for the BLA (BLA: 275.0 ± 24.44 µA) (Kim et al. 2013). Furthermore, Deng et al. (Deng, Xiao, and Wang 2016) showed that 1 mW of blue light could elicit a 60% freezing response, with 2 mW triggering flight behavior within a latency of 0.6 seconds.

(17) In Fig 2 G-J, how many animals are being used per group and how was the sequence of the experiments performed? This is very important for replicability.

A total of three rats were utilized for the robot testing experiments depicted in Fig. 2 G-J. The experimental sequence for these animals consisted of successive pre-stimulation, stimulation, post-stimulation, and robot sessions. We have updated the manuscript to include this information.

(18) For the photostimulation of PAG neurons in Figs. 2 and 3, the authors need to clarify if the same parameters of laser stimulation used during the anesthetized recordings were also used during the behavioral tests. Also, the wavelength corresponding to the blue laser should be 473 nm instead of 437 nm.

We thank the reviewer for identifying the error. We confirm that the opto-stimulation parameters (473 nm, 10-ms pulse width, 2 s duration) were consistently applied across both anesthetized recordings and behavioral tests. This consistency has been explicitly stated in the revised manuscript to ensure clarity regarding our experimental approach.

(19) In Fig. 3I, how was the representative trials selected? Instead of picking up the most representative trials, the authors should demonstrate the response of the cell during the entire session.

In response to the critique, we clarify that the color-coded PETH shown in Fig. 3I represents averaged BLA activity across a comprehensive set of trials. This includes 8 pre-stimulation, 10 stimulation, and 8 post-stimulation trials for the robot-activated sessions, with a similar distribution for non-stimulated sessions. This approach was chosen to provide a representative overview of the cell's response throughout the entire session. To address the request for more detailed data, we have added traditional PETHs to the revised manuscript (see Fig. S3H), which depict the cell's response across all trials.

(20) Fig 4 D should demonstrate a colabeling between the anterograde PAG fibers in the PVT and the retrogradely labeled neurons from BLA instead of PAG fibers only.

We wish to clarify that Fig. 4D is intended to show the distribution of dPAG terminals within the midline thalamic nuclei, as noted in prior research (Krout and Loewy 2000). Although dPAG terminals are distributed throughout the midline thalamus, our observations have specifically highlighted a notable increase in c-Fos expression within the paraventricular nucleus of the thalamus (PVT) in rats subjected to the robotic predator stimulus, in contrast to those in the foraging-only control condition (Fig. 4E). Addressing the reviewer's point, we direct attention to Fig. 4G, which includes images labeled "Robot-experienced" and "Merge." This figure demonstrates a subset of PVT neurons that were retrogradely labeled with CTB injected into the BLA, anterogradely labeled with AAV injected into the dPAG, and activated (as indicated by c-Fos expression) in response to the robotic predator. This provides specific colabeling evidence between anterograde PAG fibers in the PVT and retrogradely labeled neurons from the BLA, directly addressing the critique.

(21) The resolution of the cFos images is very low and makes it hard to appreciate.

We have updated Figs. 4F and 4G with high-resolution versions to ensure the details are more clearly visible. Furthermore, should there be a need for even greater clarity, we are prepared to supply the images as TIFF files, which are known for preserving high image quality.

Reviewer 2:

(1) The text is clearly written, and I appreciated the inclusion of interesting citations, such as the one about paintings by cavemen. The authors also do a good job of discussing the underlying theoretical framework and the figures are easy to understand. Although the topic is very interesting, the amount of novel work is somewhat low. Figure 1 shows that dPAG cells are activated by the predator, and this has been shown by many prior reports. Similarly, Figure 2 shows that dPAG activation creates defensive responses, and this too has been shown by many prior reports.

We appreciate the reviewer’s positive remarks. We acknowledge the rich body of research documenting dPAG neuronal activation by various predator cues such as odors (e.g., fox urine) (Lu et al. 2023), and scenarios involving anesthetized or spontaneously moving rat/cat predators, either physically partitioned or harness-restrained (Bindi et al. 2022, Deng, Xiao, and Wang 2016, Esteban Masferrer et al. 2020). Nevertheless, our study distinguishes itself by examining dPAG neuronal responses to a robotic predator, uniquely designed to replicate consistent looming motions across multiple trials and subjects within an environment that simulates natural foraging conditions, inclusive of a safe nest (cf. Choi and Kim, 2010). This approach allowed us to not only reveal the immediate activation of dPAG neurons in response to a rapidly approaching predator but also to explore the consequent fleeing behavior towards safety, thereby providing new insights into the dPAG's role in mediating goal-directed defensive responses in a more ecologically-relevant setting. Furthermore, our investigation extends beyond these findings to assess the impact of dPAG activation on BLA neuronal responses and their functional connectivity during predator-prey interactions, offering a fresh perspective on the neural circuits that support survival behaviors in animals when confronted with naturalistic threats.

(2) The results in Figure 3 are novel and interesting, but the characterization of BLA activity is incomplete. For example, what are the percentages of BLA cells that are inhibited or activated by all major behaviors observed? These behaviors include approach to pellet, escape from robot, freezing, stretch-attend postures, etc. These same analyses should also be added to dPAG activity in Figure 1. How does BLA single cell encoding of these behaviors relate to their responsivity to dPAG stimulation? And, finally, it is unclear what is the significance of BLA correlated synchronous firing. Is the animal more or less likely to be performing certain behaviors when correlated BLA firing occurs?

Our analysis, as presented in Figs. 3I, 3K, and S3D-F, selectively focused on BLA cell responses during distinct behaviors such as approaching a pellet and escaping from the robot. These behaviors were selected because their precise temporal markers allow for accurate correlation with BLA cell activity, building on the findings of our previous research (Kim et al. 2018, Kong et al. 2021).

The robot's motion, programmed to advance a fixed distance before retreating to its starting position, is designed to repeatedly elicit foraging, thus facilitating analysis of neural changes during conflict situations involving food approach and predator avoidance. However, this also leads to the rapid diminution of freezing and stretch-attend postures inside the nest as animals quickly adapt to the robot's movement pattern, rendering a time-stamped analysis of these behaviors unfeasible under our experimental conditions. While the inclusion of these behaviors in our analysis would be insightful, especially in extended interaction scenarios where the robot advances to the nest opening and remains before returning in a less predictable manner, such conditions would likely reduce foraging behavior due to increased fear, deviating from our study's primary objective of elucidating the interactions between the dorsal periaqueductal gray (dPAG) and the basolateral amygdala (BLA) functions.

Regarding the significance of BLA correlated synchronous firing, our findings, particularly in Figures 3M-O and S4, demonstrate significant synchronous activity among BLA neuronal pairs during encounters with the robot, as opposed to pre-stim, stim, and post-stim sessions. This synchrony is notably prominent among neurons responsive to dPAG stimulation, indicating that BLA neurons involved in processing dPAG signals may play a crucial role in enhancing BLA network coherence to effectively manage predatory threat information (pg. 13).

(3) In Figure 4, the authors identify the PVT as a potential region that can mediate dPAG to BLA communication via anatomical tracing. However, functional assays are missing. For example, if the PVT is inhibited chemogenetically, does this result in a smaller number of BLA cells that are activated by dPAG stimulation? Does activation of the dPAG-PVT or the PVT-BLA projections cause defensive behaviors? Functionally showing that the dPAG-PVT-BLA circuit controls defensive actions would be a major advance in the field and would greatly enhance the significance of this paper. It would also provide an anatomical substrate to support the view that the BLA is downstream of the dPAG, which was first demonstrated by the authors in their elegant 2013 PNAS paper.

We appreciate the reviewer’s constructive critique and valuable suggestions on the necessity for functional validation of the dPAG-PVT-BLA circuit's involvement in mediating defensive behaviors. In light of these comments, we have carefully considered and included a discussion on the importance of these proposed experiments as a direction for future research in our manuscript revision (also see response to Reviewer 1’s critique #5).

Our initial work in 2013 (Kim et al. 2013) laid the groundwork for identifying BLA neurons responsive to dPAG stimulation and suggested the PVT as a potential relay in this neural circuit. Recognizing the limitations of our current study, which does not include direct functional assays, we have adjusted our manuscript to convey the speculative aspect of the dPAG-PVT-BLA circuit’s role more accurately. Moreover, we have enriched our discussion by citing relevant studies that lend support to our proposed circuit mechanism. These references serve to place our findings within the broader context of existing research and highlight the imperative for subsequent studies to empirically confirm the functional significance of the dPAG-PVT-BLA pathway in driving defensive behaviors.

Reviewer 3:

(1) The Introduction refers to a negative feedback amygdala-dPAG from a study of the Johansen group, but in this case, the authors were referring to the ventrolateral and not the dorsal PAG.

We thank the reviewer for pointing out the need to distinguish between the dPAG and vPAG regions in our introduction. While Johansen et al. (2010) investigated the roles of PAG (including both dPAG and vPAG regions; see their Supplementary Figs. 4, 5, and 10), the differentiation between their specific contributions to the amygdala's negative feedback mechanism was not explicitly detailed in their initial publication. This distinction was further elaborated upon in later work by the same group (Yeh, Ozawa, and Johansen 2021), which specifically illuminated the dPAG's role in conditioned fear memory formation and its neural pathways to the PVT that influence fear learning. To reflect this nuanced understanding, we have revised our introduction (pg. 3): “In parallel, Johansen et al. (2010) found that pharmacological inhibition of the PAG, encompassing both dPAG and vPAG regions, diminishes the behavioral and neural responses in the amygdala elicited by periorbital shock US, thereby impairing the acquisition of auditory FC.”

(2) In the experiments recording dPAG in response to the predator threat, the authors mentioned cells activated by the predator threat, referred to as "robot cells." Were these cells inhibited in response to threat?

In the Result and Materials and Methods sections, we report that 23.4% (22 out of 94) of dPAG neurons, termed “robot cells,” showed a significant increase in firing rates (z > 3) within a latency of less than 500 ms during exposure to the looming robot threat, but not during the pre- and post-robot sessions. These cells are highlighted in Figures 1E-G. In contrast, we identified only a single unit exhibiting a decrease in activity (z-score < -3) in response to the robot threat. Given the overwhelming prevalence of cells with excitatory responses to the threat, our discussions and analyses have primarily centered on these excited cells. Nevertheless, to ensure a full depiction of our observations, we have included data on the inhibited unit in the revised manuscript, specifically in Figure S1E.

(3) The authors claim that tetrodes were implanted in the dorsal PAG; however, the electrodes' tips shown in the figures are positioned more ventrally in the lateral PAG (see Figures 1B, S5A).

The PAG is anatomically organized into dorsomedial (dmPAG), dorsolateral (dlPAG), lateral (lPAG), and ventrolateral (vlPAG) columns along the rostro-caudal axis of the aqueduct. The designation "dorsal PAG" (dPAG) traditionally encompasses the dmPAG, dlPAG, and lPAG regions, a classification supported by extensive track-tracing, neurochemical, and immunohistochemical evidence (e.g., (Bandler, Carrive, and Zhang 1991, Bandler and Keay 1996, Carrive 1993)). As Bandler and Shipley (Bandler and Shipley 1994) summarized, “These findings suggest that what has been traditionally called the 'dorsal PAG' (a collective term for regions dorsal and lateral to the aqueduct), consists of three anatomically distinct longitudinal columns: dorsomedial and lateral columns…and a dorsolateral column…" Similarly, Schenberg et al. (Schenberg et al. 2005) clarified in their review that, “According to this parcellation...the defensive behaviors (freezing, flight or fight) and aversion-related responses (switch-off behavior) were ascribed to the DMPAG, DLPAG, and LPAG (usually named the ‘dorsal’ PAG).” In our study, electrode placements were strictly within these specified dPAG regions. The electrode tip locations depicted in Figures 1B and S5A correspond with the -6.04 mm template (left panel below) from Paxinos & Watson’s atlas (Paxinos and Watson 1998), situated anteriorly to the emergence of the  vlPAG (right panel below). To enhance clarification in our manuscript, we provide a detailed definition of the dPAG that includes the dmPAG, dlPAG,  and lPAG, and support our electrode placement rationale with references to established literature (pg. 5).

Author response image 1.

(4) It would be nice to include a series of observations applying inhibitory tools (i.e., optogenetic photo inhibition) in the dPAG and BLA and see how they affect the behavioral responses in the 'approach food-avoid predator' paradigm. Moreover, it would be interesting to explore how inhibiting the dPAG to PVT pathway influences the flee response during the robot surge.

We appreciate the suggestion to explore the effects of optogenetic inhibition in the dPAG and BLA on behavioral responses within the 'approach food-avoid predator' paradigm, as well as the potential impact of inhibiting the dPAG to PVT pathway on flee responses during robot surge incidents. As mentioned in our response to Reviewer 1’s critique #5, the application of optogenetic inhibition necessitates transfecting, quantifying, and photoinhibiting a comprehensive set of dPAG neurons activated by predatory threats. This approach is more viable in future studies that can leverage transgenic mouse models for their genetic tractability. Following the Joint Public Review’s recommendations, we have revised our manuscript to ensure a more measured interpretation of our data, carefully balancing the evidence from tracer studies against the limitations of our current methodology.

Furthermore, referencing Reviewer 1’s critique #9, it is important to consider that various invasive techniques can yield different behavioral outcomes. For instance, research by Olveczky and colleagues (Otchy et al. 2015) demonstrated that acute manipulations (i.e., optogenetic and muscimol inactivation) and chronic surgical ablation of the same brain circuit can produce distinct effects in rats and finches. Despite these methodological constraints, our collective results from lesion, inactivation, electrical stimulation (Kim et al. 2013), optostimulation, and single-unit recording (present) studies cohesively suggest that the dPAG functions upstream of the BLA in processing predatory threat signals.

(5) The authors should also examine whether 'synaptic' appositions exist between the anterogradely labeled terminals from the dPAG and the double labeled CTB and cFOS neurons in the PVT.

We appreciate the suggestion to investigate the presence of synaptic appositions, which could potentially offer valuable insights into the synaptic connections and functional interactions within this neural circuit. However, due to the specialized nature of electron microscopy required for these examinations and the extensive resources it entails, this line of inquiry falls beyond the scope of our current study. We hope to address this aspect in future studies, where we can dedicate the necessary resources and expertise to conducting these intricate analyses.

(6) It is odd to see the projection fields shown in Fig. 4D, where the projection to the PVT looks much sparser compared to other targets in the thalamus and hypothalamus. If the projection to the PVT has such an important function, why does it seem so weak? This should be discussed. Also, because the projection to the PVT seems sparse, the authors should consider alternative paths like the one involving the cuneiform nucleus. The cuneiform nucleus is an important region responding to looming shadows with strong bidirectional links to the dorsolateral periaqueductal gray, providing strong projections to the rostral PVT.

The perceived scarcity of the dPAG-PVT pathway might not reflect its functional significance accurately. The PVT's small size could make its projections appear less dense in broad anatomical studies. To address this, we have updated Figure 4D with a high-resolution image that offers a detailed view of the PVT region. This enhancement (refer to the updated Fig. 4, bottom) more accurately depicts the projection density within the PVT. It is also critical to consider that the functional impact of neural pathways is not solely dependent on the quantity of projecting neurons. For instance, work by Deisseroth and colleagues (Rajasethupathy et al. 2015) has shown that even relatively sparse monosynaptic projections from the anterior cingulate cortex to the hippocampus can exert significant effects on neural circuit dynamics. Additionally, we have expanded our discussion to consider the potential roles of other circuits, such as the cuneiform nucleus, in driving the behavioral responses observed in our study (pg. 15): “Given the recent significance attributed to the superior colliculus in detecting innate visual threats (Lischinsky and Lin 2019, Wei et al. 2015, Zhou et al. 2019) and the cuneiform nucleus in the directed flight behavior of mice (Bindi et al. 2023, Tsang et al. 2023), further exploration into the communication between these structures and the dPAG-BLA circuitry is warranted.”

(7) Finally, in the Discussion, it would be nice to comment on how the BLA mediates flee responses. Which pathways are likely involved?

This excellent suggestion has been incorporated in the discussion (pg. 15): “Future studies will also need to delineate the downstream pathways emanating from the BLA that orchestrate goal-directed flight responses to external predatory threats as well as internal stimulations from the dPAG/BLA circuit. Potential key structures include the dorsal/posterior striatum, which has been associated with avoidance behaviors in response to airpuff in head-fixed mice (Menegas et al. 2018) and flight reactions triggered by auditory looming cues (Li et al. 2021). Additionally, the ventromedial hypothalamus (VMH) has been implicated in flight behaviors in mice, evidenced by responses to the presence of a rat predator (Silva et al. 2013) and upon optogenetic activation of VMH Steroidogenic factor 1 (Kunwar et al. 2015) or the VMH-anterior hypothalamic nucleus pathway (Wang, Chen, and Lin 2015). Investigating the indispensable role of these structures in flight behavior could involve lesion or inactivation studies. Such interventions are anticipated to inhibit flight behaviors elicited by amygdala stimulation and predatory threats, confirming their critical involvement. Conversely, activating these structures in subjects with an inactivated or lesioned amygdala, which would typically inhibit fear responses to external threats (Choi and Kim 2010), is expected to induce fleeing behavior, further elucidating their functional significance.”

Adamantidis, A., S. Arber, J. S. Bains, E. Bamberg, A. Bonci, G. Buzsaki, J. A. Cardin, R. M. Costa, Y. Dan, Y. Goda, A. M. Graybiel, M. Hausser, P. Hegemann, J. R. Huguenard, T. R. Insel, P. H. Janak, D. Johnston, S. A. Josselyn, C. Koch, A. C. Kreitzer, C. Luscher, R. C. Malenka, G. Miesenbock, G. Nagel, B. Roska, M. J. Schnitzer, K. V. Shenoy, I. Soltesz, S. M. Sternson, R. W. Tsien, R. Y. Tsien, G. G. Turrigiano, K. M. Tye, and R. I. Wilson. 2015. "Optogenetics: 10 years after ChR2 in neurons--views from the community."  Nat Neurosci 18 (9):1202-12. doi: 10.1038/nn.4106.

Amano, K., T. Tanikawa, H. Kawamura, H. Iseki, M. Notani, H. Kawabatake, T. Shiwaku, T. Suda, H. Demura, and K. Kitamura. 1982. "Endorphins and pain relief. Further observations on electrical stimulation of the lateral part of the periaqueductal gray matter during rostral mesencephalic reticulotomy for pain relief."  Appl Neurophysiol 45 (1-2):123-35.

Bagley, E. E., and S. L. Ingram. 2020. "Endogenous opioid peptides in the descending pain modulatory circuit."  Neuropharmacology 173:108131. doi: 10.1016/j.neuropharm.2020.108131.

Bandler, R., P. Carrive, and S. P. Zhang. 1991. "Integration of somatic and autonomic reactions within the midbrain periaqueductal grey: viscerotopic, somatotopic and functional organization."  Prog Brain Res 87:269-305. doi: 10.1016/s0079-6123(08)63056-3.

Bandler, R., and K. A. Keay. 1996. "Columnar organization in the midbrain periaqueductal gray and the integration of emotional expression."  Prog Brain Res 107:285-300. doi: 10.1016/s0079-6123(08)61871-3.

Bandler, R., and M. T. Shipley. 1994. "Columnar organization in the midbrain periaqueductal gray: modules for emotional expression?"  Trends Neurosci 17 (9):379-89. doi: 10.1016/0166-2236(94)90047-7.

Bindi, R. P., C. C. Guimaraes, A. R. de Oliveira, F. F. Melleu, M. A. X. de Lima, M. V. C. Baldo, S. C. Motta, and N. S. Canteras. 2023. "Anatomical and functional study of the cuneiform nucleus: A critical site to organize innate defensive behaviors."  Ann N Y Acad Sci 1521 (1):79-95. doi: 10.1111/nyas.14954.

Bindi, R. P., R. G. O. Maia, F. Pibiri, M. V. C. Baldo, S. L. Poulter, C. Lever, and N. S. Canteras. 2022. "Neural correlates of distinct levels of predatory threat in dorsal periaqueductal grey neurons."  Eur J Neurosci 55 (6):1504-1518. doi: 10.1111/ejn.15633.

Cameron, A. A., I. A. Khan, K. N. Westlund, and W. D. Willis. 1995. "The efferent projections of the periaqueductal gray in the rat: a Phaseolus vulgaris-leucoagglutinin study. II. Descending projections."  J Comp Neurol 351 (4):585-601. doi: 10.1002/cne.903510408.

Cannon, J. T., G. J. Prieto, A. Lee, and J. C. Liebeskind. 1982. "Evidence for opioid and non-opioid forms of stimulation-produced analgesia in the rat."  Brain Res 243 (2):315-21. doi: 10.1016/0006-8993(82)90255-4.

Carrive, P, and M. M. Morgan. 2012. "Periaqueductal Gray." In The Human Nervous System, edited by J. K.; Paxinos Mai, G., 367-400. London: Academic Press.

Carrive, P. 1993. "The periaqueductal gray and defensive behavior: functional representation and neuronal organization."  Behav Brain Res 58 (1-2):27-47. doi: 10.1016/0166-4328(93)90088-8.

Choi, E. A., P. Jean-Richard-Dit-Bressel, C. W. G. Clifford, and G. P. McNally. 2019. "Paraventricular Thalamus Controls Behavior during Motivational Conflict."  J Neurosci 39 (25):4945-4958. doi: 10.1523/JNEUROSCI.2480-18.2019.

Choi, E. A., and G. P. McNally. 2017. "Paraventricular Thalamus Balances Danger and Reward."  J Neurosci 37 (11):3018-3029. doi: 10.1523/JNEUROSCI.3320-16.2017.

Choi, J. S., and J. J. Kim. 2010. "Amygdala regulates risk of predation in rats foraging in a dynamic fear environment."  Proc Natl Acad Sci U S A 107 (50):21773-7. doi: 10.1073/pnas.1010079108.

De Franceschi, G., T. Vivattanasarn, A. B. Saleem, and S. G. Solomon. 2016. "Vision Guides Selection of Freeze or Flight Defense Strategies in Mice."  Curr Biol 26 (16):2150-4. doi: 10.1016/j.cub.2016.06.006.

De Oca, B. M., J. P. DeCola, S. Maren, and M. S. Fanselow. 1998. "Distinct regions of the periaqueductal gray are involved in the acquisition and expression of defensive responses."  J Neurosci 18 (9):3426-32. doi: 10.1523/JNEUROSCI.18-09-03426.1998.

Deng, H., X. Xiao, and Z. Wang. 2016. "Periaqueductal Gray Neuronal Activities Underlie Different Aspects of Defensive Behaviors."  J Neurosci 36 (29):7580-8. doi: 10.1523/JNEUROSCI.4425-15.2016.

Engelke, D. S., X. O. Zhang, J. J. O'Malley, J. A. Fernandez-Leon, S. Li, G. J. Kirouac, M. Beierlein, and F. H. Do-Monte. 2021. "A hypothalamic-thalamostriatal circuit that controls approach-avoidance conflict in rats."  Nat Commun 12 (1):2517. doi: 10.1038/s41467-021-22730-y.

Esteban Masferrer, M., B. A. Silva, K. Nomoto, S. Q. Lima, and C. T. Gross. 2020. "Differential Encoding of Predator Fear in the Ventromedial Hypothalamus and Periaqueductal Grey."  J Neurosci 40 (48):9283-9292. doi: 10.1523/JNEUROSCI.0761-18.2020.

Fanselow, M. S. 1998. "Pavlovian conditioning, negative feedback, and blocking: mechanisms that regulate association formation."  Neuron 20 (4):625-7. doi: 10.1016/s0896-6273(00)81002-8.

Fields, H. L. 2000. "Pain modulation: expectation, opioid analgesia and virtual pain."  Prog Brain Res 122:245-53. doi: 10.1016/s0079-6123(08)62143-3.

Gross, C. T., and N. S. Canteras. 2012. "The many paths to fear."  Nat Rev Neurosci 13 (9):651-8. doi: 10.1038/nrn3301.

Herry, C., and J. P. Johansen. 2014. "Encoding of fear learning and memory in distributed neuronal circuits."  Nat Neurosci 17 (12):1644-54. doi: 10.1038/nn.3869.

Kim, E. J., O. Horovitz, B. A. Pellman, L. M. Tan, Q. Li, G. Richter-Levin, and J. J. Kim. 2013. "Dorsal periaqueductal gray-amygdala pathway conveys both innate and learned fear responses in rats."  Proc Natl Acad Sci U S A 110 (36):14795-800. doi: 10.1073/pnas.1310845110.

Kim, E. J., M. S. Kong, S. G. Park, S. J. Y. Mizumori, J. Cho, and J. J. Kim. 2018. "Dynamic coding of predatory information between the prelimbic cortex and lateral amygdala in foraging rats."  Sci Adv 4 (4):eaar7328. doi: 10.1126/sciadv.aar7328.

Kim, J. J., J. S. Choi, and H. J. Lee. 2016. "Foraging in the face of fear: Novel strategies for evaluating amygdala functions in rats." In Living without an amygdala, edited by D. G. Amaral and R. Adolphs, 129-148. The Guilford Press.

Kim, J. J., R. A. Rison, and M. S. Fanselow. 1993. "Effects of amygdala, hippocampus, and periaqueductal gray lesions on short- and long-term contextual fear."  Behav Neurosci 107 (6):1093-8. doi: 10.1037//0735-7044.107.6.1093.

Kong, M. S., E. J. Kim, S. Park, L. S. Zweifel, Y. Huh, J. Cho, and J. J. Kim. 2021. "'Fearful-place' coding in the amygdala-hippocampal network."  Elife 10. doi: 10.7554/eLife.72040.

Krout, K. E., and A. D. Loewy. 2000. "Periaqueductal gray matter projections to midline and intralaminar thalamic nuclei of the rat."  J Comp Neurol 424 (1):111-41. doi: 10.1002/1096-9861(20000814)424:1<111::aid-cne9>3.0.co;2-3.

Kunwar, P. S., M. Zelikowsky, R. Remedios, H. Cai, M. Yilmaz, M. Meister, and D. J. Anderson. 2015. "Ventromedial hypothalamic neurons control a defensive emotion state."  Elife 4. doi: 10.7554/eLife.06633.

Lefler, Y., D. Campagner, and T. Branco. 2020. "The role of the periaqueductal gray in escape behavior."  Curr Opin Neurobiol 60:115-121. doi: 10.1016/j.conb.2019.11.014.

Li, Z., J. X. Wei, G. W. Zhang, J. J. Huang, B. Zingg, X. Wang, H. W. Tao, and L. I. Zhang. 2021. "Corticostriatal control of defense behavior in mice induced by auditory looming cues."  Nat Commun 12 (1):1040. doi: 10.1038/s41467-021-21248-7.

Lischinsky, J. E., and D. Lin. 2019. "Looming Danger: Unraveling the Circuitry for Predator Threats."  Trends Neurosci 42 (12):841-842. doi: 10.1016/j.tins.2019.10.004.

Lu, B., P. Fan, M. Li, Y. Wang, W. Liang, G. Yang, F. Mo, Z. Xu, J. Shan, Y. Song, J. Liu, Y. Wu, and X. Cai. 2023. "Detection of neuronal defensive discharge information transmission and characteristics in periaqueductal gray double-subregions using PtNP/PEDOT:PSS modified microelectrode arrays."  Microsyst Nanoeng 9:70. doi: 10.1038/s41378-023-00546-8.

Magierek, V., P. L. Ramos, N. G. da Silveira-Filho, R. L. Nogueira, and J. Landeira-Fernandez. 2003. "Context fear conditioning inhibits panic-like behavior elicited by electrical stimulation of dorsal periaqueductal gray."  Neuroreport 14 (12):1641-4. doi: 10.1097/00001756-200308260-00020.

McNally, G. P., J. P. Johansen, and H. T. Blair. 2011. "Placing prediction into the fear circuit."  Trends Neurosci 34 (6):283-92. doi: 10.1016/j.tins.2011.03.005.

Menegas, W., K. Akiti, R. Amo, N. Uchida, and M. Watabe-Uchida. 2018. "Dopamine neurons projecting to the posterior striatum reinforce avoidance of threatening stimuli."  Nat Neurosci 21 (10):1421-1430. doi: 10.1038/s41593-018-0222-1.

Morgan, M. M., P. K. Whitney, and M. S. Gold. 1998. "Immobility and flight associated with antinociception produced by activation of the ventral and lateral/dorsal regions of the rat periaqueductal gray."  Brain Res 804 (1):159-66. doi: 10.1016/s0006-8993(98)00669-6.

Otchy, T. M., S. B. Wolff, J. Y. Rhee, C. Pehlevan, R. Kawai, A. Kempf, S. M. Gobes, and B. P. Olveczky. 2015. "Acute off-target effects of neural circuit manipulations."  Nature 528 (7582):358-63. doi: 10.1038/nature16442.

Paxinos, G., and C. Watson. 1998. The Rat Brain in Stereotaxic Coordinates. San Diego: Academic Press.

Rajasethupathy, P., S. Sankaran, J. H. Marshel, C. K. Kim, E. Ferenczi, S. Y. Lee, A. Berndt, C. Ramakrishnan, A. Jaffe, M. Lo, C. Liston, and K. Deisseroth. 2015. "Projections from neocortex mediate top-down control of memory retrieval."  Nature 526 (7575):653-9. doi: 10.1038/nature15389.

Ressler, R. L., and S. Maren. 2019. "Synaptic encoding of fear memories in the amygdala."  Curr Opin Neurobiol 54:54-59. doi: 10.1016/j.conb.2018.08.012.

Schenberg, L. C., R. M. Povoa, A. L. Costa, A. V. Caldellas, S. Tufik, and A. S. Bittencourt. 2005. "Functional specializations within the tectum defense systems of the rat."  Neurosci Biobehav Rev 29 (8):1279-98. doi: 10.1016/j.neubiorev.2005.05.006.

Silva, B. A., C. Mattucci, P. Krzywkowski, E. Murana, A. Illarionova, V. Grinevich, N. S. Canteras, D. Ragozzino, and C. T. Gross. 2013. "Independent hypothalamic circuits for social and predator fear."  Nat Neurosci 16 (12):1731-3. doi: 10.1038/nn.3573.

Tsang, E., C. Orlandini, R. Sureka, A. H. Crevenna, E. Perlas, I. Prankerd, M. E. Masferrer, and C. T. Gross. 2023. "Induction of flight via midbrain projections to the cuneiform nucleus."  PLoS One 18 (2):e0281464. doi: 10.1371/journal.pone.0281464.

Vianna, D. M., and M. L. Brandao. 2003. "Anatomical connections of the periaqueductal gray: specific neural substrates for different kinds of fear."  Braz J Med Biol Res 36 (5):557-66. doi: 10.1590/s0100-879x2003000500002.

Walker, D. L., and M. Davis. 1997. "Involvement of the dorsal periaqueductal gray in the loss of fear-potentiated startle accompanying high footshock training."  Behav Neurosci 111 (4):692-702. doi: 10.1037//0735-7044.111.4.692.

Wang, L., I. Z. Chen, and D. Lin. 2015. "Collateral pathways from the ventromedial hypothalamus mediate defensive behaviors."  Neuron 85 (6):1344-58. doi: 10.1016/j.neuron.2014.12.025.

Wei, P., N. Liu, Z. Zhang, X. Liu, Y. Tang, X. He, B. Wu, Z. Zhou, Y. Liu, J. Li, Y. Zhang, X. Zhou, L. Xu, L. Chen, G. Bi, X. Hu, F. Xu, and L. Wang. 2015. "Processing of visually evoked innate fear by a non-canonical thalamic pathway."  Nat Commun 6:6756. doi: 10.1038/ncomms7756.

Yeh, L. F., T. Ozawa, and J. P. Johansen. 2021. "Functional organization of the midbrain periaqueductal gray for regulating aversive memory formation."  Mol Brain 14 (1):136. doi: 10.1186/s13041-021-00844-0.

Yilmaz, M., and M. Meister. 2013. "Rapid innate defensive responses of mice to looming visual stimuli."  Curr Biol 23 (20):2011-5. doi: 10.1016/j.cub.2013.08.015.

Zhou, Z., X. Liu, S. Chen, Z. Zhang, Y. Liu, Q. Montardy, Y. Tang, P. Wei, N. Liu, L. Li, R. Song, J. Lai, X. He, C. Chen, G. Bi, G. Feng, F. Xu, and L. Wang. 2019. "A VTA GABAergic Neural Circuit Mediates Visually Evoked Innate Defensive Responses."  Neuron 103 (3):473-488 e6. doi: 10.1016/j.neuron.2019.05.027.

eLife assessment

This valuable work describes a new protein factor that is required for filamentous phage assembly. Convincing evidence is provided for the binding of PSB15 to the packaging signal of the single-stranded DNA, Trx, and cardiolipin, and a mechanism for how the phage DNA is targeted to the assembly site in the bacterial inner membrane is presented. The work will be of interest to microbiologists.

Reviewer #1 (Public Review):

Summary:

This work describes a new protein factor required for filamentous phage assembly. The protein PSB15 binds to the packaging signal of the ssDNA, Trx and cardiolipin. A mechanism how the phage DNA is targeted to the assembly site in the bacterial inner membrane is discussed.

Strengths:

The work describes a clever way to detect factors required for phage propagation by looking at the plaque size of pseudorevertants that arise after infection of a phage with a directed mutation in the packaging signal. This led to the detection of a phage protein expressed from ORF9, the PSB15.

The authors convincingly show that PSB15 is expressed in infected cells and can complement a phage with a mutated orf9.

Weaknesses:

Given the fact that the phage LF-UK is not well explored, many open questions should be mentioned in the introduction. For the study, it is important to know if the phageLF-UK has a mimick or homolog of gV and gXI, and if not, whether PSB15 could take their role.

I am not convinced of the proposition of their term "checkpoint". The truth is that the authors do not know the real purpose of PSB15. I do not see an advantage for a checkpoint that only adds an additional step to enter the phage assembly site. There must be a biochemical reason for the action of PSB15. Looking at Figure 7, the step from "Release" to "Loading" is just adding many unknowns, e.g. how to transfer the DNA, how to dispose of PSB15 and Trx? Also, in the previous step are three question marks that do not add any solid information.

The in vivo study of subcellular localization is very questionable. Why is there a single fluorescent dot if there are thousands of PSB15 molecules expressed in the cell? I have my doubts that the conclusions the authors make here are correct and meaningful. The movies do not add anything significant.

Reviewer #2 (Public Review):

Secretion of the prototypical F-associated filamentous phage (Ff) of E. coli depends on the selective binding of a hairpin (the packaging signal, PS) by two phage encoded protein, pVII and pIX. PVII and pIX target the PS to IM channels formed by pI and pIV. However, integrative filamentous phages lack a homologue of pIX and pIV, and many of them also lack a homologue of pVII, raising questions on the assembly and secretion of new phages. In the manuscript, Yueh et al. present the identification of a phage-encoded protein, PSB15, which binds to the PS signal of a Xanthomonas integrative filamentous phage, ΦLf-UK. They showed that PSB15 is required for viral assembly and is conserved in several other integrative filamentous phages. They further analyzed how PSB15 binds to PS and demonstrated that it associates to the IM, which targets phage DNA to it. Finally, they show that thioredoxin, the only host protein that was found to be essential for Ff secretion, interacts with PSB15 and releases the PSB15-PS complex from the IM. These results are important because they elucidate a major step in the secretion of integrative filamentous phage, and the role of thioredoxin on filamentous phage secretion in general.

I found the data and interpretation convincing. However, the presentation and description are confusing in places because the reader has to juggle between figures. A scheme depicting what is known and unknown in the integration of Ff phages and interactive filamentous phages in the introduction would be useful to the general reader.

eLife assessment

This study presents important data describing cell states of olfactory ensheathing cells, and how these cell states may relate to repair after spinal cord injury. While the overall framework used for characterizing these cells is solid, the quantification and contextualization of results are incomplete, given that measurements, significance statistics, and discussion of both previous work and experimental methods that would be necessary to support several claims are not provided. With more thorough quantification and discussion, this work will be of interest to stem cell biologists and spinal cord injury researchers.

Joint Public Review:

Summary

This manuscript explores the transcriptomic identities of olfactory ensheathing cells (OECs), glial cells that support life-long axonal growth in olfactory neurons, as they relate to spinal cord injury repair. The authors show that transplantation of cultured, immunopurified rodent OECs at a spinal cord injury site can promote injury-bridging axonal regrowth. They then characterize these OECs using single-cell RNA sequencing, identifying five subtypes and proposing functional roles that include regeneration, wound healing, and cell-cell communication. They identify one progenitor OEC subpopulation and also report several other functionally relevant findings, notably, that OEC marker genes contain mixtures of other glial cell type markers (such as for Schwann cells and astrocytes), and that these cultured OECs produce and secrete Reelin, a regrowth-promoting protein that has been disputed as a gene product of OECs.

This manuscript offers an extensive, cell-level characterization of OECs, supporting their potential therapeutic value for spinal cord injury and suggesting potential underlying repair mechanisms. The authors use various approaches to validate their findings, providing interesting images that show the overlap between sprouting axons and transplanted OECs, and showing that OEC marker genes identified using single-cell RNA sequencing are present in vivo, in both olfactory bulb tissue and spinal cord after OEC transplantation.

Despite the breadth of information presented, however, further quantification of results and explanation of experimental approaches would be needed to support some of the authors' claims. Additionally, a more thorough discussion is needed to contextualize their findings relative to previous work.

(1) Important quantification is lacking for the data presented. For example, multiple figures include immunohistochemistry or immunocytochemistry data (Figures 1, 5, 6), but they are presented without accompanying measures like fractions of cells labeled or comparisons against controls. As a result, for axons projecting via OEC bridges in Figure 1, it is unclear how common these bridges are in the presence or absence of OECs. For Figure 6., it is unclear whether cells having an alternative OEC morphology coincide with progenitor OEC subtype marker genes to a statistically significant degree. Similar quantification is missing in other types of data such as Western blot images (Fig. 9) and OEC marker gene data (for which p-values are not reported; Table S2).

The addition of quantitative measures and, where appropriate, statistical comparisons with p-values or other significance measures, would be important for supporting the authors' claims and more rigorously conveying the results.

(2) Some aspects of the experimental design that are relevant to the interpretation of the results are not explained. For example, OECs appear to be collected from only female rats, but the potential implications of this factor are not discussed.

Additionally, it is unclear from the manuscript to what degree immunopurified cells are OECs as opposed to other cell types. The antibody used to retain OECs, nerve growth factor receptor p75 (Ngfr-p75), can also be expressed by non-OEC olfactory bulb cell types including astrocytes [1-3]. The possible inclusion of Ngfr-p75-positive but non-OEC cell types in the OEC culture is not sufficiently addressed. Such non-OEC cell types are also not distinguished in the analysis of single-cell RNA sequencing data (only microglia, fibroblasts, and OECs are identified; Figure 2). Thus, it is currently unclear whether results related to the OEC subtype may have been impacted by these experimental factors.

(3) The introduction, while well written, does not discuss studies showing no significant effect of OEC implantation after spinal cord injury. The discussion also fails to sufficiently acknowledge this variability in the efficacy of OEC implantation. This omission amplifies bias in the text, suggesting that OECs have significant effects that are not fully reflected in the literature. The introduction would need to be expanded to properly address the nuance suggested by the literature regarding the benefits of OECs after spinal cord injury. Additionally, in the discussion, relating the current study to previous work would help clarify how varying observations may relate to experimental or biological factors.

(a) Cragnolini, A.B. et al., Glia, (2009), doi: 10.1002/glia.20857.<br /> (b) Vickland H. et al., Brain Res., (1991), doi: 10.1016/0006-8993(91)91659-O.<br /> (c) Ung K. et al., Nat Commun., (2021), doi: 10.1038/s41467-021-25444-3.

eLife assessment

This study presents valuable research comparing three different species of extant cartilaginous fishes and describes new data on ratfish. The methods are convincing although the reviewers noted that standardized methods are essential when comparing numerical datasets. This study would be of interest to skeletal biologists working on the evolution of chondrichthyan skeletons.

Reviewer #1 (Public Review):

Summary:

It seems as if the main point of the paper is about the new data related to rat fish although your title is describing it as extant cartilaginous fishes and you bounce around between the little skate and ratfish. So here's an opportunity for you to adjust the title to emphasize ratfish is given the fact that leader you describe how this is your significant new data contribution. Either way, the organization of the paper can be adjusted so that the reader can follow along the same order for all sections so that it's very clear for comparative purposes of new data and what they mean. My opinion is that I want to read, for each subheading in the results, about the the ratfish first because this is your most interesting novel data. Then I want to know any confirmation about morphology in little skate. And then I want to know about any gaps you fill with the cat shark. (It is ok if you keep the order of "skate, ratfish, then shark, but I think it undersells the new data).

Strengths:

The imagery and new data availability for ratfish are valuable and may help to determine new phylogenetically informative characters for understanding the evolution of cartilaginous fishes. You also allude to the fossil record.

Opportunities:

I am concerned about the statement of ratfish paedomorphism because stage 32 and 33 were not statistically significantly different from one another (figure and prior sentences). So, these ratfish TMDs overlap the range of both 32 and 33. I think you need more specimens and stages to state this definitely based on TMD. What else leads you to think these are paedomorphic? Right now they are different, but it's unclear why. You need more outgroups.

Your headings for the results subsection and figures are nice snapshots of your interpretations of the results and I think they would be better repurposed in your abstract, which needs more depth.

Historical literature is more abundant than what you've listed. Your first sentence describes a long fascination and only goes back to 1990. But there are authors that have had this fascination for centuries and so I think you'll benefit from looking back. Especially because several of them have looked into histology and development of these fishes.

I agree that in the past 15 years or so a lot more work has been done because it can be done using newer technologies and I don't think your list is exhaustive. You need to expand this list and history which will help with your ultimate comparative analysis without you needed to sample too many new data yourself.

I'd like to see modifications to figure 7 so that you can add more continuity between the characters, illustrated in figure 7 and the body of the text. Generally Holocephalans are the outgroup to elasmobranchs - right now they are presented as sister taxa with no ability to indicate derivation. Why isn't the catshark included in this diagram?

In the last paragraph of the introduction, you say that "the data argue" and I admit, I am confused. Whose data? Is this a prediction or results or summary of other people's work? Either way, could be clarified to emphasize the contribution you are about to present.

Reviewer #2 (Public Review):

General comment:

This is a very valuable and unique comparative study. An excellent combination of scanning and histological data from three different species is presented. Obtaining the material for such a comparative study is never trivial. The study presents new data and thus provides the basis for an in-depth discussion about chondrichthyan mineralised skeletal tissues. I have, however, some comments. Some information is lacking and should be added to the manuscript text. I also suggest changes in the result and the discussion section of the manuscript.

Introduction:

The reader gets the impression almost no research on chondrichthyan skeletal tissues was done before the 2010 ("last 15 years", L45). I suggest to correct that and to cite also previous studies on chondrichthyan skeletal tissues, this includes studies from before 1900.

Material and Methods:

Please complete L473-492: Three different Micro-CT scanners were used for three different species? ScyScan 117 for the skate samples. Catshark different scanner, please provide full details. Chimera Scncrotron Scan? Please provide full details for all scanning protocols.

TMD is established in the same way in all three scanners? Actually not possible. Or, all specimens were scanned with the same scanner to establish TMD? If so please provide the protocol.

Please complete L494 ff: Tissue embedding medium and embedding protocol is missing. Specimens have been decalcified, if yes how? Have specimens been sectioned non-decalcified or decalcified?

Please complete L506 ff: Tissue embedding medium and embedding protocol is missing. Description of controls are missing.

