Reviewer #1 (Public Review):

The overall tone of the rebuttal and lack of responses on several questions was surprising. Clearly, the authors did not appreciate the phrase 'no smoking gun' and provided a lengthy repetition of the fair argument about 'ticking boxes' on the classic list of criteria. They also make repeated historical references that descriptions of neurotransmitters include many papers, typically over decades, e.g. in the case of ACh and its discovery by Sir Henry Dale. While I empathize with the authors' apparent frustration (I quote: '...accept the reality that Rome was not built in a single day and that no transmitter was proven by a one single paper') I am a bit surprised at the complete brushing away of the argument, and in fact the discussion. In the original paper, the notion of a receptor was mentioned only in a single sentence and all three reviewers brought up this rather obvious question. The historical comparisons are difficult: Of course many papers contribute to the identification of a neurotransmitter, but there is a much higher burden of proof in 2023 compared to the work by Otto Loewi and Sir Henry Dale: most, if not all, currently accepted neurotransmitter have a clear biological function at the level of the brain and animal behavior or function - and were in fact first proposed to exist based on a functional biological experiment (e.g. Loewi's heart rate change). This, and the isolation of the chemical that does the job, were clear, unquestionable 'smoking guns' a hundred years ago. Fast forward 2023: Creatine has been carefully studied by the authors to tick many of the boxes for neurotransmitters, but there is no clear role for its function in an animal. The authors show convincing effects upon K+ stimulation and electrophysiological recordings that show altered neuronal activity using the slc6a8 and agat mutants as well as Cr application - but, as has been pointed out by other reviewers, these effects are not a clear-cut demonstration of a chemical transmitter function, however many boxes are ticked. The identification of a role of a neurotransmitter for brain function and animal behavior has reasonably more advanced possibilities in 2023 than a hundred years ago - and e.g. a discussion of approaches for possible receptor candidates should be possible.

Again, I reviewed this positively and agree that a lot of cumulative data are great to be put out there and allow the discovery to be more broadly discussed and tested. But I have to note, that the authors simply respond with the 'Rome was not built in a single day' statement to my suggestions on at least 'have some lead' how to approach the question of a receptor e.g. through agonists or antagonists (while clearly stating 'I do not think the publication of this manuscript should not be made dependent' on this). Similarly, in response to reviewer 2's concerns about a missing receptor, the authors' only (may I say snarky) response is ' We have deleted this sentence, though what could mediate postsynaptic responses other than receptors?' The bullet point by reviewer 3 ' • No candidate receptor for creatine has been identified postsynaptically.' is the one point by that reviewer that is simply ignored by the authors completely. Finally, I note that my reivew question on the K stimulation issues (e.g. 35 neurons that simply did not respond at all) was: ' Response: To avoid the disadvantage of K stimulation, we also performed optogenetic experiments recently and obtained encouraging preliminary results.' No details, not data - no response really.

In sum, I find this all a bit strange and the rebuttal surprising - all three reviewers were supportive and have carefully listed points of discussion that I found all valid and thoughtful. In response, the authors selectively responded scientifically to some experimental questions, but otherwise simply rather non-scientifically dismissed questions with 'Rome was not built in a day'-type answers, or less. I my view, the authors have disregarded the review process and the effort of three supportive reviewers, which should be part of the permanent record of this paper.

Reviewer #2 (Public Review):

Bian et al studied creatine (Cr) in the context of central nervous system (CNS) function. They detected Cr in synaptic vesicles purified from mouse brains with anti-Synaptophysin using capillary electrophoresis-mass spectrometry. Cr levels in the synaptic vesicle fraction was reduced in mice lacking the Cr synthetase AGAT, or the Cr transporter SLC6A8. They provide evidence for Cr release within several minutes after treating brain slices with KCl. This KCl-induced Cr release was partially calcium dependent and was attenuated in slices obtained from AGAT and SLC6A8 mutant mice. Cr application also decreased the excitability of cortical pyramidal cells in one third of the cells tested. Finally, they provide evidence for SLC6A8-dependent Cr uptake into synaptosomes, and ATP-dependent Cr loading into synaptic vesicles. Based on these data, the authors propose that Cr may act as neurotransmitter in the CNS.

Strengths:

1. A major strength of the paper is the broad spectrum of tools used to investigate Cr.<br /> 2. The study provides evidence that Cr is present in/loaded into synaptic vesicles.

Weaknesses (resubmission):

1. There is no significant decrease in Cr content pulled down by anti-Syp in AGAT-/- mice when normalized to IgG controls. Hence, blocking AGAT activity/Cr synthesis does not affect Cr levels in the synaptic vesicle fraction, arguing against a Cr enrichment.<br /> 2. There is no difference in KCl-induced Cr release between SLC6A8-/Y and SLC6A8+/Y when normalizing the data to the respective controls. Thus, the data are not consistent with the idea that depolarization-induced Cr release requires SLC6A8.<br /> 3. The rationale of grouping the excitability data into responders and non-responders is not convincing because the threshold of 10% decrease in AP rate is arbitrary. The data do therefore not support the conclusion that Cr reduces neuronal excitability.

Reviewer #3 (Public Review):

SUMMARY:

The manuscript by Bian et al. promotes the idea that creatine is a new neurotransmitter. The authors conduct an impressive combination of mass spectrometry (Fig. 1), genetics (Figs. 2, 3, 6), biochemistry (Figs. 2, 3, 8), immunostaining (Fig. 4), electrophysiology (Figs. 5, 6, 7), and EM (Fig. 8) in order to offer support for the hypothesis that creatine is a CNS neurotransmitter.

STRENGTHS:

There are many strengths to this study.

• The combinatorial approach is a strength. There is no shortage of data in this study.<br /> • The careful consideration of specific criteria that creatine would need to meet in order to be considered a neurotransmitter is a strength.<br /> • The comparison studies that the authors have done in parallel with classical neurotransmitters is helpful.<br /> • Demonstration that creatine has inhibitory effects is another strength.<br /> • The new genetic mutations for Slc6a8 and AGAT are strengths and potentially incredibly helpful for downstream work.

WEAKNESSES:

• Some data are indirect. Even though Slc6a8 and AGAT are helpful sentinels for the presence of creatine, they are not creatine themselves. Of note, these molecules themselves are not essential for making the case that creatine is a neurotransmitter.<br /> • Regarding Slc6a8, it seems to work only as a reuptake transporter - not as a transporter into SVs. Therefore, we do not know what the transporter into the TVs is.<br /> • Puzzlingly, Slc6a8 and AGAT are in different cells, setting up the complicated model that creatine is created in one cell type and then processed as a neurotransmitter in another. This matter will likely need to be resolved in future studies.<br /> • No candidate receptor for creatine has been identified postsynaptically. This will likely need to be resolved in future studies.<br /> • Because no candidate receptor has been identified, it is important to fully consider other possibilities for roles of creatine that would explain these observations other than it being a neurotransmitter? There is some attention to this in the Discussion.

There are several criteria that define a neurotransmitter. The authors nicely delineated many criteria in their discussion, but it is worth it for readers to do the same with their own understanding of the data.

By this reviewer's understanding (and combining some textbook definitions together) a neurotransmitter: 1) must be present within the presynaptic neuron and stored in vesicles; 2) must be released by depolarization of the presynaptic terminal; 3) must require Ca2+ influx upon depolarization prior to release; 4) must bind specific receptors present on the postsynaptic cell; 5) exogenous transmitter can mimic presynaptic release; 6) there exists a mechanism of removal of the neurotransmitter from the synaptic cleft.

For a paper to claim that the published work has identified a new neurotransmitter, several of these criteria would be met - and the paper would acknowledge in the discussion which ones have not been met. For this particular paper, this reviewer finds that condition 1 is clearly met.

Conditions 2 and 3 seem to be met by electrophysiology, but there are caveats here. High KCl stimulation is a blunt instrument that will depolarize absolutely everything in the prep all at once and could result in any number of non-specific biological reactions as a result of K+ rushing into all neurons in the prep. Moreover, the results in 0 Ca2+ are puzzling. For creatine (and for the other neurotransmitters), why is there such a massive uptick in release, even when the extracellular saline is devoid of calcium?

Condition 4 is not discussed in detail at all. In the discussion, the authors elide the criterion of receptors specified by Purves by inferring that the existence of postsynaptic responses implies the existence of receptors. True, but does it specifically imply the existence of creatinergic receptors? This reviewer does not think that is necessarily the case. The authors should be appropriately circumspect and consider other modes of inhibition that are induced by activation or potentiation of other receptors (e.g., GABAergic or glycinergic).

Condition 5 may be met, because authors applied exogenous creatine and observed inhibition. However, this is tough to know without understanding the effects of endogenous release of creatine. if they were to test if the absence of creatine caused excess excitation (at putative creatinergic synapses), then that would be supportive of the same. Nicely, Ghirardini et al., 2023 study cited by the reviewers does provide support for this exact notion in pyramidal neurons.

For condition 6, the authors made a great effort with Slc6a8. This is a very tough criterion to understand or prove for many synapses and neurotransmitters.

In terms of fundamental neuroscience, the story should be impactful. There are certainly more neurotransmitters out there than currently identified and by textbook criteria, creatine seems to be one of them taking all of the data in this study and others into account.

eLife assessment

This study presents valuable observations on the role of creatine (Cr) in the context of synaptic transmission. Overall, the data are solid in support of the conclusion that Cr is present in synaptic vesicles, although the evidence for Cr release and Cr-dependent modulation of neuronal function was considered incomplete. The work will be of general interest to the field of neuroscience.

Reviewer #1 (Public Review):

Bolumar et al. isolated and characterized EV subpopulations, apoptotic bodies (AB), Microvesicles (MV), and Exosomes (EXO), from endometrial fluid through the female menstrual cycle. By performing DNA sequencing, they found the MVs contain more specific DNA sequences than other EVs, and specifically, more mtDNA were encapsulated in MVs. They also found a reduction of mtDNA content in the human endometrium at the receptive and post-receptive period that is associated with an increase in mitophagy activity in the cells, and a higher mtDNA content in the secreted MVs was found at the same time. Last, they demonstrated that the endometrial Ishikawa cell-derived EVs could be taken by the mouse embryos and resulted in altered embryo metabolism.

This is a very interesting study and is the first one demonstrating the direct transmission of maternal mtDNA to embryos through EVs.

eLife assessment

This manuscript reports important results on the potential influence of maternally derived extracellular vesicles on embryo metabolism. The study combines convincing techniques for isolating different subtypes of EV, DNA sequencing, embryo culture, and respiration assays performed on human endometrial samples and mouse embryos. These findings set the stage for in-depth studies to elucidate the role of EV contents in embryo energetics and further enhance our understanding on maternal-fetal communication during peri-implantation development.

Reviewer #2 (Public Review):

In Bolumar, Moncayo-Arlandi et al. the authors explore whether endometrium-derived extracellular vesicles contribute DNA to embryos and therefore influence embryo metabolism and respiration. The manuscript combines techniques for isolating different populations of extracellular vesicles, DNA sequencing, embryo culture, and respiration assays performed on human endometrial samples and mouse embryos.

Vesicle isolation is technically difficult and therefore collection from human samples is commendable. Also, the influence of maternally derived DNA on the bioenergetics of embryos is unknown and therefore novel.