Results:

L147: It is valuable and interesting to compare the degree of mineralisation in individuals from the three different species. It appears, however, not possible to provide numerical data for Tissue Mineral Density (TMD). First requirement, all specimens must be scanned with the same scanner and the same calibration values. This in not stated in the M&M section. But even if this was the case, all specimens derive from different sample locations and have, been preserved differently. Type of fixation, extension of fixation time in formalin, frozen, unfrozen, conditions of sample storage, age of the samples, and many more parameters, all influence TMD values. Likewise the relative age of the animals (adult is not the same as adult) influences TMD. One must assume different sampling and storage conditions and different types of progression into adulthood. Thus, the observation of different degrees of mineralisation is very interesting but I suggest not to link this observation to numerical values.

Parts of the results are mixed with discussion. Sometimes, a result chapter also needs a few references but this result chapter is full of references.

Based on different protocols, the staining characteristics of the tissue are analysed. This is very good and provides valuable additional data. The authors should inform the not only about the staining (positive of negative) abut also about the histochemical characters of the staining. L218: "fast green positive" means what? L234: "marked by Trichrome acid fuchsin" means what? And so on, see also L237, L289, L291<br /> Discussion

Please completely remove figure 7, please adjust and severely downsize the discussion related to figure 7. It is very interesting and valuable to compare three species from three different groups of elasmobranchs. Results of this comparison also validate an interesting discussion about possible phylogenetic aspects. This is, however, not the basis for claims about the skeletal tissue organisation of all extinct and extant members of the groups to which the three species belong. The discussion refers to "selected representatives" (L364), but how representative are the selected species? Can there be a extant species that represents the entire large group, all sharks, rays or chimeras? Are the three selected species basal representatives with a generalist life style?

Please completely remove the discussion about paedomorphosis in chimeras (already in the result section). This discussion is based on a wrong idea about the definition of paedomorphosis. Paedomorphosis can occur in members of the same group. Humans have paedormorphic characters within the primates, Ambystoma mexicanum is paedormorphic within the urodeals. Paedomorphosis does not extend to members of different vertebrate branches. That elasmobranchs have a developmental stage that resembles chimera vertebra mineralisation does not define chimera vertebra centra as paedomorphic. Teleost have a herocercal caudal fin anlage during development, that does not mean the heterocercal fins in sturgeons or elasmobranchs are paedomorphic characters.

L432-435: In times of Gadow & Abott (1895) science had completely wrong ideas bout the phylogenic position of chondrichthyans within the gnathostomes. It is curious that Gadow & Abott (1895) are being cited in support of the paedomorphosis claim.

The SCPP part of the discussion is unrelated to the data obtained by this study. Kawaki & WEISS (2003) describe a gene family (called SCPP) that control Ca-binding extracellular phosphoproteins in enamel, in bone and dentine, in saliva and in milk. It evolved by gene duplication and differentiation. They date it back to a first enamel matrix protein in conodonts (Reif 2006). Conodonts, a group of enigmatic invertebrates have mineralised structures but these structure are neither bone nor mineralised cartilage. Cat fish (6 % of all vertebrate species) on the other hand, have bone but do not have SCPP genes (Lui et al. 206). Other calcium binding proteins, such as osteocalcin, were initially believed to be required for mineralisation. It turned out that osteocalcin is rather a mineralisation inhibitor, at best it regulates the arrangement collagen fiber bundles. The osteocalcin -/- mouse has fully mineralised bone. As the function of the SCPP gene product for bone formation is unknown, there is no need to discuss SCPP genes. It would perhaps be better to finish the manuscript with summery that focuses on the subject and the methodology of this nice study.

eLife assessment

This useful study reports that epididymal proteins are required for embryogenesis after fertilization. The data presented are generally convincing, but the study is incomplete because it does not investigate in detail how those proteins cause DNA fragmentation and compromised embryonic development. This work will be of interest to reproductive biologists and andrologists.

Reviewer #1 (Public Review):

Summary:

The main observation that the sperm from CRISP proteins 1 and 3 KO lines are post-fertilization less developmentally competent is convincing. However, the molecular characterization of the mechanism that leads to these defects and the temporal appearance of the defects requires additional studies.

Strengths:

The generation of these double mutant mice is valuable for the field. Moreover, the fact that the double mutant line of Crisp 1 and 3 is phenotypically different from the Crisp 1 and 4 line suggests different functions of these epididymis proteins. The methods used to demonstrate that developmental defects are largely due to post-fertilization defects are also a considerable strength. The initial characterization of these sperm has altered intracellular Ca2+ levels, and increased rates of DNA fragmentation are valuable.

Weaknesses:

The study is mechanistically incomplete because there is no direct demonstration that the absence of these proteins alters the epididymal environment and fluid, wherein during the passage through the epididymis the sperm become affected. Also, a direct demonstration of how the proteins in question cause or lead to DNA damage and increased Ca2+ requires further characterization.

Reviewer #2 (Public Review):

The authors showed that CRISP1 and CRISP3, secreted proteins in the epididymis, are required for early embryogenesis after fertilization through DNA integrity in cauda epididymal sperm. This paper is the first report showing that the epididymal proteins are required for embryogenesis after fertilization. However, some data in this paper (Table 1 and Figure 2A) are overlapped in a published paper (Curci et al., FASEB J, 34,15718-15733, 2020; PMID: 33037689). Furthermore, the authors did not address why the disruption of CRISP1/3 leads to these phenomena (the increased level of the intracellular Ca2+ level and impaired DNA integrity in sperm) with direct evidence. Therefore, if the authors can address the following comments to improve the paper's novelty and clarification, this paper may be worthwhile to readers.

Author response

Reviewer #1 (Public Review):

The authors aimed to investigate if 2-hydroxybutyrate (2HB), a metabolite induced by exercise, influences physiological changes, particularly metabolic alterations post-exercise training. They treated young mice and cultured myoblasts with 2HB, conducted exercise tests, metabolomic profiling, gene expression analysis, and knockdown experiments to understand 2HB's mechanisms. Their findings indicate that 2HB enhances exercise tolerance, boosts branch chain amino acid (BCAA) enzyme gene expression in skeletal muscles, and increases oxidative capacity. They also highlight the role of SIRT4 in these effects. This study establishes 2HB, once considered a waste product, as a regulator of exercise-induced metabolic processes. The study's strength lies in its consistent results across in vitro, in vivo, and ex vivo analyses.

The authors propose a mechanism in which 2HB inhibits BCAA breakdown, raises NAD+/NADH ratio, activates SIRT4, increases ADP ribosylation, and controls gene expression.

However, some questions remain unclear based on these findings:

This study focused on the effects of short-term exercise (1 or 5 bouts of treadmill running) and short-term 2HB treatment (1 or 4 days of treatment). Adaptations to exercise training typically occur progressively over an extended period. It's important to investigate the effects of long-term 2HB treatment and whether extended combined 2HB treatment and exercise training have independent, synergistic, or antagonistic effects.

We agree with the reviewer that investigation of longer-term 2HB treatment may potentially yield interesting findings with more implications to exercise physiology. To investigate the effects of 2HB treatment against or in combination with a progressive exercise training protocol would require an experiment duration between 4 to 12 weeks, based on previous studies (Systematic Review by Massett et al., Frontiers in Physiology, 2021, 10.3389/fphys.2021.782695). However, our experience with these types of experiments is that such a pursuit would require a breadth of work beyond the scope of this current study. For instance, if there were evidence of weakened effect of 2HB over time, one may be compelled to investigate other organs such as the liver to find signs of metabolic adaptation to the exogenous metabolite. If there were additive or synergistic effects on exercise performance, one may be compelled to investigate changes to the cardiovascular system in addition to the skeletal muscle. Additional questions would be raised around the skeletal muscle as well, including assessment of structural and fibre-type changes. Further, these additional mechanisms would need to be characterized in a time course fashion. Rather, we view the scope of the current study to be the acute response to 2HB as an initial report on mechanistic effects of 2HB.

Exercise training leads to significant mitochondrial changes, including increased mitochondrial biogenesis in skeletal muscle. It would be valuable to compare the impact of 2HB treatment on mitochondrial content and oxidative capacity in treated mice to that in exercised mice.

We agree with the author that it is of interest to investigate how 2HB may affect mitochondrial biogenesis. However, our preliminary findings were that 2HB-treated MEFs, C2C12s, and mouse soleus muscles showed no change in PGC1α gene expression after four days of treatment (data not shown). As a follow-up assessment of mitochondrial protein expression, although not specific to mtDNA derived genes, we quantified the expression of the respiratory chain proteins in cells and soleus muscle and found no effect of 2HB treatment (SFig. 5,6). At this stage we conclude that there is not evidence of 2HB modifying mitochondrial biogenesis in this time frame and that further investigation would be best suited to a follow-up study such as one interested in long-term exercise training.

The authors demonstrate that 2-ketobutyrate (2KB) can serve as an oxidative fuel, suggesting a role for the intact BCAA catabolic pathway. However, it's puzzling that the knockout of BCKDHA, a subunit crucial for the second step of BCAA catabolism, did not result in changes in oxidative capacity in cultured myoblasts.

While we report the BCKDH complex to be dispensable for 2KB oxidation it is important to note that previous studies have reported the following: (1) that 2KB is a viable substrate for BCKDH, (2) that 2KB is a viable substrate for pyruvate dehydrogenase, and (3) that pyruvate dehydrogenase is also dispensable for 2KB oxidation (see Steele et al., J Nutr., 114: 701-710, and Paxton et al. Biochem J., 234:295-303). Collectively, these data have led previous studies to conclude that BCKDH and pyruvate dehydrogenase are redundant for the first step of 2KB oxidation, with a preference for BCKDH. The flux through either may depend upon the metabolic environment. The aim for figure 3C was to determine whether the BCAA degradation pathway was required for 2KB oxidation. We conclude that this pathway is required, first at the step of PCC.

While these past studies were mentioned in paragraph 2 of the discussion, in light of the reviewer’s comment we have expanded this paragraph. We have added language to explain that future research interested in the presented 2HB mechanism should carefully consider BCKDH and PDH expression in the cell or tissue of interest, as the metabolism of 2KB is quite central to the presented mechanism.

Nevertheless, this innovative model of metabolic signaling during exercise will serve as a valuable reference for informing future.

Reviewer #2 (Public Review):

Summary:

The manuscript entitled "A 2-HB-mediated feedback loop regulates muscular fatigue" by the Johnson group reports interesting findings with implications for the health benefits of exercise. The authors use a combination of metabolic/biochemical in vivo and in vitro assays to delineate a metabolic route triggered by 2-HB (a relatively stable metabolite induced by exercise in humans and mice) that controls branched-chain amino transferase enzymes and mitochondrial oxidative capacity. Mechanistically, the author shows that 2-HB is a direct inhibitor of BCAT enzymes that in turn control levels of SIRT4 activity and ADp-ribosylation in the nucleus targeting C/EBP transcription factor, affecting BCAA oxidation genes (see Fig 4i in the paper). Overall, these are interesting and novel observations and findings with relevance to human exercise, with the potential implication of using these metabolites to mimic exercise benefits, or conditions or muscular fatigue that occurs in different human chronic diseases including rheumatic diseases or long COVID.

Weaknesses:

There are several experiments/comments that will strengthen the manuscript-

(1) A final model in Figure 6 integrating the exercise/mechanistic findings, expanding on Fig 4i) will clarify the findings.

We appreciate the reviewer’s suggestion to incorporate the exercise findings into a summary figure. However, upon internal review we find that such a figure is too similar to Fig 4i to warrant a new diagram.

(2) In some of the graphs, statistics are missing (e.g Fig 6G).

Some figures are included primarily for the reader to visualize the data while statistical comparison is conducted in a separate figure, for example Fig 2D-G. However, we have revised the figure legends to ensure that statistical comparisons are described for all appropriate figures, including Fig 6G identified by the reviewer.

(3) The conclusions on SIRT4 dependency should be carefully written, as it is likely that this is only one potential mechanism, further validation with mouse models would be necessary.

We appreciate the reviewers feedback and take the point well that a NAD-dependent mechanism will likely stimulate other sirtuins, which are often in fact expressed at greater levels than SIRT4. To reflect this comment in the manuscript we have altered paragraph 5 of the discussion to now focus on sirtuins. We briefly discuss SIRT4 and highlight the need for future consideration of other sirtuins, perhaps particularly mitochondrial sirtuins.

(4) One of the needed experiments to support the oxidative capacity effects that could be done in cultured cells, is the use of radiosotope metabolites including BCCAs to determine the ability to produce CO2. Alternatively or in combination metabolite flux using isotopes would be useful to strengthen the current results.

We appreciate the suggestion from the reviewer and we will look to conduct such an experiment in our follow-up work.

We sincerely thank the reviewers for their input on this study as their suggestions have led to an improved manuscript for the version of record. The reviewer comments are well taken and we are glad that they will be present alongside the final manuscript to provide an important perspective on the work.

Author response:

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

Public Reviews:

Reviewer #1 (Public Review):

Summary:

Using a cross-modal sensory selection task in head-fixed mice, the authors attempted to characterize how different rules reconfigured representations of sensory stimuli and behavioral reports in sensory (S1, S2) and premotor cortical areas (medial motor cortex or MM, and ALM). They used silicon probe recordings during behavior, a combination of single-cell and population-level analyses of neural data, and optogenetic inhibition during the task.

Strengths:

A major strength of the manuscript was the clarity of the writing and motivation for experiments and analyses. The behavioral paradigm is somewhat simple but well-designed and wellcontrolled. The neural analyses were sophisticated, clearly presented, and generally supported the authors' interpretations. The statistics are clearly reported and easy to interpret. In general, my view is that the authors achieved their aims. They found that different rules affected preparatory activity in premotor areas, but not sensory areas, consistent with dynamical systems perspectives in the field that hold that initial conditions are important for determining trial-based dynamics.

Weaknesses:

The manuscript was generally strong. The main weakness in my view was in interpreting the optogenetic results. While the simplicity of the task was helpful for analyzing the neural data, I think it limited the informativeness of the perturbation experiments. The behavioral read-out was low dimensional -a change in hit rate or false alarm rate- but it was unclear what perceptual or cognitive process was disrupted that led to changes in these read-outs. This is a challenge for the field, and not just this paper, but was the main weakness in my view. I have some minor technical comments in the recommendations for authors that might address other minor weaknesses.

I think this is a well-performed, well-written, and interesting study that shows differences in rule representations in sensory and premotor areas and finds that rules reconfigure preparatory activity in the motor cortex to support flexible behavior.

Reviewer #2 (Public Review):

Summary:

Chang et al. investigate neuronal activity firing patterns across various cortical regions in an interesting context-dependent tactile vs visual detection task, developed previously by the authors (Chevee et al., 2021; doi: 10.1016/j.neuron.2021.11.013). The authors report the important involvement of a medial frontal cortical region (MM, probably a similar location to wM2 as described in Esmaeili et al., 2021 & 2022; doi: 10.1016/j.neuron.2021.05.005; doi: 10.1371/journal.pbio.3001667) in mice for determining task rules.

Strengths:

The experiments appear to have been well carried out and the data well analysed. The manuscript clearly describes the motivation for the analyses and reaches clear and well-justified conclusions. I find the manuscript interesting and exciting!

Weaknesses:

I did not find any major weaknesses.

Reviewer #3 (Public Review):

This study examines context-dependent stimulus selection by recording neural activity from several sensory and motor cortical areas along a sensorimotor pathway, including S1, S2, MM, and ALM. Mice are trained to either withhold licking or perform directional licking in response to visual or tactile stimulus. Depending on the task rule, the mice have to respond to one stimulus modality while ignoring the other. Neural activity to the same tactile stimulus is modulated by task in all the areas recorded, with significant activity changes in a subset of neurons and population activity occupying distinct activity subspaces. Recordings further reveal a contextual signal in the pre-stimulus baseline activity that differentiates task context. This signal is correlated with subsequent task modulation of stimulus activity. Comparison across brain areas shows that this contextual signal is stronger in frontal cortical regions than in sensory regions. Analyses link this signal to behavior by showing that it tracks the behavioral performance switch during task rule transitions. Silencing activity in frontal cortical regions during the baseline period impairs behavioral performance.

Overall, this is a superb study with solid results and thorough controls. The results are relevant for context-specific neural computation and provide a neural substrate that will surely inspire follow-up mechanistic investigations. We only have a couple of suggestions to help the authors further improve the paper.

(1) We have a comment regarding the calculation of the choice CD in Fig S3. The text on page 7 concludes that "Choice coding dimensions change with task rule". However, the motor choice response is different across blocks, i.e. lick right vs. no lick for one task and lick left vs. no lick for the other task. Therefore, the differences in the choice CD may be simply due to the motor response being different across the tasks and not due to the task rule per se. The authors may consider adding this caveat in their interpretation. This should not affect their main conclusion.

We thank the Reviewer for the suggestion. We have discussed this caveat and performed a new analysis to calculate the choice coding dimensions using right-lick and left-lick trials (Fig. S3h) on page 8. 

“Choice coding dimensions were obtained from left-lick and no-lick trials in respond-to-touch blocks and right-lick and no-lick trials in respond-to-light blocks. Because the required lick directions differed between the block types, the difference in choice CDs across task rules (Fig. S4f) could have been affected by the different motor responses. To rule out this possibility, we did a new version of this analysis using right-lick and left-lick trials to calculate the choice coding dimensions for both task rules. We found that the orientation of the choice coding dimension in a respond-to-touch block was still not aligned well with that in a respond-to-light block (Fig. S4h;  magnitude of dot product between the respond-to-touch choice CD and the respond-to-light choice CD, mean ± 95% CI for true vs shuffled data: S1: 0.39 ± [0.23, 0.55] vs 0.2 ± [0.1, 0.31], 10 sessions; S2: 0.32 ± [0.18, 0.46] vs 0.2 ± [0.11, 0.3], 8 sessions; MM: 0.35 ± [0.21, 0.48] vs 0.18 ± [0.11, 0.26], 9 sessions; ALM: 0.28 ± [0.17, 0.39] vs 0.21 ± [0.12, 0.31], 13 sessions).”

We also have included the caveats for using right-lick and left-lick trials to calculate choice coding dimensions on page 13.

“However, we also calculated choice coding dimensions using only right- and left-lick trials. In S1, S2, MM and ALM, the choice CDs calculated this way were also not aligned well across task rules (Fig. S4h), consistent with the results calculated from lick and no-lick trials (Fig. S4f). Data were limited for this analysis, however, because mice rarely licked to the unrewarded water port (# of licksunrewarded port  / # of lickstotal , respond-to-touch: 0.13, respond-to-light: 0.11). These trials usually came from rule transitions (Fig. 5a) and, in some cases, were potentially caused by exploratory behaviors. These factors could affect choice CDs.”

(2) We have a couple of questions about the effect size on single neurons vs. population dynamics. From Fig 1, about 20% of neurons in frontal cortical regions show task rule modulation in their stimulus activity. This seems like a small effect in terms of population dynamics. There is somewhat of a disconnect from Figs 4 and S3 (for stimulus CD), which show remarkably low subspace overlap in population activity across tasks. Can the authors help bridge this disconnect? Is this because the neurons showing a difference in Fig 1 are disproportionally stimulus selective neurons?

We thank the Reviewer for the insightful comment and agree that it is important to link the single-unit and population results. We have addressed these questions by (1) improving our analysis of task modulation of single neurons  (tHit-tCR selectivity) and (2) examining the relationship between tHit-tCR selective neurons and tHit-tCR subspace overlaps.  

Previously, we averaged the AUC values of time bins within the stimulus window (0-150 ms, 10 ms bins). If the 95% CI on this averaged AUC value did not include 0.5, this unit was considered to show significant selectivity. This approach was highly conservative and may underestimate the percentage of units showing significant selectivity, particularly any units showing transient selectivity. In the revised manuscript, we now define a unit as showing significant tHit-tCR selectivity when three consecutive time bins (>30 ms, 10ms bins) of AUC values were significant. Using this new criterion, the percentage of tHittCR selective neurons increased compared with the previous analysis. We have updated Figure 1h and the results on page 4:

“We found that 18-33% of neurons in these cortical areas had area under the receiver-operating curve (AUC) values significantly different from 0.5, and therefore discriminated between tHit and tCR trials (Fig. 1h; S1: 28.8%, 177 neurons; S2: 17.9%, 162 neurons; MM: 32.9%, 140 neurons; ALM: 23.4%, 256 neurons; criterion to be considered significant: Bonferroni corrected 95% CI on AUC did not include 0.5 for at least 3 consecutive 10-ms time bins).”

Next, we have checked how tHit-tCR selective neurons were distributed across sessions. We found that the percentage of tHit-tCR selective neurons in each session varied (S1: 9-46%, S2: 0-36%, MM:25-55%, ALM:0-50%). We examined the relationship between the numbers of tHit-tCR selective neurons and tHit-tCR subspace overlaps. Sessions with more neurons showing task rule modulation tended to show lower subspace overlap, but this correlation was modest and only marginally significant (r= -0.32, p= 0.08, Pearson correlation, n= 31 sessions). While we report the percentage of neurons showing significant selectivity as a simple way to summarize single-neuron effects, this does neglect the magnitude of task rule modulation of individual neurons, which may also be relevant. 

In summary, the apparent disconnect between the effect sizes of task modulation of single neurons and of population dynamics could be explained by (1) the percentages of tHit-tCR selective neurons were underestimated in our old analysis, (2) tHit-tCR selective neurons were not uniformly distributed among sessions, and (3) the percentages of tHit-tCR selective neurons were weakly correlated with tHit-tCR subspace overlaps. 

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

For the analysis of choice coding dimensions, it seems that the authors are somewhat data limited in that they cannot compare lick-right/lick-left within a block. So instead, they compare lick/no lick trials. But given that the mice are unable to initiate trials, the interpretation of the no lick trials is a bit complicated. It is not clear that the no lick trials reflect a perceptual judgment about the stimulus (i.e., a choice), or that the mice are just zoning out and not paying attention. If it's the latter case, what the authors are calling choice coding is more of an attentional or task engagement signal, which may still be interesting, but has a somewhat different interpretation than a choice coding dimension. It might be worth clarifying this point somewhere, or if I'm totally off-base, then being more clear about why lick/no lick is more consistent with choice than task engagement.

We thank the Reviewer for raising this point. We have added a new paragraph on page 13 to clarify why we used lick/no-lick trials to calculate choice coding dimensions, and we now discuss the caveat regarding task engagement.  

“No-lick trials included misses, which could be caused by mice not being engaged in the task. While the majority of no-lick trials were correct rejections (respond-to-touch: 75%; respond-to-light: 76%), we treated no-licks as one of the available choices in our task and included them to calculate choice coding dimensions (Fig. S4c,d,f). To ensure stable and balanced task engagement across task rules, we removed the last 20 trials of each session and used stimulus parameters that achieved similar behavioral performance for both task rules (Fig. 1d; ~75% correct for both rules).”

In addition, to address a point made by Reviewer 3 as well as this point, we performed a new analysis to calculate choice coding dimensions using right-lick vs left-lick trials. We report this new analysis on page 8:

“Choice coding dimensions were obtained from left-lick and no-lick trials in respond-to-touch blocks and right-lick and no-lick trials in respond-to-light blocks. Because the required lick directions differed between the block types, the difference in choice CDs across task rules (Fig. S4f) could have been affected by the different motor responses. To rule out this possibility, we did a new version of this analysis using right-lick and left-lick trials to calculate the choice coding dimensions for both task rules. We found that the orientation of the choice coding dimension in a respond-to-touch block was still not aligned well with that in a respond-to-light block (Fig. S4h;  magnitude of dot product between the respond-to-touch choice CD and the respond-to-light choice CD, mean ± 95% CI for true vs shuffled data: S1: 0.39 ± [0.23, 0.55] vs 0.2 ± [0.1, 0.31], 10 sessions; S2: 0.32 ± [0.18, 0.46] vs 0.2 ± [0.11, 0.3], 8 sessions; MM: 0.35 ± [0.21, 0.48] vs 0.18 ± [0.11, 0.26], 9 sessions; ALM: 0.28 ± [0.17, 0.39] vs 0.21 ± [0.12, 0.31], 13 sessions).” 

We added discussion of the limitations of this new analysis on page 13:

“However, we also calculated choice coding dimensions using only right- and left-lick trials. In S1, S2, MM and ALM, the choice CDs calculated this way were also not aligned well across task rules (Fig. S4h), consistent with the results calculated from lick and no-lick trials (Fig. S4f). Data were limited for this analysis, however, because mice rarely licked to the unrewarded water port (# of licksunrewarded port  / # of lickstotal , respond-to-touch: 0.13, respond-to-light: 0.11). These trials usually came from rule transitions (Fig. 5a) and, in some cases, were potentially caused by exploratory behaviors. These factors could affect choice CDs.”

The authors find that the stimulus coding direction in most areas (S1, S2, and MM) was significantly aligned between the block types. How do the authors interpret that finding? That there is no major change in stimulus coding dimension, despite the change in subspace? I think I'm missing the big picture interpretation of this result.

That there is no significant change in stimulus coding dimensions but a change in subspace suggests that the subspace change largely reflects a change in the choice coding dimensions.

As I mentioned in the public review, I thought there was a weakness with interpretation of the optogenetic experiments, which the authors generally interpret as reflecting rule sensitivity. However, given that they are inhibiting premotor areas including ALM, one might imagine that there might also be an effect on lick production or kinematics. To rule this out, the authors compare the change in lick rate relative to licks during the ITI. What is the ITI lick rate? I assume pretty low, once the animal is welltrained, in which case there may be a floor effect that could obscure meaningful effects on lick production. In addition, based on the reported CI on delta p(lick), it looks like MM and AM did suppress lick rate. I think in the future, a task with richer behavioral read-outs (or including other measurements of behavior like video), or perhaps something like a psychological process model with parameters that reflect different perceptual or cognitive processes could help resolve the effects of perturbations more precisely.

Eighteen and ten percent of trials had at least one lick in the ITI in respond-to-touch and  respond-tolight blocks, respectively. These relatively low rates of ITI licking could indeed make an effect of optogenetics on lick production harder to observe. We agree that future work would benefit from more complex tasks and measurements, and have added the following to make this point (page 14):

“To more precisely dissect the effects of perturbations on different cognitive processes in rule-dependent sensory detection, more complex behavioral tasks and richer behavioral measurements are needed in the future.”

Reviewer #2 (Recommendations For The Authors):

I have the following minor suggestions that the authors might consider in revising this already excellent manuscript :

(1) In addition to showing normalised z-score firing rates (e.g. Fig 1g), I think it is important to show the grand-average mean firing rates in Hz.

We thank the Reviewer for the suggestion and have added the grand-average mean firing rates as a new supplementary figure (Fig. S2a). To provide more details about the firing rates of individual neurons, we have also added to this new figure the distribution of peak responses during the tactile stimulus period (Fig. S2b).

(2) I think the authors could report more quantitative data in the main text. As a very basic example, I could not easily find how many neurons, sessions, and mice were used in various analyses.

We have added relevant numbers at various points throughout the Results, including within the following examples:

Page 3: “To examine how the task rules influenced the sensorimotor transformation occurring in the tactile processing stream, we performed single-unit recordings from sensory and motor cortical areas including S1, S2, MM and ALM (Fig. 1e-g, Fig. S1a-h, and Fig. S2a; S1: 6 mice, 10 sessions, 177 neurons, S2: 5 mice, 8 sessions, 162 neurons, MM: 7 mice, 9 sessions, 140 neurons, ALM: 8 mice, 13 sessions, 256 neurons).”

Page 5: “As expected, single-unit activity before stimulus onset did not discriminate between tactile and visual trials (Fig. 2d; S1: 0%, 177 neurons; S2: 0%, 162 neurons; MM: 0%, 140 neurons; ALM: 0.8%, 256 neurons). After stimulus onset, more than 35% of neurons in the sensory cortical areas and approximately 15% of neurons in the motor cortical areas showed significant stimulus discriminability (Fig. 2e; S1: 37.3%, 177 neurons; S2: 35.2%, 162 neurons; MM: 15%, 140 neurons; ALM: 14.1%, 256 neurons).”

Page 6: “Support vector machine (SVM) and Random Forest classifiers showed similar decoding abilities

(Fig. S3a,b; medians of classification accuracy [true vs shuffled]; SVM: S1 [0.6 vs 0.53], 10 sessions, S2

[0.61 vs 0.51], 8 sessions, MM [0.71 vs 0.51], 9 sessions, ALM [0.65 vs 0.52], 13 sessions; Random

Forests: S1 [0.59 vs 0.52], 10 sessions, S2 [0.6 vs 0.52], 8 sessions, MM [0.65 vs 0.49], 9 sessions, ALM [0.7 vs 0.5], 13 sessions).”

Page 6: “To assess this for the four cortical areas, we quantified how the tHit and tCR trajectories diverged from each other by calculating the Euclidean distance between matching time points for all possible pairs of tHit and tCR trajectories for a given session and then averaging these for the session (Fig. 4a,b; S1: 10 sessions, S2: 8 sessions, MM: 9 sessions, ALM: 13 sessions, individual sessions in gray and averages across sessions in black; window of analysis: -100 to 150 ms relative to stimulus onset; 10 ms bins; using the top 3 PCs; Methods).” 

Page 8: “In contrast, we found that S1, S2 and MM had stimulus CDs that were significantly aligned between the two block types (Fig. S4e; magnitude of dot product between the respond-to-touch stimulus CDs and the respond-to-light stimulus CDs, mean ± 95% CI for true vs shuffled data: S1: 0.5 ± [0.34, 0.66] vs 0.21 ± [0.12, 0.34], 10 sessions; S2: 0.62 ± [0.43, 0.78] vs 0.22 ± [0.13, 0.31], 8 sessions; MM: 0.48 ± [0.38, 0.59] vs 0.24 ± [0.16, 0.33], 9 sessions; ALM: 0.33 ± [0.2, 0.47] vs 0.21 ± [0.13, 0.31], 13 sessions).”  Page 9: “For respond-to-touch to respond-to-light block transitions, the fractions of trials classified as respond-to-touch for MM and ALM decreased progressively over the course of the transition (Fig. 5d; rank correlation of the fractions calculated for each of the separate periods spanning the transition, Kendall’s tau, mean ± 95% CI: MM: -0.39 ± [-0.67, -0.11], 9 sessions, ALM: -0.29 ± [-0.54, -0.04], 13 sessions; criterion to be considered significant: 95% CI on Kendall’s tau did not include 0).

Page 11: “Lick probability was unaffected during S1, S2, MM and ALM experiments for both tasks, indicating that the behavioral effects were not due to an inability to lick (Fig. 6i, j; 95% CI on Δ lick probability for cross-modal selection task: S1/S2 [-0.18, 0.24], 4 mice, 10 sessions; MM [-0.31, 0.03], 4 mice, 11 sessions; ALM [-0.24, 0.16], 4 mice, 10 sessions; Δ lick probability for simple tactile detection task: S1/S2 [-0.13, 0.31], 3 mice, 3 sessions; MM [-0.06, 0.45], 3 mice, 5 sessions; ALM [-0.18, 0.34], 3 mice, 4 sessions).”

(3) Please include a clearer description of trial timing. Perhaps a schematic timeline of when stimuli are delivered and when licking would be rewarded. I may have missed it, but I did not find explicit mention of the timing of the reward window or if there was any delay period.

We have added the following (page 3): 

“For each trial, the stimulus duration was 0.15 s and an answer period extended from 0.1 to 2 s from stimulus onset.”

(4) Please include a clear description of statistical tests in each figure legend as needed (for example please check Fig 4e legend).

We have added details about statistical tests in the figure legends:

Fig. 2f: “Relationship between block-type discriminability before stimulus onset and tHit-tCR discriminability after stimulus onset for units showing significant block-type discriminability prior to the stimulus. Pearson correlation: S1: r = 0.69, p = 0.056, 8 neurons; S2: r = 0.91, p = 0.093, 4 neurons; MM: r = 0.93, p < 0.001, 30 neurons; ALM: r = 0.83, p < 0.001, 26 neurons.” 

Fig. 4e: “Subspace overlap for control tHit (gray) and tCR (purple) trials in the somatosensory and motor cortical areas. Each circle is a subspace overlap of a session. Paired t-test, tCR – control tHit: S1: -0.23, 8 sessions, p = 0.0016; S2: -0.23, 7 sessions, p = 0.0086; MM: -0.36, 5 sessions, p = <0.001; ALM: -0.35, 11 sessions, p < 0.001; significance: ** for p<0.01, *** for p<0.001.”  

Fig. 5d,e: “Fraction of trials classified as coming from a respond-to-touch block based on the pre-stimulus population state, for trials occurring in different periods (see c) relative to respond-to-touch → respondto-light transitions. For MM (top row) and ALM (bottom row), progressively fewer trials were classified as coming from the respond-to-touch block as analysis windows shifted later relative to the rule transition. Kendall’s tau (rank correlation): MM: -0.39, 9 sessions; ALM: -0.29, 13 sessions. Left panels: individual sessions, right panels: mean ± 95% CI. Dash lines are chance levels (0.5). e, Same as d but for respond-to-light → respond-to-touch transitions. Kendall’s tau: MM: 0.37, 9 sessions; ALM: 0.27, 13 sessions.”

Fig. 6: “Error bars show bootstrap 95% CI. Criterion to be considered significant: 95% CI did not include 0.”

(5) P. 3 - "To examine how the task rules influenced the sensorimotor transformation occurring in the tactile processing stream, we performed single-unit recordings from sensory and motor cortical areas including S1, S2, MM, and ALM using 64-channel silicon probes (Fig. 1e-g and Fig. S1a-h)." Please specify if these areas were recorded simultaneously or not.

We have added “We recorded from one of these cortical areas per session, using 64-channel silicon probes.”  on page 3.  

(6) Figure 4b - Please describe what gray and black lines show.

The gray traces are the distance between tHit and tCR trajectories in individual sessions and the black traces are the averages across sessions in different cortical areas. We have added this information on page 6 and in the Figure 4b legend. 

Page 6: “To assess this for the four cortical areas, we quantified how the tHit and tCR trajectories diverged from each other by calculating the Euclidean distance between matching time points for all possible pairs of tHit and tCR trajectories for a given session and then averaging these for the session (Fig. 4a,b; S1: 10 sessions, S2: 8 sessions, MM: 9 sessions, ALM: 13 sessions, individual sessions in gray and averages across sessions in black; window of analysis: -100 to 150 ms relative to stimulus onset; 10 ms bins; using the top 3 PCs; Methods).

Fig. 4b: “Distance between tHit and tCR trajectories in S1, S2, MM and ALM. Gray traces show the time varying tHit-tCR distance in individual sessions and black traces are session-averaged tHit-tCR distance (S1:10 sessions; S2: 8 sessions; MM: 9 sessions; ALM: 13 sessions).”

(7) In addition to the analyses shown in Figure 5a, when investigating the timing of the rule switch, I think the authors should plot the left and right lick probabilities aligned to the timing of the rule switch time on a trial-by-trial basis averaged across mice.

We thank the Reviewer for suggesting this addition. We have added a new figure panel to show the probabilities of right- and left-licks during rule transitions (Fig. 5a).

Page 8: “The probabilities of right-licks and left-licks showed that the mice switched their motor responses during block transitions depending on task rules (Fig. 5a, mean ± 95% CI across 12 mice).” 

(8) P. 12 - "Moreover, in a separate study using the same task (Finkel et al., unpublished), high-speed video analysis demonstrated no significant differences in whisker motion between respond-to-touch and respond-to-light blocks in most (12 of 14) behavioral sessions.". Such behavioral data is important and ideally would be included in the current analysis. Was high-speed videography carried out during electrophysiology in the current study?

Finkel et al. has been accepted in principle for publication and will be available online shortly. Unfortunately we have not yet carried out simultaneous high-speed whisker video and electrophysiology in our cross-modal sensory selection task.

Reviewer #3 (Recommendations For The Authors):

(1) Minor point. For subspace overlap calculation of pre-stimulus activity in Fig 4e (light purple datapoints), please clarify whether the PCs for that condition were constructed in matched time windows. If the PCs are calculated from the stimulus period 0-150ms, the poor alignment could be due to mismatched time windows.

We thank the Reviewer for the comment and clarify our analysis here. We previously used timematched windows to calculate subspace overlaps. However, the pre-stimulus activity was much weaker than the activity during the stimulus period, so the subspaces of reference tHit were subject to noise and we were not able to obtain reliable PCs. This caused the subspace overlap values between the reference tHit and control tHit to be low and variable (mean ± SD, S1:  0.46± 0.26, n = 8 sessions, S2: 0.46± 0.18, n = 7 sessions, MM: 0.44± 0.16, n = 5 sessions, ALM: 0.38± 0.22, n = 11 sessions).  Therefore, we used the tHit activity during the stimulus window to obtain PCs and projected pre-stimulus and stimulus activity in tCR trials onto these PCs. We have now added a more detailed description of this analysis in the Methods (page 32). 

“To calculate the separation of subspaces prior to stimulus delivery, pre-stimulus activity in tCR trials (100 to 0 ms from stimulus onset) was projected to the PC space of the tHit reference group and the subspace overlap was calculated. In this analysis, we used tHit activity during stimulus delivery (0 to 150 ms from stimulus onset) to obtain reliable PCs.”   

We acknowledge this time alignment issue and have now removed the reported subspace overlap between tHit and tCR during the pre-stimulus period from Figure 4e (light purple). However, we think the correlation between pre- and post- stimulus-onset subspace overlaps should remain similar regardless of the time windows that we used for calculating the PCs. For the PCs calculated from the pre-stimulus period (-100 to 0 ms), the correlation coefficient was 0.55 (Pearson correlation, p <0.01, n = 31 sessions). For the PCs calculated from the stimulus period (0-150 ms), the correlation coefficient was 0.68 (Figure 4f, Pearson correlation, p <0.001, n = 31 sessions). Therefore, we keep Figure 4f.  

(2) Minor point. To help the readers follow the logic of the experiments, please explain why PPC and AMM were added in the later optogenetic experiment since these are not part of the electrophysiology experiment.

We have added the following rationale on page 9.

“We recorded from AMM in our cross-modal sensory selection task and observed visually-evoked activity (Fig. S1i-k), suggesting that AMM may play an important role in rule-dependent visual processing. PPC contributes to multisensory processing51–53 and sensory-motor integration50,54–58.  Therefore, we wanted to test the roles of these areas in our cross-modal sensory selection task.”

(3) Minor point. We are somewhat confused about the timing of some of the example neurons shown in figure S1. For example, many neurons show visually evoked signals only after stimulus offset, unlike tactile evoked signals (e.g. Fig S1b and f). In addition, the reaction time for visual stimulus is systematically slower than tactile stimuli for many example neurons (e.g. Fig S1b) but somehow not other neurons (e.g. Fig S1g). Are these observations correct?

These observations are all correct. We have a manuscript from a separate study using this same behavioral task (Finkel et al., accepted in principle) that examines and compares (1) the onsets of tactile- and visually-evoked activity and (2) the reaction times to tactile and visual stimuli. The reaction times to tactile stimuli were slightly but significantly shorter than the reaction times to visual stimuli (tactile vs visual, 397 ± 145 vs 521 ± 163 ms, median ± interquartile range [IQR], Tukey HSD test, p = 0.001, n =155 sessions). We examined how well activity of individual neurons in S1 could be used to discriminate the presence of the stimulus or the response of the mouse. For discriminability for the presence of the stimulus, S1 neurons could signal the presence of the tactile stimulus but not the visual stimulus. For discriminability for the response of the mouse, the onsets for significant discriminability occurred earlier for tactile compared with visual trials (two-sided Kolmogorov-Smirnov test, p = 1x10-16, n = 865 neurons with DP onset in tactile trials, n = 719 neurons with DP onset in visual trials).

eLife assessment

This important work advances our understanding of how brains flexibly gate actions in different contexts, a topic of great interest to the broader field of systems neuroscience. Recording neural activity from several sensory and motor cortical areas along a sensorimotor pathway, the authors found that preparatory activity in motor cortical areas of the mouse depends on the context in which an action will be carried out, consistent with previous theoretical and experimental work. Furthermore, the authors provide causal evidence that these changes support flexible gating of actions. The carefully carried out experiments were analyzed using state-of-the-art methodology and provide convincing conclusions.

Reviewer #1 (Public Review):

Summary:

Using a cross-modal sensory selection task in head-fixed mice, the authors attempted to characterize how different rules reconfigured representations of sensory stimuli and behavioral reports in sensory (S1, S2) and premotor cortical areas (medial motor cortex or MM, and ALM). They used silicon probe recordings during behavior, a combination of single-cell and population-level analyses of neural data, and optogenetic inhibition during the task.

Strengths:

A major strength of the manuscript was the clarity of the writing and motivation for experiments and analyses. The behavioral paradigm is somewhat simple but well-designed and well-controlled. The neural analyses were sophisticated, clearly presented, and generally supported the authors' interpretations. The statistics are clearly reported and easy to interpret. In general, my view is that the authors achieved their aims. They found that different rules affected preparatory activity in premotor areas, but not sensory areas, consistent with dynamical systems perspectives in the field that hold that initial conditions are important for determining trial-based dynamics.

I think this is a well-performed, well-written and interesting study that shows differences in rule representations in sensory and premotor areas, and finds that rules reconfigure preparatory activity in motor cortex to support flexible behavior.

Reviewer #2 (Public Review):

Summary:

Chang et al. investigated neuronal activity firing patterns across various cortical regions in an interesting context-dependent tactile vs visual detection task, developed previously by the authors (Chevee et al., 2021; doi: 10.1016/j.neuron.2021.11.013). The authors report the important involvement of a medial frontal cortical region (MM, probably a similar location to wM2 as described in Esmaeili et al., 2021 & 2022; doi: 10.1016/j.neuron.2021.05.005; doi: 10.1371/journal.pbio.3001667) in mice for determining task rules.

Strengths:

The experiments appear to have been well carried out and the data well analysed. The manuscript clearly describes the motivation for the analyses and reaches clear and well-justified conclusions. I find the manuscript interesting and exciting!

Weaknesses:

I did not find any major weaknesses.

Reviewer #3 (Public Review):

Summary:

This study examines context-dependent stimulus selection by recording neural activity from several sensory and motor cortical areas along a sensorimotor pathway, including S1, S2, MM, and ALM. Mice are trained to either withhold licking or perform directional licking in response to visual or tactile stimulus. Depending on the task rule, the mice must respond to one stimulus modality while ignoring the other. Neural activity to the same tactile stimulus is modulated by task in all the areas recorded, with significant activity changes in a subset of neurons and population activity occupying distinct activity subspaces. Recordings further reveal a contextual signal in the pre-stimulus baseline activity that differentiates task context. This signal is correlated with subsequent task modulation of neural activity. Comparison across brain areas shows that this contextual signal is stronger in frontal cortical regions than sensory regions. Analyses link this signal to behavior by showing that it tracks the behavioral performance switch during task rule transitions. Silencing activity in frontal cortical regions during the baseline period impairs behavioral performance.

Strengths:

This is a carefully done study with solid results and thorough controls. The authors identify a contextual signal in baseline neural activity that predicts rule-dependent decision-related activity. The comprehensive characterization across a sensorimotor pathway is another strength. Analyses and perturbation experiments link this contextual signal to animals' behavior. The results provide a neural substrate that will surely inspire follow-up mechanistic investigations.

Weaknesses:

None. The authors have further improved the manuscript during the revision with additional analyses.

Impact:

This study reports an important neural signature for context-dependent decision-making that has important implications for mechanisms of context-dependent neural computation in general.

eLife assessment

This fundamental study provides insights into the interplay of endogenous orienting and the planning of goal-directed gaze shifts (saccades). Using an elegant experimental protocol and detailed analyses of the time course of saccadic choices, the authors provide compelling evidence for independent mechanisms that guide early, reflexive eye movements and later, voluntary gaze shifts. This work will be of interest to neuroscientists and psychologists working on vision and motor control and to those researching decision-making across disciplines.