Reviewer #2 (Public Review):

This study investigates how genes in the Gr28 family of gustatory receptors function in the taste system of Drosophila larvae. Gr28 genes are intriguing because they have been implicated in taste as well as other functions, such as sensing temperature and ultraviolet light. This study makes several new findings. First, the authors show that four Gr28 genes are expressed in putative taste neurons, and these neurons can be largely divided into subsets that express Gr28a versus Gr28bc. The authors then demonstrate that these two neuronal subsets drive opposing behaviors (attraction versus avoidance) when activated. The avoidance-promoting neurons respond to bitter compounds and are required for bitter avoidance, and Gr28bc and Gr28ba were specifically implicated in bitter detection in these cells. Together, these findings provide insight into the complexity of taste receptor expression and function in Drosophila, even within a single receptor subfamily.

The conclusions are well-supported by the experimental data. Strengths of the paper include the use of precise genetic tools, thorough analyses of expression patterns, carefully validated behavioral assays, and well-controlled functional imaging experiments. The role of Gr28bc neurons is more thoroughly explored than that of Gr28a neurons. However, a previous study from the same lab (Mishra et al., 2018) showed that Gr28a neurons detect RNA and ribose, which are attractive to larvae. Presumably this is the attractive response that is being recapitulated upon artificial activation of Gr28a neurons.

Reviewer #1 (Public Review):

Ahn and Amrein characterize the expression of members of the Gr28 family of gustatory receptors in taste neurons in the Drosophila melanogaster larva, define the behaviorally-relevant ligands for these receptors, and use chemogenetic experiments to show, strikingly, that different neurons have opposite behavioral responses to the chemogenetic ligand. They go on to show what neurons need to be silenced to lose responses to bitters, and very nicely show what subunits of the Gr28 bitter receptors are necessary and sufficient for responses to bitters. This is a nice piece of work, rigorously carried out, that tackles the neurons and receptors that drive innate responses to tastants in Drosophila larvae.

The authors have revised the paper to address all of my recommendations. The new cartoons are extremely clear and I appreciate the more measured language when discussing the hypothetical structure and stoichiometry of the functional GR complex.

Reviewer #1 (Public Review):

Summary:<br /> The study examines the role of release site clearance in synaptic transmission during repetitive activity under physiological conditions in two types of central synapses, calyx of Held and hippocampal CA1 synapses. After the acute block of endocytosis by pharmacology, deeper synaptic depression or less facilitation was observed in two types of synapses. Acute block of CDC42 and actin polymerization, which possibly inhibits the activity of Intersectin, affected synaptic depression at the calyx synapse, but not at CA1 synapses. The data suggest an unexpected, fast role of the site clearance in counteracting synaptic depression.

Strengths:<br /> The study uses an acute block of the molecular targets with pharmacology together with precise electrophysiology. The experimental results are clear-cut and convincing. The study also examines the physiological roles of the site clearance using action potential-evoked transmission at physiological Ca and physiological temperature at mature animals. This condition has not been examined.

Weaknesses:<br /> Pharmacology may have some off-target effects, though acute manipulation should be appreciated. Although this is a hard question and difficult to address experimentally, reagents may affect synaptic vesicle mobilization to the release sites directly in addition to blocking endocytosis.

Reviewer #3 (Public Review):

General comments:

(1) While Dynasore and Pitstop-2 may impede release site clearance due to an arrest of membrane retrieval, neither Latrunculin-B nor ML-141 specifically acts on AZ scaffold proteins. Interference with actin polymerization may have a number of consequences many of which may be unrelated to release site clearance. Therefore, neither Latrunculin-B nor ML-141 can be considered suitable tools for specifically identifying the role of AZ scaffold proteins (i.e. ELKS family proteins, Piccolo, Bassoon, α-liprin, Unc13, RIM, RBP, etc) in release site clearance which was defined as one of the principal aims of this study.

(2) Initial EPSC amplitudes more than doubled in the presence of Dynasor at hippocampal SC->CA1 synapses (Figure S2). This unexpected result raises doubts about the specificity of Dynasor as a tool to selectively block SV endocytosis.

(3) In this study, the application of Dynasore and Pitstop-2 strongly decreases 100 Hz steady-state release at calyx synapses while - quite unexpectedly - strongly accelerates recovery from depression. A previous study found that genetic ablation of dynamin-1 actually enhanced 300 Hz steady-state release while only little affecting recovery from depression (Mahapatra et al., 2016). A similar scenario holds for the Latrunculin-B effects: In this study, Latrunculin-B strongly increased steady-state depression while in Babu et al. (2020), Latrunculin-B did not affect steady-state depression. In Mahapatra et al. (2016), Latrunculin-B marginally enhanced steady-state depression. The authors need to make a serious attempt to explain all these seemingly contradicting results.

(4) The experimental conditions need to be better specified. It is not clear which recordings were obtained in 1.3 mM and which (if any?) in 2 mM external Ca. It is also unclear whether 'pooled data' are presented (obtained from control recordings and from separate recordings after pre-incubation with the respective drugs), or whether the data actually represent 'before'/'after' comparisons obtained from the same synapses after washing in the respective drugs. The exact protocol of drug application (duration of application/pre-incubation?, measurements after wash-out or in the continuous presence of the drugs?) needs to be clearly described in the methods and needs to be briefly mentioned in Results and/or Figure legends.

(5) The authors compare results obtained in calyx with those obtained in SC->CA1 synapses which they considered examples for 'fast' and 'slow' synapses, respectively. There is little information given to help readers understand why these two synapse types were chosen, what the attributes 'fast' and 'slow' refer to, and how that may matter for the questions studied here. I assume the authors refer to the maximum frequency these two synapse types are able to transmit rather than to EPSC kinetics?

(6) Strong presynaptic stimuli such as those illustrated in Figures 1B and C induce massive exocytosis. The illustrated Cm increase of 2 to 2.5 pF represents a fusion of 25,000 to 30,000 SVs (assuming a single SV capacitance of 80 aF) corresponding to a 12 to 15% increase in whole terminal membrane surface (assuming a mean terminal capacitance of ~16 pF). Capacitance measurements can only be considered reliable in the absence of marked changes in series and membrane conductance. Since the data shown in Figs. 1 and 3 are central to the argumentation, illustration of the corresponding conductance traces is mandatory. Merely mentioning that the first 450 ms after stimulation were skipped during analysis is insufficient.

(7) It is essential for this study to preclude a contamination of the results with postsynaptic effects (AMPAR saturation and desensitization). AMPAR saturation limits the amplitudes of initial responses in EPSC trains and hastens the recovery from depression due to a 'ceiling effect'. AMPAR desensitization occludes paired-pulse facilitation and reduces steady-state responses during EPSC trains while accelerating the initial recovery from depression. The use of, for example, 1 mM kynurenic acid in the bath is a well-established strategy to attenuate postsynaptic effects at calyx synapses. All calyx EPSC recordings should have been performed under such conditions. Otherwise, recovery time courses and STP parameters are likely contaminated by postsynaptic effects. Since the effects of AMPAR saturation on EPSC_1 and desensitization on EPSC_ss may partially cancel each other, an unchanged relative STD in the presence of kynurenic acid is not necessarily a reliable indicator for the absence of postsynaptic effects. The use of kynurenic acid in the bath would have had the beneficial side effect of massively improving voltage-clamp conditions. For the typical values given in this MS (10 nA EPSC, 3 MOhm Rs) the expected voltage escape is ~30 mV corresponding to a change in driving force of 30 mV/80 mV=38%, i.e. initial EPSCs in trains are likely underestimated by 38%. Such large voltage escape usually results in unclamped INa(V) which was suppressed in this study by routinely including 2 mM QX-314 in the pipette solution. That approach does, however, not reduce the voltage escape.

(8) In the Results section (pages 7 and 8), the authors analyze the time course into STD during 100 Hz trains in the absence and presence of drugs. In the presence of drugs, an additional fast component is observed which is absent from control recordings. Based on this observation, the authors conclude that '... the mechanisms operate predominantly at the beginning of synaptic depression'. However, the consequences of blocking or slowing site clearing are expected to be strongly release-dependent. Assuming a probability of <20% that a fusion event occurs at a given release site, >80% of the sites cannot be affected at the arrival of the second AP even by a total arrest of site clearance simply because no fusion has yet occurred. That number decreases during a train according to (1-0.2)^n, where n is the number of the AP, such that after 10 APs, ~90% of the sites have been used and may potentially be unavailable for new rounds of release after slowing site clearance. Perhaps, the faster time course into STD in the presence of the drugs isn't related to site clearance?

(9) In the Discussion (page 10), the authors present a calculation that is supposed to explain the reduced size of the second calyx EPSC in a 100 Hz train in the presence of Dynasore or Pitstop-2. Does this calculation assume that all endocytosed SVs are immediately available for release within 10 ms? Please elaborate.

(10) It is not clear, why the bafilomycin/folimycin data is presented in Fig. S5. The data is also not mentioned in the Discussion. Either explain the purpose of these experiments or remove the data.

(11) The scheme in Figure 7 is not very helpful.

eLife assessment:

This important study combines a comparative approach in different synapses with experiments that show how synaptic vesicle endocytosis in nerve terminals regulates short-term plasticity. The data presented support the conclusions and make a convincing case for fast endocytosis as necessary for rapid vesicle recruitment to active zones. Some aspects of the description of the data and analysis are however incomplete and would benefit from a more rigorous approach. With more discussion of methods and analysis, this paper would be of great interest to neurobiologists and biophysicists working on synaptic vesicle recycling and short-term plasticity mechanisms.

Reviewer #2 (Public Review):

Summary:<br /> In this manuscript, Mahapatra and Takahashi report on the physiological consequences of pharmacologically blocking either clathrin and dynamin function during compensatory endocytosis or of the cortical actin scaffold both in the calyx of Held synapse and hippocampal boutons in acute slice preparations

Strengths:<br /> Although many aspects of these pharmacological interventions have been studied in detail during the past decades, this is a nice comprehensive and comparative study, which reveals some interesting differences between a fast synapse (Calyx of Held) tuned to reliably transmit at several 100 Hz and a more slow hippocampal CA1 synapse. In particular, the authors find that acute disturbance of the synaptic actin network leads to a marked frequency-dependent enhancement of synaptic depression in the Calyx, but not in the hippocampal synapse. This striking difference between both preparations is the most interesting and novel finding.

Weaknesses:<br /> Unfortunately, however, these findings concerning the different consequences of actin depolymerization are not sufficiently discussed in comparison to the literature. My only criticism concerns the interpretation of the ML 141 and Lat B data. With respect to the Calyx data, I am missing a detailed discussion of the effects observed here in light of the different RRP subpools SRP and FRP. This is very important since Lee et al. (2012, PNAS 109 (13) E765-E774) showed earlier that disruption of actin inhibits the rapid transition of SRP SVs to the FRP at the AZ. The whole literature on this important concept is missing. Likewise, the role of actin for the replacement pool at a cerebellar synapse (Miki et al., 2016) is only mentioned in half a sentence. There is quite some evidence that actin is important both at the AZ (SRP to FRP transition, activation of replacement pool) and at the peri-active zone for compensatory endocytosis and release site clearance. Both possible underlying mechanisms (SRP to FRP transition or release site clearance) should be better dissected.

eLife assessment

This study provides a comprehensive whole genome transcriptomic analysis of three small mammals, including Peromyscus leucopus, after exposure to endotoxin lipopolysaccharide. The authors find that the inflammatory response of the three species is complex and that P. leucopus responds differently compared to mice and rats. The data are convincing and constitute an important advance in our understanding of inflammatory responses in animals that serve as reservoirs for relevant pathogens.

Reviewer #1 (Public Review):

Summary:

o A well-executed series of experiments that will likely be of immense interest to (a) vector-borne disease researchers and (b) gram-negative sepsis/bacteremia researchers. The study uses comparative transcriptomics to begin probing what makes Peromyscus leucopus a unique host for numerous pathogens across the tree of life. Authors responded well to concerns raised in peer review and have produced an excellent second version of the manuscript.