Reviewer #1 (Public Review):

Summary:

The classical pro/antisaccade task has become a valuable diagnostic tool in neurology and psychiatry (Antoniades et al., 2013, Vision Res). Although it is well-established that antisaccades require substantially longer latencies than prosaccades, the exact attentional mechanisms underlying these differences are not yet fully elucidated. This study investigates the separate influences of exogenous and endogenous attention on saccade generation. These two mechanisms are often confounded in classical pro/antisaccade tasks. In the current study, the authors build on their previous work using an urgent choice task (Salinas et al., 2019, eLife) to time-resolve the influences of exogenous and endogenous factors on saccade execution. The key contribution of the current study is to show that, when controlling for exogenous capture, antisaccades continue to require longer processing times. This longer processing time may be explained by a coupling between endogenous attention and saccade motor plans.

Strengths:

In the classical pro/antisaccade task the direction of exogenous capture (caused by the presentation of the cue) is typically congruent with the direction of prosaccades and incongruent with antisaccades. A key strength of the current study is the introduction of different experimental conditions that control for the effects of exogenous capture on saccade generation. In particular, Experiments 3 and 4 provide strong evidence for two independent (exogenous and endogenous) mechanisms that guide saccadic choices, acting at different times. Differences in timing for pro and antisaccades during the endogenous phase were consistent and independent of whether the exogenous capture biased early saccades toward the correct prosaccade direction or toward the correct antisaccade directions.

As in previous studies by the same group (Salinas et al., 2019, eLife; Goldstein et al., 2023, eLife), the detailed analysis of the time course of goal-directed saccades allowed the authors to determine the exact, additional time of 30 ms that is necessary to generate a correct antisaccade versus prosaccade.

Overall, the manuscript is very well written, and the data are presented clearly.

Weaknesses:

The main research question could be defined more clearly. In the abstract and at some points throughout the manuscript, the authors indicate that the main purpose of the study was to assess whether the allocation of endogenous attention requires saccade planning [e.g., ll.3-5 or ll.247-248]. While the data show a coupling between endogenous attention and saccades, they do not point to a specific direction of this coupling (i.e., whether endogenous attention is necessary to successfully execute a saccade plan or whether a saccade plan necessarily accompanies endogenous attention).

Some of the analyses were performed only on subgroups of the participants. The reporting of these subgroup analyses is transparent and data from all participants are reported in the supplementary figures. Still, these subgroup analyses may make the data appear more consistent, compared to when data is considered across all participants. For instance, the exogenous capture in Experiments 1 and 2 appears much weaker in Figure 2 (subgroup) than Figure S3 (all participants). Moreover, because different subgroups were used for different analyses, it is often difficult to follow and evaluate the results. For instance, the tachometric curves in Figure 2 (see also Figure 3 and 4) show no motor bias towards the cue (i.e., performance was at ~50% for rPTs <75 ms). I assume that the subsequent analyses of the motor bias were based on a very different subgroup. In fact, based on Figure S2, it seems that the motor bias was predominantly seen in the unreliable participants. Therefore, I often found the figures that were based on data across all participants (Figures 7 and S3) more informative to evaluate the overall pattern of results.

Reviewer #2 (Public Review):

Goldstein et al. provide a thorough characterization of the interaction of attention and eye movement planning. These processes have been thought to be intertwined since at least the development of the Premotor Theory of Attention in 1987, and their relationship has been a continual source of debate and research for decades. Here, Goldstein et al. capitalize on their novel urgent saccade task to dissociate the effects of endogenous and exogenous attention on saccades towards and away from the cue. They find that attention and eye movements are, to some extent, linked to one another but that this link is transient and depends on the nature of the task. A primary strength of the work is that the researchers are able to carefully measure the timecourse of the interaction between attention and eye movements in various well-controlled experimental conditions. As a result, the behavioral interplay of two forms of attention (endogenous and exogenous) is illustrated at the level of tens of milliseconds as they interact with the planning and execution of saccades towards and away from the cued location. Overall, the results allow the authors to make meaningful claims about the time course of visual behavior, attention, and the potential neural mechanisms at a timescale relevant to everyday human behavior.

Reviewer #3 (Public Review):

Summary and overall evaluation:

Human vision is inherently limited so that only a small part of a visual scene can be perceived at a given moment. To address this limitation, the visual system has evolved a number of strategies and mechanisms that work in concert. First, humans move their eyes using saccadic eye movements. This allows us to place the high-resolution region in the center of the eye's retina (the fovea centralis) on objects of interest so that these are sampled with high acuity. Second, salient, conspicuous stimuli that appear abruptly and/or differ strongly from the other stimuli in the scene, seem to automatically attract ("exogenous") attention, so that a large share of the neuronal "resources" for visual processing is devoted to the stimuli, which improves the perception of the stimuli. Third, stimuli that are important for the current task and the current behavioral goals can be prioritized by attention mechanisms ("endogenous" attention), which also secures their allocated share of processing resources and helps them be perceived. It is well-established that eye movements are closely linked to the mechanisms of attention (for a review, see Carrasco, 2011, cited in the manuscript). However, it is still unclear what role voluntary, endogenous attention plays in the control of saccadic eye movements.

The present study used an experimental procedure involving time-pressure for responding, in order to uncover how the control of saccades by exogenous and endogenous attention unfolds over time. The findings of the study indicate that saccade planning was indeed influenced by the locus of endogenous attention, but that this influence was short-lasting and could be overcome quickly. Taken together, the present findings reveal new dynamics between endogenous attention and eye movement control, and lead the way for studying them using experiments under time pressure.

The results provided by the present study advance our understanding of vision, eye movements, and their control by brain mechanisms for attention. In addition, they demonstrate how tasks involving time pressure can be used to study the dynamics of cognitive processes. Therefore, the present study seems highly important not only for vision science, but also for psychology, (cognitive) neuroscience, and related research fields more generally.

Strengths:

The experiments of the study are performed with great care and rigor and the data is analyzed thoroughly and comprehensively. Overall, the results support the authors' conclusions, so I have only minor comments (see below). Taken together, the findings seem important for a wide community of researchers in vision science, psychology, and neuroscience.

Weaknesses (minor points):

(1) In this experimental paradigm, participants must decide where to saccade based on the color of the cue in the visual periphery (they should have made a prosaccade toward a green cue and an antisaccade away from a magenta cue). Thus, irrespective of whether the cue signaled that a prosaccade or an antisaccade was to be made, the identity of the cue was always essential for the task (as the authors explain on p. 5, lines 129-138). Also, the location where the cue appeared was blocked, and thus known to the participants in advance, so that endogenous attention could be directed to the cue at the beginning of a trial (e.g., p. 5, lines 129-132). These aspects of the experimental paradigm differ from the classic prosaccade/antisaccade paradigm (e.g. Antoniades et al., 2013, Vision Research). In the classic paradigm, the identity of the cues does not have to be distinguished to solve the task, since there is only one stimulus that should be looked at (prosaccade) or away from (antisaccade), and whether a prosaccade or antisaccade was required is constant across a block of trials. Thus, in contrast to the present paradigm, in the classic paradigm, the participants do not know where the cue is about to appear, but they know whether to perform a prosaccade or an antisaccade based on the location of the cue.

The present paradigm keeps the location of the cue constant in a block of trials by intention, because this ensures that endogenous attention is allocated to its location and is not overpowered by the exogenous capture of attention that would happen when a single stimulus appeared abruptly in the visual field. Thus, the reason for keeping the location of the cue constant seems convincing. However, I wondered what consequences the constant location would have for the task representations that persist across the task and govern how attention is allocated. In the classic paradigm, there is always a single stimulus that captures attention exogenously (as it appears abruptly). In a prosaccade block, participants can prioritize the visual transient caused by the stimulus, and follow it with a saccade to its coordinates. In an antisaccade block, following the transient with a saccade would always be wrong, so that participants could try to suppress the attention capture by the transient, and base their saccade on the coordinates of the opposite location. Thus, in prosaccade and antisaccade blocks, the task representations controlling how visual transients are processed to perform the task differ. In the present task, prosaccades and antisaccades cannot be distinguished by the visual transients. Thus, such a situation could favor endogenous attention and increase its influence on saccade planning, even though saccade planning under more naturalistic conditions would be dominated by visual transients. I suggest discussing how this (and vice versa the emphasis on visual transients in the classic paradigm) could affect the generality of the presented findings (e.g., how does this relate to the interpretation that saccade plans are obligatorily coupled to endogenous attention? See, Results, p. 10, lines 306-308, see also Deubel & Schneider, 1996, Vision Research).

(2) Discussion (p. 16, lines 472-475): The authors suppose that "It is as if the exogenous response was automatically followed by a motor bias in the opposite direction. Perhaps the oculomotor circuitry is such that an exogenous signal can rapidly trigger a saccade, but if it does not, then the corresponding motor plan is rapidly suppressed regardless of anything else.". I think this interesting point should be discussed in more detail. Could it also be that instead of suppression, other currently active motor plans were enhanced? Would this involve attention? Some attention models assume that attention works by distributing available (neuronal) processing resources (e.g., Desimone & Duncan, 1995, Annual Review of Neuroscience; Bundesen, 1990, Psychological Review; Bundesen et al., 2005, Psychological Review) so that the information receiving the largest share of resources results in perception and is used for action, but this happens without the active suppression of information.

(3) Methods, p. 19, lines 593-596: It is reported that saccades were scored based on their direction. I think more information should be provided to understand which eye movements entered the analysis. Was there a criterion for saccade amplitude? I think it would be very helpful to provide data on the distributions of saccade amplitudes or on their accuracy (e.g. average distance from target) or reliability (e.g. standard deviation of landing points). Also, it is reported that some data was excluded from the analysis, and I suggest reporting how much of the data was excluded. Was the exclusion of the data related to whether participants were "reliable" or "unreliable" performers?

(4) Results, p. 9, lines 262-266: Some data analyses are performed on a subset of participants that met certain performance criteria. The reasons for this data selection seem convincing (e.g. to ensure empirical curves were not flat, line 264). Nevertheless, I suggest to explain and justify this step in more detail. In addition, if not all participants achieved an acceptable performance and data quality, this could also speak to the experimental task and its difficulty. Thus, I suggest discussing the potential implications of this, in particular, how this could affect the studied mechanisms, and whether it could limit the presented findings to a special group within the studied population.

Author response:

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

In response to Reviewer #2, we agree with the reviewer that it needs to be noted that not all forms of recognition are the same and have added the following: "However, we note that not all forms of recognition are the same; researchers may prefer to have their work featured instead of personal stories or critiques of the scientific environment."

---

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

We thank both reviewers for their detailed comments and insightful suggestions. Below we summarize our responses to each concern in addition to the edits within the manuscript.

We would also like to add a clarification to the eLife assessment, it states “This important bibliometric analysis shows that authors of scientific papers whose names suggest they are female or East Asian get quoted less often in news stories about their work.” We show that individuals with names predicted to be from women or East Asian name origins are less likely to be quoted or mentioned in Nature’s scientific news stories than expected by publication demographics. In this study, we did not compare the level of coverage of a scientific article by the demographics of the authors of the article.

Reviewer #1

The article is not so clearly structured, which makes it hard to follow. A better framing, contextualization, and conceptualization of their analysis would help the readers to better understand the results. There are some unclear definitions and wrong wording of key concepts.

We have adapted our wording in the text and added a more detailed discussion which hopefully makes the paper easier to comprehend. These changes are described in the context of your reviewer's suggestions and addressed in the next section.

Language use: Male/Female refers to sex, not to gender.

We have now updated the language throughout the text. Thank you for pointing this out.

Regional disparities are not the same as names' origin. While the first might relate to the academic origin of authors, inferred from their institutional belonging, the latter reflects the authors' inferred identity. Ethnic identities and the construction of prejudice against specific populations need proper contextualization.

We have added better contextualization in the manuscript and reworded the section in our results and discussion to clarify that we are analyzing disparities related to perceived ethnicity and not regions. We also added the following text to the results section “In our analysis, we use name origin as an estimate for the perceived ethnicity of a primary source by a journalist. Our prediction is not intended to assign ethnicity to an individual, but to be used broadly as a tool to quantify representational differences in a journalist's sociologically constructed perception of a primary source's ethnicity.” We also added the following text to our Discussion: “Our use of name origins is a proxy for a journalist's or referring scholarly peer’s potential perceptions of the ethnicity of a primary source as signaled by an individual's name. We do not intend to assign an identity to an individual, but to generate a broad metric to measure possible bias for particular ethnicities during journalists' primary source gathering.”

It would be helpful to have a clear definition of what are quotes, mentions, and citations. For me, it was not so clear and made understanding the results more difficult.

We added the following text to the results section Extracted Data Used for Analysis: “Quoted names are any names that were attached to a quote within the article. Mentioned names are any names that were stated within the article. Cited names are all author names of a scientific paper that was cited in the news article.”

The comparison against Nature published research articles is not perfect because journalists will also cover articles not published in Nature. If for example, the gender representation in the quoted articles is not the same between Nature journals and other journals, then this source of inequality would be missing (e.g. if the journalists are biased against women, but not as much when they published in Nature, because they are also biased towards Nature articles). Also, the gender representation among Nature authors could not be the same as in general. Nevertheless, this seems to be a fair benchmark, especially if the authors did not have access to other more comprehensive databases. But a statement of limitations including these potential issues would be good to have.

To add better context to the generalizability of our work, we added the following text to our discussion: “Furthermore, the news articles present on "www.nature.com" are intended for a very specific readership that may not be reflective of more broad scientific news outlets. In a separate analysis, we took a cursory look into a comparison with The Guardian and found similar disparities in gender and name origin. However, it is not clear which publications should be used as a comparator for science-related articles in The Guardian, and difficult to compare relative rates of representation. While other science news outlets may not have a direct comparator, it would be useful to take a broad comparison across multiple science news outlets to compare against one another. Our existing pipeline could be easily applied to other science news outlets and identify if there exists a consistent pattern of disparity regardless of the intended readership.”

"we select the highest probability origin for each name as the resultant assignment". Threshold based approaches for race/ethnicity name-based inference have been criticized by the literature as they might reproduce biases (see Kozlowski, D., Murray, D. S., Bell, A., Hulsey, W., Larivière, V., Monroe-White, T., & Sugimoto, C. R. (2022). Avoiding bias when inferring race using name-based approaches. Plos one, 17(3), e0264270.). The authors could use the full distribution of probabilities over names instead of selecting one. The formulae proposed (3-5) could be easily adapted to this change.

We thank the author for pointing this out. We have updated our analysis to use the probabilities instead of hard assignments. Figure 3 and formulae 3-5 have been updated. While we observe a slight shift in the calculated values, the overall trends are unchanged.

Is it possible to make an analysis that intersects both name origin and gender? I am not sure if the sample size would allow for this, but if some other dimensions were collapsed, it would be very important to show what happens at the intersection of these two dimensions of discrimination.

We agree that identifying any differences in quotation patterns at the intersection of gender and name origin would be very useful to identify. To address this, we added supplemental table 5. This table identifies the number of quotes per predicted name origin and gender over all years and article types. In this table, we don’t see a significant difference in gender distribution across predicted name origins.

Given a larger sample size, we would be able to better identify more subtle differences, but at this sample size, we cannot make more detailed inferences. Additionally, this also addresses a QC-issue, where predicted gender accuracy varies by name origin, specifically East Asian name origin. From our data, we don’t see a large difference in proportions across any name origin. We added the following text to the results section to incorporate this analysis:

“However, it should be noted that the error rate varies by name origin with the largest decrease in performance on names with an Asian origin [@doi:10.7717/peerj-cs.156;@doi:10.5195/jmla.2021.1252]

. In our analysis, we did not observe a large difference in names predicted to come from a man or woman between predicted East Asian and other name origins (Table 5). “

The use of vocabulary should be more homogeneous. For example, in page 13 the authors start to use the concepts of over/under enrichment, which appeared before in a title but was not used.

The text has been updated to remove all mentions of “over/under enrichment” with “over/under representation”

In the discussions section, it would be important to see as a statement of limitations the problems that automatic origin and gender inference have.

We thank the reviewer for this suggestion. We have added the following paragraph to our discussion.

Computational tools enabled us to automatically analyze thousands of articles to identify existing disparities by gender and name origin, but these tools are not without limitations. Our tools are unable to identify non-binary people and rely on gender predictors that are known to have region-specific biases, with the largest decrease in performance on names of an Asian origin [@doi:10.7717/peerj-cs.156;@doi:10.5195/jmla.2021.1252]. Furthermore, name origin is only a proxy for externally perceived racial or ethnic origins of a source or author and is not as accurate as self-identified race or ethnicity. Self-identification better captures the lived experience of an individual that computational estimates from a name can not capture. This is highlighted in our inability to distinguish between Black and White people from the US by their names. As the collection of demographic data by publication outlets grows, we believe this will enable a more fine-grained and accurate analysis of disparities in scientific journalism.

Figures 2a and 3a show that the affiliations of authors and their countries was going to be used in this analysis. Yet, this section is not present in the article. I would encourage the authors to add this to the analysis as it would show important patterns, and to intersect the dimensions of gender, name origin and country.

We were interested in using this analysis in our work, but unfortunately the sample size of cited works in each country was too small to make inferences. If this work was extended to larger scientific outlets to include larger corpora such as The Guardian or New York Times, we think one could be able to make more robust inferences. Since our work only focuses on Nature, we decided not to include this analysis. However, we do include a section in our discussion for future work.

“As a proxy for measuring possible geographical bias of a journalist, we attempted to identify if there was any geographical bias of cited authors. To do this, we identified the affiliation of each cited author and identified their affiliated country. Unfortunately, we could not robustly extract a large enough number of cited authors from different countries to make any conclusive statements. Expanding our work to other science journalism outlets could help identify possible ways in which geographic region, genders, and perceived ethnicity interact and affect scientific visibility of specific groups. While we are unable to identify that journalists have a specific geographical bias, having reporters explicitly focused on specific regional sources will broaden coverage of international opinions in science.”

It is not clear at that point what column dependence means.

The abstract has been updated to state, “Gender disparity in Nature quotes was dependent on the article type.”

Reviewer #2

We thank the reviewer for their very detailed and insightful suggestions regarding our analysis and the key caveats that needed better contextualization in our analysis. We went through each major point the reviewer brought up below and included any additional text that was needed.

In some cases, the manuscript lacks consistency in terminology, and uses word choice that is strange (e.g., "enrichment" and "depletion" when discussion representation).

We thank the review for pointing this out, we have removed all instances of depletion/enrichment for over/under-representation

Caveats to Claim 1. So while Claim 1 holds, it does not hold for all comparator sets and for all years. I don't think this is critical of the paper-the authors do discuss the trend in Claim 2-but interpretation of this claim should take care of these caveats, and readers should consider the important differences in first and last authorship.

We thank the reviewer for their detailed feedback on this section. We have added the missing contextualization of our results. In the results section, I changed the figure caption to: “Speakers predicted to be men are sometimes overrepresented in quotes, but this depends on the year and article type.” Added the following paragraph “When considering the relative proportion of authors and speakers predicted to be men, we only find a slight over-representation of men. This overrepresentation is dependent on the authorship position and the year. Before 2010, quotes predicted as from men are overrepresented in comparison to both first and last authors, but between 2010 and 2017 quotes predicted from men are only overrepresented in comparison for first authors. In 2020, we find a slight over-representation of quotes predicted to be from women relative to first and last authors, but still severely under-represented when considering the general population. The choice of comparison between first and last authors can reveal different aspects of the current state of academia. While this does not hold in all scientific fields, first authors are typically early career scientists and last authors are more senior scientists. It has also been shown that early career scientists tend to be more diverse than senior scientists [@doi:10.7554/eLife.60829; @doi:10.1096/fj.201800639]. Since we find that quotes are only slightly more likely to come from a last author, it is reasonable to compare the relative rate of predicted quotes from men to either authorship position. Comparison with last authorships may reveal more how gender bias currently exists whereas comparison with early career scientists may reveal bias in comparison to a future, more possibly diverse academic environment. We hope that increased representation and recognition of women in science, even beyond what is observed in authorship, can increase the proportion of women first and last authors such that it better reflects the general population.”

Generalizability to other contexts of science journalism:

We thank the reviewer for their feedback on the generalizability of our work. We have now added the following text to our discussion to provide the reader with a better context of our results: “To articles presented on "www.nature.com" are intended for a very specific readership that may not be reflective of more broad scientific news outlets. In a separate analysis, we took a cursory look into a comparison with The Guardian and found very similar disparities in gender and name origin. However, it is not clear which publications should be used as a comparator for science-related articles in The

Guardian, and difficult to compare relative rates of representation. While other science news outlets may not have a direct comparator, it would be useful to take a broad comparison across multiple science news outlets to compare against one another. Our existing pipeline could be easily applied to other science news outlets and identify if there exists a consistent pattern of disparity regardless of the intended readership. ”

Shallow discussion:

The authors highlight gender parity in career features, but why exactly is there gender parity in this format

We thank the reviewer for encouraging us to better contextualize our findings in the broader discourse. We have now added several sections to our Discussion. To address gender parity, we have added the following text: “This finding, coupled with the near equal number of articles written by journalists predicted to be men or women, argues for more diversity in topical coverage. "Career Feature" articles highlight current topics relevant to working scientists and frequently highlight systemic issues with the scientific environment. This column allows space for marginalized people to critique the current state of affairs in science or share their personal stories. This type of content encourages the journalist to seek out a diverse set of primary sources. Including more content that is not primarily focused on recent publications, but all topics surrounding the practice of science, can serve as an additional tool to rapidly achieve gender parity in journalistic recognition.”

Representation in quotations varies by first and last author, most certainly as a result of the academic division of labor in the life sciences. However, what does it say about the scientific quotation that it appears first authors are more often to be quoted? Does this mean that the division of labor is changing such that the first authors are the lead scientists? Or does it imply that senior authors are being skipped over, or giving away their chance to comment on a study to the first author?

We thank the reviewer for asking bringing up these important questions. We have added better context to our first author analysis in our discussion. We have included the following two sections to address this. Also, we want to state that we find last authors to be slightly more quoted than first authors, as depicted in Fig. 2d., with first author quotation percentage largely appearing below the red line. We included this text in a response above and include it again here for convenience.

“Before 2010, quotes predicted as from men are overrepresented in comparison to both first and last authors, but between 2010 and 2017 quotes predicted from men are only overrepresented in comparison for first authors. In 2020, we find a slight over-representation of quotes predicted to be from women relative to first and last authors, but still severely under-represented when considering the general population. The choice of comparison between first and last authors can reveal different aspects of the current state of academia. While this does not hold in all scientific fields, first authors are typically early career scientists and last authors are more senior scientists. It has also been shown that early career scientists tend to be more diverse than senior scientists [@doi:10.7554/eLife.60829; @doi:10.1096/fj.201800639]. Since we find that quotes are only slightly more likely to come from a last author, it is reasonable to compare the relative rate of predicted quotes from men to either authorship position. Comparison with last authorships may reveal more how gender bias currently exists whereas comparison with early career scientists may reveal bias in comparison to a future, more possibly diverse academic environment. We hope that increased representation and recognition of women in science, even beyond what is observed in authorship, can increase the proportion of women first and last authors such that it better reflects the general population.”

“In our analysis, we also find that there are more first authors with predicted East Asian name origin than last authors. This is in contrast to predicted Celtic/English and European name origins.

Furthermore, we see that the amount of first author people with predicted East Asian name origins is increasing at a much faster rate than quotes are increasing. If this mismatched rate of representation continues, this could lead to an increasingly large erasure of early career scientists with East Asian name origins. As noted before, focusing on increasing engagement with early career scientists can help to reduce the growing disparity of public visibility of scientists with East Asian name origins.”

What might be the downstream impacts on the public stemming from the under-representation of scientists with East Asian names? According to Figure 3d, not only are East Asian names under-represented in quotations, but they are becoming more under-represented over time as they appear as authors in a greater number of Nature publications; Those with European names are proportionately represented in quotations given their share of authors in Nature. Why might this be, especially seeing as Anglo names are heavily over-represented?

To address this point, we have added the following text to our discussion: “In our analysis, we also find that there are more first authors with predicted East Asian name origin than last authors. This is in contrast to predicted Celtic/English and European name origins. Furthermore, the amount of first author people with predicted East Asian name origins is increasing at a much faster rate than quotes are increasing. If this mismatched rate of representation continues, this could lead to an increasingly large erasure of early career scientists with East Asian name origins. As noted before, focusing on increasing engagement with early career scientists can help to reduce the growing disparity of public visibility of scientists with East Asian name origins.”

I am very confused by Figure 1B. It mixes the counts of News-related items with (non-Springer) research articles in a single stacked bar plot which makes determining the quantity of either difficult. I would advise splitting them out

Figure 1B has been updated, and the News and Research articles have been separated.

When querying the first 2000 or so results from the SpringerNature API, are the authors certain that they are getting a random sample of papers?

These papers were the first 200 English language "Journal" papers returned by the Springer Nature API for each month, resulting in 2400 papers per year from 2005 through 2020. These papers are the first 200 papers published each month by a Springer Nature journal, which may not be completely random, but we believe to be a reasonably representative sample. Furthermore, the Springer Nature comparator set is being used as an additional comparator to the complete set of all Nature research papers used in our analyses.

In all figures: the authors use capital letters to indicate panels in the caption, but lowercase letters in the figure itself and in the main text. This should be made consistent.

This has been updated.

In all figures: the authors should make the caption letter bold in the figure captions, which makes it much easier to find descriptions of specific panels

This has been updated.

In the section "coreNLP": the authors mention "co-reference resolution" but without really remarking why it is being used. This is an issue throughout the methods-the authors describe what method they are using but either they don't mention why they are using that method until later, or else not at all.

We have added better reasoning behind our coreNLP selected methods: “We used the standard set of annotaters: tokenize, ssplit, pos, lemma, ner, parse, coref, and additionally the quote annotator. These perform text tokenization, sentence splitting, part of speech recognition, lemmatization, named entity recoginition, division of sentences into constituent phrases, co-reference resolution, and identification of quoted entities, respectively. We used the "statistical" algorithm to perform coreference resolution for speed. Each of these aspects is required to identify the names of quoted or mentioned speakers and identify any of their associated pronouns. All results were output to json format for further downstream processing.”

We included a better description of scrapy: “Scrapy is a tool that applies user-defined rules to follow hyperlinks on webpages and return the information contained on each webpage.

We used Scrapy to extract all web pages containing news articles and extract the text.”

We also included our motivation for bootstrapping: “We used the boostrap method to construct confidence intervals for each of our calculated statistics.”

In the section "Name Formatting for Gender Prediction in Quotes or Mentions", genderizeR is mentioned before an introduction to the tool

We added the following text to provide context: “Even though genderizeR, the computational method used to predict the name's gender, only uses the first name to make the gender prediction, identifying the full name gives us greater confidence that we correctly identified the first name. “

In the section "Name Formatting for Gender Prediction of Authors", you state that you exclude papers with only one author. How many papers is this? I assume few, in Nature, but if not I can imagine gender differences based on who writes first-authored papers.

We find that the number excluded is roughly 7% of all papers, which is consistent across Nature and Springer Nature (1113/15013 for cited springer articles, 2899/42155 for random springer articles, 955/12459 for nature authors). We have added the following text to the manuscript for better context: “Roughly 7% of all papers were estimated to be by a single author and removed from this analysis.: 1113/15013 for cited Springer articles, 2899/42155 for random Springer articles, 955/12459 for Nature research articles.”

In "Name Origin Analysis", for the in-text reference to Equation 3: include the prefix "Eq." or similar to mark this as referencing the equation and not something else

This has been updated.

The use of the word "enrichment" in reference to the representation of East Asian authors is strange and does not fit the colloquial definition of the term. I suggest just using a simpler term like "representation" instead

Similarly, the authors use the word "depletion" to reflect the lower rate of quotes to scientists with East-Asian names, but I feel a simpler word would be more appropriate.

We thank the reviewer for this suggestion, all instances of “enrichment/depletion” have been replaced with “over/under representation”

The authors claim in Figure 2d that there is a steady increase in the rate of first author citations, however, this graph is not convincing. It appears to show much more noise than anything resembling a steady change.

We have reworded our figure description to state that there is a consistent bias towards quoting last authors. Our figure description now states: “Panel d shows a consistent but slight bias towards quoting the last author of a cited article than the first author over time.”

Supplemental Figures 1b and 1c do not seem to be mentioned in the main text, and I struggle to see their relevance.

We thank the reviewer for identifying this error; these subpanels have been removed.

Reviewer #2 (Public Review):

This manuscript illustrates the power of "combined" research, incorporating a range of tools, both old and new to answer a question. This thorough approach identifies a novel target in a well-established signalling pathway and characterises a new player in Drosophila CNS development.

Largely, the experiments are carried out with precision, meeting the aims of the project, and setting new targets for future research in the field. It was particularly refreshing to see the use of multi-omics data integration and Targeted DamID (TaDa) findings to triage scRNA-seq data. Some of the TaDa methodology was unorthodox, however, this does not affect the main finding of the study. The authors (in the revised manuscript) have appropriately justified their TaDa approaches and mentioned the caveats in the main text.

Their discovery of Spar as a neuropeptide precursor downstream of Alk is novel, as well as its ability to regulate activity and circadian clock function in the fly. Spar was just one of the downstream factors identified from this study, therefore, the potential impact goes beyond this one Alk downstream effector.

Reviewer #3 (Public Review):

Summary:

The receptor tyrosine kinase Anaplastic Lymphoma Kinase (ALK) in humans is nervous system expressed and plays an important role as an oncogene. A number of groups have been studying ALK signalling in flies to gain mechanistic insight into its various roles. In flies, ALK plays a critical role in development, particularly embryonic development and axon targeting. In addition, ALK was also shown to regulate adult functions including sleep and memory. In this manuscript, Sukumar et al., used a suite of molecular techniques to identify downstream targets of ALK signalling. They first used targeted DamID, a technique that involves a DNA methylase to RNA polymerase II, so that GATC sites in close proximity to PolII binding sites are marked. They performed these experiments in wild type and ALK loss of function mutants (using an Alk dominant negative ALkDN), to identify Alk responsive loci. Comparing these loci with a larval single cell RNAseq dataset identified neuroendocrine cells as an important site of Alk action. They further combined these TaDa hits with data from RNA seq in Alk Loss and Gain of Function manipulations to identify a single novel target of Alk signalling - a neuropeptide precursor they named Sparkly (Spar) for its expression pattern. They generated a mutant allele of Spar, raised an antibody against Spar, and characterised its expression pattern and mutant behavioural phenotypes including defects in sleep and circadian function.

Strengths:

The molecular biology experiments using TaDa and RNAseq were elegant and very convincing. The authors identified a novel gene they named Spar. They also generated a mutant allele of Spar (using CrisprCas technology) and raised an antibody against Spar. These experiments are lovely, and the reagents will be useful to the community. The paper is also well written, and the figures are very nicely laid out making the manuscript a pleasure to read.

Weaknesses:

The manuscript has improved very substantially in revision. The authors have clearly taken the comments on board in good faith.

Editors' note: The authors have satisfactorily addressed the concerns raised in the previous rounds of review. These were related to the unconventional analysis of the TaDa data, the addition of other means of down regulated gene function, and the nature of analyses of behavioural data.

Author response:

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

Point-by-point response to concerns raised by reviewer #3:

The manuscript has improved very substantially in revision. The authors have clearly taken the comments on board in good faith. Yet, some small concerns remain around the behavioural analysis.

In Fig. 8H and H' average sleep/day is ~100. Is this minutes of sleep? 100 min/day is far too low, is it a typo?

The numbers for sleep bouts are also too low to me e.g. in Fig 9 number of sleep bouts avg around 4.

In their response to reviewers the authors say these errors were fixed, yet the figures appear not to have been changed. Perhaps the old figures were left in inadvertently?

Indeed this correction was somehow missed and we thank the reviewer for noticing this. We have now corrected Fig 8H-H’ and Fig 9D.  

The circadian anticipatory activity analyses could also be improved. The standard in the field is to perform eduction analyses and quantify anticipatory activity e.g. using the method of Harrisingh et al. (PMID: 18003827). This typically computed as the ratio of activity in the 3hrs preceding light transition to activity in the 6hrs preceding light transition.

In their response to reviewers, the authors have revised their anticipation analyses by quantifying the mean activity in the 6 hrs preceding light transition. However, in the method of Harrisingh et al., anticipation is the ratio of activity in the 3hrs preceding light transition to activity in the 6hrs preceding light transition. Simply computing the activity in the 6hrs preceding light transition does not give a measure of anticipation, determining the ratio is key.

We acknowledge the importance of obtaining accurate results in our analysis, therefore we have re-evaluated the anticipation activity by measuring the ratio of the mean activity in the 3h preceding light transition over the activity in the 6h preceding light transition. We have reported the data as percentages in Fig 8F-G and modified the figure legends accordingly.

Reviewer #1 (Public Review):

Olszyński and colleagues present data showing variability from canonical "aversive calls", typically described as long 22 kHz calls rodents emit in aversive situations. Similarly long but higher-frequency (44 kHz) calls are presented as a distinct call type, including analyses both of their acoustic properties and animals' responses to hearing playback of these calls. While this work adds an intriguing and important reminder, namely that animal behavior is often more variable and complex than perhaps we would like it to be, there is some caution warranted in the interpretation of these data.

The exclusive use of males is a major concern lacking adequate justification and should be disclosed in the title and abstract to ensure readers are aware of this limitation. With several reported sex differences in rat vocal behaviors this means caution should be exercised when generalizing from these findings. The occurrence of an estrus cycle in typical female rats is not justification for their exclusion. Note also that male rodents experience great variability in hormonal states as well, distinguishing between individuals and within individuals across time. The study of endocrinological influences on behavior can be separated from the study of said behavior itself, across all sexes. Similarly, concerns about needing to increase the number of animals when including all sexes are usually unwarranted (see Shansky [2019] and Phillips et al. [2023]).

Regarding the analysis where calls were sorted using DBSCAN based on peak frequency and duration, my comment on the originally reviewed version stands. It seems that the calls are sorted by an (unbiased) algorithm into categories based on their frequency and duration, and because 44kHz calls differ by definition on frequency and duration the fact that the algorithm sorts them as a distinct category is not evidence that they are "new calls [that] form a separate, distinct group". I appreciate that the authors have softened their language regarding the novelty and distinctness of these calls, but the manuscript contains several instances where claims of novelty and specificity (e.g. the subtitle on line 193) is emphasized beyond what the data justifies.

The behavioral response to call playback is intriguing, although again more in line with the hypothesis that these are not a distinct type of call but merely represent expected variation in vocalization parameters. Across the board animals respond rather similarly to hearing 22 kHz calls as they do to hearing 44 kHz calls, with occasional shifts of 44 kHz call responses to an intermediate between appetitive and aversive calls. This does raise interesting questions about how, ethologically, animals may interpret such variation and integrate this interpretation in their responses. However, the categorical approach employed here does not address these questions fully.

I appreciate the amendment in discussing the idea of arousal being the key determinant for the increased emission of 44kHz, and the addition of other factors. Some of the items in this list, such as annoyance/anger and disgust/boredom, don't really seem to fit the data. I'm not sure I find the idea that rats become annoyed or disgusted during fear conditioning to be a particularly compelling argument. As such the list appears to be a collection of emotion-related words, with unclear potential associations with the 44kHz calls.

Later in the Discussion the authors argue that the 44kHz aversive calls signal an increased intensity of a negative valence emotional state. It is not clear how the presented arguments actually support this. For example, what does the elongation of fear conditioning to 10 trials have to do with increased negative emotionality? Is there data supporting this relationship between duration and emotion, outside anthropomorphism? Each of the 6 arguments presented seems quite distant from being able to support this conclusion.

In sum, rather than describing the 44kHz long calls as a new call type, it may be more accurate to say that sometimes aversive calls can occur at frequencies above 22 kHz. Individual and situational variability in vocalization parameters seems to be expected, much more so than all members of a species strictly adhering to extremely non-variable behavioral outputs.

[Editors' note: The reviewer agrees that the additional analysis has ruled out the possibility that the calls are due to fatigue.]

Author response:

eLife assessment 

This important study provides evidence for a combination of the latest generation of Oxford Nanopore Technology long reads with state-of-the art variant callers enabling bacterial variant discovery at accuracy that matches or exceeds the current "gold standard" with short reads. The evidence supporting the claims of the authors is convincing, although the inclusion of a larger number of reference genomes would further strengthen the study. The work will be of interest to anyone performing sequencing for outbreak investigations, bacterial epidemiology, or similar studies. 

We thank the editor and reviewers for the accurate summary and positive assessment. We address the comment about increasing the number of reference genomes in the response to reviewer 2.

Public Reviews: 

Reviewer #1 (Public Review): 

Summary: 

The authors assess the accuracy of short variant calling (SNPs and indels) in bacterial genomes using Oxford Nanopore reads generated on R10.4 flow cells from a very similar genome (99.5% ANI), examining the impact of variant caller choice (three traditional variant callers: bcftools, freebayes, and longshot, and three deep learning based variant callers: clair3, deep variant, and nano caller), base calling model (fast, hac and sup) and read depth (using both simplex and duplex reads). 

Strengths: 

Given the stated goal (analysis of variant calling for reads drawn from genomes very similar to the reference), the analysis is largely complete and results are compelling. The authors make the code and data used in their analysis available for re-use using current best practices (a computational workflow and data archived in INSDC databases or Zenodo as appropriate). 

Weaknesses: 

While the medaka variant caller is now deprecated for diploid calling, it is still widely used for haploid variant calling and should at least be mentioned (even if the mention is only to explain its exclusion from the analysis). 

We agree that this would be an informative addition to the study and will add it to the benchmarking.

Appraisal: 

The experiments the authors engaged in are well structured and the results are convincing. I expect that these results will be incorporated into "best practice" bacterial variant calling workflows in the future. 

Thank you for the positive appraisal.

Reviewer #2 (Public Review): 

Summary: 

Hall et al describe the superiority of ONT sequencing and deep learning-based variant callers to deliver higher SNP and Indel accuracy compared to previous gold-standard Illumina short-read sequencing. Furthermore, they provide recommendations for read sequencing depth and computational requirements when performing variant calling. 

Strengths: 

The study describes compelling data showing ONT superiority when using deep learning-based variant callers, such as Clair3, compared to Illumina sequencing. This challenges the paradigm that Illumina sequencing is the gold standard for variant calling in bacterial genomes. The authors provide evidence that homopolymeric regions, a systematic and problematic issue with ONT data, are no longer a concern in ONT sequencing. 

Weaknesses: 

(1) The inclusion of a larger number of reference genomes would have strengthened the study to accommodate larger variability (a limitation mentioned by the authors). 

Our strategic selection of 14 genomes—spanning a variety of bacterial genera and species, diverse GC content, and both gram-negative and gram-positive species (including M. tuberculosis, which is neither)—was designed to robustly address potential variability in our results. Moreover, all our genome assemblies underwent rigorous manual inspection as the quality of the true genome sequences is the foundation this research is built upon. Given this, the fundamental conclusions regarding the accuracy of variant calls would likely remain unchanged with the addition of more genomes.  However, we do acknowledge that a substantially larger sample size, which is beyond the scope of this study, would enable more fine-grained analysis of species differences in error rates.

(2) In Figure 2, there are clearly one or two samples that perform worse than others in all combinations (are always below the box plots). No information about species-specific variant calls is provided by the authors but one would like to know if those are recurrently associated with one or two species. Species-specific recommendations could also help the scientific community to choose the best sequencing/variant calling approaches.

Thank you for highlighting this observation. The precision, recall, and F1 scores for each sample and condition can be found in Supplementary Table S4. We will investigate the samples that consistently perform below expectation to determine if this is associated with specific species, which may necessitate tailored recommendations for those species. Additionally, we will produce a species-segregated version of Figure 2 for a clearer interpretation and will place it in the supplementary materials.

(3) The authors support that a read depth of 10x is sufficient to achieve variant calls that match or exceed Illumina sequencing. However, the standard here should be the optimal discriminatory power for clinical and public health utility (namely outbreak analysis). In such scenarios, the highest discriminatory power is always desirable and as such an F1 score, Recall and Precision that is as close to 100% as possible should be maintained (which changes the minimum read sequencing depth to at least 25x, which is the inflection point).

We agree that the highest discriminatory power is always desirable for clinical or public health applications. In which case, 25x is probably a better minimum recommendation. However, we are also aware that there are resource-limited settings where parity with Illumina is sufficient. In these cases, 10x depth from ONT would provide sufficient data.

The manuscript currently emphasises the latter scenario, but we will revise the text to clearly recommend 25x depth as a conservative aim in settings where resources are not a constraint, ensuring the highest possible discriminatory power for applications like outbreak analysis.

(4) The sequencing of the samples was not performed with the same Illumina and ONT method/equipment, which could have introduced specific equipment/preparation artefacts that were not considered in the study. See for example https://academic.oup.com/nargab/article/3/1/lqab019/6193612. 

To our knowledge, there is no evidence that sequencing on different ONT machines or barcoding kits leads to a difference in read characteristics or accuracy. To ensure consistency and minimise potential variability, we used the same ONT flowcells for all samples and performed basecalling on the same Nvidia A100 GPU. We will update the methods to emphasise this.

For Illumina and ONT, the exact machines used for which samples will be added as a supplementary table. We will also add a comment about possible Illumina error rate differences in the ‘Limitations’ section of the Discussion.

In summary, while there may be specific equipment or preparation artifacts to consider, we took steps to minimise these effects and maintain consistency across our sequencing methods.

Reviewer #3 (Public Review): 

Hall et al. benchmarked different variant calling methods on Nanopore reads of bacterial samples and compared the performance of Nanopore to short reads produced with Illumina sequencing. To establish a common ground for comparison, the authors first generated a variant truth set for each sample and then projected this set to the reference sequence of the sample to obtain a mutated reference. Subsequently, Hall et al. called SNPs and small indels using commonly used deep learning and conventional variant callers and compared the precision and accuracy from reads produced with simplex and duplex Nanopore sequencing to Illumina data. The authors did not investigate large structural variation, which is a major limitation of the current manuscript. It will be very interesting to see a follow-up study covering this much more challenging type of variation. 

We fully agree that investigating structural variations (SVs) would be a very interesting and important follow-up. Identifying and generating ground truth SVs is a nontrivial task and we feel it deserves its own space and study. We hope to explore this in the future.

In their comprehensive comparison of SNPs and small indels, the authors observed superior performance of deep learning over conventional variant callers when Nanopore reads were basecalled with the most accurate (but also computationally very expensive) model, even exceeding Illumina in some cases. Not surprisingly, Nanopore underperformed compared to Illumina when basecalled with the fastest (but computationally much less demanding) method with the lowest accuracy. The authors then investigated the surprisingly higher performance of Nanopore data in some cases and identified lower recall with Illumina short read data, particularly from repetitive regions and regions with high variant density, as the driver. Combining the most accurate Nanopore basecalling method with a deep learning variant caller resulted in low error rates in homopolymer regions, similar to Illumina data. This is remarkable, as homopolymer regions are (or, were) traditionally challenging for Nanopore sequencing. 

Lastly, Hall et al. provided useful information on the required Nanopore read depth, which is surprisingly low, and the computational resources for variant calling with deep learning callers. With that, the authors established a new state-of-the-art for Nanopore-only variant, calling on bacterial sequencing data. Most likely these findings will be transferred to other organisms as well or at least provide a proof-of-concept that can be built upon. 

As the authors mention multiple times throughout the manuscript, Nanopore can provide sequencing data in nearly real-time and in remote regions, therefore opening up a ton of new possibilities, for example for infectious disease surveillance. 

However, the high-performing variant calling method as established in this study requires the computationally very expensive sup and/or duplex Nanopore basecalling, whereas the least computationally demanding method underperforms. Here, the manuscript would greatly benefit from extending the last section on computational requirements, as the authors determine the resources for the variant calling but do not cover the entire picture. This could even be misleading for less experienced researchers who want to perform bacterial sequencing at high performance but with low resources. The authors mention it in the discussion but do not make clear enough that the described computational resources are probably largely insufficient to perform the high-accuracy basecalling required. 

We have provided runtime benchmarks for basecalling in Supplementary Figure S16 and detailed these times in Supplementary Table S7. In addition, we state in the Results section (P10 L228-230) “Though we do note that if the person performing the variant calling has received the raw (pod5) ONT data, basecalling also needs to be accounted for, as depending on how much sequencing was done, this step can also be resource-intensive.”

Even with super-accuracy basecalling considered, our analysis shows that variant calling remains the most resource-intensive step for Clair3, DeepVariant, FreeBayes, and NanoCaller. Therefore, the statement “the described computational resources are probably largely insufficient to perform the high-accuracy basecalling required”, is incorrect. However, we will endeavour to make the basecalling component and considerations more prominent in the Results and Discussion.

eLife assessment

This important study provides evidence for a combination of the latest generation of Oxford Nanopore Technology long reads with state-of-the art variant callers enabling bacterial variant discovery at accuracy that matches or exceeds the current "gold standard" with short reads. The evidence supporting the claims of the authors is convincing, although the inclusion of a larger number of reference genomes would further strengthen the study. The work will be of interest to anyone performing sequencing for outbreak investigations, bacterial epidemiology, or similar studies.

Reviewer #1 (Public Review):

Summary:

The authors assess the accuracy of short variant calling (SNPs and indels) in bacterial genomes using Oxford Nanopore reads generated on R10.4 flow cells from a very similar genome (99.5% ANI), examining the impact of variant caller choice (three traditional variant callers: bcftools, freebayes, and longshot, and three deep learning based variant callers: clair3, deep variant, and nano caller), base calling model (fast, hac and sup) and read depth (using both simplex and duplex reads).

Strengths:

Given the stated goal (analysis of variant calling for reads drawn from genomes very similar to the reference), the analysis is largely complete and results are compelling. The authors make the code and data used in their analysis available for re-use using current best practices (a computational workflow and data archived in INSDC databases or Zenodo as appropriate).

Weaknesses:

While the medaka variant caller is now deprecated for diploid calling, it is still widely used for haploid variant calling and should at least be mentioned (even if the mention is only to explain its exclusion from the analysis).

Appraisal:

The experiments the authors engaged in are well structured and the results are convincing. I expect that these results will be incorporated into "best practice" bacterial variant calling workflows in the future.

Reviewer #2 (Public Review):

Summary:

Hall et al describe the superiority of ONT sequencing and deep learning-based variant callers to deliver higher SNP and Indel accuracy compared to previous gold-standard Illumina short-read sequencing. Furthermore, they provide recommendations for read sequencing depth and computational requirements when performing variant calling.

Strengths:

The study describes compelling data showing ONT superiority when using deep learning-based variant callers, such as Clair3, compared to Illumina sequencing. This challenges the paradigm that Illumina sequencing is the gold standard for variant calling in bacterial genomes. The authors provide evidence that homopolymeric regions, a systematic and problematic issue with ONT data, are no longer a concern in ONT sequencing.

Weaknesses:

(1) The inclusion of a larger number of reference genomes would have strengthened the study to accommodate larger variability (a limitation mentioned by the authors).

(2) In Figure 2, there are clearly one or two samples that perform worse than others in all combinations (are always below the box plots). No information about species-specific variant calls is provided by the authors but one would like to know if those are recurrently associated with one or two species. Species-specific recommendations could also help the scientific community to choose the best sequencing/variant calling approaches.

(3) The authors support that a read depth of 10x is sufficient to achieve variant calls that match or exceed Illumina sequencing. However, the standard here should be the optimal discriminatory power for clinical and public health utility (namely outbreak analysis). In such scenarios, the highest discriminatory power is always desirable and as such an F1 score, Recall and Precision that is as close to 100% as possible should be maintained (which changes the minimum read sequencing depth to at least 25x, which is the inflection point).

(4) The sequencing of the samples was not performed with the same Illumina and ONT method/equipment, which could have introduced specific equipment/preparation artefacts that were not considered in the study. See for example https://academic.oup.com/nargab/article/3/1/lqab019/6193612.

Reviewer #3 (Public Review):

Hall et al. benchmarked different variant calling methods on Nanopore reads of bacterial samples and compared the performance of Nanopore to short reads produced with Illumina sequencing. To establish a common ground for comparison, the authors first generated a variant truth set for each sample and then projected this set to the reference sequence of the sample to obtain a mutated reference. Subsequently, Hall et al. called SNPs and small indels using commonly used deep learning and conventional variant callers and compared the precision and accuracy from reads produced with simplex and duplex Nanopore sequencing to Illumina data. The authors did not investigate large structural variation, which is a major limitation of the current manuscript. It will be very interesting to see a follow-up study covering this much more challenging type of variation.

In their comprehensive comparison of SNPs and small indels, the authors observed superior performance of deep learning over conventional variant callers when Nanopore reads were basecalled with the most accurate (but also computationally very expensive) model, even exceeding Illumina in some cases. Not surprisingly, Nanopore underperformed compared to Illumina when basecalled with the fastest (but computationally much less demanding) method with the lowest accuracy. The authors then investigated the surprisingly higher performance of Nanopore data in some cases and identified lower recall with Illumina short read data, particularly from repetitive regions and regions with high variant density, as the driver. Combining the most accurate Nanopore basecalling method with a deep learning variant caller resulted in low error rates in homopolymer regions, similar to Illumina data. This is remarkable, as homopolymer regions are (or, were) traditionally challenging for Nanopore sequencing.

Lastly, Hall et al. provided useful information on the required Nanopore read depth, which is surprisingly low, and the computational resources for variant calling with deep learning callers. With that, the authors established a new state-of-the-art for Nanopore-only variant, calling on bacterial sequencing data. Most likely these findings will be transferred to other organisms as well or at least provide a proof-of-concept that can be built upon.

As the authors mention multiple times throughout the manuscript, Nanopore can provide sequencing data in nearly real-time and in remote regions, therefore opening up a ton of new possibilities, for example for infectious disease surveillance.

However, the high-performing variant calling method as established in this study requires the computationally very expensive sup and/or duplex Nanopore basecalling, whereas the least computationally demanding method underperforms. Here, the manuscript would greatly benefit from extending the last section on computational requirements, as the authors determine the resources for the variant calling but do not cover the entire picture. This could even be misleading for less experienced researchers who want to perform bacterial sequencing at high performance but with low resources. The authors mention it in the discussion but do not make clear enough that the described computational resources are probably largely insufficient to perform the high-accuracy basecalling required.

eLife assessment

The study, from the group that pioneered migrasome, describes a novel vaccine platform derived from this newly discovered organelle. Using these cleverly engineered migrasomes – that behave like natural migrasomes – as a novel vaccine platform has the potential to overcome obstacles such as cold chain issues for vaccines like messenger RNA. Although the findings are important with practical implications for the vaccine technology, and the evidence, based on appropriate and validated methodology is convincing and is in line with current state-of-the-art, there are some critical issues that need to be addressed. These include a head-to-head comparison with proven vaccine platforms, for example, a SARS-CoV-2 mRNA vaccine or an adjuvanted recombinant spike protein.

Reviewer #1 (Public Review):

Summary:

Winged seeds or ovules from the Devonian are crucial to understanding the origin and early evolutionary history of wind dispersal strategy. Based on exceptionally well-preserved fossil specimens, the present manuscript documented a new fossil plant taxon (new genus and new species) from the Famennian Series of Upper Devonian in eastern China and demonstrated that three-winged seeds are more adapted to wind dispersal than one-, two- and four-winged seeds by using mathematical analysis.

Strengths:

The manuscript is well organised and well presented, with superb illustrations. The methods used in the manuscript are appropriate.

Weaknesses:

I would only like to suggest moving the "Mathematical analysis of wind dispersal of ovules with 1-4 wings" section from the supplementary information to the main text, leaving the supplementary figures as supplementary materials.

eLife assessment

This useful manuscript describes the second earliest known winged ovule without a capule in the Famennian of Late Devonian. Using solid mathematical analysis, the authors demonstrate that three-winged seeds are more adapted to wind dispersal than one-, two- and four-winged seeds. The manuscript will help the scientific community to understand the origin and early evolutionary history of wind dispersal strategy of early land plants.

Reviewer #2 (Public Review):

Summary:

This manuscript described the second earliest known winged ovule without a capule in the Famennian of Late Devonian. Using Mathematical analysis, the authors suggest that the integuments of the earliest ovules without a cupule, as in the new taxon and Guazia, evolved functions in wind dispersal.

Strengths:

The new ovule taxon's morphological part is convincing. It provides additional evidence for the earliest winged ovules, and the mathematical analysis helps to understand their function.

Weaknesses:

The discussion should be enhanced to clarify the significance of this finding. What is the new advance compared with the Guazia finding? The authors can illustrate the character transformations using a simplified cladogram. The present version of the main text looks flat.

eLife assessment

This important study reports the deep evolutionary conservation of a core genetic program regulating spermatogenesis in flies, mice, and humans. The data presented are supportive of the main conclusion and generally convincing. This work will be of interest to evolutionary and reproductive biologists.

Reviewer #1 (Public Review):

Summary:

By combining an analysis of the evolutionary age of the genes expressed in male germ cells, a study of genes associated with spermatocyte protein-protein interaction networks and functional experiments in Drosophila, Brattig-Correia and colleagues provide evidence for an ancient origin of the genetic program underlying metazoan spermatogenesis. This leads to identifying a relatively small core set of functional interactions between deeply conserved gene expression regulators, whose impairment is then shown to be associated with cases of human male infertility.

Strengths:

In my opinion, the work is important for three different reasons. First, it shows that, even though reproductive genes can evolve rapidly and male germ cells display a significant level of transcriptional noise, it is still possible to obtain convincing evidence that a conserved core of functionally interacting genes lies at the basis of the male germ transcriptome. Second, it reports an experimental strategy that could also be applied to gene networks involved in different biological problems. Third, the authors make a compelling case that, due to its effects on human spermatogenesis, disruption of the male germ cell orthoBackbone can be exploited to identify new genetic causes of infertility.

Weaknesses:

The main strength of the general approach followed by the authors is, inevitably, also a weakness. This is because a study rooted in comparative biology is unlikely to identify newly emerged genes that may adopt key roles in processes such as species-specific gamete recognition. Additionally, using a TPM >1 threshold for protein-coding transcripts may exclude genes, such as those encoding proteins required for gamete fusion, which are thought to be expressed at a very low level. Although these considerations raise the possibility that the chosen approach may miss information that, depending on the species, could be potentially highly functionally important, this by no means reduces its value in identifying genes belonging to the conserved genetic program of spermatogenesis.

Reviewer #2 (Public Review):

Summary:

This is a tour de force study that aims to understand the genetic basis of male germ cell development across three animal species (human, mouse, and flies) by performing a genetic program conservation analysis (using phylostratigraphy and network science) with a special emphasis on genes that peak or decline during mitosis-to-meiosis. This analysis, in agreement with previous findings, reveals that several genes active during and before meiosis are deeply conserved across species, suggesting ancient regulatory mechanisms. To identify critical genes in germ cell development, the investigators integrated clinical genetics data, performing gene knockdown and knockout experiments in both mice and flies. Specifically, over 900 conserved genes were investigated in flies, with three of these genes further studied in mice. Of the 900 genes in flies, ~250 RNAi knockdowns had fertility phenotypes. The fertility phenotypes for the fly data can be viewed using the following browser link: https://pages.igc.pt/meionav. The scope of target gene validation is impressive. Below are a few minor comments.

(1) In Supplemental Figure 2, it is notable that enterocyte transcriptomes are predominantly composed of younger genes, contrasting with the genetic age profile observed in brain and muscle cells. This difference is an intriguing observation and it would be curious to hear the author's comments.

(2) Regarding the document, the figures provided only include supplemental data; none of the main text figures are in the full PDF.

(3) Lastly, it would be great to section and stain mouse testis to classify the different stages of arrest during meiosis for each of the mouse mutants in order to compare more precisely to flies.

This paper serves as a vital resource, emphasizing that only through the analysis of hundreds of genes can we prioritize essential genes for germ cell development. its remarkable that about 60% of conserved genes have no apparent phenotype during germ cell development.

Strengths:

The high-throughput screening was conducted on a conserved network of 920 genes expressed during the mitosis-to-meiosis transition. Approximately 250 of these genes were associated with fertility phenotypes. Notably, mutations in 5 of the 250 genes have been identified in human male infertility patients. Furthermore, 3 of these genes were modeled in mice, where they were also linked to infertility. This study establishes a crucial groundwork for future investigations into germ cell development genes, aiming to delineate their essential roles and functions.

Weaknesses:

The fertility phenotyping in this study is limited, yet dissecting the mechanistic roles of these proteins falls beyond its scope. Nevertheless, this work serves as an invaluable resource for further exploration of specific genes of interest.

eLife assessment

This important study reports the developmental dynamics and molecular markers of the rete ovarii during ovarian development. However, the data supporting the main conclusions remain incomplete. This study will be of interest to developmental and reproductive biologists.

Reviewer #1 (Public Review):

Summary:

The manuscript by Anbarcia et al. re-evaluates the function of the enigmatic Rete Ovarii (RO), a structure that forms in close association with the mammalian ovary. The RO has generally been considered a functionless structure in the adult ovary. This manuscript follows up on a previous study from the lab that analyzed ovarian morphogenesis using high-resolution microscopy (McKey et al., 2022). The present study adds finer details to RO development and possible function by (1) identifying new markers for OR sub-regions (e.g. GFR1a labels the connecting rete) suggesting that the sub-regions are functionally distinct, (2) showing that the OR sub-regions are connected by a luminal system that allows transport of material from the extra-ovarian rete (EOR) to the inter-ovarian rete (IOG), (3) identifies proteins that are secreted into the OR lumen and that may regulate ovarian homeostasis, and finally, (4) better defines how the vasculature, nervous, and immune system integrates with the OR.

Strengths:

The data is beautifully presented and convincing. They show that the RO is composed of three distinct domains that have unique gene expression signatures and thus likely are functionally distinct.

Weaknesses:

It is not always clear what the novel findings are that this manuscript is presenting. It appears to be largely similar to the analysis done by McKey et al. (2022) but with more time points and molecular markers. The novelty of the present study's findings needs to be better articulated.

Reviewer #2 (Public Review):

A large number of ovarian experiments have been conducted - especially in morphological and molecular biology studies - specifically removing the ovarian membrane. This experiment is a good supplement to existing knowledge and plays an important role in early ovarian development and the regulation of ovarian homeostasis during the estrous cycle. There are also innovations in research ideas and methods, which will meet the requirements of experimental design and provide inspiration for other researchers.

This reviewer did not identify any major issues with the article. However, the following points could be further clarified:

(1) Is there any comparative data on the proteomics of RO and rete testis in early development? With some molecular markers also derived from rete testis, it would be better to provide the data or references.

(2) Although the size of RO and its components is quite small and difficult to operate, the researchers in this article had already been able to perform intracavitary injection of EOR and extract EOR or CR for mass spectrometry analysis. Therefore, can EOR, CR, or IOR be damaged or removed, providing further strong evidence of ovarian development function?

(3) Although IOR is shown on the schematic diagram, it cannot be observed in the immunohistochemistry pictures in Figure 1 and Figure 3. The authors should provide a detailed explanation.

Reviewer #3 (Public Review):

Summary:

The rete ovarii (RO) has long been disregarded as a non-functional structure within the ovary. In their study, Anbarci and colleagues have delineated the markers and developmental dynamics of three distinct regions of the RO - the intraovarian rete (IOR), the extraovarian rete (EOR), and the connecting rete (CR). Notably focusing on the EOR, the authors presented evidence illustrating that the EOR forms a convoluted tubular structure culminating in a dilated tip. Intriguingly, microinjections into this tip revealed luminal flow towards the ovary containing potentially secreted functional proteins. Additionally, the EOR cells exhibit associations with vasculature, macrophages, and neuronal projections, proposing the notion that the RO may play a functional role in ovarian development during critical ovariogenesis stages. By identifying marker genes within the RO, the authors have also suggested that the RO could serve as a potential structure linking the ovary with the neuronal system.

Strengths:

Overall, the reviewer commends the authors for their systematic research on the RO, shedding light on this overlooked structure in developing ovaries. Furthermore, the authors have proposed a series of hypotheses that are both captivating and scientifically significant, with the potential to reshape our understanding of ovarian development through future investigations.

Weaknesses:

There is a lack of conclusive data supporting many conclusions in the manuscript. Therefore, the paper's overall conclusions should be moderated until functional validations are conducted.

eLife assessment

The authors combined human genetic analysis with zebrafish experiments to produce evidence that alleles that impair the function of EPHA4 cause idiopathic scoliosis (IS), a common spinal deformity. The significance of the findings is important because the cellular and molecular mechanisms that contribute to IS remain poorly understood. The human genetic data are quite convincing whereas the zebrafish data, although supportive, are incomplete.

Joint Public Review:

Summary:

Idiopathic scoliosis (IS) is a common spinal deformity. Various studies have linked genes to IS, but underlying mechanisms are unclear such that we still lack understanding of the causes of IS. The current manuscript analyzes IS patient populations and identifies EPHA4 as a novel associated gene, finding three rare variants in EPHA4 from three patients (one disrupting splicing and two missense variants) as well as a large deletion (encompassing EPHA4) in a Waardenburg syndrome patient with scoliosis. EPHA4 is a member of the Eph receptor family. Drawing on data from zebrafish experiments, the authors argue that EPHA4 loss of function disrupts the central pattern generator (CPG) function necessary for motor coordination.

Strengths:

The main strength of this manuscript is the human genetic data, which provides convincing evidence linking EPHA4 variants to IS. The loss of function experiments in zebrafish strongly support the conclusion that EPHA4 variants that reduce function lead to IS.

Weaknesses:

The conclusion that disruption of CPG function causes spinal curves in the zebrafish model is not well supported. The authors' final model is that a disrupted CPG leads to asymmetric mechanical loading on the spine and, over time, the development of curves. This is a reasonable idea, but currently not strongly backed up by data in the manuscript. Potentially, the impaired larval movements simply coincide with, but do not cause, juvenile-onset scoliosis. Support for the authors' conclusion would require independent methods of disrupting CPG function and determining if this is accompanied by spine curvature. At a minimum, the language of the manuscript could be toned down, with the CPG defects put forward as a potential explanation for scoliosis in the discussion rather than as something this manuscript has "shown". An additional weakness of the manuscript is that the zebrafish genetic tools are not sufficiently validated to provide full confidence in the data and conclusions.

eLife assessment

This work is important because it attempts to elucidate how immune cells migrate across the blood brain barrier. The authors developed a convincing framework to visualize, recognize and track the movement of different immune cells across primary human and mouse brain microvascular endothelial cells without the need for fluorescence-based imaging using microfluidic devices. The data gathered are solid, and this work will be of interest to the cancer biology, immunology and medical therapeutics fields.

Reviewer #1 (Public Review):

Summary:

It is evident that studying leukocyte extravasation in vitro is a challenge. One needs to include physiological flow, culture cells and isolate primary immune cells. Timing is of utmost importance and a reproducible setup essential. Extra challenges are met when extravasation kinetics in different vascular beds is required, e.g., across the blood-brain barrier. In this study, the authors describe a reliable and reproducible method to analyze leukocyte TEM under physiological flow conditions, including this analysis. That the software can also detect reverse TEM is a plus.

Strengths:

It is quite a challenge to get this assay reproducible and stable, in particular as there is flow included. Also for the analysis, there is currently no clear software analysis program, and many labs have their own methods. This paper gives the opportunity to unify the data and results obtained with this assay under label-free conditions. This should eventually lead to more solid and reproducible results.

Also, the comparison between manual and software analysis is appreciated.

Weaknesses:

The authors stress that it can be done in BBB models, but I would argue that it is much more broadly applicable. This is not necessarily a weakness of the study but more an opportunity to strengthen the method. So I would encourage the authors to rewrite some parts and make it more broadly applicable.

Reviewer #2 (Public Review):

Summary:

This paper develops an under-flow migration tracker to evaluate all the steps of the extravasation cascade of immune cells across the BBB. The algorithm is useful and has important applications.

Strengths:

Algorithm is almost as accurate as manual tracking and importantly saves time for researchers.

Weaknesses:

Applicability can be questioned because the device used is 2D and physiological biology is in 3D. Comparisons to other automated tools was not performed by the authors.

Reviewer #3 (Public Review):

Summary:

The authors aimed to establish a faster and more efficient method of tracking steps of T-cell extravasation across the blood brain barrier. The authors developed a framework to visualize, recognize and track the movement of different immune cells across primary human and mouse brain microvascular endothelial cells without the need for fluorescence-based imaging. The authors succinctly describe the basic requirements for tracking in the introduction followed by an in-depth account of the execution.

Weaknesses and Strengths:

Materials & methods and results:

(1) The methods section also lacks details of the microfluidic device that the authors talk about in the paper. Under physiological sheer stress, the T-cells detach from the pMBMEC monolayer, and are hence unable to be detected; however, this observation requires an explanation pertaining to the reason of occurrence and potential solutions to circumvent it to ensure physiologically relevant experimental parameters.

(2) The author describes a method for debris exclusion using UFMTrack that eliminates objects of <30 pixels in size from analysis based on a mean pixel size of 400 for T lymphocytes. However, this mean pixel size appears to stem from in-vitro activated CD8 T cells, which rapidly grow and proliferate upon stimulation. In line with this, activated lymphocytes exhibit increased cytoplasmic area, making them appear less dense or "brighter" by phase microscopy compared to naïve lymphocytes, which are relatively compact and subsequently appear dimmer. Given this, it is not clear whether UFMTrack is sufficiently trained to identify naïve human lymphocytes in circulating blood, nor smaller, murine lymphocytes. Analysis of each lymphocyte subtype in terms of pixel size and intensity would be beneficial to strengthen the claim that UFMTrack can identify each of these populations. Additionally, demonstrating that UFMTrack can correctly characterize the behavior of naïve versus activated lymphocytes isolated from murine and human sources would strengthen the claim that UFMTrack can be broadly applied to study lymphocyte dynamics in diverse models without additional training

(3) Average precision was compared to the analysis of UFMTrack but it is unclear how average precision was calculated. This information should have been included in the methods section

(4) CD4 and CD8 T cells exhibit distinct biology and interaction kinetics driven in part by their MHC molecule affinity and distinct receptor expression profiles. Thus, it is unclear why two distinct mechanisms of endothelial cell activation are needed to see differences between the populations.

(5) The BMECs are barrier tissues but were cultured on µdishes in this study. To study the transmigration of T-cells across the endothelium, the model would have been more relevant on a semi-permeable membrane instead of a closed surface.

(6) Methods are provided for the isolation and expansion of human effector and memory CD4+ T cells. However, there is no mention of specific CD4+ T cell populations used for analysis with UFMTrack, nor a clear breakdown of tracking efficiency for each subpopulation. Further, there is no similar method for the isolation of CD8+ T cell compartments. A clear breakdown of the performance efficiency of UFMTrack with each cell population investigated in this study would provide greater insight into the software's performance with regard to tracking the behavior and movement of distinct immune populations.

(7) The results section is quite extensive and discusses details of establishment of the framework while highlighting both the pros and cons of the different aspects of the process, for example the limitation of the two models, 2D and 2D+T were highlighted well. However, the results section includes details which may be more fitting in the methods section.

(8) A few statements in the results section lacked literary support, which was not provided in the discussion either, such as support for increased variance of T-cell instantaneous speed on stimulated vs non-stimulated pMBMECs. Another example is the enhancement of cytokine stimulation directed T-cell movement on the pMBMECs that the authors observed but failed to relay the physiological relevance of it. The authors don't provide enough references for developments in the field prior to their work which form the basis and need for this technology.

(9) The rationale for use of OT-1 and 2D2-derived murine lymphocytes is unclear here. The OT-1 model has been generated to study antigen-specific CD8+ T cell responses, while the 2D2 model has been generated to recapitulate CD4 T cell-specific myelin oligodendrocyte glycoprotein (MOG) responses.

Figures and text:

(1) There are certain discrepancies and misarrangement of figures and text. For example, discussion of the effect of sheer flow on T cell attachment as part of the introduction in figure 1 and then mentioning it in the text again in the results section as part of figure 4 is repetitive.

(2) Section IV, subsection 1 of the results section, refers to 'data acquisition section above' in line 279, however the said section is part of materials and methods which is provided towards the end of the manuscript.

(3) There are figures in the manuscript that have not been referenced in the results section, for example, figure 3A and B. Figure 1 hasn't been addressed until subsection 7 of materials and methods

(4) A lack of significance but an observed trend of increased variance of T cell instantaneous speed is reported in line 296-298; however, the graph (figure 4G) shows a significant change in instantaneous speed between non-stimulated and TNFα-stimulated systems. This is misleading to the readers.

(5) The authors talk about three beginner experimentors testing the manual T cell tracking process but figure 5 only showcases data from two experimentors without stating the reason for excluding experimentor 1.

Discussion:

(1) While the discussion captures the major takeaways from the paper, it lacks relevant supporting references to relate the observation to physiological conditions and applicability.

(2) The discussion lacks connection to the results since the figures were not referenced while discussing an observed trend

(3) The authors briefly looked into mouse and human BMECs and their individual interaction with T-cells, but don't discuss the differences between the two, if any, that challenged their framework.

(4) Even though though the imaging tool relies on difference in appearance for detection, the authors talk about lack of feasibility in detecting transmigration of BMDMs due to their significantly different appearance. The statement lacks a problem solving approach to discuss how and why this was the case.

Relevance to the field:

Utilizing the framework provided by the authors, the application can be adapted and/or utilized for visualizing a range of different cell types, provided they are different in appearance. However, this would require extensive changes to the script and won't be adaptable in its current form.

eLife assessment

This fundamental study provides a modeling regime that provides new insight into the energy-preservation parameters among schooling fish. The strength of the evidence supporting observations such as distilled dynamics between leading and lagging schooling fish which are derived from emergent properties is convincing. Overall, the study provides exciting insights into energetic coupling with respect to group swimming dynamics. Some potential improvements to strengthen the study include clarification regarding degrees of freedom and parameter ranges in the model.

Reviewer #1 (Public Review):

Summary:

The study seeks to establish accurate computational models to explore the role of hydrodynamic interactions on energy savings and spatial patterns in fish schools. Specifically, the authors consider a system of (one degree-of-freedom) flapping airfoils that passively position themselves with respect to the streamwise direction, while oscillating at the same frequency and amplitude, with a given phase lag and at a constant cross-stream distance. By parametrically varying the phase lag and the cross-stream distance, they systematically explore the stability and energy costs of emergent configurations. Computational findings are leveraged to distill insights into universal relationships and clarify the role of the wake of the leading foil.

Strengths:

(1) The use of multiple computational models (computational fluid dynamics, CFD, for full Navier-Stokes equations and computationally efficient inviscid vortex sheet, VS, model) offers an extra degree of reliability of the observed findings and backing to the use of simplified models for future research in more complex settings.

(2) The systematic assessment of the stability and energy savings in multiple configurations of pairs and larger ensembles of flapping foils is an important addition to the literature.

(3) The discovery of a linear phase-distance relationship in the formation attained by pairs of flapping foils is a significant contribution, which helps compare different experimental observations in the literature.

(4) The observation of a critical size effect for in-line formations of larger, above which cohesion and energetic benefits are lost at once, is a new discovery in the field.

Weaknesses:

(1) The extent to which observations on one-degree-of-freedom flapping foils could translate to real fish schools is presently unclear so some of the conclusions on live fish schools are likely to be overstated and would benefit from some more biological framing.

(2) The analysis of non-reciprocal coupling is not as novel as the rest of the study and potentially not as convincing due to the chosen linear metric of interaction (that is, the flow agreement).

Overall, this is a rigorous effort on a critical topic: findings of the research can offer important insight into the hydrodynamics of fish schooling, stimulating interdisciplinary research at the interface of computational fluid mechanics and biology.

Reviewer #2 (Public Review):

The document "Mapping spatial patterns to energetic benefits in groups of flow-coupled swimmers" by Heydari et al. uses several types of simulations and models to address aspects of stability of position and power consumption in few-body groups of pitching foils. I think the work has the potential to be a valuable and timely contribution to an important subject area. The supporting evidence is largely quite convincing, though some details could raise questions, and there is room for improvement in the presentation. My recommendations are focused on clarifying the presentation and perhaps spurring the authors to assess additional aspects:

(1) Why do the authors choose to set the swimmers free only in the propulsion direction? I can understand constraining all the positions/orientations for investigating the resulting forces and power, and I can also understand the value of allowing the bodies to be fully free in x, y, and their orientation angle to see if possible configurations spontaneously emerge from the flow interactions. But why constrain some degrees of freedom and not others? What's the motivation, and what's the relevance to animals, which are fully free?

(2) The model description in Eq. (1) and the surrounding text is confusing. Aren't the authors computing forces via CFD or the VS method and then simply driving the propulsive dynamics according to the net horizontal force? It seems then irrelevant to decompose things into thrust and drag, and it seems irrelevant to claim that the thrust comes from pressure and the drag from viscous effects. The latter claim may in fact be incorrect since the body has a shape and the normal and tangential components of the surface stress along the body may be complex.

(3) The parameter taudiss in the VS simulations takes on unusual values such as 2.45T, making it seem like this value is somehow very special, and perhaps 2.44 or 2.46 would lead to significantly different results. If the value is special, the authors should discuss and assess it. Otherwise, I recommend picking a round value, like 2 or 3, which would avoid distraction.

(4) Some of the COT plots/information were difficult to interpret because the correspondence of beneficial with the mathematical sign was changing. For example, DeltaCOT as introduced on p. 5 is such that negative indicates bad energetics as compared to a solo swimmer. But elsewhere, lower or more negative COT is good in terms of savings. Given the many plots, large amounts of data, and many quantities being assessed, the paper needs a highly uniform presentation to aid the reader.

(5) I didn't understand the value of the "flow agreement parameter," and I didn't understand the authors' interpretation of its significance. Firstly, it would help if this and all other quantities were given explicit definitions as complete equations (including normalization). As I understand it, the quantity indicates the match of the flow velocity at some location with the flapping velocity of a "ghost swimmer" at that location. This does not seem to be exactly relevant to the equilibrium locations. In particular, if the match were perfect, then the swimmer would generate no relative flow and thus no thrust, meaning such a location could not be an equilibrium. So, some degree of mismatch seems necessary. I believe such a mismatch is indeed present, but the plots such as those in Figure 4 may disguise the effect. The color bar is saturated to the point of essentially being three tones (blue, white, red), so we cannot see that the observed equilibria are likely between the max and min values of this parameter.

(6) More generally, and related to the above, I am favorable towards the authors' attempts to find approximate flow metrics that could be used to predict the equilibrium positions and their stability, but I think the reasoning needs to be more solid. It seems the authors are seeking a parameter that can indicate equilibrium and another that can indicate stability. Can they clearly lay out the motivation behind any proposed metrics, and clearly present complete equations for their definitions? Further, is there a related power metric that can be appropriately defined and which proves to be useful?

(7) Why do the authors not carry out CFD simulations on the larger groups? Some explanations should be given, or some corresponding CFD simulations should be carried out. It would be interesting if CFD simulations were done and included, especially for the in-line case of many swimmers. This is because the results seem to be quite nuanced and dependent on many-body effects beyond nearest-neighbor interactions. It would certainly be comforting to see something similar happen in CFD.

(8) Related to the above, the authors should discuss seemingly significant differences in their results for long in-line formations as compared to the CFD work of Peng et al. [48]. That work showed apparently stable groups for numbers of swimmers quite larger than that studied here. Why such a qualitatively different result, and how should we interpret these differences regarding the more general issue of the stability of tandem groups?

(9) The authors seem to have all the tools needed to address the general question about how dynamically stable configurations relate to those that are energetically optimal. Are stable solutions optimal, or not? This would seem to have very important implications for animal groups, and the work addresses closely related topics but seems to miss the opportunity to give a definitive answer to this big question.

(10) Time-delay particle model: This model seems to construct a simplified wake flow. But does the constructed flow satisfy basic properties that we demand of any flow, such as being divergence-free? If not, then the formulation may be troublesome.

Author response:

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

eLife assessment

This is a valuable study that describes the effects of T. pallidum on neural development by applying single-cell RNA sequencing to an iPSC-derived brain organoid model. The evidence supporting the claims of the authors is solid, although further evidence to understand the differences in infection rates would strengthen the conclusions of the study. In particular, the conclusions would be strengthened by validating infection efficiency as this can impact the interpretation of single-cell sequencing results, and how these metrics affect organoid size as well as comparison with additional infectious agents. Furthermore, additional validations of downstream effectors are not adequate and could be improved. 

Thank you very much for your valuable comments. Since we used the organoid model for the first time to investigate the effects of T. pallidum on brain development, the study design is not perfect. As you have accurately mentioned, the results of the paper do not have more in-depth details, especially to verify the infection rate of T. pallidum. Your valuable comments will be very useful for us for carrying out further research. In addition, the downstream effector validation is inadequate, so we performed an analysis of single-cell sequencing data to strengthen our view in the revised manuscript (See Figure 5F for a description in current manuscript).

Public Reviews:

Reviewer #1 (Public Review):

Summary:

This is an interesting study by Xu et al showing the effects of infection with the Treponema pallidum virus (which causes syphilis disease) on neuronal development using iPSC-derived human brain organoids as a model and single-cell RNA sequencing. This work provides an important insight into the impact of the virus on human development, bridging the gap between the phenomena observed in studies using animal models as well as non-invasive human studies showing developmental abnormalities in fetuses infected with the virus in utero through maternal vertical transmission.

Using single-cell RNAseq in combination with qPCR and immunofluorescence techniques, the authors show that T. pallidum infected organoids are smaller in size, in particular during later growth stages, contain a larger number of undifferentiated neuronal lineage cells, and exhibit decreased numbers of specific neuronal subcluster, which the authors have identified as undifferentiated hindbrain neurons.

The study is an important first step in understanding how T. pallidum affects human neuronal development and provides important insight into the potential mechanisms that underlie the neurodevelopmental abnormalities observed in infected human fetuses. Several important weaknesses have also been noted, which need to be addressed to strengthen the study's conclusions.

Strengths:

(1) The study is well written, and the data quality is good for the most part.

(2) The study provides an important first step in utilizing human brain organoids to study the impact of T. pallidum infection on neuronal development.

(3) The study's conclusions may provide important insight to other researchers focused on studying how viral infections impact neuronal development. 

Thank you very much for your positive feedback. Below, you will find our detailed responses to your concerns, addressed point-by-point. I once again sincerely appreciate your time and effort in reviewing our manuscript.

Weaknesses:

(1) It is unclear how T. pallidum infection was validated in the organoids. If not all cells are infected, this could have important implications for the study's conclusions, in particular the single-cell RNAseq experiments. Were only cells showing the presence of the virus selected for sequencing? A detailed description of how infection was validated and the process of selection of cells for RNAseq would strongly support the study's conclusions. 

Thank you for your valuable comment. We completely agree with your point. Exploring the infection rate of T. pallidum to brain organoids is a key factor that must be considered. We selected pluripotent stem cell-derived brain organoids to simulate the process of foetal brain neurodevelopment and cultured them mixed with T. pallidum to mimic T. pallidum invading brain tissue. Since brain organoids are three-dimensional structures formed by nerve cell aggregation, T. pallidum invades organoids from the periphery to the center of the organoids gradually. T. pallidum acts on organoids long enough to increase the infection rates; however, the pathogen is selective in invading human cells. If we only select cells present in T. pallidum for sequencing, the authenticity of simulating "real world" infections is somewhat weakened. To better carry out this study, selecting cells from intact organoids for sequencing, without eliminating cells without T. pallidum, can better simulate the effect of T. pallidum infection on the nervous system. Of course, we should also set up a blank control group.

(2) The authors show that T. pallidum infection results in impaired development of hindbrain neurons. How does this finding compare to what has already been shown in animal studies? Is a similar deficit in this brain region observed with this specific virus? It would be useful to strengthen the study's conclusions if the authors added a discussion about the observed deficits in hindbrain neuronal development, and prior literature on similar studies conducted in animal models or human patients. Does T. pallidum preferentially target these neurons, or is this a limitation of the current organoid model system? 

Thank you for your valuable comments. The finding that T. pallidum infection results in impaired development of hindbrain neurons has not been verified in animal experiments. Of course, it is better to further validate the findings in organoid studies through animal experiments. Unfortunately, due to the technical challenges, mature animal models have not been developed for the study of congenital syphilis. Although our team has been working on the development of animal models of congenital neurosyphilis, the current progress is still not satisfactory. After struggling hard in this field for many years, we decided to attempt to utilize human brain organoids instead of animal models to study the impact of T. pallidum infection on neuronal development.

We also checked prior literature on similar studies that have referred to the content in human patients. Dan Doherty et al. reported that patients with pontocerebellar hypoplasia develop microcephaly at birth or over time after birth (PMID: 23518331). Based on your constructive suggestions, we have added some content related to hindbrain to the “Discussion” section.

Our study found that T. pallidum could inhibit the differentiation of subNPC1B in brain organoids, thereby reducing the differentiation from subNPC1B to hindbrain neurons, and ultimately affecting the development and maturation of hindbrain neurons during pregnancy. Based on our results, T. pallidum does not preferentially target hindbrain neurons. Of course, there are limitations to the current organoid model system, see the "Limitations" section.

PMID: 23518331- Dan Doherty et al, Midbrain and hindbrain malformations: advances in clinical diagnosis, imaging, and genetics.

Revision in the “Discussion” section, line 343-352:

“The vertebrate hindbrain contains a complex network of dedicated neural circuits that play an essential role in controlling many physiological processes and behaviors, including those related to the cerebellum, pons, and medulla oblongata (Shoja et al., 2018). Patients with pontocerebellar hypoplasia represent the less severe end of the spectrum with early hyperreflexia, developmental delay, and feeding problems, eventually developing spasticity and involuntary movements in childhood, while some patients represent the severe end of the spectrum characterised by polyhydramnios, severe hyperreflexia, contracture, and early death from central respiratory failure. Patients with pontocerebellar hypoplasia develop microcephaly at birth or over time after birth (Doherty et al., 2013).”

(3) The authors show that T. pallidum-infected organoids are smaller in size by measuring organoid diameter during later stages of organoid growth, with no change during early stages. Does that represent insufficient infection at the early stages? Is this due to increased cell death or lack of cell division in the infected organoids? Experiments using IHC to quantify levels of cleaved caspase and/or protein markers for cell proliferation would be able to address these questions. 

Thank you for your valuable suggestion. The concentration of T. pallidum in patients with syphilis was generally very low (PMID: 21752804, 35315702, 33099614). In this study, a low concentration of T. pallidum was applied to brain organoids to simulate early foetal transmission of syphilis. Nerve cells mainly establish intercellular connections to form brain organoids in the way of adhesion, which can easily cause organoids to divide and die if treated with a high concentration of T. pallidum. Furthermore, based on your suggestions, we performed additional immunostaining analyses to verify the apoptosis of brain organoids infected by T. pallidum. Cleaved caspase 3 (clCASP3) staining showed that the number of apoptotic cells increased following T. pallidum infection; however, the proportion of apoptotic cells in both groups of brain organoids was very low (Figure supplement 2) (N=12 organoids, each group from three independent bioreactors), which would be not enough to affect the results of the experiment, thereby suggesting that neural differentiation and development of brain organoids were mainly inhibited following T. pallidum infection (rather than promoting organoid apoptosis).

PMID: 21752804-- Craig Tipple et al, Getting the measure of syphilis: qPCR to better understand early infection.

PMID: 35315702-- Cuini Wang et al, Quantified Detection of Treponema pallidum DNA by PCR Assays in Urine and Plasma of Syphilis Patients.

PMID: 33099614—Cuini Wang et al, A New Specimen for Syphilis Diagnosis: Evidence by High Loads of Treponema pallidum DNA in Saliva.

Revision in the “Results” section, line 105-108:

“… cleaved caspase 3 (clCASP3) staining showed that the number of apoptotic cells increased significantly following T. pallidum infection, but the proportion of apoptotic cells in both groups of brain organoids was very low (Figure supplement 2) (N=12 organoids, each group from three independent bioreactors) …”

Revision in the “Materials and methods” section, line 446-447:

“…anti-cleaved caspase 3 (rabbit, 1:100, Cell Signaling Technology, 9661S),”

Revision in the “Supplementary File” section, line 78-81:

Author response image 1.

The number of clCASP3+ cells in the microscopic field of brain organoids. A nonparametric t-test was used to evaluate the statistical differences between the two groups. (**: P < 0.01).

(4) In Figure 1D authors show differences in rosette-like structure in the infected organoids. The representative images do not appear to be different in any of the discussed components (e.g., the sox2 signal looks fairly similar between the two conditions). No quantification of these structures was presented. Authors should provide quantification or a more representative image to support their statement. 

Thank you for your valuable suggestion. I have quantified the neural rosette structure and compared the number of intact rosette-like structures between the two groups (See Figure 1D for a description in current manuscript).

(5) The IHC images shown in Figures 3E, G, and Figure 4E look very similar between the two conditions despite the discussed decrease in the text. A more suitable representative image should be presented, or the analysis should be amended to reflect the observed results. 

Thank you for your valuable suggestion. I have replaced more representative images in Figure 3E, G, and Figure 4E in the manuscript.

Reviewer #2 (Public Review):

Summary:

This study provides an important overview of infectious etiology for neurodevelopment delay.

Strengths:

Strong RNA evaluation.

Weaknesses:

The study lacks an overview of other infectious agents. The study should address the epigenetic contributors (PMID: 36507115) and the role of supplements in improving outcomes (PMID: 27705610). 

Addressing the above - with references included - is recommended. 

Thank you for your valuable comment. Our research is mainly inspired by other infectious agents, such as Zika virus; there are many descriptions of Zika virus in the “Discussion” section of the manuscript to better describe and demonstrate our point of view (See pages 12–13). I was unable to retrieve the article (PMID: 36507115), kindly help in confirming the PMID number. I will be very grateful if you can provide the full text. Secondly, I have carefully read the article (PMID: 27705610), which is a very rich and comprehensive review, and summarised and cited it in appropriate places in our manuscript.

Revision in the “Discussion- limitation” section, line 375-379:

“First, although several recent protocols have made use of growth factors to promote further neuronal maturation and survival (Lucke-Wold et al., 2018), the organoid culture scheme needs to be further improved owing to the lower percentage of mature neurons and the challenge of cell necrosis within the organoids at this stage in day 55 organoids.”

Reviewer #3 (Public Review): 

This article is the first report to study the effects of T. pallidum on the neural development of an iPSC-derived brain organoid model. The study indicates that T. pallidum inhibits the differentiation of subNPC1B neurons into hindbrain neurons, hence affecting brain organoid neurodevelopment. Additionally, the TCF3 and notch signaling pathways may be involved in the inhibition of the subNPC1B-hindbrain neuron differentiation axis. While the majority of the data in this study support the conclusions, there are still some questions that need to be addressed and data quality needs to be improved. The study provides valuable insights for future investigations into the mechanisms underlying congenital neurodevelopment disability. 

I sincerely appreciate your comments on our paper. The comments have helped us greatly improve the quality of our paper. Thank you for your time and constructive critique.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors): 

Paired t-test analysis is not appropriate if two distinct groups are compared. 

I sincerely apologize for our presentation. We used a nonparametric t-test to compare the two groups. I have confirmed and corrected the statistical method description of this manuscript (Revision in the “Materials and methods” section (line 553-555) and “Figures-legend” section (line 789-790, 817-818, 829-830) in current manuscript).

Reviewer #3 (Recommendations For The Authors): 

(1) Can the authors explain why the mean size of organoids infected with T. pallidum is smaller?