Strengths:

o Use of outbred M. musculus is a commendable choice for the studies here.<br /> o Use of both LPS and B. hermsii allows analysis of multiple different signaling pathways that may differ between the species.<br /> o Upload of analyzed data onto Dryad is appreciated.

Weaknesses:

o None noted beyond the authors own limitation discussion section

Reviewer #2 (Public Review):

Milovic, Duong, and Barbour investigate the inflammatory response of three species of small mammals (P. leucopus, M. musculus, and R. norvegicus) to endotoxin lipopolysaccharide (LPS) injection via genome-wide transcriptomics from blood samples. Understanding the inflammation response of P. leucopus is of importance as they are a reservoir for several pathogens. The study is a thorough, controlled, well researched analysis that will be valuable for designing and interpreting future studies. The authors discuss the limitations of the data and the potential directions. Clearly P. leucopus respond differently to the LPS exposure which is very interesting and opens the door for numerous other comparative studies.

The conclusions of the manuscript are thoughtful and supported by the data. The authors addressed my questions about mouse numbers, sex differences, and the presentation of Nos2 and Arg1 data.

eLife assessment

This study presents important findings regarding the local dynamics at the anion binding site in the SLC26 transporter prestin that is responsible for electromotility in outer hair cells. The authors reveal critical differences to homologous proteins and thereby provide insight into prestin's unique function. The evidence is generally convincing, although the interpretations concerning the mechanistic basis for voltage sensitivity would benefit from orthogonal evidence.

Reviewer #1 (Public Review):

The manuscript by Lin, Sosnick et al investigates the functional conformational dynamics of two members of the SLC26 family of anion transporters (Prestin and SLC26A9). A key aspect of the work is that the authors use HDX-MS to convincingly identify that the folding of the unstable anion binding site is related to the fast electromechanical changes that are important for the function of Prestin. In good apparent agreement, such folding-related changes upon anion binding are absent in the related non-piezoelectric SLC26A9 that does not exhibit similar electro-motile transport. Overall, I find the work very interesting and generally well carried out - and it should be of considerable interest to researchers studying transmembrane transporters or just membrane proteins in general.

Reviewer #2 (Public Review):

In this manuscript, Xiaoxuan Lin and colleagues provide new insights into the dynamics of prestin using H/D exchange coupled with mass spectrometry. The authors aim to reveal how local changes in folding upon anion binding sustain the unique electro-transduction capabilities of prestin.

Prestin is an unusual member of the SLC26 family, that changes its cross-sectional area in the membrane upon binding of a chloride ion. In contrast to SLC26 homologs, prestin is not an anion transporter per se but requires an anion to sense voltage. Binding of Cl- at a conserved binding site located between the end of TM3 and TM10 drives the displacement of a conserved arginine (R399), that causes major conformational changes, transmitting the voltage sensing into a mechanical force exerted on the membrane.

Cryo-EM structures are available for the protein bound to various anions, including Cl-, but these structures do not explain how a conserved couple of positive (R399) and negative (the Cl- anion) charge pair transforms voltage sensitivity into mechanical changes in the membrane. To address this challenge, the authors explore local dynamics of the anion binding site and compare it with that of a "real" anion transporter SLC26A9. The authors make a convincing case that the differences in local dynamics they measure are the molecular basis for voltage sensing and its translation into electromotility.

Practically the authors make a thorough HDX-MS investigation of prestin in the presence of different anions Cl-, SO4-, salicylate as well as in the apo form, and provide insight mostly on local dynamics of the anion binding site. The experiments are well-designed and conducted and their quality and reproducibility allows for quantitative interpretation by deriving ΔΔG values of changes in dynamics at specific sites. Furthermore, the authors show by comparing the apo condition with Cl- bound condition that the absence of Cl- causes fraying of the TM3 and TM10 helices. They deduce that Cl- binding allows for directional helix structuration, leading to local structural changes that cause a rearrangement of the charge configuration at the anion binding site that lays the molecular basis for voltage sensitivity. They demonstrate based on a detailed analysis of their HDX data that such helix fraying is a specific feature of the binding site and differs from the cooperative unfolding happening elsewhere on the prestin.

However, the main question that the authors are addressing is how voltage sensitivity translates at the molecular level in the requirement for a negative-positive charge pair. The interpretation that the binding site instability observed only for prestin is a feature required for this voltage dependent is a bit speculative. Could other lines of evidence support the claim that the charge ion gap is reduced upon Cl- binding and that this leads to cross-section area expansion? An obvious option that comes to mind is MD simulations There are differences in time-scale between HDX and simulations, but the propensity for H-bond destabilization can be quantified even at short timescales. It might be that such data is already available out there but it should be explicit in the discussion. The discussion section itself is a bit narrow in scope at the moment. Discussing the data in the context of the available structures would help the non-specialist reader.

Reviewer #3 (Public Review):

Synopsis:<br /> The lack of visualizing the dynamic nature of biomolecules is a major weakness of crystallography or electron microscopy to study structure-function relationship of proteins. Such a challenge can be exemplified by the case of prestin, which shares high structural similarity to SLC26A9 anion transporter but is not an ion transporter. In this study, Lin et al aimed to use hydrogen-deuterium exchange and mass spectrometry (HDX-MS) to investigate the mobility of prestin and its response to anions. The authors exploited the nature of anion-dependent folding of this type of transporter to systematically analyze the mobility of transmembrane helices of both transporters by HDX. The authors found that the anion-binding helices engage in the stabilization of the anion-binding site. When stripped from Cl-, the site exposes to the transporter's extracellular side. More importantly, the authors narrowed down TM3 and TM10 with experimental data supporting the notion of R399's unique role in prestin's function. The results thus provide a working model of how the charged residue works in conjunction with the cooperativity of helix unfolding at the anion-binding site to drive the electromotive force of prestin.