Thank you for your valuable comment. In our study, T. pallidum infection resulted in brain organisational changes in neural rosette-like structures resembling the proliferative regions of the human ventricular zone and caused fewer and incomplete rosette-like structures. Next, the ventricular zone is also the main area where neural progenitor cells (NPCs) reside (PMID: 33838105); our results showed that the proportion of neural progenitor cells (NPC)1 was reduced after T. pallidum infection. Rosette-like structure size changes owing to NPC depletion. Therefore, the mean size of organoids infected with T. pallidum is smaller.

Revision in the “Results” section, line 101-104:

“T. pallidum infection resulted in brain organisational changes in neural rosette-like structures resembling the proliferative regions of the human ventricular zone where NPC reside (Krenn et al., 2021), and caused fewer and incomplete rosette-like structures (P < 0.01) (Figure 1D)”

(2) Why was the target gene for qRT-PCR validation selected to be HOXA5、HOXC5、HOXA4?

Thank you for your valuable comment. The qRT-PCR experiment was selected here to verify the analysis results of the scRNA-seq. HOX family genes are key factors controlling early hindbrain development, which are expressed in the hindbrain region during the gastrulation stage of early embryonic development and persist into the nerve cell stage, and are essential for the correct induction of hindbrain development and segmentation (PMID: 2571936, 1983472, 1673098, 15930115). Therefore, we selected the HOX family gene for verification.

PMID: 2571936-WILKINSON D G, et al. Segmental expression of Hox-2 homoeobox- containing genes in the developing mouse hindbrain.

PMID: 1983472-- FROHMAN M A, et al. Isolation of the mouse Hox-2.9 gene; analysis of embryonic expression suggests that positional information along the anterior-posterior axis is specified by mesoderm.

PMID: 1673098--MURPHY P, et al. Expression of the mouse labial-like homeobox-containing genes, Hox 2.9 and Hox 1.6, during segmentation of the hindbrain.

PMID: 15930115-- MCNULTY C L, et al. Knockdown of the complete Hox paralogous group 1 leads to dramatic hindbrain and neural crest defects.

(3) Why was qRT-PCR not employed in other experimental validations, but solely to validate early neural-specific transcription factor changes?

Thank you for your valuable comment. The qRT-PCR experiment was selected to validate early neural-specific transcription factor changes, indicating the reliability of the scRNA-seq. Then, validated scRNA-seq data were used to analyze for other neuro-specific gene differences, such as violin plots and heatmap showing differentially expressed genes (Figure 4D and Figure 5B, C). Of course, we also tested it with other experiments, such as immunohistochemistry and flow cytometric screening.

(4) The authors found that T. pallidum might reduce the differentiation from subNPC1B to hindbrain neurons by inhibiting subNPC1B differentiation in brain organoids. Why were the subNPC1B-specific markers declining?

Thank you for your valuable comment. scRNA-seq is aimed at complete brain organoids. Cluster analysis of cell types of organoids is performed according to specific marker genes of different cells. The decrease in the expression of marker genes of certain cell groups indicates that the cell proportion of such cell groups in the whole organoids is reduced. We analysed organoids following T. pallidum infection, uniform manifold approximation and projection (UMAP), and clustering of the NPC1 population demonstrated that T. pallidum downregulated the number of subNPC1B population. Therefore, the results demonstrated a decrease in the subNPC1B -specific markers.

(5) In comparison to the other figures, Figure 5E letter size is excessively small and ambiguous.

Thanks for your valuable comments, I have adjusted Figure 5E letter size.

(6) Figure 5E shows that TCF3, more than one gene, is specifically enriched in subNPC1B of the T. pallidum group. It is best to confirm the impact of the other gene. 

Thank you for raising this key issue that we had not addressed properly in our previous version of the manuscript; we have added further analytical data. The SCENIC analysis found that the transcriptional activity of 52 genes has significantly changed after T. pallidum infection. Furthermore, GO analyses demonstrated that 27 transcription factors were significantly enriched in four key pathways of neural differentiation and development. TCF3 is the sole transcription factor present in all four terms simultaneously, speculating that TCF3 is the key transcription factor for the inhibition of subNPC1B-hindbrain neuron differentiation caused by T. pallidum.

Revision in the “Results” section, line 261-273:

“Next, the single-cell regulatory network inference and clustering (SCENIC) analysis for the subNPC1B subcluster was performed to assess the differences in the transcriptional activity of the transcription factors between the two groups and found that the transcriptional activity of 52 genes significantly changed after T. pallidum infection (Figure 5E). Furthermore, GO analyses demonstrated that 27 transcription factors were significantly enriched in key pathways of neural differentiation and development in response to nervous system development, positive regulation of sequence-specific DNA-binding transcription factor activity, positive regulation of neuronal differentiation, and DNA templated transcription regulation. Remarkably, transcription factor 3 (TCF3) is the sole transcription factor present in all four terms simultaneously (Figure 5F), speculating that TCF3 is the key transcription factor for the inhibition of subNPC1B-hindbrain neuron differentiation caused by T. pallidum.”

Revision in the “Materials and methods” section, line 540-543:

“The Sankey diagram was created using SankeyMATIC (https://sankeymatic.com/) (Zhang et al., 2023), which was used to characterize the interactions between differential transcription factors and neural differentiation and development.”

Revision in the “Figure and Figure Legend” section, line 832, 842-844:

Author response image 2.

Sankey diagram showing the correspondence between differential transcription factors and neural differentiation and development.

(7) Are there other experiments demonstrating that TCF3 is a key transcription factor for the inhibition of subNPC1B-hindbrain neuron differentiation caused by T. pallidum? 

Thank you for your valuable comment. In the previous experiment, we attempted to select a subNPC1B subcluster by flow sorting to verify the relevant molecular mechanism. Due to the small proportion of subNPC1B subcluster in the whole organoids, the selected cells were in a poor state and could not reach the number of cells required for the experiment. However, we used scRNA-seq data to further identify TCF3 as a key transcription factor that inhibits subNPC1B - hindbrain neuron differentiation induced by T. pallidum. The relevant results and descriptions of the analysis are detailed in the revised manuscript, please see our response to point (6) above.

eLife assessment

This is a valuable study that describes the effects of T. pallidum on neural development by applying single-cell RNA sequencing to an iPSC-derived brain organoid model. The evidence supporting the claims of the authors is solid, although further evidence to understand the differences in infection rates would strengthen the conclusions of the study. In particular, the conclusions would be strengthened by validating infection efficiency as this can impact the interpretation of single-cell sequencing results, and how these metrics affect organoid size as well as comparison with additional infectious agents. Furthermore, additional functional validations of downstream effectors could be insightful.

Reviewer #1 (Public Review):

Summary:

This is an interesting study by Xu et al showing the effects of infection with the Treponema pallidum virus (which causes syphilis disease) on neuronal development using iPSC-derived human brain organoids as a model and single-cell RNA sequencing. This work provides an important insight into the impact of the virus on human development, bridging the gap between the phenomena observed in studies using animal models as well as non-invasive human studies showing developmental abnormalities in fetuses infected with the virus in utero through maternal vertical transmission.

Using single-cell RNAseq in combination with qPCR and immunofluorescence techniques, the authors show that T. pallidum infected organoids are smaller in size, in particular during later growth stages, contain a larger number of undifferentiated neuronal lineage cells, and exhibit decreased numbers of specific neuronal subcluster, which the authors have identified as undifferentiated hindbrain neurons.

The study is an important first step in understanding how T. pallidum affects human neuronal development and provides important insight into the potential mechanisms that underlie the neurodevelopmental abnormalities observed in infected human fetuses.

Strengths:

(1) The study is well written, and the data quality is good for the most part.

(2) The study provides an important first step in utilizing human brain organoids to study the impact of T. pallidum infection on neuronal development.

(3) The study's conclusions may provide important insight to other researchers focused on studying how viral infections impact neuronal development.

Reviewer #3 (Public Review):

This article is the first report to study the effects of T. pallidum on the neural development of an iSPC-derived brain organoid model. The study indicates that T. pallidum inhibits the differentiation of subNPC1B neurons into hindbrain neurons, hence affecting brain organoid neurodevelopment. Additionally, the TCF3 and notch signaling pathways may be involved in the inhibition of the subNPC1B-hindbrain neuron differentiation axis. While the majority of the data in this study support the conclusions, there are still some questions that need to be addressed and data quality needs to be improved. The study provides valuable insights for future investigations into the mechanisms underlying congenital neurodevelopment disability.

Author response:

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

Reviewer #1 (Public Review):

Summary:

The authors use the innovative CRISPRi method to uncover regulators of cell density and volume in neutrophils. The results show that cells require NHE activity during chemoattractant-driven cell migration. Before migration occurs, cells also undergo a rapid cell volume increase. These results indicate that water flux, driven by ion channels, appears to play a central role in neutrophil migration. The paper is very well written and clear. I suggest adding some discussion about the role of actin in the process, but this is not essential.

Strengths

The novel use of CRIPSPi to uncover cell density regulators is very novel. Some of the uncovered molecules were known before, e.g. discussed in Li & Sun, Frontiers in Cell and Developmental Biology, 2021. Others are more interesting, for example PI3K-gamma. The use of caged fMLP is also nice.

We thank the reviewer for their positive appraisal of our work and have pursued their suggestions for improving our paper in this revision.

Weaknesses

One area of investigation that seems to be absent is mentioned in the introduction. I.e., actin is expected to play a role in regulating cell volume increase. Did the authors perform any experiments with LatA? What was seen there? Do cells still migrate with LatA, or is a different interplay seen? The role of PI3K is interesting, and maybe somewhat related to actin. But this may be a different line of inquiry for the future.

We agree that we could have done a better job explicitly investigating the role of actin dynamics in volume changes. Towards this end, by using Latrunculin B to depolymerize actin, we find that the volume increase in suspension is not affected (Figure 1 – supplemental figure 2A). In our FxM single cell volume measurements of adherent cells, we similarly observed unhindered swelling following latrunculin treatment. These data indicate that actin is dispensable for chemoattractant-induced cell swelling (Figure 1 – supplemental figure 2B) . There was a minor apparent reduction in the final volume reached with the Latrunculin-treated cells as measured by FxM, but this likely reflects minor uptake of the excluded dye following Latrunculin treatment rather than an actual change in final volume. This conclusion is reinforced by the change in 2D footprint area being well modeled by the 2D projection of an isotropically expanding sphere (Figure 1 – supplemental figure 2C) . Latrunculin treatment completely abolishes migration, as is expected for unconfined migration on fibronectin (Figure 1 – supplemental figure 2D-E) . The second Reviewer also wanted us to dig deeper on the role of PI3K-gamma, so we expanded our analysis of this hit (Figure 3 – supplemental figure 1B-D; Figure 4 – supplemental figure 1D-G) .

Author response image 1.

Chemoattractant-induced swelling, but not motility, is independent of actin polymerization. (A) Human primary neutrophils were incubated with DMSO or Latrunculin B, activated with 20 nM fMLP, and then volume responses were measured using electronic sizing via a Coulter counter. Latrunculin treatment did not alter cell swelling, indicating that actin polymerization is dispensable for the chemoattractant-induced volume increase. (B) Similar results were obtained using the FxM assay, showing that Latrunculin-treated cells are capable of swelling after stimulation. (C) The Latrunculin-treated cells also increase their footprints, albeit less so than control cells, but this is within the range of what would be expected for this degree of chemoattractant-induced volume increase (modeled by a sphere expanding an equivalent volume). (D) Single cell tracks of primary human neutrophils responding to acute chemoattractant stimulation. Both panels show 15 minutes of tracks with the tracks prior (left) and the 15 minutes post (right) uncaging the chemoattractant. The scale bar is 50 microns. The top panels show the large increase in motility displayed by control cells, while the Latrunculin-treated cells (bottom panels) fail to move. (E) Latrunculin-treated cells consistently fail to move in response to chemoattractant-stimulation. (F) Representative single cell volume traces show that Latrunculin-treated cells (black) lack short-term volume fluctuations but persistently maintain an elevated volume following chemoattractant stimulation. Control cells (blue) exhibit short-term volume fluctuations. (G) The lack of short-term volume fluctuations following latrunculin treatment is borne out across the population, with the coefficient of variation in the volume for single cells (post-swelling) being dramatically lower in Latrunculin-treated cells, suggesting that these short term volume fluctuations depend on actin-based motility.

Author response image 2.

Additional validation of swelling screen hits. (A) Mixed WT and CRISPR KO dHL-60 populations post-stimulation show that CA2 (black) and PI3Ky (green) KO both fail to decrease their densities as much as the WT (cyan) population following chemoattractant stimulation. Cells with negative control guides (light gray) have normal volume responses. All tubes were fractionated and aligned on the fraction containing the median of the WT population. Negative values indicate a fraction with a higher density than WT. (B) To validate the perturbations to cell swelling observed with FxM, primary human neutrophils were stimulated in suspension, and their volumes were measured using a Coulter counter. 20 nM fMLP was added at the 0 minute mark. Shaded regions represent the 95% confidence intervals. (C) PI3Kγ inhibition blocks the chemoattractant-induced volume change in primary human neutrophils, as assayed by FxM. (D) PI3Kγ inhibition also blocked the chemoattractant-drive shape change in human primary neutrophils, as measured by the change in footprint area in FxM (E) The coefficient of variation in volume for control (cyan) and iNHE1 (gold) inhibited human primary neutrophils undergoing chemokinesis are comparable, suggesting that the volume fluctuations are unchanged in moving cells upon NHE1 and PI3Kγ inhibition despite the different baseline volumes.

Author response image 3.

Additional validation of motility phenotypes. (A-D) Single cell tracks of primary human neutrophils responding to acute chemoattractant stimulation. Both panels show tracks of cells 15 minutes prior (left) versus 15 minutes post (right) uncaging the chemoattractant. The scale bar is 50 microns. Color saturation indicates time with tracks progressing from gray to full color. (A) Control cells show a large increase in movement upon uncaging, (B) NHE1 inhibited cells also initiate movement but to a lesser degree, (C) hypo-osmotic shock rescues the NHE1 motility defect. (D) PI3Kγ leads to a large fraction of cells failing to initiate movement. (E) PI3Kγ inhibition showed near complete blockage of the chemoattractant-induced motility increase in primary human neutrophils. (F) Control neutrophils (blue) show an increased angular alignment upon stimulation as their motility becomes directional. NHE1-inhibition (gold, iNHE1) has very little effect on this process, while PI3Kγ inhibition (green) leads to a reduction in this alignment at the population level. (G) For the PI3Kγ inhibited cells that start migrating, the migration-induced volume fluctuations are comparable to iNHE1 and control cells. The top panel shows the track of a representative migrating PI3Kγ inhibited cell and the bottom panel, its corresponding volume normalized to the pre-stimulation volume. The scale bar is 50 microns.

 

Reviewer #2 (Public Review):

Nagy et al investigated the role of volume increase and swelling in neutrophils in response to the chemoattractant. Authors show that following chemoattractant response cells lose their volume slightly owing to the cell spreading phase and then have a relatively rapid increase in the cell volume that is concomitant with cell migration. The authors performed an impressive genome-wide CRISPR screen and buoyant density assay to identify the regulators of neutrophil swelling. This assay showed that stimulating cells with chemoattractant fMLP led to an increase in the cell volume that was abrogated with the FPR1 receptor knockout. The screen revealed a cascade that could potentially be involved in cell swelling including NHE1 (sodium-proton antiporter) and PI3K. NHE1 and PI3K are required for chemoattractant-induced swelling in human primary neutrophils. Authors also suggest slightly different functions of NHE1 and PI3K activity where PI3K is also required to maintain chemoattractant-induced cell shape changes. The authors convincingly show that chemoattractant-induced cell swelling is linked to cell migration and NHE1 is required for swelling at the later stages of swelling since the cells at the early point work on low-volume and low-velocity regime. Interestingly, the authors also show that lack of swelling in NHE1-inhibited cells could be rescued by mild hypo-osmotic swelling strengthening the argument that water influx followed chemoattractant stimulation is important for potentiation for migration.

The conclusions of this paper are mostly well supported by data and are pretty convincing, but some aspects of image acquisition and data analysis need to be clarified and extended.

We thank the reviewer for their positive appraisal of our work and pursued their suggestions for improving our paper in this revision.

Weaknesses

(1) It would really help if the authors could add the missing graph for the footprint area when cells are treated with Latranculin. Graph S1F for volume changes with Lat treatment should be compared with DMSO-treated controls.

We agree that the Latrunculin condition merits more thorough investigation. To this end, we compared the volume response of human primary neutrophils to chemoattractant addition for Latrunculin B treated cells versus DMSO controls in suspension and show that there is no difference in swelling (Figure 1 – supplemental figure 2A) . This is additionally confirmed with FxM measurements with a slight undershooting of the final volume likely due to minor uptake of the excluded dye by Latrunculin treated cells (Figure 1 – supplemental figure 2B) . We have also included the requested footprint area changes in the Latrunculin treated cells as compared to controls (Figure 1 – supplemental figure 2C) . The treated cell footprints increase much less than the controls, and this is likely due to a lack of active cell spreading in the Latrunculin treated cells. The increase in footprint area observed following latrunculin treatment is within the range of what would be expected for the 2D projection of an isotropically expanding sphere fitted to the Latrunculin volume data (salmon line).

Author response image 4.

Chemoattractant-induced swelling, but not motility, is independent of actin polymerization. (A) Human primary eutrophils were incubated with DMSO or Latrunculin B, activated with 20 nM fMLP, and then volume responses were measured using electronic sizing via a Coulter counter. Latrunculin treatment did not alter cell swelling, indicating that actin polymerization is dispensable for the chemoattractant-induced volume increase. (B) Similar results were obtained using the FxM assay, showing that Latrunculin-treated cells are capable of swelling after stimulation. (C) The Latrunculin-treated cells also increase their footprints, albeit less so than control cells, but this is within the range of what would be expected for this degree of chemoattractant-induced volume increase (modeled by a sphere expanding an equivalent volume).

(2) The authors show inhibition of NHE1 blocked cell swelling using Coulter counter, a similar experiment should be done with PI3K inhibitions especially since they see PI3K inhibition impact chemoattractant-induced cell shape change.

Good idea. PI3Ky inhibition led to a substantial reduction in the chemoattractant-driven swelling in suspension showing the critical role of PI3K in the swelling of human primary neutrophils (Figure 3 – supplemental figure 1B) .

Author response image 5.

Additional validation of swelling screen hits. (B) To validate the perturbations to cell swelling observed with FxM, primary human neutrophils were stimulated in suspension, and their volumes were measured using a Coulter counter. 20 nM fMLP was added at the 0 minute mark. Shaded regions represent the 95% confidence intervals.

(3) It would be more convincing visually if the authors could also include the movie of cell spreading (footprint) and then mobility with PI3K inhibition.

Included as suggested. We agree this is a more compelling way to present the data (Figure 4 – supplemental figure 1A-D,G)

Author response image 6.

Additional validation of motility phenotypes. (A-D) Single cell tracks of primary human neutrophils responding to acute chemoattractant stimulation. Both panels show tracks of cells 15 minutes prior (left) versus 15 minutes post (right) uncaging the chemoattractant. The scale bar is 50 microns. Color saturation indicates time with tracks progressing from gray to full color. (A) Control cells show a large increase in movement upon uncaging. (D) PI3Kγ leads to a large fraction of cells failing to initiate movement. (E) PI3Kγ inhibition showed near complete blockage of the chemoattractant-induced motility increase in primary human neutrophils. (G) For the PI3Kγ inhibited cells that start migrating, the migration-induced volume fluctuations are comparable to iNHE1 and control cells. The top panel shows the track of a representative migrating PI3Kγ inhibited cell and the bottom panel, its corresponding volume normalized to the pre-stimulation volume. The scale bar is 50 microns.

(4) It is not clear how cell spreading and later volume increase are linked to overall mobility of neutrophils. Are authors suggesting that cell spreading is not required for cell mobility in neutrophils?

We did not mean to imply that cell spreading is not required for neutrophil motility. We take advantage of the fact that we can inhibit cell swelling without inhibiting spreading to investigate the specific role of swelling on migration ( Figure 4) . Conversely, cell spreading on a substrate is not required for chemoattractant-induced cell swelling, as chemoattractant-induced swelling occurs in latrunculin-treated cells (Figure 1 – supplemental figure 2A-C) . However, these latrunculin-treated cells are not able to migrate, at least not in the context studied here (Figure 1 – supplemental figure 2 D-E) . Cell spreading and swelling are likely both critical contributors to neutrophil motility, but their relative importance is dependent on the migratory context. The single cell volume fluctuation analysis indicates that migration-associated spreading and shape changes have large impacts on cell volume ( Figure 1 F) . These fluctuations are asynchronous, obscuring their observation at the population level, but the single cell traces clearly demonstrate them and their correlation with movement.

( 5) Volume fluctuations associated with motility were impacted by NHE1 inhibition at the baselines, what about PI3K inhibitions? Does that impact the actual fluctuations?

PI3K inhibition causes a significant fraction of cells to stop migrating (Figure 4 – supplemental figure 1D) , but among those that do move, they are still able to fluctuate in volume (Figure 4 – supplemental figure 1G) .

Author response image 7.

Additional validation of motility phenotypes. (G) For the PI3Kγ inhibited cells that start migrating, the migration-induced volume fluctuations are comparable to iNHE1 and control cells. The top panel shows the track of a representative migrating PI3Kγ inhibited cell and the bottom panel, its corresponding volume normalized to the pre-stimulation volume. The scale bar is 50 microns.

In contrast, latrunculin abolishes the volume fluctuations that normally accompany migration (Figure 1 – supplemental figure 2F-G) . These data suggest that movement/spreading itself is the driver of the rapid volume fluctuations. In contrast, the sustained volume increase following chemoattractant stimulation is independent of shape change and still occurs in latrunculin-treated cells.

Author response image 8.

Chemoattractant-induced swelling, but not motility, is independent of actin polymerization. (F) Representative single cell volume traces show that Latrunculin-treated cells (black) lack short-term volume fluctuations but persistently maintain an elevated volume following chemoattractant stimulation. Control cells (blue) exhibit short-term volume fluctuations. (G) The lack of short-term volume fluctuations following latrunculin treatment is borne out across the population, with the coefficient of variation in the volume for single cells (post-swelling) being dramatically lower in Latrunculin-treated cells, suggesting that these short term volume fluctuations depend on actin-based motility.

(6) It would really help if the authors compared similar analyses and drew conclusions from that, for example, it is unclear what the authors mean by they found no change in the angular persistence of WT and NHE1 inhibited cells which is in contrast to PI3K inhibition since they do not really have an analysis for angular persistence in PI3K inhibited cells. (S4A and S4B).

Thanks for catching this oversight in these experiments that we previously performed but neglected to include in the initial submission. We now include plots for angular persistence, velocity, and footprint size for the PI3K-gamma-inhibited cells. The results show that PI3K-gamma inhibition interferes both with swelling (Figure 3 – supplemental figure 1B-D) and motility (Figure 4 – supplemental figure 1D-F) , which aligns with its role upstream of the other hits identified in our screen.

Author response image 9.

Additional validation of motility phenotypes. (A-D) Single cell tracks of primary human neutrophils responding to acute chemoattractant stimulation. Both panels show tracks of cells 15 minutes prior (left) versus 15 minutes post (right) uncaging the chemoattractant. The scale bar is 50 microns. Color saturation indicates time with tracks progressing from gray to full color. (A) Control cells show a large increase in movement upon uncaging, (B) NHE1 inhibited cells also initiate movement but to a lesser degree, (C) hypo-osmotic shock rescues the NHE1 motility defect. (D) PI3Kγ leads to a large fraction of cells failing to initiate movement. (E) PI3Kγ inhibition showed near complete blockage of the chemoattractant-induced motility increase in primary human neutrophils. (F) Control neutrophils (blue) show an increased angular alignment upon stimulation as their motility becomes directional. NHE1-inhibition (gold, iNHE1) has very little effect on this process, while PI3Kγ inhibition (green) leads to a reduction in this alignment at the population level. (G) For the PI3Kγ inhibited cells that start migrating, the migration-induced volume fluctuations are comparable to iNHE1 and control cells. The top panel shows the track of a representative migrating PI3Kγ inhibited cell and the bottom panel, its corresponding volume normalized to the pre-stimulation volume. The scale bar is 50 microns.

eLife assessment

This fundamental study significantly advances our understanding of the role of water influx and swelling on neutrophil migration in response to chemoattractant. The evidence supporting the conclusions, based on a genome-wide CRISPR screen and high quality cellular observations, is compelling. This paper will be of interest to cell biologists and biophysicists working on cell migration.

Reviewer #1 (Public Review):

Summary:

The authors use innovative CRISPRi method to uncover regulators of cell density and volume in neutrophils. The results show that cells require NHE activity during chemoattractant-driven cell migration. Before migration occurs, cells also undergo a rapid cell volume increase. These results indicate that water flux, driven by ion channels, appears to play a central role in neutrophil migration. The paper is very well written and clear. The revised version has addressed all of my questions.

Reviewer #2 (Public Review):

Nagy et al investigated the role of volume increase and swelling in neutrophils in response to the chemoattractant. Authors show that following chemoattractant response cells lose their volume slightly owing to the cell spreading phase and then have a relatively rapid increase in the cell volume that is concomitant with cell migration. Authors performed an impressive genome-wide CRISPR screen and buoyant density assay to identify the regulators of neutrophil swelling. This assay showed that stimulating cells with chemoattractant fMLP lead to an increase in the cell volume that was abrogated with the FPR1 receptor knockout. The screen revealed a cascade that could potentially be involved cell swelling including NHE1 (sodium-proton antiporter) and PI3K. NHE1 and PI3K is required for chemoattractant-induced swelling in human primary neutrophils. Authors also suggest slightly different functions of NHE1 and PI3K activity where PI3K is also required for maintain chemoattractant-induced cell shape changes. Authors convincingly show that chemoattractant induced cell swelling is linked to cell migration and NHE1 is required for swelling at the later stages of swelling since the cells at the early point work on low-volume and low-velocity regime. Interesting authors also show that lack of swelling in NHE1 inhibited cells could be rescued by mild hypo-osmotic swelling strengthening the argument that water influx followed chemoattractant stimulation is important for potentiation for migration.

The conclusions of this paper are mostly well supported by data and is pretty convincing

Reviewer #1 (Public Review):

The revised manuscript "Diffusive lensing as a mechanism of intracellular transport and compartmentalization" is very similar to the original manuscript. The main difference between the revised and the original manuscript is that the authors have removed the reference to viscosity gradient and instead talk of diffusivity gradient. With this change the manuscript the analysis and claims in the manuscript are much more aligned. The manuscript, as the original version, explores the role of spatially varying diffusion constant in three scenarios:

(i) Spatial localization of non-particles<br /> (ii) Clustering in presence of inter-particle interactions<br /> (iii) Moment analysis for non-interacting particles in space with discrete patches of inhomogeneous diffusivity.

Since the manuscript has not changed much the strengths and weaknesses, in my opinion, remain similar to that of the original manuscript.

Strengths: The implications of a heterogeneous environment on phase separation and reaction kinetics in cells are under-explored. This makes the general theme of this manuscript relevant and interesting.

Weaknesses: The central part of the paper "diffusive lensing", i.e., particles localizing in the region of low diffusion constant is not new. Some of the papers authors cite already show that. The parts on phase separation and frap analysis that could provide new results are not rigorous enough for a theory paper.

I reiterate some of my comments from the original version that are valid for the revised version as well.

My main criticism was not to say that some convention should be used or some not. But instead, the main point was to say that just because there is spatial diffusion constant that does not mean there will be a spatial gradient of particles. From the authors response to my comments, it is clear that they understand the subtilties around it and are aware of the relevant papers. However, a reader not familiar with this discussion may work under the impression that if there if there is a spatialy varying diffusion constant in cell there will be an accumulation of particles in the region of low diffusivity but that may not always be the case. Moreover, localisation of particles in the region of low diffusivity has been reported in many different context. Some of the papers that the author cite already show that. For example, in Rupprecht et al. 2018 non-isothermal interpretation is applied to the dynamics of objects inside cells.

Given that the central result is not new. The paper could still be of general interest to the biophysics community if the follow up sections (ii) Clustering in presence of inter-particle interactions and (iii) Moment analysis for non-interacting particles in space with discrete patches of inhomogeneous diffusivity were analysed rigorously.

Reviewer #2 (Public Review):

Summary:

The authors study through theory and simulations the diffusion of microscopic particles, and aim to account for the effects of inhomogeneous viscosity and diffusion - in particular regarding the intracellular environment. They propose a mechanism, termed "Diffusive lensing", by which particles are attracted towards low-diffusivity regions where they remain trapped. To obtain these results, the authors rely on agent-based simulations using custom rules performed within the Ito stochastic calculus convention, without drift. They acknowledge the fact that this convention does not describe equilibrium systems, and that their results would not hold at equilibrium - and discard these facts by invoking the facts that cells are out-of-equilibrium. Finally, they show some applications of their findings, in particular enhanced clustering in the low-diffusivity regions. The authors conclude that as inhomogeneous diffusion is ubiquitous in life, so must their mechanism be, and hence it must be important.

Strengths:

The article is well-written, clearly intelligible, its hypotheses are stated relatively clearly and the models and mathematical derivations are compatible with these hypotheses. In the appendices, the authors connect their findings to known results for classic stochastic differential equation formalisms.

Weaknesses:

This study is, in my opinion, deeply flawed. The main problem lies in the hypotheses, in particular the choice of considering drift-less dynamics in the Ito convention. It is regrettable that the authors choose to use agent-based custom simulations with little physical motivation, rather than a well-established stochastic differential equations framework.

Indeed, stochastic conventions are a notoriously tricky business, but they are both mathematically and physically well-understood and do not result in any "dilemma" [some citations in the article, such as (Lau and Lubensky) and (Volpe and Wehr), make an unambiguous resolution of these]. In the continuous-time limit, conventions are not an intrinsic, fixed property of a system, but a choice of writing; however, whenever going from one to another, one must include a corresponding "spurious drift" that compensates the effect of this change - a mathematical subtlety that is omitted in the article (except in a quick note in the appendix): in the presence of diffusive gradients, if the drift is zero in one convention, it will thus be non-zero in another. It is well established that for equilibrium systems obeying fluctuation-dissipation, the spurious drift vanishes in the anti-Ito stochastic convention; more precisely one can write in the anti-Ito convention

dx/dt = - D(x)/kT grad U(x) + sqrt(2D(x)) dW

with D(x) the diffusion, kT the thermal energy (which is space-independent at equilibrium), and dW a d-dimensional Wiener process. Equivalently one can write in the Ito convention:

dx/dt = - D(x)/kT grad U(x) + sqrt(2D(x)) dW + div D(x) (*)

where the latter term is the spurious drift arising from convention change. This ensures that the diffusion gradients do not induce currents and probability gradients, and thus that the steady-state PDF is the Gibbs measure (this form has been confirmed experimentally, for instance, for colloidal particles near walls, that have strong diffusivity gradients despite not having significant forces). It generalizes to near-equilibrium systems with non-conservative forces and/or temperature gradient in the form:

dx/dt = F(x) + sqrt(2D(x)) dW + div D(x) (**)

where the drift field F(x) encodes these forces. In some cases, it has been shown through careful microscopic analysis that one can have effectively a different form for the last term, namely

dx/dt = F(x) + sqrt(2D(x)) dW + alpha div D(x)

where alpha is a "convention parameter" that would be =1 at equilibrium. For instance, in the Volpe and Wehr review this can occur through memory effects in robotic dynamics, or through strong fluctuation-dissipation breakdown. In a near-equilibrium system, this should be strongly justified, as the continuous-time dynamics with alpha \neq 1 and drift F would be indistinguishable from one with alpha = 1 and drift F + (1-alpha) div D: the authors would have the burden of proving that the observed (absence of) drift is indeed due to alpha\neq 1, rather than to much more common force fields F(x).

Here, without further motivation than the statement that cells are out-of-equilibrium, drifts are arbitrarily set to zero in the Ito convention, which is in (**) the equivalent to adding a force with drift $-div D$ exactly compensating the spurious drift. It is the effects of this arbitrary force that are studied in the article. The fact that it results in probability gradients is trivial once formulated this way (and in no way is this new - many of the references, for instance Volpe and Wehr, mention this). Enhanced clustering is also a trivial effect of this probability gradient (the local concentration is increased by this force field, so phase separation can occur). As a side note the "neighbor sensing" scheme to describe interactions is itself very peculiar and not physically motivated - it violates stochastic thermodynamics laws too, as detailed balance is apparently not respected. There again, the authors have chosen to disregard a century of stochastic thermodynamics in favor of a non-justified unphysical custom rule.

The authors make no further justification of their choice of driftless Ito simulations than the fact that cells are out-of-equilibrium, leaving the feeling that this is a detail. They make mentions of systems (eg glycogen, prebiotic environment) for which (near-)equilibrium physics should mostly prevail, and of fluctuation dissipation ("Diffusivity varies inversely with viscosity", in the introduction). Yet the "phenomenon" they discuss is entirely reliant on an undiscussed mechanism by which these assumptions would be completely violated (the citations they make for this - Gnesotto '18 and Phillips '12 - are simply discussions of the fact that cells are out-of-equilibrium, not on any consequences on the convention).

Finally, while inhomogeneous diffusion is ubiquitous, the strength of this effect in realistic conditions is not discussed. Even in the most "optimistic" case where alpha=0 would make sense (knowing that in the cellular context we are discussing thermal systems immersed in water and if energy consumption and metabolism were stopped alpha would relax back to 1), the equation (*) above shows that having zero ito drift is equivalent to having a potential countering the spurious drift, with value

U(x) = kT log(D(x) / D0 )

[I have assumed isotropic diffusion for simplicity here, so the div is replaced by a grad]. This means that the diffusion contrasts logarithmically compare to the chemical potential ones -- for instance a major diffusion difference of 100x is equivalent to 4.6kT in potential energy, a relatively modest effect. To prove that the authors' effect of "diffusive lensing" is involved in such a system, one would thus have to<br /> 1) observe strong spatial variations of the diffusion coefficient (this is doable, and was done before), AND<br /> 2) show that there is an enrichment of the diffusing species in the low-diffusion region inversely proportional to the diffusion, AND<br /> 3) show that this enrichment cannot be attributed to mild differences in potential energy, for instance by showing that if nonequilibrium energy consumption stops, the concentration fully homogenizes while the diffusion gradients remain.

If the authors were to successfully show all that in an experimental system, or design a theoretical framework where these effects convincingly emerge from physically realistic microscopic dynamical rules, they would have indeed discovered a new phenomenon. In contrast, the current article only demonstrates the well-known fact that when using arbitrary dynamical rules in heterogeneous diffusion simulations, one can get concentration gradients.

Author response:

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

eLife assessment

The authors discuss an effect, "diffusive lensing", by which particles would accumulate in high-viscosity regions, for instance in the intracellular medium. To obtain these results, the authors rely on agent-based simulations using custom rules performed with the Ito stochastic calculus convention. The "lensing effect" discussed is a direct consequence of the choice of the Ito convention without spurious drift which has been discussed before and is likely to be inadequate for the intracellular medium, causing the presented results to likely have little relevance for biology.

We thank the editors and the reviewers for their consideration of our manuscript. We argue in this rebuttal and revision that our results and conclusions are in fact likely to have relevance for biology. While we use the Itô convention for ease of modeling considering its non-anticipatory nature upon discretization (see (Volpe and Wehr 2016) for the discretization schemes), we refer to Figure S1B to emphasize that diffusive lensing occurs not only under the Itô convention but across a wide parameter space. Indeed, it is absent only in the normative isothermal convention; note that even a stochastic differential equation conforming to the isothermal convention may be reformulated into the Itô convention by adding suitable drift terms, allowing for diffusive lensing to be seen even in case of the isothermal convention. We note in particular that the choice of the convention is a highly context-dependent one (Sokolov 2010); there is not a universally correct choice, and one can obtain stochastic differential equations consistent with Ito or Stratonovich interpretations in different regimes. Lastly, space-dependent diffusivity is now an experimentally well-recognized feature of the cellular interior, as noted in our references and as discussed further later in this response. This fact points towards the potential relevance of our model for subcellular diffusion.

In our revised preprint, we have made changes to the text and minor changes to figures to address reviewer concerns.

Responses to the Reviewers

We thank the reviewers for their feedback and address the issues they raised in this rebuttal and in the revised manuscript. The central point that the reviewers raise concerns the validity of the drift-less Itô interpretation in modeling potential nonequilibrium types of subcellular transport arising from space-dependent diffusivity. If the drift term were considered, the resulting stochastic differential equation stochastic differential equation (SDE) is equivalent to one arising from the isothermal interpretation of heterogeneous diffusivity (Volpe and Wehr 2016), wherein no diffusive lensing is seen (as shown in Fig. S1B). That is, the isothermal interpretation and the drift-comprising Itô SDE produce the same uniform steady-state particle densities.

While we agree with the reviewers that for a given interpretation, equivalent stochastic differential equations (SDEs) arising from other interpretations may be drawn, we disagree with the generalization that all types of subcellular diffusion conform to the isothermal interpretation. That is, there is no reason why any and all instances of nonequilibrium subcellular particle diffusion must be modeled using isothermal-conforming SDEs (such as the drift-comprising Itô SDE, for instance). We refer to (Sokolov 2010) which prescribes choosing a convention in a context-dependent manner. In this regard, we disagree with the second reviewer’s characterization of making such a choice merely a “choice of writing” considering that it is entirely dependent on the choice of microscopic parameters, as detailed in the discussion section of the manuscript. The following references have also been added to the manuscript: the reference from the first reviewer (Kupferman et al. 2004) proposes a prescription for choosing an appropriate convention based upon comparing the noise correlation time and the particle relaxation time. The reference notes that the Itô convention is appropriate when the particle relaxation time is large when compared to the noise correlation time and the Stratonovich convention is appropriate in the converse scenario. In (Rupprecht et al. 2018), active noise is considered and the resulting Fokker-Planck equation conforms to the Stratonovich convention when thermal noise was negligible. The related reference, (Vishen et al. 2019) compares three timescales: those of particle relaxation, noise correlation and viscoelastic relaxation, to make the choice. Indeed, as noted in the manuscript, lensing is seen in all but one interpretation (without drift additions); only its magnitude is altered by the interpretation/choice of the drift term. The appendix has been modified to include a subsection on the interchangeability of the conventions.

Separately, with regards to the discussion on anomalous diffusion, the section on mean squared displacement calculation has been amended to avoid confusing our model with canonical anomalous diffusion which considers the anomalous exponent; how the anomalous exponent varies with space-dependent diffusivity offers an interesting future area of study.

Responses to specific reviewer comments appear below.

Reviewer #1 (Public Review):

The manuscript "Diffusive lensing as a mechanism of intracellular transport and compartmentalization", explores the implications of heterogeneous viscosity on the diffusive dynamics of particles. The authors analyze three different scenarios:

(i)   diffusion under a gradient of viscosity,

(ii)  clustering of interacting particles in a viscosity gradient, and

(iii) diffusive dynamics of non-interacting particles with circular patches of heterogeneous viscous medium.

The implications of a heterogeneous environment on phase separation and reaction kinetics in cells are under-explored. This makes the general theme of this manuscript very relevant and interesting. However, the analysis in the manuscript is not rigorous, and the claims in the abstract are not supported by the analysis in the main text.

Following are my main comments on the work presented in this manuscript:

(a) The central theme of this work is that spatially varying viscosity leads to position-dependent diffusion constant. This, for an overdamped Langevin dynamics with Gaussian white noise, leads to the well-known issue of the interpretation of the noise term.

The authors use the Ito interpretation of the noise term because their system is non-equilibrium.

One of the main criticisms I have is on this central point. The issue of interpretation arises only when there are ill-posed stochastic dynamics that do not have the relevant timescales required to analyze the noise term properly. Hence, if the authors want to start with an ill-posed equation it should be mentioned at the start. At least the Langevin dynamics considered should be explicitly mentioned in the main text. Since this work claims to be relevant to biological systems, it is also of significance to highlight the motivation for using the ill-posed equation rather than a well-posed equation. The authors refer to the non-equilibrium nature of the dynamics but it is not mentioned what non-equilibrium dynamics to authors have in mind. To properly analyze an overdamped Langevin dynamics a clear source of integrated timescales must be provided. As an example, one can write the dynamics as Eq. (1) \dot x = f(x) + g(x) \eta , which is ill-defined if the noise \eta is delta correlated in time but well-defined when \eta is exponentially correlated in time. One can of course look at the limit in which the exponential correlation goes to a delta correlation which leads to Eq. (1) interpreted in Stratonovich convention. The choice to use the Ito convention for Eq. (1) in this case is not justified.

We thank the reviewer for detailing their concerns with our model’s assumptions. We have addressed them in the common rebuttal.

(b) Generally, the manuscript talks of viscosity gradient but the equations deal with diffusion which is a combination of viscosity, temperature, particle size, and particle-medium interaction. There is no clear motivation provided for focus on viscosity (cytoplasm as such is a complex fluid) instead of just saying position-dependent diffusion constant. Maybe authors should use viscosity only when talking of a context where the existence of a viscosity gradient is established either in a real experiment or in a thought experiment.

The manuscript has been amended to use only “diffusivity” to avoid confusion.

(c) The section "Viscophoresis drives particle accumulation" seems to not have new results. Fig. 1 verifies the numerical code used to obtain the results in the later sections. If that is the case maybe this section can be moved to supplementary or at least it should be clearly stated that this is to establish the correctness of the simulation method. It would also be nice to comment a bit more on the choice of simulation methods with changing hopping sizes instead of, for example, numerically solving stochastic ODE.

The main point of this section and of Fig. 1 is the diffusive lensing effect itself: the accumulation of particles in lower-diffusivity areas. To the best of our knowledge, diffusive lensing has not been reported elsewhere as a specific outcome of non-isothermal interpretations of diffusion, with potential relevance to nonequilibrium subcellular motilities. The simulation method has been fully described in the Methods section, and the code has also been shared (see Code Availability).

A minor comment, the statement "the physically appropriate convention to use depends upon microscopic parameters and timescale hierarchies not captured in a coarse-grained model of diffusion." is not true as is noted in the references that authors mention, a correct coarse-grained model provides a suitable convention (see also Phys. Rev. E, 70(3), 036120., Phys. Rev. E, 100(6), 062602.).

This has been addressed in the common rebuttal.

(d) The section "Interaction-mediated clustering is affected by viscophoresis" makes an interesting statement about the positioning of clusters by a viscous gradient. As a theoretical calculation, the interplay between position-dependent diffusivity and phase separation is indeed interesting, but the problem needs more analysis than that offered in this manuscript. Just a plot showing clustering with and without a gradient of diffusion does not give enough insight into the interplay between density-dependent diffusion and position-dependent diffusion. A phase plot that somehow shows the relative contribution of the two effects would have been nice. Also, it should be emphasized in the main text that the inter-particle interaction is through a density-dependent diffusion constant and not a conservative coupling by an interaction potential.

The density-dependence has been added from the Methods to the main text. The goal of the work is to present lensing as a natural outcome of the parameter choices we make and present its effects as they relate to clustering and commonly used biophysical methods to probe dynamics within cells. A dense sampling of the phase space and how it is altered as a function of diffusivity, and the subsequent interpretation, lie beyond the scope of the present work but offer exciting future directions of study.

(e) The section "In silico microrheology shows that viscophoresis manifests as anomalous diffusion" the authors show that the MSD with and without spatial heterogeneity is different. This is not a surprise - as the underlying equations are different the MSD should be different.

The goal here is to compare and contrast the ways in which homogeneous and heterogeneous diffusion manifest in simulated microrheology measurements. We hope that an altered saturation MSD, as is observed in our simulations, provokes interest in considering lensing while modeling experimental data.

There are various analogies drawn in this section without any justification:

(i) "the saturation MSD was higher than what was seen in the homogeneous diffusion scenario possibly due to particles robustly populating the bulk milieu followed by directed motion into the viscous zone (similar to that of a Brownian ratchet, (Peskin et al., 1993))."

In case of i), the Brownian ratchet is invoked as a model to explain directed accumulation. We have removed this analogy to avoid confusion as it is not delved into further over the course of our work.

(ii) "Note that lensing may cause particle displacements to deviate from a Gaussian distribution, which could explain anomalous behaviors observed both in our simulations and in experiments in cells (Parry et al., 2014)." Since the full trajectory of the particles is available, it can be analyzed to check if this is indeed the case.

This has been addressed in the common rebuttal.

(f) The final section "In silico FRAP in a heterogeneously viscous environment ... " studies the MSD of the particles in a medium with heterogeneous viscous patches which I find the most novel section of the work. As with the section on inter-particle interaction, this needs further analysis.

We thank the reviewer for their appreciation. In presenting these three sections discussing the effects of diffusive lensing, we intend to broadly outline the scope of this phenomenon in influencing a range of behaviors. Exploring the directions further comprise promising future directions of study that lie beyond the scope of this manuscript.

To summarise, as this is a theory paper, just showing MSD or in silico FRAP data is not sufficient. Unlike experiments where one is trying to understand the systems, here one has full access to the dynamics either analytically or in simulation. So just stating that the MSD in heterogeneous and homogeneous environments are not the same is not sufficient. With further analysis, this work can be of theoretical interest. Finally, just as a matter of personal taste, I am not in favor of the analogy with optical lensing. I don't see the connection.

We value the reviewer’s interest in investigating the causes underlying the differences in the MSDs and agree that it represents a promising future area of study. The main point of this section of the manuscript was to make a connection to experimentally measurable quantities.

Reviewer #2 (Public Review):

Summary:

The authors study through theory and simulations the diffusion of microscopic particles and aim to account for the effects of inhomogeneous viscosity and diffusion - in particular regarding the intracellular environment. They propose a mechanism, termed "Diffusive lensing", by which particles are attracted towards high-viscosity regions where they remain trapped. To obtain these results, the authors rely on agent-based simulations using custom rules performed with the Ito stochastic calculus convention, without spurious drift. They acknowledge the fact that this convention does not describe equilibrium systems, and that their results would not hold at equilibrium - and discard these facts by invoking the fact that cells are out-of-equilibrium. Finally, they show some applications of their findings, in particular enhanced clustering in the high-viscosity regions. The authors conclude that as inhomogeneous diffusion is ubiquitous in life, so must their mechanism be, and hence it must be important.

Strengths:

The article is well-written, and clearly intelligible, its hypotheses are stated relatively clearly and the models and mathematical derivations are compatible with these hypotheses.

We thank the reviewer for their appreciation.

Weaknesses:

The main problem of the paper is these hypotheses. Indeed, it all relies on the Ito interpretation of the stochastic integrals. Stochastic conventions are a notoriously tricky business, but they are both mathematically and physically well-understood and do not result in any "dilemma" [some citations in the article, such as (Lau and Lubensky) and (Volpe and Wehr), make an unambiguous resolution of these]. Conventions are not an intrinsic, fixed property of a system, but a choice of writing; however, whenever going from one to another, one must include a "spurious drift" that compensates for the effect of this change - a mathematical subtlety that is entirely omitted in the article: if the drift is zero in one convention, it will thus be non-zero in another in the presence of diffusive gradients. It is well established that for equilibrium systems obeying fluctuation-dissipation, the spurious drift vanishes in the anti-Ito stochastic convention (which is not "anticipatory", contrarily to claims in the article, are the "steps" are local and infinitesimal). This ensures that the diffusion gradients do not induce currents and probability gradients, and thus that the steady-state PDF is the Gibbs measure. This equilibrium case should be seen as the default: a thermal system NOT obeying this law should warrant a strong justification (for instance in the Volpe and Wehr review this can occur through memory effects in robotic dynamics, or through strong fluctuation-dissipation breakdown). In near-equilibrium thermal systems such as the intracellular medium (where, although out-of-equilibrium, temperature remains a relevant and mostly homogeneous quantity), deviations from this behavior must be physically justified and go to zero when going towards equilibrium.

Considering that the physical phenomena underlying diffusion span a range of timescales (particle relaxation, noise, environmental correlation, et cetera), we disagree with the assertion that all types of subcellular diffusion processes can be modeled as occurring at thermal equilibrium: for example, one can easily imagine memory effects arising in the presence of an appropriate hierarchy of timescales. We have added references that describe in more detail the way in which the comparison of timescales can dictate the applicability of different conventions. We also refer the referee to the common rebuttal section of our response in which we discuss factors that govern the choice of the interpretation. The adiabatic elimination arguments highlighted in (Kupferman et al. 2004) provide a clear description of how relevant particle and environment-related timescales can inform the choice of stochastic calculus to use.

With regards to the use of the term “anticipatory” to refer to the isothermal interpretation, we refer to the comment in (Volpe and Wehr 2016) of the Itô interpretation “not looking into the future”. In any case, whether anticipatory or otherwise, the interpretation’s effect on our model remains unchanged, as highlighted in the section in the Appendix on the conversion between different conventions; this section has been added to minimize confusion about the effects of the choice of convention on lensing.

Here, drifts are arbitrarily set to zero in the Ito convention (the exact opposite of the equilibrium anti-Ito), which is the equilibrium equivalent to adding a force (with drift $- grad D$) exactly compensating the spurious drift. If we were to interpret this as a breakdown of detailed balance with inhomogeneous temperature, the "hot" region would be effectively at 4x higher temperature than the cold region (i.e. 1200K) in Fig 1A.

Our work is based on existing observations of space-dependent diffusivity in cells (Garner et al., 2023; Huang et al., 2021; Parry et al., 2014; Śmigiel et al., 2022; Xiang et al., 2020). These papers support a definitive model for the existence of space-dependent diffusivity without invoking space-dependent temperature.

It is the effects of this arbitrary force (exactly compensating the Ito spurious drift) that are studied in the article. The fact that it results in probability gradients is trivial once formulated this way (and in no way is this new - many of the references, for instance, Volpe and Wehr, mention this).

Addressed in the common rebuttal.

Enhanced clustering is also a trivial effect of this probability gradient (the local concentration is increased by this force field, so phase separation can occur). As a side note the "neighbor sensing" scheme to describe interactions is very peculiar and not physically motivated - it violates stochastic thermodynamics laws too, as the detailed balance is apparently not respected.

The neighbor-sensing scheme used here is just one possible model of an effective attractive potential between particles. Other models that lead to density-dependent attraction between particles should also provide qualitatively similar results as ours; this offers an interesting prospect for future research.

Finally, the "anomalous diffusion" discussion is at odds with what the literature on this subject considers anomalous (the exponent does not appear anomalous).

This has been addressed in the common rebuttal, and the relevant part of the manuscript has been modified to avoid confusion.

The authors make no further justification of their choice of convention than the fact that cells are out-of-equilibrium, leaving the feeling that this is a detail. They make mentions of systems (eg glycogen, prebiotic environment) for which (near-)equilibrium physics should mostly prevail, and of fluctuation-dissipation ("Diffusivity varies inversely with viscosity", in the introduction). Yet the "phenomenon" they discuss is entirely reliant on an undiscussed mechanism by which these assumptions would be completely violated (the citations they make for this - Gnesotto '18 and Phillips '12 - are simply discussions of the fact that cells are out-of-equilibrium, not on any consequences on the convention).

Finally, while inhomogeneous diffusion is ubiquitous, the strength of this effect in realistic conditions is not discussed (this would be a significant problem if the effect were real, which it isn't). Gravitational attraction is also an ubiquitous effect, but it is not important for intracellular compartmentalization.

The manuscript text has been supplemented with additional references that detail the ways in which the comparison of timescales can dictate how one can apply different conventions. We refer the reviewer to the common rebuttal section of our response where we detail factors that dictate the choice of the convention to use. As previously noted, the adiabatic elimination arguments highlighted in (Kupferman et al., 2004) provide a prescription for how different timescales are to be considered in deciding the choice of stochastic calculus to use.

With regards to the strength of space-dependent diffusivity in subcellular milieu, various measurements of heterogeneous diffusivity have been made both across different model systems and via different modalities, as cited in our manuscript. (Garner et al. 2023) used single-particle tracking to determine over 100-fold variability in diffusivity within individual S. pombe cells. Single-molecule measurements in (Xiang et al. 2020) and (Śmigiel et al. 2022) reveal an order-of-magnitude variation in tracer diffusion in mammalian cells and multi-fold variation in E. coli cytoplasm respectively. Fluorescence correlation spectroscopy measurements in (Huang et al. 2022) have found a two-fold increase in short-range diffusion of protein-sized tracers in X. laevis extracts. We have also added a reference to a study that uses 3D single particle tracking in the cytosol of a multinucleate fungus, A. gossypii, to identify regions of low-diffusivity near nuclei and hyphal tips (McLaughlin et al. 2020). Many of these references deploy particle tracking and investigate how mesoscale-sized particles (i.e. tracers spanning biologically relevant size scales) are directly impacted by space-dependent diffusivity. Therefore, we base our model on not only space-dependent diffusivity being a well-recognized feature of the cellular interior, but also on these observations pertaining to mesoscale-sized particles’ motion along relevant timescales.

These measurements are also relevant to the reviewer’s question about the strength of the effect, which depends directly on the variability in diffusivity: for ten- or a hundred-fold diffusivity variations, the effect would be expected to be significant. In case of using the Itô convention directly, the contrast in concentration gradient is, in fact, that of the diffusivity gradient.

To conclude, the "diffusive lensing" effect presented here is not a deep physical discovery, but a well-known effect of sticking to the wrong stochastic convention.

As detailed in the various responses above, we respectfully disagree with the notion that there exists a singular correct stochastic convention that is applicable for all cases of subcellular heterogeneous diffusion. Further, as detailed in (Volpe and Wehr 2016) and as detailed in the Appendix, it is possible to convert between conventions and that an isothermal-abiding stochastic differential equation may be suitably altered, by means of adding a drift term, to an Itô-abiding stochastic differential equation; therefore, one can observe diffusive lensing without discarding the isothermal convention if the latter were modified. Indeed, it is only the driftless (or canonical) isothermal convention that does not allow for diffusive lensing.

eLife assessment

This fundamental study reports differential expression of key genes in full-term placenta between Tibetans and Han Chinese at high elevations, which are more pronounced in the placenta of male fetus than in female fetus. The gene expression data were collected and analyzed using solid and validated methodology, although there is limited support for hypoxia-specific responses due to a lack of low-altitude samples. Several of the placental genes found in this study have been previously reported to show signatures of positive selection in Tibetans, pointing to a potential mechanism of how human populations adapt to high elevation by mitigating the negative effects of low oxygen on fetal growth. The work will be of interest to evolutionary and population geneticists as well as researchers working on human hypoxic response.

Joint Public Review:

This manuscript by Yue et al. aims to understand the molecular mechanisms underlying the better reproductive outcomes of Tibetans at high altitude by characterizing the transcriptome and histology of full-term placenta of Tibetans and compare them to those Han Chinese at high elevations.

The approach is innovative, and the data collected are valuable for testing hypotheses regarding the contribution of the placenta to better reproductive success of populations that adapted to hypoxia. The authors identified hundreds of differentially expressed genes (DEGs) between Tibetans and Han, including the EPAS1 gene that harbors the strongest signals of genetic adaptation. The authors also found that such differential expression is more prevalent and pronounced in the placentas of male fetuses than those of female fetuses, which is particularly interesting, as it echoes with the more severe reduction in birth weight of male neonates at high elevation observed by the same group of researchers (He et al., 2022).

Comments on latest version:

The revised manuscript has incorporated the suggested changes and weakened conclusions regarding natural selection. Limitations of the study are also clearly stated in the Discussion section.

Author response:

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

Public Review:

This manuscript by Yue et al. aims to understand the molecular mechanisms underlying the better reproductive outcomes of Tibetans at high altitude by characterizing the transcriptome and histology of full-term placenta of Tibetans and compare them to those Han Chinese at high elevations.

The approach is innovative, and the data collected are valuable for testing hypotheses regarding the contribution of the placenta to better reproductive success of populations that adapted to hypoxia. The authors identified hundreds of differentially expressed genes (DEGs) between Tibetans and Han, including the EPAS1 gene that harbors the strongest signals of genetic adaptation. The authors also found that such differential expression is more prevalent and pronounced in the placentas of male fetuses than those of female fetuses, which is particularly interesting, as it echoes with the more severe reduction in birth weight of male neonates at high elevation observed by the same group of researchers (He et al., 2022).

This revised manuscript addressed several concerns raised by reviewers in last round. However, we still find the evidence for natural selection on the identified DEGs--as a group--to be very weak, despite more convincing evidence on a few individual genes, such as EPAS1 and EGLN1.

The authors first examined the overlap between DEGs and genes showing signals of positive selection in Tibetans and evaluated the significance of a larger overlap than expected with a permutation analysis. A minor issue related to this analysis is that the p-value is inflated, as the authors are counting permutation replicates with MORE genes in overlap than observed, yet the more appropriate way is counting replicates with EQUAL or MORE overlapping genes. Using the latter method of p-value calculation, the "sex-combined" and "female-only" DEGs will become non-significantly enriched in genes with evidence of selection, and the signal appears to solely come from male-specific DEGs. A thornier issue with this type of enrichment analysis is whether the condition on placental expression is sufficient, as other genomic or transcriptomic features (e.g., expression level, local sequence divergence level) may also confound the analysis.

According to the suggested methods, we counted the replicates with equal or more overlapping genes than observed (≥4 for the “combined” set; ≥9 for the “male-only” set; ≥0 for the “female-only” set). We found that the overlaps between DEGs and TSNGs were significantly enriched only in the “male-only” set (p-value < 1e-4, counting 0 time from 10,000 permutations), but not in the “female-only” set (p-value = 1, counting 10,000 time from 10,000 permutations), or “combined” set (p-value = 0.0603, counting 603 time from 10,000 permutations) (see Table R1 below).

We updated this information in the revised manuscript, including Results, Methods, and Figure S9.

Author response table 1.

Permutation analysis of the overlapped genes between DEGs and TSNGs.

The authors next aimed to detect polygenic signals of adaptation of gene expression by applying the PolyGraph method to eQTLs of genes expressed in the placenta (Racimo et al 2018). This approach is ambitious but problematic, as the method is designed for testing evidence of selection on single polygenic traits. The expression levels of different genes should be considered as "different traits" with differential impacts on downstream phenotypic traits (such as birth weight). As a result, the eQTLs of different genes cannot be naively aggregated in the calculation of the polygenic score, unless the authors have a specific, oversimplified hypothesis that the expression increase of all genes with identified eQTL will improve pregnancy outcome and that they are equally important to downstream phenotypes. In general, PolyGraph method is inapplicable to eQTL data, especially those of different genes (but see Colbran et al 2023 Genetics for an example where the polygenic score is used for testing selection on the expression of individual genes).

We would recommend removal of these analyses and focus on the discussion of individual genes with more compelling evidence of selection (e.g., EPAS1, EGLN1).

According to the suggestion, we removed these analyses in the revised manuscript.

eLife assessment

This study aggregates across five fMRI datasets and reports that a network of brain areas previously associated with response inhibition processes, including several in the basal ganglia, are more active on failed stop than successful stop trials. This study is valuable as a well-powered investigation of fMRI measures of stopping. However, evidence for the authors' conclusions regarding the role of subcortical nodes in stopping is incomplete, due to the limitations in the fMRI analysis.

Reviewer #1 (Public Review):

This study is one in a series of excellent papers by the Forstmann group focusing on the ability of fMRI to reliably detect activity in small subcortical nuclei - in this case, specifically those purportedly involved in the hyper- and indirect inhibitory basal ganglia pathways. I have been very fond of this work for a long time, beginning with the demonstration of De Hollander, Forstmann et al. (HBM 2017) of the fact that 3T fMRI imaging (as well as many 7T imaging sequences) do not afford sufficient signal to noise ratio to reliably image these small subcortical nuclei. This work has done a lot to reshape my view of seminal past studies of subcortical activity during inhibitory control, including some that have several thousand citations.

Comments on revised version:

This is my second review of this article, now entitled "Multi-study fMRI outlooks on subcortical BOLD responses in the stop-signal paradigm" by Isherwood and colleagues.

The authors have been very responsive to the initial round of reviews.

I still think it would be helpful to see a combined investigation of the available 7T data, just to really drive the point home that even with the best parameters and a multi-study sample size, fMRI cannot detect any increases in BOLD activity on successful stop compared to go trials. However, I agree with the authors that these "sub samples still lack the temporal resolution seemingly required for looking at the processes in the SST."

As such, I don't have any more feedback.

Reviewer #2 (Public Review):

This work aggregates data across 5 openly available stopping studies (3 at 7 tesla and 2 at 3 tesla) to evaluate activity patterns across the common contrasts of Failed Stop (FS) > Go, FS > stop success (SS), and SS > Go. Previous work has implicated a set of regions that tend to be positively active in one or more of these contrasts, including the bilateral inferior frontal gyrus, preSMA, and multiple basal ganglia structures. However, the authors argue that upon closer examination, many previous papers have not found subcortical structures to be more active on SS than FS trials, bringing into question whether they play an essential role in (successful) inhibition. In order to evaluate this with more data and power, the authors aggregate across five datasets and find many areas that are *more* active for FS than SS, including bilateral preSMA, GPE, thalamus, and VTA. They argue that this brings into question the role of these areas in inhibition, based upon the assumption that areas involved in inhibition should be more active on successful stop than failed stop trials, not the opposite as they observed.

Since the initial submission, the authors have improved their theoretical synthesis and changed their SSRT calculation method to the more appropriate integration method with replacement for go omissions. They have also done a better job of explaining how these fMRI results situate within the broader response inhibition literature including work using other neuroscience methods.

They have also included a new Bayes Factor analysis. In the process of evaluating this new analysis, I recognized the following comments that I believe justify additional analyses and discussion:

First, if I understand the author's pipeline, for the ROI analyses it is not appropriate to run FSL's FILM method on the data that were generated by repeating the same time series across all voxels of an ROI. FSL's FILM uses neighboring voxels in parts of the estimation to stabilize temporal correlation and variance estimates and was intended and evaluated for use on voxelwise data. Instead, I believe it would be more appropriate to average the level 1 contrast estimates over the voxels of each ROI to serve as the dependent variables in the ROI analysis.

Second, for the group-level ROI analyses there seems to be inconsistencies when comparing the z-statistics (Figure 3) to the Bayes Factors (Figure 4) in that very similar z-statistics have very different Bayes Factors within the same contrast across different brain areas, which seemed surprising (e.g., a z of 6.64 has a BF of .858 while another with a z of 6.76 has a BF of 3.18). The authors do briefly discuss some instances in the frequentist and Bayesian results differ, but they do not ever explain by similar z-stats yield very different bayes factors for a given contrast across different brain areas. I believe a discussion of this would be useful.

Third, since the Bayes Factor analysis appears to be based on repeated measures ANOVA and the z-statistics are from Flame1+2, the BayesFactor analysis model does not pair with the frequentist analysis model very cleanly. To facilitate comparison, I would recommend that the same repeated measures ANOVA model should be used in both cases. My reading of the literature is that there is no need to be concerned about any benefits of using Flame being lost, since heteroscedasticity does not impact type I errors and will only potentially impact power (Mumford & Nichols, 2009 NeuroImage).

Fourth, though frequentist statistics suggest that many basal ganglia structures are significantly more active in the FS > SS contrast (see 2nd row of Figure 3), the Bayesian analyses are much more equivocal, with no basal ganglia areas showing Log10BF > 1 (which would be indicative of strong evidence). The authors suggest that "the frequentist and Bayesian analyses are monst in line with one another", but in my view, this frequentist vs. Bayesian analysis for the FS > SS contrast seems to suggest substantially different conclusions. More specifically, the frequentist analyses suggest greater activity in FS than SS in most basal ganglia ROIs (all but 2), but the Bayesian analysis did not find *any* basal ganglia ROIs with strong evidence for the alternative hypothesis (or a difference), and several with more evidence for the null than the alternative hypothesis. This difference between the frequentist and Bayesian analyses seems to warrant discussion, but unless I overlooked it, the Bayesian analyses are not mentioned in the Discussion at all. In my view, the frequentist analyses are treated as the results, and the Bayesian analyses were largely ignored.

Overall, I think this paper makes a useful and mostly solid contribution to the literature. I have made some suggestions for adjustments and clarification of the neuroimaging pipeline and Bayesian analyses that I believe would strengthen the work further.

Author response:

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

Reviewer #1: 

This is my first review of the article entitled "The canonical stopping network: Revisiting the role of the subcortex in response inhibition" by Isherwood and colleagues. This study is one in a series of excellent papers by the Forstmann group focusing on the ability of fMRI to reliably detect activity in small subcortical nuclei - in this case, specifically those purportedly involved in the hyper- and indirect inhibitory basal ganglia pathways. I have been very fond of this work for a long time, beginning with the demonstration of De Hollander, Forstmann et al. (HBM 2017) of the fact that 3T fMRI imaging (as well as many 7T imaging sequences) do not afford sufficient signal to noise ratio to reliably image these small subcortical nuclei. This work has done a lot to reshape my view of seminal past studies of subcortical activity during inhibitory control, including some that have several thousand citations.

In the current study, the authors compiled five datasets that aimed to investigate neural activity associated with stopping an already initiated action, as operationalized in the classic stop-signal paradigm. Three of these datasets are taken from their own 7T investigations, and two are datasets from the Poldrack group, which used 3T fMRI.

The authors make six chief points: 

(1) There does not seem to be a measurable BOLD response in the purportedly critical subcortical areas in contrasts of successful stopping (SS) vs. going (GO), neither across datasets nor within each individual dataset. This includes the STN but also any other areas of the indirect and hyperdirect pathways.

(2) The failed-stop (FS) vs. GO contrast is the only contrast showing substantial differences in those nodes.

(3) The positive findings of STN (and other subcortical) activation during the SS vs. GO contrast could be due to the usage of inappropriate smoothing kernels.

(4) The study demonstrates the utility of aggregating publicly available fMRI data from similar cognitive tasks. 

(5) From the abstract: "The findings challenge previous functional magnetic resonance (fMRI) of the stop-signal task" 

(6) and further: "suggest the need to ascribe a separate function to these networks." 

I strongly and emphatically agree with points 1-5. However, I vehemently disagree with point 6, which appears to be the main thrust of the current paper, based on the discussion, abstract, and - not least - the title.

To me, this paper essentially shows that fMRI is ill-suited to study the subcortex in the specific context of the stop-signal task. That is not just because of the issues of subcortical small-volume SNR (the main topic of this and related works by this outstanding group), but also because of its limited temporal resolution (which is unacknowledged, but especially impactful in the context of the stop-signal task). I'll expand on what I mean in the following.

First, the authors are underrepresenting the non-fMRI evidence in favor of the involvement of the subthalamic nucleus (STN) and the basal ganglia more generally in stopping actions. 

- There are many more intracranial local field potential recording studies that show increased STN LFP (or even single-unit) activity in the SS vs. FS and SS vs. GO contrast than listed, which come from at least seven different labs. Here's a (likely non-exhaustive) list of studies that come to mind:

Ray et al., NeuroImage 2012 <br /> Alegre et al., Experimental Brain Research 2013 <br /> Benis et al., NeuroImage 2014 <br /> Wessel et al., Movement Disorders 2016 <br /> Benis et al., Cortex 2016 <br /> Fischer et al., eLife 2017 <br /> Ghahremani et al., Brain and Language 2018 <br /> Chen et al., Neuron 2020 <br /> Mosher et al., Neuron 2021 <br /> Diesburg et al., eLife 2021 

- Similarly, there is much more evidence than cited that causally influencing STN via deep-brain stimulation also influences action-stopping. Again, the following list is probably incomplete: 

Van den Wildenberg et al., JoCN 2006 <br /> Ray et al., Neuropsychologia 2009 <br /> Hershey et al., Brain 2010 <br /> Swann et al., JNeuro 2011 <br /> Mirabella et al., Cerebral Cortex 2012 <br /> Obeso et al., Exp. Brain Res. 2013 <br /> Georgiev et al., Exp Br Res 2016 <br /> Lofredi et al., Brain 2021 <br /> van den Wildenberg et al, Behav Brain Res 2021 <br /> Wessel et al., Current Biology 2022 

- Moreover, evidence from non-human animals similarly suggests critical STN involvement in action stopping, e.g.: 

Eagle et al., Cerebral Cortex 2008 <br /> Schmidt et al., Nature Neuroscience 2013 <br /> Fife et al., eLife 2017 <br /> Anderson et al., Brain Res 2020 

Together, studies like these provide either causal evidence for STN involvement via direct electrical stimulation of the nucleus or provide direct recordings of its local field potential activity during stopping. This is not to mention the extensive evidence for the involvement of the STN - and the indirect and hyperdirect pathways in general - in motor inhibition more broadly, perhaps best illustrated by their damage leading to (hemi)ballism. 

Hence, I cannot agree with the idea that the current set of findings "suggest the need to ascribe a separate function to these networks", as suggested in the abstract and further explicated in the discussion of the current paper. For this to be the case, we would need to disregard more than a decade's worth of direct recording studies of the STN in favor of a remote measurement of the BOLD response using (provably) sub ideal imaging parameters. There are myriads of explanations of why fMRI may not be able to reveal a potential ground-truth difference in STN activity between the SS and FS/GO conditions, beginning with the simple proposition that it may not afford sufficient SNR, or that perhaps subcortical BOLD is not tightly related to the type of neurophysiological activity that distinguishes these conditions (in the purported case of the stop-signal task, specifically the beta band). But essentially, this paper shows that a specific lens into subcortical activity is likely broken, but then also suggests dismissing existing evidence from superior lenses in favor of the findings from the 'broken' lens. That doesn't make much sense to me.

Second, there is actually another substantial reason why fMRI may indeed be unsuitable to study STN activity, specifically in the stop-signal paradigm: its limited time resolution. The sequence of subcortical processes on each specific trial type in the stop-signal task is purportedly as follows: at baseline, the basal ganglia exert inhibition on the motor system. During motor initiation, this inhibition is lifted via direct pathway innervation. This is when the three trial types start diverging. When actions then have to be rapidly cancelled (SS and FS), cortical regions signal to STN via the hyperdirect pathway that inhibition has to be rapidly reinstated (see Chen, Starr et al., Neuron 2020 for direct evidence for such a monosynaptic hyperdirect pathway, the speed of which directly predicts SSRT). Hence, inhibition is reinstated (too late in the case of FS trials, but early enough in SS trials, see recordings from the BG in Schmidt, Berke et al., Nature Neuroscience 2013; and Diesburg, Wessel et al., eLife 2021). 

Hence, according to this prevailing model, all three trial types involve a sequence of STN activation (initial inhibition), STN deactivation (disinhibition during GO), and STN reactivation (reinstantiation of inhibition during the response via the hyperdirect pathway on SS/FS trials, reinstantiation of inhibition via the indirect pathway after the response on GO trials). What distinguishes the trial types during this period is chiefly the relative timing of the inhibitory process (earliest on SS trials, slightly later on FS trials, latest on GO trials). However, these temporal differences play out on a level of hundreds of milliseconds, and in all three cases, processing concludes well under a second overall. To fMRI, given its limited time resolution, these activations are bound to look quite similar. 

Lastly, further building on this logic, it's not surprising that FS trials yield increased activity compared to SS and GO trials. That's because FS trials are errors, which are known to activate the STN (Cavanagh et al., JoCN 2014; Siegert et al. Cortex 2014) and afford additional inhibition of the motor system after their occurrence (Guan et al., JNeuro 2022). Again, fMRI will likely conflate this activity with the abovementioned sequence, resulting in a summation of activity and the highest level of BOLD for FS trials. 

In sum, I believe this study has a lot of merit in demonstrating that fMRI is ill-suited to study the subcortex during the SST, but I cannot agree that it warrants any reappreciation of the subcortex's role in stopping, which are not chiefly based on fMRI evidence. 

We would like to thank reviewer 1 for their insightful and helpful comments. We have responded point-by-point below and will give an overview of how we reframed the paper here.  

We agree that there is good evidence from other sources for the presence of the canonical stopping network (indirect and hyperdirect) during action cancellation, and that this should be reflected more in the paper. However, we do not believe that a lack of evidence for this network during the SST makes fMRI ill-suited for studying this task, or other tasks that have neural processes occurring in quick succession. What we believe the activation patterns of fMRI reflect during this task, is the large of amount of activation caused by failed stops. That is, that the role of the STN in error processing may be more pronounced that its role in action cancellation. Due to the replicability of fMRI results, especially at higher field strengths, we believe the activation profile of failed stop trials reflects a paramount role for the STN in error processing. Therefore, while we agree we do not provide evidence against the role of the STN in action cancellation, we do provide evidence that our outlook on subcortical activation during different trial types of this task should be revisited. We have reframed the article to reflect this, and discuss points such as fMRI reliability, validity and the complex overlapping of cognitive processes in the SST in the discussion. Please see all changes to the article indicated by red text.

A few other points: 

- As I said before, this team's previous work has done a lot to convince me that 3T fMRI is unsuitable to study the STN. As such, it would have been nice to see a combination of the subsamples of the study that DID use imaging protocols and field strengths suitable to actually study this node. This is especially true since the second 3T sample (and arguably, the Isherwood_7T sample) does not afford a lot of trials per subject, to begin with.

Unfortunately, this study already comprises of the only 7T open access datasets available for the SST. Therefore, unless we combined only the deHollander_7T and Miletic_7T subsamples there is no additional analysis we can do for this right now. While looking at just the sub samples that were 7T and had >300 trials would be interesting, based on the new framing of the paper we do not believe it adds to the study, as the sub samples still lack the temporal resolution seemingly required for looking at the processes in the SST.

- What was the GLM analysis time-locked to on SS and FS trials? The stop-signal or the GO-signal? 

SS and FS trials were time-locked to the GO signal as this is standard practice. The main reason for this is that we use contrasts to interpret differences in activation patterns between conditions. By time-locking the FS and SS trials to the stop signal, we are contrasting events at different time points, and therefore different stages of processing, which introduces its own sources of error. We agree with the reviewer, however, that a separate analysis with time-locking on the stop-signal has its own merit, and now include results in the supplementary material where the FS and SS trials are time-locked to the stop signal as well.

- Why was SSRT calculated using the outdated mean method? 

We originally calculated SSRT using the mean method as this was how it was reported in the oldest of the aggregated studies. We have now re-calculated the SSRTs using the integration method with go omission replacement and thank the reviewer for pointing this out. Please see response to comment 3.

- The authors chose 3.1 as a z-score to "ensure conservatism", but since they are essentially trying to prove the null hypothesis that there is no increased STN activity on SS trials, I would suggest erring on the side of a more lenient threshold to avoid type-2 error. 

We have used minimum FDR-corrected thresholds for each contrast now, instead of using a blanket conservative threshold of 3.1 over all contrasts. The new thresholds for each contrast are shown in text. Please see below (page 12):

“The thresholds for each contrast are as follows: 3.01 for FS > GO, 2.26 for FS > SS and 3.1 for SS > GO.”

- The authors state that "The results presented here add to a growing literature exposing inconsistencies in our understanding of the networks underlying successful response inhibition". It would be helpful if the authors cited these studies and what those inconsistencies are. 

We thank reviewer 1 for their detailed and thorough evaluation of our paper. Overall, we agree that there is substantial direct and indirect evidence for the involvement of the cortico-basal-ganglia pathways in response inhibition. We have taken the vast constructive criticism on board and agree with the reviewer that the paper should be reframed. We would like to thank the reviewer for the thoroughness of their helpful comments aiding the revising of the paper.

(1) I would suggest reframing the study, abstract, discussion, and title to reflect the fact that the study shows that fMRI is unsuitable to study subcortical activity in the SST, rather than the fact that we need to question the subcortical model of inhibition, given the reasons in my public review.

We agree with the reviewer that the article should be reframed and not taken as direct evidence against the large sum of literature pointing towards the involvement of the cortico-basal-ganglia pathway in response inhibition. We have significantly rewritten the article in light of this.

(2) I suggest combining the datasets that provide the best imaging parameters and then analyzing the subcortical ROIs with a more lenient threshold and with regressors time-locked to the stop-signals (if that's not already the case). This would make the claim of a null finding much more impactful. Some sort of power analysis and/or Bayes factor analysis of evidence for the null would also be appreciated. 

Instead of using a blanket conservative threshold of 3.1, we instead used only FDR-corrected thresholds. The threshold level is therefore different for each contrast and noted in the figures. We have also added supplementary figures including the group-level SPMs and ROI analyses when the FS and SS trials were time-locked to the stop signal instead of the GO signal (Supplementary Figs 4 & 5). But as mentioned above, due to the difference in time points when contrasting, we believe that time-locking to the GO signal for all trial types makes more sense for the main analysis.

We have now also computed BFs on the first level ROI beta estimates for all contrasts using the BayesFactor package as implemented in R. We add the following section to the methods and updated the results section accordingly (page 8):

“In addition to the frequentist analysis we also opted to compute Bayes Factors (BFs) for each contrast per ROI per hemisphere. To do this, we extracted the beta weights for each individual trial type from our first level model. We then compared the beta weights from each trial type to one another using the ‘BayesFactor’ package as implement in R (Morey & Rouder, 2015). We compared the full model comprising of trial type, dataset and subject as predictors to the null model comprising of only the dataset and subject as predictor. The datasets and subjects were modeled as random factors. We divided the resultant BFs from the full model by the null model to provide evidence for or against a significant difference in beta weights for each trial type. To interpret the BFs, we used a modified version of Jeffreys’ scale (Jeffreys, 1939; Lee & Wagenmakers, 2014).”

(3) I suggest calculating SSRT using the integration method with the replacement of Go omissions, as per the most recent recommendation (Verbruggen et al., eLife 2019).

We agree we should have used a more optimal method for SSRT estimation. We have replaced our original estimations with that of the integration method with go omissions replacement, as suggested and adapted the results in table 3.

We have also replaced text in the methods sections to reflect this (page 5):

“For each participant, the SSRT was calculated using the mean method, estimated by subtracting the mean SSD from median go RT (Aron & Poldrack, 2006; Logan & Cowan, 1984).”

Now reads:

“For each participant, the SSRT was calculated using the integration method with replacement of go omissions (Verbruggen et al., 2019), estimated by integrating the RT distribution and calculating the point at which the integral equals p(respond|signal). The completion time of the stop process aligns with the nth RT, where n equals the number of RTs in the RT distribution of go trials multiplied by the probability of responding to a signal.”

Reviewer #2:

This work aggregates data across 5 openly available stopping studies (3 at 7 tesla and 2 at 3 tesla) to evaluate activity patterns across the common contrasts of Failed Stop (FS) > Go, FS > stop success (SS), and SS > Go. Previous work has implicated a set of regions that tend to be positively active in one or more of these contrasts, including the bilateral inferior frontal gyrus, preSMA, and multiple basal ganglia structures. However, the authors argue that upon closer examination, many previous papers have not found subcortical structures to be more active on SS than FS trials, bringing into question whether they play an essential role in (successful) inhibition. In order to evaluate this with more data and power, the authors aggregate across five datasets and find many areas that are *more* active for FS than SS, specifically bilateral preSMA, caudate, GPE, thalamus, and VTA, and unilateral M1, GPi, putamen, SN, and STN. They argue that this brings into question the role of these areas in inhibition, based upon the assumption that areas involved in inhibition should be more active on successful stop than failed stop trials, not the opposite as they observed. 

As an empirical result, I believe that the results are robust, but this work does not attempt a new theoretical synthesis of the neuro-cognitive mechanisms of stopping. Specifically, if these many areas are more active on failed stop than successful stop trials, and (at least some of) these areas are situated in pathways that are traditionally assumed to instantiate response inhibition like the hyperdirect pathway, then what function are these areas/pathways involved in? I believe that this work would make a larger impact if the author endeavored to synthesize these results into some kind of theoretical framework for how stopping is instantiated in the brain, even if that framework may be preliminary. 

I also have one main concern about the analysis. The authors use the mean method for computing SSRT, but this has been shown to be more susceptible to distortion from RT slowing (Verbruggen, Chambers & Logan, 2013 Psych Sci), and goes against the consensus recommendation of using the integration with replacement method (Verbruggen et al., 2019). Therefore, I would strongly recommend replacing all mean SSRT estimates with estimates using the integration with replacement method. 

I found the paper clearly written and empirically strong. As I mentioned in the public review, I believe that the main shortcoming is the lack of theoretical synthesis. I would encourage the authors to attempt to synthesize these results into some form of theoretical explanation. I would also encourage replacing the mean method with the integration with replacement method for computing SSRT. I also have the following specific comments and suggestions (in the approximate order in which they appear in the manuscript) that I hope can improve the manuscript: 

We would like to thank reviewer 2 for their insightful and interesting comments. We have adapted our paper to reflect these comments. Please see direct responses to your comments below. We agree with the reviewer that some type of theoretical synthesis would help with the interpretability of the article. We have substantially reworked the discussion and included theoretical considerations behind the newer narrative. Please see all changes to the article indicated by red text.

(1) The authors say "performance on successful stop trials is quantified by the stop signal reaction time". I don't think this is technically accurate. SSRT is a measure of the average latency of the stop process for all trials, not just for the trials in which subjects successfully stop. 

Thank you for pointing this technically incorrect statement. We have replaced the above sentence with the following (page 1):

“Inhibition performance in the SST as a whole is quantified by the stop signal reaction time (SSRT), which estimates the speed of the latent stopping process (Verbruggen et al., 2019).”

(2) The authors say "few studies have detected differences in the BOLD response between FS and SS trials", but then do not cite any papers that detected differences until several sentences later (de Hollander et al., 2017; Isherwood et al., 2023; Miletic et al., 2020). If these are the only ones, and they only show greater FS than SS, then I think this point could be made more clearly and directly. 

We have moved the citations to the correct place in the text to be clearer. We have also rephrased this part of the introduction to make the points more direct (page 2).

“In the subcortex, functional evidence is relatively inconsistent. Some studies have found an increase in BOLD response in the STN in SS > GO contrasts (Aron & Poldrack, 2006; Coxon et al., 2016; Gaillard et al., 2020; Yoon et al., 2019), but others have failed to replicate this (Bloemendaal et al., 2016; Boehler et al., 2010; Chang et al., 2020; B. Xu et al., 2015). Moreover, some studies have actually found higher STN, SN and thalamic activation in failed stop trials, not successful ones (de Hollander et al., 2017; Isherwood et al., 2023; Miletić et al., 2020).”

(3) Unless I overlooked it, I don't believe that the author specified the criterion that any given subject is excluded based upon. Given some studies have significant exclusions (e.g., Poldrack_3T), I think being clear about how many subjects violated each criterion would be useful. 

This is indeed interesting and important information to include. We have added the number of participants who were excluded for each criterion. Please see added text below (page 4):

“Based on these criteria, no subjects were excluded from the Aron_3T dataset. 24 subjects were excluded from the Poldrack_3T dataset (3 based on criterion 1, 9 on criterion 2, 11 on criterion 3, and 8 on criterion 4). Three subjects were excluded from the deHollander_7T dataset (2 based on criterion 1 and 1 on criterion 2). Five subjects were excluded from the Isherwood_7T dataset (2 based on criterion 1, 1 on criterion 2, and 2 on criterion 4). Two subjects were excluded from the Miletic_7T dataset (1 based on criterion 2 and 1 on criterion 4). Note that some participants in the Poldrack_3T study failed to meet multiple inclusion criteria.”

(4) The Method section included very exhaustive descriptions of the neuroimaging processing pipeline, which was appreciated. However, it seems that much of what is presented is not actually used in any of the analyses. For example, it seems that "functional data preprocessing" section may be fMRIPrep boilerplate, which again is fine, but I think it would help to clarify that much of the preprocessing was not used in any part of the analysis pipeline for any results. For example, at first blush, I thought the authors were using global signal regression, but after a more careful examination, I believe that they are only computing global signals but never using them. Similarly with tCompCor seemingly being computed but not used. If possible, I would recommend that the authors share code that instantiates their behavioral and neuroimaging analysis pipeline so that any confusion about what was actually done could be programmatically verified. At a minimum, I would recommend more clearly distinguishing the pipeline steps that actually went into any presented analyses.

We thank the reviewer for finding this inconsistency. The methods section indeed uses the fMRIprep boilerplate text, which we included so to be as accurate as possible when describing the preprocessing steps taken. While we believe leaving the exact boilerplate text that fMRIprep gives us is the most accurate method to show our preprocessing, we have adapted some of the text to clarify which computations were not used in the subsequent analysis. As a side-note, for future reference, we’d like to add that the fmriprep authors expressly recommend users to report the boilerplate completely and unaltered, and as such, we believe this may become a recurring issue (page 7).

“While many regressors were computed in the preprocessing of the fMRI data, not all were used in the subsequent analysis. The exact regressors used for the analysis can be found above. For example, tCompCor and global signals were calculated in our generic preprocessing pipeline but not part of the analysis. The code used for preprocessing and analysis can be found in the data and code availability statement.”

(5) What does it mean for the Poldrack_3T to have N/A for SSD range? Please clarify. 

Thank you for pointing out this omission. We had not yet found the possible SSD range for this study. We have replaced this value with the correct value (0 – 1000 ms).

(6) The SSD range of 0-2000ms for deHollander_7T and Miletic_7T seems very high. Was this limit ever reached or even approached? SSD distributions could be a useful addition to the supplement. 

Thank you for also bringing this mistake to light. We had accidentally placed the max trial duration in these fields instead of the max allowable SSD value. We have replaced the correct value (0 – 900 ms).

(7) The author says "In addition, median go RTs did not correlate with mean SSRTs within datasets (Aron_3T: r = .411, p = .10, BF = 1.41; Poldrack_3T: r = .011, p = .91, BF = .23; deHollander_7T: r = -.30, p = .09, BF = 1.30; Isherwood_7T: r = .13, p = .65, BF = .57; Miletic_7T: r = .37, p = .19, BF = 1.02), indicating independence between the stop and go processes, an important assumption of the horse-race model (Logan & Cowan, 1984)." However, the independent race model assumes context independence (the finishing time of the go process is not affected by the presence of the stop process) and stochastic independence (the duration of the go and stop processes are independent on a given trial). This analysis does not seem to evaluate either of these forms of independence, as it correlates RT and SSRT across subjects, so it was unclear how this analysis evaluated either of the types of independence that are assumed by the independent race model. Please clarify or remove. 

Thank you for this comment. We realize that this analysis indeed does not evaluate either context or stochastic independence and therefore we have removed this from the manuscript.

(8) The RTs in Isherwood_7T are considerably slower than the other studies, even though the go stimulus+response is the same (very simple) stimulus-response mapping from arrows to button presses. Is there any difference in procedure or stimuli that might explain this difference? It is the only study with a visual stop signal, but to my knowledge, there is no work suggesting visual stop signals encourage more proactive slowing. If possible, I think a brief discussion of the unusually slow RTs in Isherwood_7T would be useful. 

We have included the following text in the manuscript to reflect this observed difference in RT between the Isherwood_7T dataset and the other datasets (page 9).

“Longer RTs were found in the Isherwood_7T dataset in comparison to the four other datasets. The only difference in procedure in the Isherwood_7T dataset is the use of a visual stop signal as opposed to an auditory stop signal. This RT difference is consistent with previous research, where auditory stop signals and visual go stimuli have been associated with faster RTs compared to unimodal visual presentation (Carrillo-de-la-Peña et al., 2019; Weber et al., 2024). The mean SSRTs and probability of stopping are within normal range, indicating that participants understood the task and responded in the expected manner.”