Strengths:<br /> The use of HDX-MS to probe the dynamic nature of prestin is a major strength of this study, which provides experimental evidence revealing the global and local differences in the folding events between prestin and SLC26A9. The mass experimental data led to the identification of TM3 and TM10 as the primary contributors to the folding changes, as well as a calculation of ΔΔG of ~2.4 kcal/mol, within the thermodynamic range of the dipole between the two helices. The latter also suggests the role of R399 as previously speculated in cryo-EM structures.

This study went further to dissect the cooperativity during the folding and unfolding events on TM3, in which the authors observed a helix fraying at the anion-binding site and cooperative unfolding at the distal lipid-facing helices. This provides strong evidence of why prestin can undergo fast electromechanical rearrangement.

Weakness:<br /> The authors tried to investigate the allostery by probing the intermediate folding/unfolding states by using sulfate or salicylate in the absence of chloride. Sulfate-bound proteins appear in an apo state earlier than normal chloride binding, and salicylate treatment led to a stable TMD state with slower HDX. It is unclear from the data (Fig 4) how the allostery works without titrating chloride ions into the reaction. The sulfate or salicylate experiments seem to show two extreme folding events outside the normal chloride conditions.

TM3 and TM10 contribute to the anion-binding site together, and the authors beautifully showed the cooperativity of TM3. Does TM10 show the same cooperativity in prestin and SLC26A9? In addition, it is unclear whether the folding model at the anion-binding helices (Fig. 5B) remains the same when expressing prestin on live cells, such as thermodynamic data derived from electrophysiology studies.

The authors observed increased stability upon chloride binding at the subunit interface in the cytosol for both prestin and SLC26A9 (Fig 1). How does this similarity in the cytosolic region contribute to the differential mechanisms as seen in the TMD in both transporters? It is unclear in this version of the manuscript.

eLife assessment

This important study provides data that challenges the standard model that binding of Type 2 Nuclear Receptors to chromatin is limited by the available pool of their common heterodimerization partner Retinoid X Receptor. The evidence supporting the conclusions is compelling, utilizing state-of-the-art single-molecule microscopy. This work will be of broad interest to cell biologists who wish to determine limiting factors in gene regulatory networks.

Reviewer #1 (Public Review):

This study provides compelling evidence that RAR, rather than its obligate dimerization partner RXR, is functionally limiting for chromatin binding. This manuscript provides a paradigm for how to dissect the complicated regulatory networks formed by dimerizing transcription factor families.

Dahal and colleagues use advanced SMT techniques to revisit the role of RXR in DNA-binding of the type-2 nuclear receptor (T2NR) RAR. The dominant consensus model for regulated DNA binding of T2NRs posits that they compete for a limited pool of RXR to form an obligate T2NR-RXR dimer. Using advanced SMT and proximity-assisted photoactivation technologies, Dahal et al. now test the effect of manipulating the endogenous pool size of RAR and RXR on heterodimerization and DNA-binding in live U2OS cells. Surprisingly, it turns out that RAR, rather than RXR, is functionally limiting for heterodimerization and chromatin binding. By inference, the relative pool size of various T2NRs expressed in a given cell, rather than RXR, is likely to determine chromatin binding and transcriptional output.

The conclusions of this study are well supported by the experimental results and provide unexpected novel insights into the functioning of the clinically important class of T2NR TFs. Moreover, the presented results show how the use of novel technologies can put long-standing theories on how transcription factors work upside down. This manuscript provides a paradigm for how to further dissect the complicated regulatory networks formed by T2NRs or other dimerizing TFs. I found this to be a complete story that does not require additional experimental work. However, I do have some suggestions for the authors to consider.

Reviewer #2 (Public Review):

Summary:<br /> In the manuscript "Surprising Features of Nuclear Receptor Interaction Networks Revealed by Live Cell Single Molecule Imaging", Dahal et al combine fast single molecule tracking (SMT) with proximity-assisted photoactivation (PAPA) to study the interaction between RARa and RXRa. The prevalent model in the nuclear receptor field suggests that type II nuclear receptors compete for a limiting pool of their partner RXRa. Contrary to this, the authors find that over-expression of RARa but not RXRa increases the fraction of RXRa molecules bound to chromatin, which leads them to conclude that the limiting factor is the abundance of RARa and not RXRa. The authors also perform experiments with a known RARa agonist, all trans retinoic acid (atRA) which has little effect on the bound fraction. Using PAPA, they show that chromatin binding increases upon dimerization of RARa and RXRa.

Strengths:<br /> In my view, the biggest strength of this study is the use of endogenously tagged RARa and RXRa cell lines. As the authors point out, most previous studies used either in vitro assays or over-expression. I commend the authors on the generation of single-cell clones of knock-in RARa-Halo and Halo-RXRa. The authors then carefully measure the abundance of each protein using FACS, which is very helpful when comparing across conditions. The manuscript is generally well written and figures are easy to follow. The consistent color-scheme used throughout the manuscript is very helpful.

Weaknesses:<br /> 1. Agonist treatment:<br /> The authors test the effect of all trans retinoic acid (atRA) on the bound fraction of RARa and RXRa and find that "These results are consistent with the classic model in which dimerization and chromatin binding of T2NRs are ligand independent." However, all the agonist treatments are done in media containing FBS. FBS is not chemically defined and has been found to have between 10 and 50 nM atRA (see references in PMID 32359651 for example). The addition of 1 nM or 100 nM atRA is unlikely to result in a strong effect since the medium already contains comparable or higher levels of agonist. To test their hypothesis of ligand-independent dimerization, the authors should deplete the media of atRA by growing the cells in a medium containing charcoal-stripped FBS for at least 24 hours before adding agonist.

2. Photobleaching and its effect on bound fraction measurements:<br /> The authors discard the first 500 to 1000 frames due to the high localization density in the initial frames. This will preferentially discard bound molecules that will bleach in the initial frames of the movie and lead to an over-estimation of the unbound fraction.