(9) When the authors included both 3T and 7T data, I thought they were preparing to evaluate the effect of magnet strength on stop networks, but they didn't do this analysis. Is this because the authors believe there is insufficient power? It seems that this could be an interesting exploratory analysis that could improve the paper.

We thank the reviewer for this interesting comment. As our dataset sample contains only two 3T and three 7T datasets we indeed believe there is insufficient power to warrant such an analysis. In addition, we wanted the focus of this paper to be how fMRI examines the SST in general, and not differences between acquisition methods. With a greater number of datasets with different imaging parameters (especially TE or resolution) in addition to field strength, we agree such an analysis would be interesting, although beyond the scope of this article.

(10) The authors evaluate smoothing and it seems that the conclusion that they want to come to is that with a larger smoothing kernel, the results in the stop networks bleed into surrounding areas, producing false positive activity. However, in the absence of a ground truth of the true contributions of these areas, it seems that an alternative interpretation of the results is that the denser maps when using a larger smoothing kernel could be closer to "true" activation, with the maps using a smaller smoothing kernel missing some true activity. It seems worth entertaining these two possible interpretations for the smoothing results unless there is clear reason to conclude that the smoothed results are producing false positive activity. 

We agree with the view of the reviewer on the interpretation of the smoothing results. We indeed cannot rule this out as a possible interpretation of the results, due to a lack of ground truth. We have added text to the article to reflect this view and discuss the types of errors we can expect for both smaller and larger smoothing kernels (page 15).

“In the absence of a ground truth, we are not able to fully justify the use of either larger or smaller kernels to analyse such data. On the one hand, aberrantly large smoothing kernels could lead to false positives in activation profiles, due to bleeding of observed activation into surrounding tissues. On the other side, too little smoothing could lead to false negatives, missing some true activity in surrounding regions. While we cannot concretely validate either choice, it should be noted that there is lower spatial uncertainty in the subcortex compared to the cortex, due to the lower anatomical variability. False positives from smoothing spatially unmatched signal, are more likely than false negatives. It may be more prudent for studies to use a range of smoothing kernels, to assess the robustness of their fMRI activation profiles.”

eLife assessment

This important study provides a new perspective on why preparatory activity occurs before the onset of movement. The authors report that when there is a cost on the inputs, the optimal inputs should start before the desired network output for a wide variety of recurrent networks. The authors present convincing evidence by combining mathematically tractable analyses in linear networks and numerical simulation in nonlinear networks.

Reviewer #1 (Public Review):

In this work, the authors investigate an important question - under what circumstances should a recurrent neural network optimised to produce motor control signals receive preparatory input before the initiation of a movement, even though it is possible to use inputs to drive activity just-in-time for movement?

This question is important because many studies across animal models have show that preparatory activity is widespread in neural populations close to motor output (e.g. motor cortex / M1), but it isn't clear under what circumstances this preparation is advantageous for performance, especially since preparation could cause unwanted motor output during a delay.

They show that networks optimised under reasonable constraints (speed, accuracy, lack of pre-movement) will use input to seed the state of the network before movement, and that these inputs reduce the need for ongoing input during the movement. By examining many different parameters in simplified models they identify a strong connection between the structure of the network and the amount of preparation that is optimal for control - namely, that preparation has the most value when nullspaces are highly observable relative to the readout dimension and when the controllability of readout dimensions is low. They conclude by showing that their model predictions are consistent with the observation in monkey motor cortex that even when a sequence of two movements is known in advance, preparatory activity only arises shortly before movement initiation.

Overall, this study provides valuable theoretical insight into the role of preparation in neural populations that generate motor output, and by treating input to motor cortex as a signal that is optimised directly this work is able to sidestep many of the problematic questions relating to estimating the potential inputs to motor cortex.

Reviewer #2 (Public Review):

This work clarifies neural mechanisms that can lead to a phenomenology consistent with motor preparation in its broader sense. In this context, motor preparation refers to activity that occurs before the corresponding movement. Another property often associated with preparatory activity is a correlation with global movement characteristics such as reach speed (Churchland et al., Neuron 2006), reach angle (Sun et al., Nature 2022), or grasp type (Meirhaeghe et al., Cell Reports 2023). Such activity has notably been observed in premotor and primary motor cortices, and it has been hypothesized to serve as an input to a motor execution circuit. The timing and mechanisms by which such 'preparatory' inputs are made available to motor execution circuits remain however unclear in general, especially in light of the presence of a 'trigger-like' signal that appears to relate to the transition from preparatory dynamics to execution activity (Kaufman et al. eNeuron 2016, Iganaki et al., Cell 2022, Zimnik and Churchland, Nature Neuroscience 2021).

The preparatory inputs have been hypothesized to fulfill one or several (non-mutually-exclusive) possible objectives. Two notable hypotheses are that these inputs could be shaped to maximize output accuracy under regularization of the input magnitude; or that they may help the flexible re-use of the neural machinery involved in the control of movements in different contexts.

Here, the authors investigate in detail how the former hypothesis may be compatible with the presence of early inputs in recurrent network models driving arm movements, and compare models to data.

Strengths:

The authors are able to deploy an in-depth evaluation of inputs that are optimized for producing an accurate output at a pre-defined time while using a regularization term on the input magnitude, in the case of movements that are thought to be controlled in a quasi-open loop fashion such as reaches.

First, the authors have identified that optimal control theory is a great framework to study this question as it provides methods to find and analyze exact solutions to this cost function in the case of models with linear dynamics. The authors not only use this framework to get an exact assessment of how much pre-movement input arises in large recurrent networks, but also give insight into the mechanisms by which it happens by dissecting in detail low-dimensional networks. The authors find that two key network properties - observability of the readout's nullspace and limited controllability - give rise to optimal inputs that are large before the start of the movement (while the corresponding network activity lies in the nullspace of the readout). Further, the authors numerically investigate the timing of optimized inputs in models with nonlinear dynamics, and find that pre-movement inputs can also arise in these more general networks. The authors also explore how some variations on their model's constraints - such as penalizing the input roughness or changing task contingencies about the go cue timing - affect their results. Finally, the authors point out some coarse-grained similarities between the pre-movement activity driven by the optimized inputs in some of the models they studied, and the phenomenology of preparation observed in the brain during single reaches and reach sequences. Overall, the authors deploy an impressive arsenal of tools and a very in-depth analysis of their models.

Limitations:

(1) Though the optimal control theory framework is ideal to determine inputs that minimize output error while regularizing the input norm or other simple input features, it cannot easily account for some other varied types of objectives - especially those that may lead to a complex optimization landscape. For instance, the reusability of parts of the circuit, sparse use of additional neurons when learning many movements, and ease of planning (especially under uncertainty about when to start the movement), may be alternative or additional reasons that could help explain the preparatory activity observed in the brain. It is interesting to note that inputs that optimize the objective chosen by the authors arguably lead to a trade-off in terms of other desirable objectives. Specifically, the inputs the authors derive are time-dependent, so a recurrent network would be needed to produce them and it may not be easy to interpolate between them to drive new movement variants. In addition, these inputs depend on the desired time of output and therefore make it difficult to plan, e.g. in circumstances when timing should be decided depending on sensory signals. Finally, these inputs are specific to the full movement chain that will unfold, so they do not permit reuse of the inputs e.g. in movement sequences of different orders. Of note, the authors have pointed out in the discussion how their framework may be extended in future work to account for some additional objectives, such as inputs' temporal smoothness or some strategies for dealing with go cue timing uncertainty.

(2) Relatedly, if the motor circuits were to balance different types of objectives, the activity and inputs occurring before each movement may be broken down into different categories that may each specialize into their own objective. For instance, previous work (Kaufman et al. eNeuron 2016, Iganaki et al., Cell 2022, Zimnik and Churchland, Nature Neuroscience 2021) has suggested that inputs occurring before the movement could be broken down into preparatory inputs 'stricto sensu' - relating to the planned characteristics of the movement - and a trigger signal, relating to the transition from planning to execution - irrespective of whether the movement is internally timed or triggered by an external event. The current work does not address which type(s) of early input may be labeled as 'preparatory' or may be thought of as a part of 'planning' computations, or whether these inputs may come from several different source circuits.

(3) While the authors rightly point out some similarities between the inputs that they derive and observed preparatory activity in the brain, notably during motor sequences, there are also some differences. For instance, while both the derived inputs and the data show two peaks during sequences, the data reproduced from Zimnik and Churchland show preparatory inputs that have a very asymmetric shape that really plummets before the start of the next movement, whereas the derived inputs have larger amplitude during the movement period - especially for the second movement of the sequence. In addition, the data show trigger-like signals before each of the two reaches. Finally, while the data show a very high correlation between the pattern of preparatory activity of the second reach in the double reach and compound reach conditions, the derived inputs appear to be more different between the two conditions. Note that the data would be consistent with separate planning of the two reaches even in the compound reach condition, as well as the re-use of the preparatory input between the compound and double reach conditions. Therefore, different motor sequence datasets - notably, those that would show even more coarticulation between submovements - may be more promising to find a tight match between the data and the author's inputs. Further analyses in these datasets could help determine whether the coarticulation could be due to simple filtering by the circuits and muscles downstream of M1, planning of movements with adjusted curvature to mitigate the work performed by the muscles while permitting some amount of re-use across different sequences, or - as suggested by the authors - inputs fully tailored to one specific movement sequence that maximize accuracy and minimize the M1 input magnitude.

(4) Though iLQR is a powerful optimization method to find inputs optimizing the author's cost function, it also has some limitations. First, given that it relies on a linearization of the dynamics at each timestep, it has a limited ability to leverage potential advantages of nonlinearities in the dynamics. Second, the iLQR algorithm is not a biologically plausible learning rule and therefore it might be difficult for the brain to learn to produce the inputs that it finds. Therefore, when observing differences between model and data, this can confound the question of whether it comes from a difference of assumed objective or a difference of optimization procedure. It remains unclear whether using alternative algorithms with different limitations - for instance, using variants of BPTT to train a separate RNN to produce the inputs in question - could impact some of the results.

(5) Under the objective considered by the authors, the amount of input occurring before the movement might be impacted by the presence of online sensory signals for closed-loop control. Even if considering that the inputs could include some sensory activity and/or that the RNN activity could represent general variables whose states can be decoded from M1, the model would not include mechanisms that process imperfect (delayed, noisy) sensory feedback to adapt the output in a trial-specific manner. It is therefore an open question whether the objective and network characteristics suggested by the authors could also explain the presence of preparatory activity before e.g. grasping movements that are thought to be more sensory-driven (Meirhaeghe et al., Cell Reports 2023).

Reviewer #3 (Public Review):

I remain enthusiastic about this study. The manuscript is well-written, logical, and conceptually clear. To my knowledge, no prior modeling study has tackled the question of 'why prepare before executing, why not just execute?' Prior studies have simply assumed, to emulate empirical findings, that preparatory inputs precede execution. They never asked why. The authors show that, when there are constraints on inputs, preparation becomes a natural strategy. In contrast, with no constraint on inputs, there is no need for preparation as one could get anything one liked just via the inputs during movement. For the sake of tractability, the authors use a simple magnitude constraint: the cost function punishes the integral of the squared inputs. Thus, if small inputs before movement can reduce the size of the inputs needed during movement, preparation is a good strategy. This occurs if (and only if) the network has strong dynamics (otherwise feeding it preparatory activity would not produce anything interesting). All of this is sensible and clarifying.

As discussed in the prior round of reviews, the central constraint that the authors use is a mathematically tractable stand-in for a range of plausible (but often trickier to define and evaluate) constraints, such as simplicity of inputs (or inputs being things that other areas could provide). The manuscript now embraces this fact more explicitly, and also gives some results showing that other constraints (such as on the derivative of activity, which is one component of complexity) can have the same effect. The manuscript also now discusses and addresses a modest weakness of the previous manuscript: the preparatory activity in their simulations is often overly complex temporally, lacking the (rough) plateau typically seen for data. Depending on your point of view, this is simply 'window dressing', but from my perspective it was important to know that their approach could yield more realistic-looking preparatory activity. Both these additions (the new constraint, and the more realistic temporal profile of preparatory activity) are added simply as supplementary figures rather than in the main text, and are brought up only in the Discussion. At first this struck me as slightly odd, but in the end I think this is appropriate. These are really Discussion-type issues, and dealing with them there makes sense. The 'different constraints' issue in particular is deep, tricky to explore for technical reasons, and could thus support a small research program. I think it is fair to talk about it thoughtfully (as the Discussion now does) and then just mention some simple results.

My remaining comments largely pertain to some subtle (but to me important) nuances at a few locations in the text. These should be easy for the authors to address, in whatever way they see fit.

Specific comments:

(1) The authors state the following on line 56: "For preparatory processes to avoid triggering premature movement, any pre-movement activity in the motor and dorsal pre-motor (PMd) cortices must carefully exclude those pyramidal tract neurons."<br /> This constraint is overly restrictive. PT neurons absolutely can change their activity during preparation in principle (and appear to do so in practice). The key constraint is looser: those changes should have no net effect on the muscles. E.g., if d is the vector of changes in PT neuron firing rates, and b is the vector of weights, then the constraint is that b'd = 0. d = 0 is one good way of doing this, but only one. Half the d's could go up and half could go down. Or they all go up, but half the b's are negative. Put differently, there is no reason the null space has to be upstream of the PT neurons. It could be partly, or entirely, downstream.<br /> In the end, this doesn't change the point the authors are making. It is still the case that d has to be structured to avoid causing muscle activity, which raises exactly the point the authors care about: why risk this unless preparation brings benefits? However, this point can be made with a more accurate motivation. This matters, because people often think that a null-space is a tricky thing to engineer, when really it is quite natural. With enough neurons, preparing in the null space is quite simple.

(2) Line 167: 'near-autonomous internal dynamics in M1'.<br /> It would be good if such statements, early in the paper, could be modified to reflect the fact that the dynamics observed in M1 may depend on recurrence that is NOT purely internal to M1. A better phrase might be 'near-autonomous dynamics that can be observed in M1'. A similar point applies on line 13. This issue is handled very thoughtfully in the Discussion, starting on line 713. Obviously it is not sensible to also add multiple sentences making the same point early on. However, it is still worth phrasing things carefully, otherwise the reader may have the wrong impression up until the Discussion (i.e. they may think that both the authors, and prior studies, believe that all the relevant dynamics are internal to M1). If possible, it might also be worth adding one sentence, somewhere early, to keep readers from falling into this hole (and then being stuck there till the Discussion digs them out).

(3) The authors make the point, starting on line 815, that transient (but strong) preparatory activity empirically occurs without a delay. They note that their model will do this but only if 'no delay' means 'no external delay'. For their model to prepare, there still needs to be an internal delay between when the first inputs arrive and when movement generating inputs arrive.

This is not only a reasonable assumption, but is something that does indeed occur empirically. This can be seen in Figure 8c of Lara et al. Similarly, Kaufman et al. 2016 noted that "the sudden change in the CIS [the movement triggering event] occurred well after (~150 ms) the visual go cue... (~60 ms latency)" Behavioral experiments have also argued that internal movement-triggering events tend to be quite sluggish relative to the earliest they could be, causing RTs to be longer than they should be (Haith et al. Independence of Movement Preparation and Movement Initiation). Given this empirical support, the authors might wish to add a sentence indicating that the data tend to justify their assumption that the internal delay (separating the earliest response to sensory events from the events that actually cause movement to begin) never shrinks to zero.

While on this topic, the Haith and Krakauer paper mentioned above good to cite because it does ponder the question of whether preparation is really necessary. By showing that they could get RTs to shrink considerably before behavior became inaccurate, they showed that people normally (when not pressured) use more preparation time than they really need. Given Lara et al, we know that preparation does always occur, but Haith and Krakauer were quite right that it can be very brief. This helped -- along with neural results -- change our view of preparation from something more cognitive that had to occur, so something more mechanical that was simply a good network strategy, which is indeed the authors current point. Working a discussion of this into the current paper may or may not make sense, but if there is a place where it is easy to cite, it would be appropriate.

Author response:

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

General response:

We thank all the reviewers for their detailed reviews.

All reviewers made a number of valuable comments, in particular by highlighting several points that would benefit from additional clarifications and discussion. We really appreciate the time and effort that went into the reviews. We have updated the paper to reflect the changes we have made in response to the reviewers' comments (largely by including more discussion regarding the model limitations and the effect of various modeling choices). We have also included several new supplementary figures (S7, S8, S9, S10) that provide further details of the model behavior, and show the effect of changing some of the terms in the cost. Below, we go through the individual comments, and highlight the places in which we have made changes to address the reviewers’ comments.

Reviewer 1:

Thank you for your review and pointing out multiple things to be discussed and clarified! Below, we go through the various limitations you pointed out and refer to the places where we have tried to address them.

(1) It's important to keep in mind that this work involves simplified models of the motor system, and often the terminology for 'motor cortex' and 'models of motor cortex' are used interchangeably, which may mislead some readers. Similarly, the introduction fails in many cases to state what model system is being discussed (e.g. line 14, line 29, line 31), even though these span humans, monkeys, mice, and simulations, which all differ in crucial ways that cannot always be lumped together.

That is a good point. We have clarified this in the text (Introduction and Discussion), to highlight the fact that our model isn’t necessarily meant to just capture M1. We have also updated the introduction to make it more clear which species the experiments which motivate our investigation were performed in.

(2) At multiple points in the manuscript thalamic inputs during movement (in mice) is used as a motivation for examining the role of preparation. However, there are other more salient motivations, such as delayed sensory feedback from the limb and vision arriving in the motor cortex, as well as ongoing control signals from other areas such as the premotor cortex.

Yes – the motivation for thalamic inputs came from the fact that those have specifically been shown to be necessary for accurate movement generation in mice. However, it is true that the inputs in our model are meant to capture any signals external to the dynamical system modeled, and as such are likely to represent a mixture of sensory signals, and feedback from other areas. We have clarified this in the Discussion, and have added this additional motivation in the Introduction.

(3) Describing the main task in this work as a delayed reaching task is not justified without caveats (by the authors' own admission: line 687), since each network is optimized with a fixed delay period length. Although this is mentioned to the reader, it's not clear enough that the dynamics observed during the delay period will not resemble those in the motor cortex for typical delayed reaching tasks.

Yes, we completely agree that the terminology might be confusing. While the task we are modeling is a delayed reaching task, it does differ from the usual setting since the network has knowledge of the delay period, and that is indeed a caveat of the model. We have added a brief paragraph just after the description of the optimal control objective to highlight this limitation.

We have also performed additional simulations using two different variants of a model-predictive control approach that allow us to relax the assumption that the go-cue time is known in advance. We show that these modifications of the optimal controller yield results that remain consistent with our main conclusions, and can in fact in some settings lead to preparatory activity plateaus during the preparation epoch as often found in monkey M1 (e.g in Elsayed et al. 2016). We have modified the Discussion to explain these results and their limitations, which are summarized in a new Supplementary Figure (S9).

(4) A number of simplifications in the model may have crucial consequences for interpretation.

a) Even following the toy examples in Figure 4, all the models in Figure 5 are linear, which may limit the generalisability of the findings.

While we agree that linear models may be too simplistic, much prior analyses of M1 data suggest that it is often good enough to capture key aspects of M1 dynamics; for example, the generative model underlying jPCA is linear, and Sussillo et al. (2015) showed that the internal activity of nonlinear RNN models trained to reproduce EMG data aligned best with M1 activity when heavily regularized; in this regime, the RNN dynamics were close to linear. Nevertheless, this linearity assumption is indeed convenient from a modeling viewpoint: the optimal control problem is more easily solved for linear network dynamics and the optimal trajectories are more consistent across networks. Indeed, we had originally attempted to perform the analyses of Figure 5 in the nonlinear setting, but found that while the results were overall similar to what we report in the linear regime, iLQR was occasionally trapped into local minimal, resulting in more variable results especially for inhibition-stabilized network in the strongly connected end of the spectrum. Finally, Figure 5 is primarily meant to explore to what extent motor preparation can be predicted from basic linear control-theoretic properties of the Jacobian of the dynamics; in this regard, it made sense to work with linear RNNs (for which the Jacobian is constant).

b) Crucially, there is no delayed sensory feedback in the model from the plant. Although this simplification is in some ways a strength, this decision allows networks to avoid having to deal with delayed feedback, which is a known component of closed-loop motor control and of motor cortex inputs and will have a large impact on the control policy.

This comment resonates well with Reviewer 3's remark regarding the autonomous nature (or not) of M1 during movement. Rather than thinking of our RNN models as anatomically confined models of M1 alone, we think of them as models of the dynamics which M1 implements possibly as part of a broader network involving “inter-area loops and (at some latency) sensory feedback”, and whose state appears to be near-fully decodable from M1 activity alone. We have added a paragraph of Discussion on this important point.

(5) A key feature determining the usefulness of preparation is the direction of the readout dimension. However, all readouts had a similar structure (random Gaussian initialization). Therefore, it would be useful to have more discussion regarding how the structure of the output connectivity would affect preparation, since the motor cortex certainly does not follow this output scheme.

We agree with this limitation of our model — indeed one key message of Figure 4 is that the degree of reliance on preparatory inputs depends strongly on how the dynamics align with the readout. However, this strong dependence is somewhat specific to low-dimensional models; in higher-dimensional models (most of our paper), one expects that any random readout matrix C will pick out activity dimensions in the RNN that are sufficiently aligned with the most controllable directions of the dynamics to encourage preparation.

We did consider optimizing C away (which required differentiating through the iLQR optimizer, which is possible but very costly), but the question inevitably arises what exactly should C be optimized for, and under what constraints (e.g fixed norm or not). One possibility is to optimize C with respect to the same control objective that the control inputs are optimized for, and constrain its norm (otherwise, inputs to the M1 model, and its internal activity, could become arbitrarily small as C can grow to compensate). We performed this experiment (new Supplementary Figure S7) and obtained a similar preparation index; there was one notable difference, namely that the optimized readout modes led to greater observability compared to a random readout; thus, the same amount of “muscle energy” required for a given movement could now be produced by a smaller initial condition. In turn, this led to smaller control inputs, consistent with a lower control cost overall.

Whilst we could have systematically optimized C away, we reasoned that (i) it is computationally expensive, and (ii) the way M1 affects downstream effectors is presumably “optimized” for much richer motor tasks than simple 2D reaching, such that optimizing C for a fixed set of simple reaches could lead to misleading conclusions. We therefore decided to stick with random readouts.

Additional comments :

(1) The choice of cost function seems very important. Is it? For example, penalising the square of u(t) may produce very different results than penalising the absolute value.

Yes, the choice of cost function does affect the results, at least qualitatively. The absolute value of the inputs is a challenging cost to use, as iLQR relies on a local quadratic approximation of the cost function. However, we have included additional experiments in which we penalized the squared derivative of the inputs (Supplementary Figure S8; see also our response to Reviewer 3's suggestion on this topic), and we do see differences in the qualitative behavior of the model (though the main takeaway, i.e. the reliance on preparation, continues to hold). This is now referred to and discussed in the Discussion section.

(2) In future work it would be useful to consider the role of spinal networks, which are known to contribute to preparation in some cases (e.g. Prut and Fetz, 1999).

(3) The control signal magnitude is penalised, but not the output torque magnitude, which highlights the fact that control in the model is quite different from muscle control, where co-contraction would be a possibility and therefore a penalty of muscle activation would be necessary. Future work should consider the role of these differences in control policy.

Thank you for pointing us to this reference! Regarding both of these concerns, we agree that the model could be greatly improved and made more realistic in future work (another avenue for this would be to consider a more realistic biophysical model, e.g. using the MotorNet library). We hope that the current Discussion, which highlights the various limitations of our modeling choices, makes it clear that a lot of these choices could easily be modified depending on the specific assumptions/investigation being performed.

Reviewer 2:

Thank you for your positive review! We very much agree with the limitations you pointed out, some of which overlapped with the comments of the other reviewers. We have done our best to address them through additional discussion and new supplementary figures. We briefly highlight below where those changes can be found.

(1) Though the optimal control theory framework is ideal to determine inputs that minimize output error while regularizing the input norm, it however cannot easily account for some other varied types of objectives especially those that may lead to a complex optimization landscape. For instance, the reusability of parts of the circuit, sparse use of additional neurons when learning many movements, and ease of planning (especially under uncertainty about when to start the movement), may be alternative or additional reasons that could help explain the preparatory activity observed in the brain. It is interesting to note that inputs that optimize the objective chosen by the authors arguably lead to a trade-off in terms of other desirable objectives. Specifically, the inputs the authors derive are time-dependent, so a recurrent network would be needed to produce them and it may not be easy to interpolate between them to drive new movement variants. In addition, these inputs depend on the desired time of output and therefore make it difficult to plan, e.g. in circumstances when timing should be decided depending on sensory signals. Finally, these inputs are specific to the full movement chain that will unfold, so they do not permit reuse of the inputs e.g. in movement sequences of different orders.

Yes, that is a good point! We have incorporated further Discussion related to this point. We have additionally included a new example in which we regularize the temporal complexity of the inputs (see also our response to Reviewer 3's suggestion on this topic), which leads to more slowly varying inputs, and may indeed represent a more realistic constraint and lead to simpler inputs that can more easily be interpolated between. We also agree that uncertainty about the upcoming go cue may play an important role in the strategy adopted by the animals. While we have not performed an extensive investigation of the topic, we have included a Supplementary Figure (S9) in which we used Model Predictive Control to investigate the effect of planning under uncertainty about the go cue arrival time. We hope that this will give the reader a better sense of what sort of model extensions are possible within our framework.

(2) Relatedly, if the motor circuits were to balance different types of objectives, the activity and inputs occurring before each movement may be broken down into different categories that may each specialize into one objective. For instance, previous work (Kaufman et al. eNeuron 2016, Iganaki et al., Cell 2022, Zimnik and Churchland, Nature Neuroscience 2021) has suggested that inputs occurring before the movement could be broken down into preparatory inputs 'stricto sensu' - relating to the planned characteristics of the movement - and a trigger signal, relating to the transition from planning to execution - irrespective of whether the movement is internally timed or triggered by an external event. The current work does not address which type(s) of early input may be labeled as 'preparatory' or may be thought of as a part of 'planning' computations.

Yes, our model does indeed treat inputs in a very general way, and does not distinguish between the different types of processes they may be composed of. This is partly because we do not explicitly model where the inputs come from, such that our inputs likely englobe multiple processes. We have added discussion related to this point.

(3) While the authors rightly point out some similarities between the inputs that they derive and observed preparatory activity in the brain, notably during motor sequences, there are also some differences. For instance, while both the derived inputs and the data show two peaks during sequences, the data reproduced from Zimnik and Churchland show preparatory inputs that have a very asymmetric shape that really plummets before the start of the next movement, whereas the derived inputs have larger amplitude during the movement period - especially for the second movement of the sequence. In addition, the data show trigger-like signals before each of the two reaches. Finally, while the data show a very high correlation between the pattern of preparatory activity of the second reach in the double reach and compound reach conditions, the derived inputs appear to be more different between the two conditions. Note that the data would be consistent with separate planning of the two reaches even in the compound reach condition, as well as the re-use of the preparatory input between the compound and double reach conditions. Therefore, different motor sequence datasets - notably, those that would show even more coarticulation between submovements - may be more promising to find a tight match between the data and the author's inputs. Further analyses in these datasets could help determine whether the coarticulation could be due to simple filtering by the circuits and muscles downstream of M1, planning of movements with adjusted curvature to mitigate the work performed by the muscles while permitting some amount of re-use across different sequences, or - as suggested by the authors - inputs fully tailored to one specific movement sequence that maximize accuracy and minimize the M1 input magnitude.

Regarding the exact shape of the occupancy plots, it is important to note that some of the more qualitative aspects (e.g the relative height of the two peaks) will change if we change the parameters of the cost function. Right now, we have chosen the parameters to ensure that both reaches would be performed at roughly the same speed (as a way to very loosely constrain the parameters based on the observed behavior). However, small changes to the hyperparameters can lead to changes in the model output (e.g one of the two consecutive reaches being performed using greater acceleration than the other), and since our biophysical model is fairly simple, changes in the behavior are directly reflected in the network activity. Essentially, what this means is that while the double occupancy is a consistent feature of the model, the exact shape of the peaks is more sensitive to hyperparameters, and we do not wish to draw any strong conclusions from them, given the simplicity of the biophysical model. However, we do agree that our model exhibits some differences with the data. As discussed above, we have included additional discussion regarding the potential existence of separate inputs for planning vs triggering the movement in the context of single reaches.

Overall, we are excited about the suggestions made by the Reviewer here about using our approach to analyze other motor sequence datasets, but we think that in order to do this properly, one would need to adopt a more realistic musculo-skeletal model (such as one provided by MotorNet).

(4) Though iLQR is a powerful optimization method to find inputs optimizing the author's cost function, it also has some limitations. First, given that it relies on a linearization of the dynamics at each timestep, it has a limited ability to leverage potential advantages of nonlinearities in the dynamics. Second, the iLQR algorithm is not a biologically plausible learning rule and therefore it might be difficult for the brain to learn to produce the inputs that it finds. It remains unclear whether using alternative algorithms with different limitations - for instance, using variants of BPTT to train a separate RNN to produce the inputs in question - could impact some of the results.

We agree that our choice of iLQR has limitations: while it offers the advantage of convergence guarantees, it does indeed restrict the choice of cost function and dynamics that we can use. We have now included extensive discussion of how the modeling choices affect our results.

We do not view the lack of biological plausibility of iLQR as an issue, as the results are agnostic to the algorithm used for optimization. However, we agree that any structure imposed on the inputs (e.g by enforcing them to be the output of a self-contained dynamical system) would likely alter the results. A potentially interesting extension of our model would be to do just what the reviewer suggested, and try to learn a network that can generate the optimal inputs. However, this is outside the scope of our investigation, as it would then lead to new questions (e.g what brain region would that other RNN represent?).

(5) Under the objective considered by the authors, the amount of input occurring before the movement might be impacted by the presence of online sensory signals for closed-loop control. It is therefore an open question whether the objective and network characteristics suggested by the authors could also explain the presence of preparatory activity before e.g. grasping movements that are thought to be more sensory-driven (Meirhaeghe et al., Cell Reports 2023).

It is true that we aren’t currently modeling sensory signals explicitly. However, some of the optimal inputs we infer may be capturing upstream information which could englobe some sensory information. This is currently unclear, and would likely depend on how exactly the model is specified. We have added new discussion to emphasize that our dynamics should not be understood as just representing M1, but more general circuits whose state can be decoded from M1.

Reviewer #2 (Recommendations For The Authors):

Additionally, thank you for pointing out various typos in the manuscript, we have fixed those!

Reviewer 3:

Thank you very much for your review, which makes a lot of very insightful points, and raises several interesting questions. In summary, we very much agree with the limitations you pointed out. In particular, the choice of input cost is something we had previously discussed, but we had found it challenging to decide on what a reasonable cost for “complexity” could be. Following your comment, we have however added a first attempt at penalizing “temporal complexity”, which shows promising behavior. We have only included those additional analyses as supplementary figures, and we have included new discussion, which hopefully highlights what we meant by the different model components, and how the model behavior may change as we vary some of our choices. We hope this can be informative for future models that may use a similar approach. Below, we highlight the changes that we have made to address your comments.

The main limitation of the study is that it focuses exclusively on one specific constraint - magnitude - that could limit motor-cortex inputs. This isn't unreasonable, but other constraints are at least as likely, if less mathematically tractable. The basic results of this study will probably be robust with regard such issues - generally speaking, any constraint on what can be delivered during execution will favor the strategy of preparing - but this robustness cuts both ways. It isn't clear that the constraint used in the present study - minimizing upstream energy costs - is the one that really matters. Upstream areas are likely to be limited in a variety of ways, including the complexity of inputs they can deliver. Indeed, one generally assumes that there are things that motor cortex can do that upstream areas can't do, which is where the real limitations should come from. Yet in the interest of a tractable cost function, the authors have built a system where motor cortex actually doesn't do anything that couldn't be done equally well by its inputs. The system might actually be better off if motor cortex were removed. About the only thing that motor cortex appears to contribute is some amplification, which is 'good' from the standpoint of the cost function (inputs can be smaller) but hardly satisfying from a scientific standpoint.

The use of a term that punishes the squared magnitude of control signals has a long history, both because it creates mathematical tractability and because it (somewhat) maps onto the idea that one should minimize the energy expended by muscles and the possibility of damaging them with large inputs. One could make a case that those things apply to neural activity as well, and while that isn't unreasonable, it is far from clear whether this is actually true (and if it were, why punish the square if you are concerned about ATP expenditure?). Even if neural activity magnitude an important cost, any costs should pertain not just to inputs but to motor cortex activity itself. I don't think the authors really wish to propose that squared input magnitude is the key thing to be regularized. Instead, this is simply an easily imposed constraint that is tractable and acts as a stand-in for other forms of regularization / other types of constraints. Put differently, if one could write down the 'true' cost function, it might contain a term related to squared magnitude, but other regularizing terms would by very likely to dominate. Using only squared magnitude is a reasonable way to get started, but there are also ways in which it appears to be limiting the results (see below).

I would suggest that the study explore this topic a bit. Is it possible to use other forms of regularization? One appealing option is to constrain the complexity of inputs; a long-standing idea is that the role of motor cortex is to take relatively simple inputs and convert them to complex time-evolving inputs suitable for driving outputs. I realize that exploring this idea is not necessarily trivial. The right cost-function term is not clear (should it relate to low-dimensionality across conditions, or to smoothness across time?) and even if it were, it might not produce a convex cost function. Yet while exploring this possibility might be difficult, I think it is important for two reasons.

First, this study is an elegant exploration of how preparation emerges due to constraints on inputs, but at present that exploration focuses exclusively on one constraint. Second, at present there are a variety of aspects of the model responses that appear somewhat unrealistic. I suspect most of these flow from the fact that while the magnitude of inputs is constrained, their complexity is not (they can control every motor cortex neuron at both low and high frequencies). Because inputs are not complexity-constrained, preparatory activity appears overly complex and never 'settles' into the plateaus that one often sees in data. To be fair, even in data these plateaus are often imperfect, but they are still a very noticeable feature in the response of many neurons. Furthermore, the top PCs usually contain a nice plateau. Yet we never get to see this in the present study. In part this is because the authors never simulate the situation of an unpredictable delay (more on this below) but it also seems to be because preparatory inputs are themselves strongly time-varying. More realistic forms of regularization would likely remedy this.

That is a very good point, and it mirrors several concerns that we had in the past. While we did focus on the input norm for the sake of simplicity, and because it represents a very natural way to regularize our control solutions, we agree that a “complexity cost” may be better suited to models of brain circuits. We have addressed this in a supplementary investigation. We chose to focus on a cost that penalizes the temporal complexity of the inputs, as ||u(t+1) - u(t)||^2. Note that this required augmenting the state of the model, making the computations quite a bit slower; while it is doable if we only penalize the first temporal derivative, it would not scale well to higher orders.

Interestingly, we did find that the activity in that setting was somewhat more realistic (see new Supplementary Figure S8), with more sustained inputs and plateauing activity. While we have kept the original model for most of the investigations, the somewhat more realistic nature of the results under that setting suggests that further exploration of penalties of that sort could represent a promising avenue to improve the model.

We also found the idea of a cost that would ensure low-dimensionality of the inputs across conditions very interesting. However, it is challenging to investigate with iLQR as we perform the optimization separately for each condition; nevertheless, it could be investigated using a different optimizer.

At present, it is also not clear whether preparation always occurs even with no delay. Given only magnitude-based regularization, it wouldn't necessarily have to be. The authors should perform a subspace-based analysis like that in Figure 6, but for different delay durations. I think it is critical to explore whether the model, like monkeys, uses preparation even for zero-delay trials. At present it might or might not. If not, it may be because of the lack of more realistic constraints on inputs. One might then either need to include more realistic constraints to induce zero-delay preparation, or propose that the brain basically never uses a zero delay (it always delays the internal go cue after the preparatory inputs) and that this is a mechanism separate from that being modeled.

I agree with the authors that the present version of the model, where optimization knows the exact time of movement onset, produces a reasonably realistic timecourse of preparation when compared to data from self-paced movements. At the same time, most readers will want to see that the model can produce realistic looking preparatory activity when presented with an unpredictable delay. I realize this may be an optimization nightmare, but there are probably ways to trick the model into optimizing to move soon, but then forcing it to wait (which is actually what monkeys are probably doing). Doing so would allow the model to produce preparation under the circumstances where most studies have examined it. In some ways this is just window-dressing (showing people something in a format they are used to and can digest) but it is actually more than that, because it would show that the model can produce a reasonable plateau of sustained preparation. At present it isn't clear it can do this, for the reasons noted above. If it can't, regularizing complexity might help (and even if this can't be shown, it could be discussed).

In summary, I found this to be a very strong study overall, with a conceptually timely message that was well-explained and nicely documented by thorough simulations. I think it is critical to perform the test, noted above, of examining preparatory subspace activity across a range of delay durations (including zero) to see whether preparation endures as it does empirically. I think the issue of a more realistic cost function is also important, both in terms of the conceptual message and in terms of inducing the model to produce more realistic activity. Conceptually it matters because I don't think the central message should be 'preparation reduces upstream ATP usage by allowing motor cortex to be an amplifier'. I think the central message the authors wish to convey is that constraints on inputs make preparation a good strategy. Many of those constraints likely relate to the fact that upstream areas can't do things that motor cortex can do (else you wouldn't need a motor cortex) and it would be good if regularization reflected that assumption. Furthermore, additional forms of regularization would likely improve the realism of model responses, in ways that matter both aesthetically and conceptually. Yet while I think this is an important issue, it is also a deep and tricky one, and I think the authors need considerable leeway in how they address it. Many of the cost-function terms one might want to use may be intractable. The authors may have to do what makes sense given technical limitations. If some things can't be done technically, they may need to be addressed in words or via some other sort of non-optimization-based simulation.

Specific comments

As noted above, it would be good to show that preparatory subspace activity occurs similarly across delay durations. It actually might not, at present. For a zero ms delay, the simple magnitude-based regularization may be insufficient to induce preparation. If so, then the authors would either have to argue that a zero delay is actually never used internally (which is a reasonable argument) or show that other forms of regularization can induce zero-delay preparation.

Yes, that is a very interesting analysis to perform, which we had not considered before! When investigating this, we found that the zero-delay strategy does not rely on preparation in the same way as is seen in the monkeys. This seems to be a reflection of the fact that our “Go cue” corresponds to an “internal” go cue which would likely come after the true, “external go cue” – such that we would indeed never actually be in the zero delay setting. This is not something we had addressed (or really considered) before, although we had tried to ensure we referred to “delta prep” as the duration of the preparatory period but not necessarily the delay period. We have now included more discussion on this topic, as well as a new Supplementary Figure S10.

I agree with the authors that prior modeling work was limited by assuming the inputs to M1, which meant that prior work couldn't address the deep issue (tackled here) of why there should be any preparatory inputs at all. At the same time, the ability to hand-select inputs did provide some advantages. A strong assumption of prior work is that the inputs are 'simple', such that motor cortex must perform meaningful computations to convert them to outputs. This matters because if inputs can be anything, then they can just be the final outputs themselves, and motor cortex would have no job to do. Thus, prior work tried to assume the simplest inputs possible to motor cortex that could still explain the data. Most likely this went too far in the 'simple' direction, yet aspects of the simplicity were important for endowing responses with realistic properties. One such property is a large condition-invariant response just before movement onset. This is a very robust aspect of the data, and is explained by the assumption of a simple trigger signal that conveys information about when to move but is otherwise invariant to condition. Note that this is an implicit form of regularization, and one very different from that used in the present study: the input is allowed to be large, but constrained to be simple. Preparatory inputs are similarly constrained to be simple in the sense that they carry only information about which condition should be executed, but otherwise have little temporal structure. Arguably this produces slightly too simple preparatory-period responses, but the present study appears to go too far in the opposite direction. I would suggest that the authors do what they can to address these issue via simulations and/or discussion. I think it is fine if the conclusion is that there exist many constraints that tend to favor preparation, and that regularizing magnitude is just one easy way of demonstrating that. Ideally, other constraints would be explored. But even if they can't be, there should be some discussion of what is missing - preparatory plateaus, a realistic condition-invariant signal tied to movement onset - under the present modeling assumptions.

As described above, we have now included two additional figures. In the first one (S8, already discussed above), we used a temporal smoothness prior, and we indeed get slightly more realistic activity plateaus. In a second supplementary figure (S9), we have also considered using model predictive control (MPC) to optimize the inputs under an uncertain go cue arrival time. There, we found that removing the assumption that the delay period is known came with new challenges: in particular, it requires the specification of a “mental model” of when the Go cue will arrive. While it is reasonable to expect that monkeys will have a prior over the go time arrival cue that will be shaped by the design of the experiment, some assumptions must be made about the utility functions that should be used to weigh this prior. For instance, if we imagine that monkeys carry a model of the possible arrival time of the go cue that is updated online, they could nonetheless act differently based on this information, for instance by either preparing so as to be ready for the earliest go cue possible or alternatively to be ready for the average go cue. This will likely depend on the exact task design and reward/penalty structure. Here, we added simulations with those two cases (making simplifying assumptions to make the problem tractable/solvable using model predictive control), and found that the “earliest preparation” strategy gives rise to more realistic plateauing activity, while the model where planning is done for the “most likely go time” does not. We suspect that more realistic activity patterns could be obtained by e.g combining this framework with the temporal smoothness cost. However, the main point we wished to make with this new supplementary figure is that it is possible to model the task in a slightly more realistic way (although here it comes at the cost of additional model assumptions). We have now added more discussion related to those points. Note that we have kept our analyses on these new models to a minimum, as the main takeaway we wish to convey from them is that most components of the model could be modified/made more realistic. This would impact the qualitative behavior of the system and match to data but – in the examples we have so far considered – does not appear to modify the general strategy of networks relying on preparation.

On line 161, and in a few other places, the authors cite prior work as arguing for "autonomous internal dynamics in M1". I think it is worth being careful here because most of that work specifically stated that the dynamics are likely not internal to M1, and presumably involve inter-area loops and (at some latency) sensory feedback. The real claim of such work is that one can observe most of the key state variables in M1, such that there are periods of time where the dynamics are reasonably approximated as autonomous from a mathematical standpoint. This means that you can estimate the state from M1, and then there is some function that predicts the future state. This formal definition of autonomous shouldn't be conflated with an anatomical definition.

Yes, that is a good point, thank you for making it so clearly! Indeed, as previous work, we do not think of our “M1 dynamics” as being internal to M1, but they may instead include sensory feedback / inter-area loops, which we summarize into the connectivity, that we chose to have dynamics that qualitatively resemble data. We have now incorporated more discussion regarding what exactly the dynamics in our model represent.

eLife assessment

The valuable findings by Dasgupta et al demonstrate the role of Sema7a in fine tuning the morphology of the microcircuit between afferent axons and sensory hair cells in the lateral line organ. The loss and gain of function evidence provides solid support for a role for Sema7a in this process. Additional work is needed to determine the role for different isoforms in Sema7a-mediated synapse formation and chemoattraction as well as cell type specificity.

Reviewer #1 (Public Review):

Dasguta et al. have dissected the role of Sema7a in fine tuning of a sensory microcircuit in the posterior lateral line organ of zebrafish. They attempt to also outline the different roles of a secreted verses membrane-bound form of Sema7a in this process. Using genetic perturbations and axonal network analysis, the authors show that loss of both Sema7a isoforms causes abnormal axon terminal structure with more bare terminals and fewer loops in contact with presynaptic sensory hair cells. Further, they show that loss of Sema7a causes decreased number and size of both the pre- and post-synapse. Finally, they show that overexpression of the secreted form of Sema7a specifically can elicit axon terminal outgrowth to an ectopic Sema7a expressing cell. Together, the analysis of Sema7a loss of function and overexpression on axon arbor structure is fairly thorough and revealed a novel role for Sema7a in axon terminal structure.

Reviewer #2 (Public Review):

In this work, Dasgupta et al. investigate the role of Sema7a in the formation of peripheral sensory circuit in the lateral line system of zebrafish. They show that Sema7a protein is present during neuromast maturation and localized, in part, to the base of hair cells (HCs). This would be consistent with pre-synaptic Sema7a mediating formation and/or stabilization of the synapse. They use sema7a loss-of-function strain to show that lateral line sensory terminals display abnormal arborization. They provide highly quantitative analysis of the lateral line terminal arborization to show that a number of specific topological parameters are affected in mutants. Next, they ectopically express a secreted form of Sema7a to show that lateral line terminals can be ectopically attracted to the source. Finally, they also demonstrate that the synaptic assembly is impaired in the sema7a mutant. Overall, the data are of high quality and properly controlled. The availability of Sema7a antibody is a big plus, as it allows to address the endogenous protein localization as well to show the signal absence in the sema7a mutant. The quantification of the arbor topology should be useful to people in the field who are looking at the lateral line as well as other axonal terminals.

Reviewer #3 (Public Review):

The data reported here demonstrate that Sema7a defines the local behavior of growing axons in the developing zebrafish lateral line. The analysis is sophisticated and convincingly demonstrates effects on axon growth and synapse architecture. Collectively, the findings point to the idea that the diffusible form of sema7a may influence how axons grow within the neuromast and that the GPI-linked form of sema7a may subsequently impact how synapses form, though additional work is needed to strongly link each form to its' proposed effect on circuit assembly.

Comments on latest version:

The authors comprehensively and appropriately addressed most of the reviewers' concerns. In particular, they added evidence that hair cells express both Sema7A isoforms, showed that membrane bound Sema7A does not have long range effects on guidance, demonstrated how axons behave close to ectopic Sema7A, and analyzed other features of the hair cells that revealed no strong phenotypes. The authors also softened the language in many, but not all places. Overall, I am satisfied with the study as a whole.

Reviewer #4 (Public Review):<br /> <br /> This study provides direct evidence showing that Sema7a plays a role in the axon growth during the formation of peripheral sensory circuits in the lateral-line system of zebrafish. This is a valuable finding because the molecules for axon growth in hair-cell sensory systems are not well understood. The majority of the experimental evidence is convincing, and the analysis is rigorous. The evidence supporting Sema7a's juxtracrine vs. secreted role and involvement in synapse formation in hair cells is less conclusive. The study will be of interest to cell, molecular and developmental biologists, and sensory neuroscientists.

Author response:

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

eLife assessment 

Dasgupta and colleagues make a valuable contribution to the understanding how the guidance factor Sema7a promotes connections between mechanosensory hair cells and afferent neurons of the zebrafish lateral line system. The authors provide solid evidence that loss of Sema7a function results in fewer contacts between hair cells and afferents through comprehensive quantitative analysis. Additional work is needed to distinguish the effects of different isoforms of Sema7a to determine whether there are specific roles of secreted and membrane bound forms. 

Public Reviews:

Reviewer #1 (Public Review):

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

Reviewer #2 (Public Review):

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

Reviewer #3 (Public Review):

The data reported here demonstrate that Sema7a defines the local behavior of growing axons in the developing zebrafish lateral line. The analysis is sophisticated and convincingly demonstrates effects on axon growth and synapse architecture. Collectively, the findings point to the idea that the diffusible form of sema7a may influence how axons grow within the neuromast and that the GPI-linked form of sema7a may subsequently impact how synapses form, though additional work is needed to strongly link each form to its' proposed effect on circuit assembly. 

The revised manuscript is significantly improved. The authors comprehensively and appropriately addressed most of the reviewers' concerns. In particular, they added evidence that hair cells express both Sema7A isoforms, showed that membrane bound Sema7A does not have long range effects on guidance, demonstrated how axons behave close to ectopic Sema7A, and analyzed other features of the hair cells that revealed no strong phenotypes. The authors also softened the language in many, but not all places. Overall, I am satisfied with the study as a whole. 

Reviewer #4 (Public Review):

This study provides direct evidence showing that Sema7a plays a role in the axon growth during the formation of peripheral sensory circuits in the lateral-line system of zebrafish. This is a valuable finding because the molecules for axon growth in hair-cell sensory systems are not well understood. The majority of the experimental evidence is convincing, and the analysis is rigorous. The evidence supporting Sema7a's juxtracrine vs. secreted role and involvement in synapse formation in hair cells is less conclusive. The study will be of interest to cell, molecular and developmental biologists, and sensory neuroscientists. 

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

In their revised manuscript, Dasgupta et al. have provided further experiments to address the role of Sema7a (sec and GPI-anchored) in regulating axon guidance in the lateral line system. Specifically, the inclusion of the heat shock controls and FM labeling to show hair cell mechanotransduction were crucial to interpretation of the results. However, there are still concerns about the specificity of the results. My primary concern is if the change in axon patterning is specifically due to loss of Sema7a in the mutant hair cells. These animals are morphologically very abnormal and, in the rebuttal, the authors state that hair cell number is reduced. This is not quantified in the manuscript and should be included. 

Thank you for this suggestion. We have included the data in the manuscript in lines 137-139, in Figure 2—figure supplement 1B, and in the source data for Figure 2 and Figure 2-figure supplements.

If there is not a function for Sema7a in hair cells themselves, why is the number reduced? 

The sema7a-/- homozygous mutants are not viable and they die by 6 dpf. The loss of Sema7A protein produce other developmental defects including brain edema and a curved body axis. We believe a slight but not significant decrease in hair cell number may arise from a minute developmental delay in the morphogenesis of the neuromast. We have accordingly quantified our data at three distinct developmental stages-at 2 dpf, 3 dpf, and 4 dpf-and have incorporated them in the revised manuscript.

Additionally, FM data should be quantified and presented in animals without a transgene in the same excitation/emission spectra for clearer interpretation of the staining.

We have quantified the intensities of labeling with FM 4-64 styryl dye from the control and the sema7a-/- mutant larvae and incorporated the data in lines 139-146, in Figure 2—figure supplement 1D, and in source data for Figure 2 and Figure 2-figure supplements. We Kept the transgenes to concurrently show the arborization phenotype, hair cell morphology, and the FM 4-64 incorporation between the genotypes. 

Rescue analysis using the myo6d promotor would allow the authors to ensure that the axon deficits can be rescued by putting Sema7a back into the sensory hair cells. Transient transgenesis could be useful for this approach and would not require the creation of a stable line. This could be done with both forms of Sema7a allowing the true assessment of whether or not the secreted and GPI-anchored form have disparate functions as claimed in lines 418424. 

Although we recognize the importance of the rescue of the sema7a-/- mutant phenotype with the sema7asec and the sema7aGPI transcripts, it is not possible for us to perform that experiment at the moment, for the first author will leave the lab next week.  However, he plans to continue work on this project as an independent investigator to dissect the individual roles of the transcript variants in specifying the pattern of sensory arborization, a project that includes generation of transcript-specific knockout animals and rescue experiments with stable transgenic fish lines. 

Other concerns:

(1) The timeline of the heat shock experiment is confusing to me and, therefore, it makes me question the specificity of those results. Based on the speed of axon outgrowth and the time necessary for transcription and translation after heat shock induction of the transgene, it is unclear to me how the axon growth defects could occur in the timeline provided. Imaging two hours after the start of the heat shock is very rapid and speaks to either an indirect effect of the transgenesis on the axon growth or a leaky promotor/induction paradigm. It is possible I am just misunderstanding the set up but, from what I could gather, the imaging is being done 2 hrs after the start of the heat shock. This should be clarified. 

The axons of the zebrafish posterior lateral line migrate relatively fast. The pioneering axons migrate at around 120 μm/hour (Sato et. al., 2010) and the follower axons migrate at almost 30-80 μm/hour (Sato et. al., 2010). The heat-shock promoter that we have utilized, hsp70l, is highly effective in inducing gene expression and subsequent protein formation within 30 to 60 mins. We believe an hour of heat shock and an hour of incubation post heat shock is sufficient to induce directed axon migration to a distance that spans from 27 μm to 140 μm. 

We strongly believe that the directed arborization of the sensory axons towards the Sema7Asec source is not due to an indirect effect of transgenesis or leaky promoter induction, as in all 18 of the injected but not heat-shocked control larvae we did not observe ectopic Sema7Asec expression, and no aberrant projection was formed from the sensory arbor network. We highlight this observation in lines 297-299 and in Figure 4E.

Sato et. al., 2010: Single-cell analysis of somatotopic map formation in the zebrafish lateral line system. Developmental Dynamics 239:2058–2065, 2010.

Similarly, it would help to clarify if t(0) in the figure is the onset of the heat shock or onset of imaging two hours after the heat shock is started. 

The t=0 hour in the Figure 4I denotes the onset of imaging two hours after the heat shock began. We have clarified this in the manuscript in lines 1155-1156.

(2) In the rebuttal, the line numbers cited do not match up with the appropriate text, I believe.

We have corrected this and updated the manuscript.

(3) Some of the supplemental figures are not mentioned in the text, or I could not find them. For example: Figure 1 supplement 2J. 

Thank you for pointing this. We have corrected the manuscript, and the new information is added in line 114.  

(4) Table 1 statistics: were these adjusted for multiple comparisons using a bonferroni correction or something similar? This is necessary for statistical significance to be meaningful. 

We did not adjust the p-values for multiple comparisons because the values correspond to only three or four statistical tests per experiment, strongly indicating the unlikelihood of erroneous significance due solely to multiple tests.

(5) Figure 1I and 1-S3 - The legend states a positive correlation between axonal signal and sema7A signal. Correlations are 0.5, 0.6, and 0.4 (2,3, 4dpf). This is not a convincing positive correlation. At best this is no to a very weak positive correlation. 

In lines 122-126 we mention that the basal association of the sensory arbors shows a positive correlation with Sema7A accumulation. We never emphasize on the strength of the correlation. However, a consistent positive correlation at three different developmental stages suggests that progressive Sema7A accumulation at the base of the hair cells may guide the sensory arbors to increasingly associate themselves with the hair cells.    

Reviewer #2 (Recommendations For The Authors):

I am a bit disappointed that the authors elected not to experimentally address the issue raised by all reviewers: whether the secreted or membrane bound isoform is active in hair cells. They rather decided to change their interpretation in the text. It is fine, given the eLife review structure. However, that would make the manuscript much stronger. Other issues were adequately addressed through textual changes as well. 

Although we recognize the importance of the rescue of the sema7a-/- mutant phenotype with the sema7asec and the sema7aGPI transcripts, it is not possible for us to perform that experiment at the moment, for the first author will leave the lab next week.  However, he plans to continue work on this project as an independent investigator to dissect the individual roles of the transcript variants in specifying the pattern of sensory arborization, a project that includes generation of transcript-specific knockout animals and rescue experiments with stable transgenic fish lines. 

Reviewer #3 (Recommendations For The Authors):

Overall, I am satisfied with the study as a whole and just have a few minor comments that remain to be addressed. 

(1) Although the authors say that they added appropriate no plasmid/heatshock-only and plasmid-only/no heatshock controls, these results need to be presented more clearly, as they are separated in the paper and only one was quantified (i.e. 100% of embryos showed no defect). Please just make it clear that no defects were observed in either control for either experiment (both secreted and membrane bound ectopic expression). 

We have clearly stated this information in lines 297-299 and 343-345.

(2) Please add a compass to Fig. 1A to indicate the orientation of the neuromast. It would also be helpful to add labels for developmental ages to all of the figures, rather than making the reader look it up in the legend. 

We have updated the Figure 1A and the corresponding figure legend in lines 882883 . We have denoted the larval age in the figure legends to keep the individual images uncluttered.  

(3) For the RT-PCR experiments in Figure 1, no negative control was included to show that supporting cell or neuronal genes are not detected in the purified hair cells and v.v. that neither isoform is detected in supporting cells or neurons. I ask only because there is a lot of immune-signal outside of the hair cells and I am curious whether that is secreted or might come from other cell types. For neurons and supporting cells, simply demonstrating absence of Sema7a overall would suffice. 

We have utilized the transgenic line Tg(myo6b:actb1-EGFP) that expresses the fluorophore GFP specifically in the hair cells of the neuromast. Unfortunately, we do not possess a transgenic line that reliably and specifically labels the support cells in the neuromast. Hence, in our sorting experiment the GFP-negative cells that are collected from the trunk segments of the larvae contain all the non-hair cells including epidermal cells, neuronal cells, and immune cells etc. Such a mixture of varied cellular identity may not serve as a reliable negative control. 

In Figure 7, we have plotted the normalized expression values of the sema7a gene in the neuromast. The plot clearly depicts that the source of Sema7A is the young and the mature hair cells, not the support cells. We further confirm this observation by

immunostaining where the Sema7A signal is highly restricted to the hair cells and not in any other cell in the neuromast (Figure 1E). Immunostaining further demonstrates that the lateral line sensory arbors also do not produce the Sema7A protein (Figure 1H; Video 1).

We agree with the reviewer that there are diverse immune cells, including macrophages in and around the neuromast. These macrophages are dynamic and possess highly ramified structure (Denans et. al., 2022). In all our Sema7A immunostainings, we never observed structures that resemble macrophages. Albeit we cannot confirm that Sema7A is not expressed in a distant immune cell, but we highly doubt that signal coming from immune cells is impacting hair cell innervation by the sensory arbors during homeostatic development.

Denans et. al., 2022: Nature Communications volume 13, Article number: 5356 (2022).

(4) In Figure 1, Supplement 4, I do not see the immunogen labeled in blue. 

We have corrected the figure legend. The immunogenic region of the Sema7A protein is now clearly denoted in the figure legend of Figure 1—figure supplement 4.

(5) In Figure 2, please add a control image as requested, as that enables direct comparison. There is ample room in the figure. 

We have updated the Figure 2 and made the suggested change.

(6) In Figure 2, Supplement 1, the FM4-64 data are not presented in a quantified fashion. Please report at least how many embryos showed reliable uptake and preferably how many hair cells per embryo showed reliable uptake. 

We have quantified the FM 4-64 intensities in control and sema7a-/- mutant larvae. The new data is added to the manuscript in lines 142-146, 577-579 , and in Figure 2—figure supplement 1D.

(7) In Figure 3, there seems to be a typo in the figure legend: "mutants in the same larvae" does not make sense to me. 

We have corrected the error. The modified statement is represented in lines 10671068.

(8) The text should refer more explicitly to the statistical tests reported in Table 1, i.e. as the results are presented. 

In lines 1105 and 1109, we clearly state the statistical tests that were performed.

(9) In Figure 6, Supplement 1, please show the raw data points not just the bar graphs

We have updated the Figure 6—figure supplement 1.

(10) Minor point: the authors state that they addressed the distance over which secreted Sema7A may act, but this was not evident to me in the text. Please make this finding clearer.

We have clarified this information in lines 310-311.

(11) Finally, the discussion contains a statement that is not supported by the data: "We have discovered dual modes of Sema7A function in vivo." They have discovered evidence that there are two isoforms, that loss of both disrupts connectivity, and that overexpression of only the secreted form can elicit growth from a distance. However, there is no direct evidence that the membrane-bound form is responsible for local effects. It is formally possible still that the phenotypes are a result of dual roles for the secreted form. It is clear that another manuscript is forthcoming that will expand on the role of the transmembrane form, but for this manuscript, the authors should make firm conclusions only about the data presented herein.

Thank you for this suggestion. We have modified the manuscript in lines 425-434.

Reviewer #4 (Recommendations For The Authors):

The authors have made significant changes to the manuscript based on the comments of the reviewers. It is now suitable for publication.

eLife assessment

This important work provides convincing data on neuronal heterogeneity in the dorsal raphe nucleus (DRN), focusing on their electrophysiological properties, morphology, and susceptibility to the neurodegeneration of noradrenaline and dopamine systems in the Parkinsonian state. These findings suggest a significant interplay between catecholaminergic systems in healthy and parkinsonian conditions, as well as neuronal structure and function. Such findings provide a strong foundation for basic scientists as well as pre-clinical researchers interested in the role of dorsal raphe neurons in Parkinson's disease.

Reviewer #1 (Public Review):

Summary:

People with Parkinson's disease often experience a variety of nonmotor symptoms, the biological bases of which remain poorly understood. Johansson et al began to study potential roles of the dorsal raphe nucleus (DRN) degeneration in the pathophysiology of neuropsychiatric symptoms in PD.

Strengths:

Boi et al validated a transgenic reporter mouse line that can reliably label dopaminergic neurons in the DRN. This brain region shows severe neurodegeneration and has been proposed to contribute to the manifestation of neuropsychiatric symptoms in PD. Using this mouse line (and others), Boi and colleagues characterized electrophysiological and morphological phenotypes of dopaminergic and serotoninergic neurons in the raphe nucleus. This study involved very careful topographical registration of recorded neurons to brain slices for post hoc immunohistochemical validation of cell identity, making it an elegant and thorough piece of work.

Of relevance to PD pathophysiology, the authors evaluated the physiological and morphological changes of DRN serotoninergic and dopaminergic neurons after a partial loss of nigrostriatal dopamine neurons, which serves as a mouse model of early parkinsonian pathology. Moreover, the authors identified a series of physiological and morphological changes of subtypes of DRN neurons that depend on nigral dopaminergic neurodegeneration, LC noradrenergic neurodegeneration, or both. Indeed this work highlights the importance of LC noradrenergic degeneration in PD pathophysiology.

Overall, this is a well-designed study with high significance to the Parkinson's research field.

Reviewer #2 (Public Review):

In this paper, Boi et al. thoroughly classified the electrophysiological and morphological characteristics of serotonergic and dopaminergic neurons in the DRN and examined the alterations of these neurons in the 6-OHDA-induced mouse PD model. Using whole-cell patch clamp recording, they found that 5-HT and dopamine (DA) neurons in the DRN are electrophysiologically distinct from each other. Additionally, they characterized distinct morphological features of 5-HT and DA neurons in the DRN. Notably, these specific features of 5-HT and DA neurons in the DRN exhibited different changes in the 6-OHDA-induced PD model. Then the authors utilized desipramine (DMI) to separate the effects of nigrostriatal DA depletion and noradrenaline (NA) depletion induced by 6-OHDA. Interestingly, protection from NA depletion by DMI pretreatment reversed the changes in 5-HT neurons, while having a minor impact on the changes in DA neurons in the DRN. These data indicate that the role of NA lesion in the altered properties of DRN 5-HT neurons by 6-OHDA is more critical than that of DA lesions.

Overall, this study provides foundational data on the 5-HT and DA neurons in the DRN and their potential involvement in PD symptoms. Given the deficits of the DRN in PD, this paper may offer insights into the cellular mechanisms underlying non-motor symptoms associated with PD.

Author response:

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

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I have no more experiment to ask but the following errors should be corrected prior.

(1) L. 183-198: Figure 3 panels were erroneously referred in several places.

This has been corrected.

(2) L.182-183: description of active/total cell numbers in main text does not match numbers in Figure 3B

This has been corrected.

(3) L.185-187: Figure 3C indicates significant changes of rheobase only between DMI+6OHDA versus 6-OHDA group. Statistical comparison between sham and DMI+6-OHDA was not provided, which may change the interpretation of the data in Figure 3B, C: "...these findings suggest that the 6-OHDA induced lesion of midbrain dopaminergic neurons evoked the increased firing of DRN5-HT neurons" (L.185-187).

We thank the reviewer for highlighting this point. Indeed, a Kruskal-Wallis test comparing all three groups revealed a significantly lower rheobase in DMI + 6-OHDA mice compared to Sham while the 6-OHDA injected group was not affected. Therefore, the increased firing of DRN5-HT neurons recorded in 6-OHDA injected mice pretreated with DMI also critically involves the noradrenergic system. This is now included in the revised results section of the manuscript (lines 190-197).

(4) L. 188: The description of "While the excitability of DRN5-HT neurons was not affected in 6-OHDA mice..." does not match the clearly increased cellular excitability shown in Figure 3G-I.

This has been corrected and we are now referring more specifically to the rheobase, which is not affected in 6-OHDA mice.

(5) Mann-Whitney tests were inappropriately used for statistics in Figures 3-6: Multiple comparisons (>=3 groups) should be performed one-way ANOVA or the Kruskal-Wallis test for nonparametric data.

We thank the reviewer for the comment. We now applied the one-way ANOVA/KruskalWallis tests and the text has been modified accordingly.

(6) It seems that the data points in some panels of Figure 4C represented a cell, but others were averaged within a mouse (Figure 4D). This needs to be clarified or corrected.

None of the data in Figure 4 was averaged within a mouse. In the the type of chosen graph (aligned dot plot) the equal data are overlapped.

Reviewer #2 (Recommendations For The Authors):

The authors' revised manuscript has addressed most of my concerns. However, I'm not convinced by the authors' claim regarding Figure 5B. It would be great if the authors at least discuss in their manuscript why the DMI pretreatment group alone, not the 6OHDA group, significantly lowers the firing rate of DRN (DA) and increases the Erest of DRN (DA), compared to the sham-lesion group. These statistically significant data are not explained at all in the revised manuscript (This effect can be explained by the neuroprotection of NA-neurons from 6-OHDA toxicity?).

We thank the reviewer for this comment. Since using a one-way ANOVA or a KruskalWallis test for comparing the three groups (as suggested by reviewer 1), the changes previously shown in Figure 5B are not significant.

Author response:

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

eLife assessment

This manuscript represents a cleanly designed experiment for assessing biological motion processing in children (mean age = 9) with and without ADHD. The group differences concerning accuracy in global and local motion processing abilities are solid, but the analyses suggesting dissociable relationships between global and local processing and social skills, age, and IQ are inconclusive. The results are useful in terms of understanding ADHD and the ontogenesis of different components of the processing of biological motion.

We thank the editors and reviewers for their valuable feedback and constructive comments. We have carefully considered each point raised by the reviewers and made the necessary revisions to the manuscript. Regarding the relationships between global and local BM processing, the accumulated evidence from previous studies has converged on the dissociation of the two BM components, e.g., while global BM processing is susceptible to learning and practice, local BM processing does not show a learning trend (Chang and Troje, 2009; Grossman et al., 2004), and the brain activations in response to local and global BM cues are different (Chang et al., 2018; Duarte et al., 2022). Nevertheless, we concurred with reviewers that the evidence for such dissociation from the current study by itself is not strong enough. Therefore, we have toned down on this point and no longer claimed the dissociation (including the title). Based on the current results, we focused our discussion on the different aspects of BM processing in children with and without ADHD.

Public Reviews:

Reviewer #1 (Public Review):

Summary:

The paper presents a nice study investigating the impairments of biological motion perception in individuals with ADHD in comparison with neurotypical controls. Motivated by the idea that there is a relationship between biological motion perception and social capabilities, the authors investigated the impairments of local and global (holistic) biological motion perception, the diagnosis status, and several additional behavioral variables that are affected in ADHS (IQ, social responsiveness, and attention / impulsivity). As well local as global biological motion perception is impaired in ADHD individuals. In addition, the study demonstrates a significant correlation between local biological motion perception skills and the social responsiveness score in the ADHD group, but not in controls. A path analysis in the ADHD group suggests that general performance in biological motion perception is influenced mainly by global biological motion perception performance and attentional and perceptual reasoning skills.

Strengths:

It is true that there exists not much work on biological motion perception and ADHD. Therefore, the presented study contributes an interesting new result to the biological motion literature, and adds potentially also new behavioral markers for this clinical group. The design of the study is straightforward and technically sound, and the drawn conclusions are supported by the presented results.

Thanks for this positive assessment of our work.

Weaknesses:

Some of the claims about the relationship between genetic factors and ADHD and the components of biological motion processing have to remain speculative at this point because genetic influences were not explicitly tested in this paper. Specifically, the hypothesis that the perception of human social interaction is critically based on a local mechanism for the detection of asymmetry in foot trajectories of walkers (this is what 'BL-local' really measures), or on the detection of live agents in cluttered scenes seems not very plausible.

Thanks for these comments. We agree that the relationship between genetic factors and BM perception remains to be further examined, as we did not test the genetic influences in this study. We have deleted relavant discussion about genetics. Based on our results, we discuss the possible mechanisms behind the relationship between local BM processing and social interaction in the revised manuscript as follows:

“As mentioned above, we found a significant negative correlation between the SRS total score and the accuracy of local BM processing, specifically in the ADHD group. This could be due to decreased visual input related to atypical local BM processing, which further impairs global BM processing. According to the two-process theory of biological motion processing61, local BM cues guide visual attention towards BM stimuli55,62. Consequently, the visual input of BM stimuli increases, facilitating the development of the ability to process global BM cues through learning21,63. The latter is a prerequisite for attributing intentions to others and facilitating social interactions with other individuals20,64,65. Thus, atypical local BM processing may contribute to impaired social interaction through altered visual inputs. Further empirical studies are required to confirm these hypotheses.” (lines 417 - 428)

Based on my last comments, now the discussion has been changed in a way that tries to justify the speculative claims by citing a lot of other speculative papers, which does not really address the problem. For example, the fact that chicks walk towards biological motion stimuli is interesting. To derive that this verifies a fundamental mechanism in human biological motion processing is extremely questionable, given that birds do not even have a cortex. Taking the argumentation of the authors serious, one would have to assume that the 'Local BM' mechanism is probably located in the mesencephalon in humans, and then would have to interact in some way with social perception differences of ADHD children. To me all this seems to make very strong (over-)claims. I suggest providing a much more modest interpretation of the interesting experimental result, based on what has been really experimentally shown by the authors and closely related other data, rather than providing lots of far-reaching speculations.

In the same direction, in my view, go claims like 'local BM is an intrinsic trait' (L. 448) , which is not only imprecise (maybe better 'mechanisms of processing of local BM cues') but also rather questionable. Likely, this' local processing of BM' is a lower level mechanisms, located probably in early and mid-levels of the visual cortex, with a possible influence of lower structures. It seems not really plausible that this is related to a classical trait variables in the sense of psychology, like personality, as seems to be suggested here. Also here I suggest a much more moderate and less speculative interpretation of the results.

We thank the reviewer for pointing out these issues. According to these comments, we have carefully revised the discussion to avoid strong (over-) claims. We have deleted the example of chicks, but substituted with more empirical studies to explain our results. We agree that the Local BM mechanism is probably located in subcortical regions in humans, which were reported by some MRI studies (Chang et al., 2018; Hirai and Senju, 2020; Loula et al., 2005). We have added some evidence that atypical local BM processing may decrease visual inputs related to social information as follows:

“According to the two-process theory of biological motion processing61, local BM cues guide visual attention towards BM stimuli55,62. Consequently, the visual input of BM stimuli increases, facilitating the development of the ability to process global BM cues through learning21,63. The latter is a prerequisite for attributing intentions to others and facilitating social interactions with other individuals20,64,65. Thus, atypical local BM processing may contribute to impaired social interaction through altered visual inputs.” (lines 421 - 427)

We have also deleted the clarims of 'local BM is an intrinsic trait' (originally L. 448) and related discussion as it was not conclusive based on the current study.

Reviewer #2 (Public Review):

Summary:

Tian et al. aimed to assess differences in biological motion (BM) perception between children with and without ADHD, as well as relationships to indices of social functioning and possible predictors of BM perception (including demographics, reasoning ability and inattention). In their study, children with ADHD showed poorer performance relative to typically developing children in three tasks measuring local, global, and general BM perception. The authors further observed that across the whole sample, performance in all three BM tasks was negatively correlated with scores on the social responsiveness scale (SRS), whereas within groups a significant relationship to SRS scores was only observed in the ADHD group and for the local BM task. Local and global BM perception showed a dissociation in that global BM processing was predicted by age, while local BM perception was not. Finally, general (local & global combined) BM processing was predicted by age and global BM processing, while reasoning ability mediated the effect of inattention on BM processing.

Strengths:

Overall, the manuscript is presented in a clear fashion and methods and materials are presented with sufficient detail so the study could be reproduced by independent researchers. The study uses an innovative, albeit not novel, paradigm to investigate two independent processes underlying BM perception. The results are novel and have the potential to have wide-reaching impact on multiple fields.

We appreciate the reviewer’s positive feedback very much.

Weaknesses:

The manuscript has greatly improved in clarity and methodological considerations in response to the review. There are only a few minor points which deserve the authors' attention:

When outlining the moviation for the current study, results from studies in ADHD and ASD are used too interchangeably. The authors use a lack of evidence for contributing (psychological/developmental) factors on BM processing in ASD to motivate the present study and refer to evidence for differences between typical and non-typical BM processing using studies in both ASD and ADHD. While there are certainly overlapping features between the two conditions/neurotypes, they are not to be considered identical and may have distinct etiologies, therefore the distinction between the two should be made clearer.

We thank the reviewer for pointing out this issue. We have removed some unnecessary citations about ASD and referred to studies about social cognition in ADHD to elaborate the motivation of this study:

“Further exploration of a diverse range of social cognitions (e.g., biological motion perception) can provide a fresh perspective on the impaired social function observed in ADHD. Moreover, recent studies have indicated that the social cognition in ADHD may vary depending on different factors at the cognitive, pathological, or developmental levels, such as general cognitive impairment5, symptoms severity8, or age5. Nevertheless, understanding how these factors relate to social cognitive dysfunction of in ADHD is still in its infancy. Bridging this gap is crucial as it can help depict the developmental trajectory of social cognition and identify effective interventions for impaired social interaction in individuals with ADHD.” (lines 53 - 62)

In the first/main analysis, is unclear to me why in the revised manuscript the authors changed the statistical method from ANOVA/ANCOVA to independent samples t-tests (unless the latter were only used for post-hoc comparisons, then this needs to be stated). Furthermore, although p-values look robust, for this analysis too it should be indicated whether and how multiple comparison problems were accounted for.

Thanks for the reviewer’s comments. According to the suggestions from reviewer #3, it may be inapposite to regard gender as a covariate in ANOVA, which may violate the assumptions of ANCOVA. To ensure that gender does not influence the results, firstly, we separated boys and girls on the plots with different coloured individual data points, and there are no signs of a gender effect in their TD group. Secondly, we use t-tests to examine the difference between TD and ADHD groups. Finally, we conducted a subsampling analysis with balanced data, and the results remained consistent.

In part 1 of the results, we aimed to compare the task accuracies between the TD and ADHD groups in three independent tasks, which assess the participants’ abilities to process three types of BM cues. We assumed that individuals with ADHD show poorer performance in three tasks compared to TD individuals. With regard to that, we consider that multiple comparisons may not be necessary.

Reviewer #3 (Public Review):

Strengths:

The authors present differences between ADHD and TD children in biological motion processing, and this question has not received as much attention as equivalent processing capabilities in autism. They use a task that appears well controlled. They raise some interesting mechanistic possibilities for differences in local and global motion processing, which are distinctions worth exploring. The group differences will therefore be of interest to those studying ADHD, as well as other developmental conditions, and those examining biological motion processing mechanisms in general.

We appreciate the reviewer’s positive assessment of this work.

Weaknesses:

The data are not strong enough to support claims about differences between global and lobal processing wrt social communication skills and age. The mechanistic possibilities for why these abilities may dissociate in such a way are interesting, but the crucial tests of differences between correlations do not present a clear picture. Further empirical work would be needed to test the authors' claims. Specifics:

The authors state frequently that it was the local BM task that related to social communication skills (SRS) and not the global tasks. However, the results section shows a correlation between SRS and all three tasks. The only difference is that when looking specifically within the ADHD group, the correlation is only significant for the local task. The supplementary materials demonstrate that tests of differences between correlations present an incomplete picture. Currently they have small samples for correlations, so this is unsurprising.

Thanks for this comment. We agree with the reviewer that the relationship between local and global processing with social communication and age needs more expirical work. Based on our results, there are only possible dissociable roles of local and global BM processing. The accumulated evidence from previous studies has converged on this dissociation, e.g., whild global BM processing is susceptible to learning and practice, local BM processing does not show a learning trend (Chang and Troje, 2009; Grossman et al., 2004), and the brain activations in response to local and global BM cues are different (Chang et al., 2018; Duarte et al., 2022). We concurred with reviewers that the evidence for such dissociation from the current study by itself is not strong enough. Therefore, we have toned down on this point and no longer emphasized the dissociation. Based on the current results, we focused our discussion on the different aspects of BM processing in children with and without ADHD. Future studies with larger sample sizes are needed to confirm this disociable relationship.

Theoretical assumptions. The authors make some statements about local vs global biological motion processing that should still be made more tentatively. They assume that local processing is specifically genetically whereas global processing is a product of experience. These data in newborn chicks are controversial and confounded - I cannot remember the specifics but I think there an upper vs lower visual field complexity difference here.

We appreciate the reviewer’s suggestion. We agree that the relationship between genetic factors and BM perception remains to be further examined as we didn’t perform any genetic analysis in the current study. Some speculative papers have been removed, so do the statement about newborn chicks given the controversial and confounded results. We have toned down our claims and povided a moderate interpretation of the results:

“Sensitivity to local BM cues emerges early in life54,55 and involves rapid processing in the subcortical regions16,56-58. As a basic pre-attentive feature23, local BM cues can guide visual attention spontaneously59,60. In contrary, the ability to process global BM cues is related to slow cortical BM processing and is influenced by many factors such as attention25,26 and visual experience21,51. As mentioned above, we found a significant negative correlation between the SRS total score and the accuracy of local BM processing, specifically in the ADHD group. This could be due to decreased visual input related to atypical local BM processing, which further impairs global BM processing. According to the two-process theory of biological motion processing61, local BM cues guide visual attention towards BM stimuli55,62. Consequently, the visual input of BM stimuli increases, facilitating the development of the ability to process global BM cues through learning21,63. The latter is a prerequisite for attributing intentions to others and facilitating social interactions with other individuals20,64,65. Thus, atypical local BM processing may contribute to impaired social interaction through altered visual inputs.” (lines 413 - 427)

“Few developmental studies have been conducted on local BM processing. The ability to process local BM cues remained stable and did not exhibit a learning trend21,25. A reasonable interpretation may be that local BM processing is a low-level mechanism, probably performed by the primary visual cortex and subcortical regions such as the superior colliculus, pulvinar, and ventral lateral nucleus14,56,61.” (lines 441- 446)

Readability. The manuscript needs very careful proofreading and correction for grammar. There are grammatical errors throughout.

Thank the reviewer for this feedback. We have performed thorough proofreading and corrected grammatical errors throughout the manuscript.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I thank the authors for their revisions that address several of the minor points that I raised in my last review. A number of requests are still not sufficiently answered:

L. 290 ff.: These model 'BM-local = age + gender etc ' is a pretty sloppy notation. I think what is meant that a GLM was used that uses the predictors genderetc. time appropriate beta_i values. This formulas should be corrected or one just says that a GLM was run with the predictors gender

The same criticism applies to these other models that follow.

This was corrected.

However, the corrected text remains sloppy: example: 'BM-locaL = ...' What exacty is 'BM-Local' the accuracy? etc. Here a precise notation shoudl be given that clearly names which variables are used here as predictors and target variables.

We appreciate the reviewer’s suggestion. We clarified which variables are used in our model and gived them precise notations:

“Three linear models were built to investigate the contributing factors: (a) ACClocal = β0 + β1 * age + β2 * gender + β3 * FIQ + β4 * QbInattention, (b) ACCglobal = β0 + β1 * age + β2 * gender + β3 * FIQ + β4 * QbInattention, and (c) ACCgeneral = β0 + β1 * age + β2 * gender + β3 * FIQ + β4 * QbInattention + β5 * ACClocal + β6 * ACCglobal. ACClocal, ACCglobal and ACCgeneral refer to the response accuracies of the three tasks in the ADHD group, and QbInattention is the standardised score for sustained attention function.” (lines 337 - 343)

All these models assume linearity of the combination of the predictors. was this assumption verified?

We referred to the previous study of BM perception in children. They found main predictor variables, including IQ (Rutherford et al., 2012; Jones et al., 2011) and age (Annaz et al., 2010; van et al., 2016), have a linear relation with the ability of BM processing.

This answer is insufficient and not convincing. Because a variable Y depends linearly on predictor A and B in some other study, this does not imply that is is also linear in predictor C, or does not show interactions with such predictors in the present study.

What is needed here is the testing of models with interaction terms and verifying that such models are not better predictors. If authors do not want to do this, they need at least to clearly point out that they made the strong assumption of linearity of their model, which might be wrong and thus be a substantial limitation of their analysis.

Thanks for the suggestion. We tried to compare each possible mode with and without relative interactions. The results showed that the change of Coefficient of Determination (R-squared, R2) between the two models was not statistically significant.

L. 296ff.: For model (b) it looks like general BM performance is strongly driven by the predictor global BM performance in the ADHD group. Does the same observation also apply to the controls?

The same phenomenon was not observed in TD children. We have briefly discussed this point in the Discussion section of the revised manuscript (lines 449 - 459).

Was such a path analysis also done for the TD subjects or not? If yes, was then also predicted that the variable BM-Global largely and directedly influences the variable BM-General? (The answer refers to the general discussion section, where no such analysis is presented, as far as I understand.)

Thank you for your comment. We also conduct a path analysis similar to that in the ADHD group. There is no statistically significant mediator effect in the TD group. Please see Figure S3 for complete statistics.

Reviewer #2 (Recommendations For The Authors):

(1) Please add public access to the data repository so data availability can be assessed.

The data analyzed during the study is available at https://osf.io/37p5s/.

(2) Lines 119-115: The differences observed in ADHD participants in the studies referenced here were relative to what group? The last sentence here also refers to two groups, and it is difficult to gather which specific groups are meant, also because the two references relate to both ADHD and ASD samples. Please clarify.

The suggestion is well taken. We have clarified the expressions accordingly:

“Specifically, compared with the typically developing (TD) group, children with ADHD showed reduced activity of motion-sensitive components (N200) while watching biological and scrambled motions, although no behavioural differences were observed. Another study found that children with ADHD performed worse in BM detection with moderate noise ratios than the TD group32.” (lines 100 - 105)

(3) Line 116: I'm not sure what is meant by 'despite initial indications' - please briefly specify/summarise here why the investigation into BM processing in ADHD is warranted.

Thank the reviewer for pointing out this issue. We rephrase this part and briefly specify “why the investigation into BM processing in ADHD is warranted”:

“Despite initial findings about atypical BM perception in ADHD, previous studies on ADHD treated BM perception as a single entity, which may have led to misleading or inconsistent findings28. Hence, it is essential to deconstruct BM processing into multiple components and motion features.” (lines 108 -111)

(4) Lines 290-293: Please complete the sentence.

Thank the reviewer for pointing out this issue. Th sentence has been completed:

“For Task 2 and 3, where children were asked to detect the presence or discriminate the facing direction of the target walker, TD group have higher accuracies than the ADHD group (Task 2 - TD: 0.70 ± 0.12, ADHD: 0.59 ± 0.12, t73 = 3.677, p < 0.001, Cohen's d = 0.861; Task 3 - TD: 0.79 ± 0.12, ADHD: 0.63 ± 0.17, t73 = 4.702, p < 0.001, Cohen's d = 1.100).” (lines 284 - 288)

Reviewer #3 (Recommendations For The Authors):

(1) Conclusions concerning differences between the local and global tasks wrt SRS and age (see above). I believe the authors need to reword throughout to reflect that the tests of differences between these crucial correlations did not present a clear picture.

We have reworded throughout the paper to reflect the inconclusiveness with regard to the relationship between local and global processing with social communication based on this study only. Future studies with larger sample sizes are needed to confirm this conclusion. The mechanism for this dissociable relationship should be validated by more psychologial tests in the future studies.

(2) I would again tone down the discussion of genetic specification of local processing, given it is highly controversial.

We thank the reviewer for pointing out the issue. We agree the point about the genetic specification of local processing remains controversial. The interpretation of results about local BM processing has been rephrased. Please refer to our response to the point #2 mentioned.

(3) The manuscript needs very careful proofreading and grammatical correction throughout.

Thanks for the suggestion to check the grammar. We have carefully proofread the manuscript to correct grammatical errors

eLife assessment

The authors use point light displays to measure biological motion (BM) perception in children (mean = 9 years) with and without ADHD, and relate it to IQ, social responsiveness scale (SRS) scores and age. They report that children with ADHD were worse at all three BM tasks, but that those tasks loading more heavily on local processing relate to social interaction skills and those loading on global processing relate to age. There are still some elements of the results that are unclear, but nevertheless, the important and solid findings extend our limited knowledge of BM perception in ADHD, as well as biological motion processing mechanisms in general.

Reviewer #2 (Public Review):

Summary:

Tian et al. aimed to assess differences in biological motion (BM) perception between children with and without ADHD, as well as relationships to indices of social functioning and possible predictors of BM perception (including demographics, reasoning ability and inattention). In their study, children with ADHD showed poorer performance relative to typically developing children in three tasks measuring local, global, and general BM perception. The authors further observed that across the whole sample, performance in all three BM tasks was negatively correlated with scores on the social responsiveness scale (SRS), whereas within groups a significant relationship to SRS scores was only observed in the ADHD group and for the local BM task. Local and global BM perception showed a dissociation in that global BM processing was predicted by age, while local BM perception was not. Finally, general (local & global combined) BM processing was predicted by age and global BM processing, while reasoning ability mediated the effect of inattention on BM processing.

Strengths:

Overall, the manuscript is presented in a clear fashion and methods and materials are presented with sufficient detail so the study could be reproduced by independent researchers. The study uses an innovative, albeit not novel, paradigm to investigate two independent processes underlying BM perception. The results are novel and have the potential to have wide-reaching impact on multiple fields.

Weaknesses:

The manuscript has improved in clarity and conceptual and methodological considerations in response to the last review. However, the reported results still provide incomplete support for the claims the authors make in the paper.

In relation to other reviewers' earlier comments, the model notation used is still not consistent and model results are reported incompletely, which make it difficult to gain a full picture of the data and how they support the authors' secondary claims. For instance, across the models in the supplementary materials, ß coefficients are only reported selectively which makes it difficult to assess the model as a whole. Furthermore, different terms (task 1, task 2 vs. BM-Local, BM-global) are used to refer to the same levels of a variable, and it is unclear which levels of a dummy variable correspond to which task, making it overall very difficult to comprehend the modelling procedure.

Reviewer #3 (Public Review):

The authors presented point light displays of human walkers to children (mean = 9 years) with and without ADHD to compare their biological motion perception abilities, and relate them to IQ, social responsiveness scale (SRS) scores and age. They report that children with ADHD were worse at all three biological motion tasks, but that those loading more heavily on local processing related to social interaction skills and global processing to age. The valuable and solid findings are informative for understanding this complex condition, as well as biological motion processing mechanisms in general. However, the correlations present a pattern that needs further examination in future studies because many of the differences between correlations are not significant.

Strengths:

The authors present differences between ADHD and TD children in biological motion processing, and this question has not received as much attention as equivalent processing capabilities in autism. They use a task that appears well controlled. They raise some interesting mechanistic possibilities for differences in local and global motion processing, which are distinctions worth exploring. The group differences will therefore be of interest to those studying ADHD, as well as other developmental conditions, and those examining biological motion processing mechanisms in general.

Weaknesses:

The data are not strong enough to support claims about differences between global and lobal processing wrt social communication skills and age. The mechanistic possibilities for why these abilities may dissociate in such a way are interesting, but the crucial tests of differences between correlations do not present a clear picture. Further empirical work would be needed to test this further. Specifics:

The authors state frequently that it was the local BM task that related to social communication skills (SRS) and not the global tasks. However, the results section shows a correlation between SRS and all three tasks. The only difference is that when looking specifically within the ADHD group, the correlation is only significant for the local task. The supplementary materials demonstrate that tests of differences between correlations present an incomplete picture. Currently they have small samples for correlations, so this is unsurprising.

Theoretical assumptions. The authors make some statements about local vs global biological motion processing that may have been made in previous studies, but would appear controversial and not definitive. E.g., that local BM processing does not improve with age and is uninfluenced by attention.