For experiments with over-expression of RAR-Halo and Halo-RXR, the authors state that the cells were pre-bleached and that these frames were used to calculate the mean intensity of the nuclei. When pre-bleaching, bound molecules will preferentially bleach before the diffusing population. This will again lead to an over-representation of the unbound fraction since this is the population that will remain relatively unaffected by the pre-bleaching. Indeed, the bound fraction for over-expressed RARa and RXRa is significantly lower than that for the corresponding knock in lines. To confirm whether this is a biological result, I suggest that the authors either reduce the amount of dye they use so that this pre-bleaching is not necessary or use the direct reactivation strategy they use for their PAPA experiments to eliminate the pre-bleaching step.

As for the measurement of the nuclear intensity, since the authors have access to multiple HaloTag dyes, they can saturate the HaloTagged proteins with a high concentration of JF646 or JFX650 to measure the mean intensity of the protein while still using the PA-JFX549 for SMT. Together, these will eliminate the need to pre-bleach or discard any frames.

3. Heterogeneous expression of the SNAP fusion proteins:<br /> The cell lines expressing SNAP tagged transgenes shown in Fig S6 have very heterogeneous expression of the SNAP proteins. While the bulk measurements done by Western blotting are useful, while doing single-cell experiments (especially with small numbers - ~20 - of cells), it is important to control for expression levels. Since these transgenic stable lines were not FACS sorted, it would be helpful for the reader to know the spread in the distribution of mean intensities of the SNAP proteins for the cells that the SMT data are presented for. This step is crucial while claiming the absence of an effect upon over-expression and can easily be done with a SNAPTag ligand such as SF650 using the procedure outlined for the over-expressed HaloTag proteins.

4. Definition of bound molecules:<br /> The authors state that molecules with a diffusion coefficient less than 0.15 um2/s are considered bound and those between 1-15 um2/s are considered unbound. Clarification is needed on how this threshold was determined. In previous publications using saSPT, the authors have used a cutoff of 0.1 um2/s (for example, PMID 36066004, 36322456). Do the results rely on a specific cutoff? A diffusion coefficient by itself is only a useful measure of normal diffusion. Bound molecules are unlikely to be undergoing Brownian motion, but the state array method implemented here does not seem to account for non-normal diffusive modes. How valid is this assumption here?

5. Movies:<br /> Since this is an imaging manuscript, I request the authors to provide representative movies for all the presented conditions. This is an essential component for a reader to evaluate the data and for them to benchmark their own images if they are to try to reproduce these findings.

6. Definition of an ROI:<br /> The authors state that "ROI of random size but with maximum possible area was selected to fit into the interior of the nuclei" while imaging. However, the readout speed of the Andor iXon Ultra 897 depends on the size of the defined ROI. If the ROI was variable for every movie, how do the authors ensure the same sampling rate?

Reviewer #3 (Public Review):

Summary:<br /> This study aims to investigate the stoichiometric effect between core factors and partners forming the heterodimeric transcription factor network in living cells at endogenous expression levels. Using state-of-the-art single-molecule analysis techniques, the authors tracked individual RARα and RXRα molecules labeled by HALO-tag knock-in. They discovered an asymmetric response to the overexpression of counter-partners. Specifically, the fact that an increase in RARα did not lead to an increase in RXRα chromatin binding is incompatible with the previous competitive core model. Furthermore, by using a technique that visualizes only molecules proximal to partners, they directly linked transcription factor heterodimerization to chromatin binding.

Strengths:<br /> The carefully designed experiments, from knock-in cell constructions to single-molecule imaging analysis, strengthen the evidence of the stoichiometric perturbation response of endogenous proteins. The novel finding that RXR, previously thought to be a target of competition among partners, is in excess provides new insight into key factors in dimerization network regulation. By combining the cutting-edge single-molecule imaging analysis with the technique for detecting interactions developed by the authors' group, they have directly illustrated the relationship between the physical interactions of dimeric transcription factors and chromatin binding. This has enabled interaction analysis in live cells that was challenging in single-molecule imaging, proving it is a powerful tool for studying endogenous proteins.

Weaknesses:<br /> As the authors have mentioned, they have not investigated the effects of other T2NRs or RXR isoforms. These invisible factors leave room for interpretation regarding the origin of chromatin binding of endogenous proteins (Recommendations 4). In the PAPA experiments, overexpressed factors are visualized, but changes in chromatin binding of endogenous proteins due to interactions with the overexpressed proteins have not been investigated. This might be tested by reversing the fluorescent ligands for the Sender and Receiver. Additionally, the PAPA experiments are likely to be strengthened by control experiments (Recommendations 5).

eLife assessment

This work presents valuable new information on the microtubule-binding mode of the microtubule kinesin-13, MCAK. The authors use quantitative single-molecule studies to propose that MCAK preferentially binds to a GDP-Pi-tubulin portion of the microtubule end. However, the evidence provided to support this claim remains incomplete and would benefit from a more rigorous methodology. Additionally, the physiological relevance of the proposed binding mode remains speculative.

Reviewer #1 (Public Review):

Major concerns:

1. Is the direct binding of MCAK to the microtubule cap important for its in vivo function?

a. The authors claim that their "study provides mechanistic insights into understanding the end-binding mechanism of MCAK". I respectfully disagree. My concern is that the paper offers limited insights into the physiological significance of direct end-binding for MCAK activity, even in vitro. The authors estimate that in the absence of other proteins in vitro, ~95% of MCAK molecules arrive at the tip by direct binding in the presence of ~ physiological ATP concentration (1 mM). In cells, however, the major end-binding pathway may be mediated by EB, with the direct binding pathway contributing little to none. This is a reasonable concern because the apparent dissociation constant measured by the authors shows that MCAK binding to microtubules in the presence of ATP is very weak (69 uM). This concern should be addressed by 1) calculating relative contributions of direct and EB-dependent pathways based on the affinities measured in this and other published papers and estimated intracellular concentrations. Although there are many unknowns about these interactions in cells, a modeling-based analysis may be revealing. 2) the recapitulation of these pathways using purifying proteins in vitro is also feasible. Ideally, some direct evidence should be provided, e.g. based on MCAK function-separating mutants (GDP-Pi tubulin binding vs. catalytic activity at the curled protofilaments) that contribution from the direct binding of MCAK to microtubule cap in EB presence is significant.

b. As mentioned in the Discussion, preferential MCAK binding to tubulins near the MT tip may enhance MCAK targeting of terminal tubulins AFTER the MCAK has been "delivered" to the distal cap via the EB-dependent mechanism. This is a different targeting mechanism than the direct MCAK-binding. However, the measured binding affinity between MCAK and GMPCPP tubulins is so weak (69 uM), that this effect is also unlikely to have any impact because the binding events between MCAK and microtubule should be extremely rare. Without hard evidence, the arguments for this enhancement are very speculative.

2. The authors do not provide sufficient justification and explanation for their investigation of the effects of different nucleotides in MCAK binding affinity. A clear summary of the nucleotide-dependent function of MCAK (introduction with references to prior affinity measurements and corresponding MCAK affinities), the justifications for this investigation, and what has been learned from using different nucleotides (discussion) should be provided. My take on these results is that by far the strongest effect on microtubule wall and tip binding is achieved by adding any adenosine, whereas differences between different nucleotides are relatively minor. Was this expected? What can be learned from the apparent similarity between ATP and AMPPNP effects in some assays (Fig 1E, 4C, etc) but not others (Fig 1D,F, etc)?

3. It is not clear why the authors decided to use these specific mutant MCAK proteins to advance their arguments about the importance of direct tip binding. Both mutants are enzymatically inactive. Both show roughly similar tip interactions, with some (minor) differences. Without a clear understanding of what these mutants represent, the provided interpretations of the corresponding results are not convincing.

4. GMPCPP microtubules are used in the current study to represent normal dynamic microtubule ends, based on some published studies. However, there is no consensus in the field regarding the structure of growing vs. GMPCPP-stabilized microtubule ends, which additionally may be sensitive to specific experimental conditions (buffers, temperature, age of microtubules, etc). To strengthen the authors' argument, Taxol-stabilized microtubules should be used as a control to test if the effects are specific. Additionally, the authors should consider the possibility that stronger MCAK binding to the ends of different types of microtubules may reflect MCAK-dependent depolymerization events on a very small scale (several tubulin rows). These nano-scale changes to tubulins and the microtubule end may lead to the accumulation of small tubulin-MCAK aggregates, as is seen with other MAPs and slowly depolymerizing microtubules. These effects for MCAK may also depend on specific nucleotides, further complicating the interpretation. This possibility should be addressed because it provides a different interpretation than presented in the manuscript.

5. It would be helpful if the authors provided microtubule polymerization rates and catastrophe frequencies for assays with dynamic microtubules and MCAK in the presence of different nucleotides. The video recordings of microtubules under these conditions are already available to the authors, so it should not be difficult to provide these quantifications. They may reveal that microtubule ends are different (or not) under the examined conditions. It would also help to increase the overall credibility of this study by providing data that are easy to compare between different labs.

6. Are there other published studies that report MCAK binding affinity to microtubules? I find it quite surprising that the authors have reported the apparent dissociation constant for MCAK as 1mM. Such a high Kd value suggests no interaction under normal conditions, given that the intracellular concentrations of most proteins are orders of magnitude lower. If this information is inaccurate, it raises questions about the accuracy of other quantifications in the study.

7. Experimental and data analysis techniques are described superficially, and in some cases, only references to the prior work by others are provided. More direct evidence for these techniques and the corresponding controls should be provided.

Reviewer #2 (Public Review):

Summary:<br /> In this manuscript, Chen et al. investigate the localization of microtubule kinesin-13 MCAK to the microtubule ends. MCAK is a prominent microtubule depolymerase whose molecular mechanisms of action have been extensively studied by a number of labs over the last ~twenty years. Here, the authors use single-molecule approaches to investigate the precise localization of MCAK on growing microtubules and conclude that MCAK preferentially binds to a GDP-Pi-tubulin portion of the microtubule end. The conclusions are speculative and not well substantiated by the data, making the impact of the study in its current form rather limited. Specifically, greater effort should be made to define the region of MCAK binding on microtubule ends, as well as its structural characteristics. Given that MCAK has been previously shown to effectively tip-track growing microtubule ends through an established interaction with EB proteins, the physiological relevance of the present study is unclear. Finally, the manuscript does not cite or properly discuss a number of relevant literature references, the results of which should be directly compared and contrasted to those presented here.

Reviewer #3 (Public Review):

The authors revisit an old question of how MCAK goes to microtubule ends, partially answered by many groups over the years. The authors seem to have omitted the literature on MCAK in the past 10-15 years. The novelty is limited due to what has previously been done on the question. Previous work showed MCAK targets to microtubule plus-ends in cells through association with EB proteins and Kif18b (work from Wordeman, Medema, Walczak, Welburn, Akhmanova) but none of their work is cited.

It is not obvious in the paper that these in vitro studies only reveal microtubule end targeting, rather than plus end targeting. MCAK diffuses on the lattice to both ends and its conformation and association with the lattice and ends has also been addressed by other groups-not cited here. I want to particularly highlight the work from Friel's lab where they identified a CDK phosphomimetic mutant close to helix4 which reduces the end preference of MCAK. This residue is very close to the one mutated in this study and is highly relevant because it is a site that is phosphorylated in vivo. This study and the mutant produced here suggest a charge-based recognition of the end of microtubules.

Here the authors analyze this MCAK recognition of the lattice and microtubule ends, with different nucleotide states of MCAK and in the presence of different nucleotide states for the microtubule lattice. The main conclusion is that MCAK affinity for microtubules varies in the presence of different nucleotides (ATP and analogs) which was partially known already. How different nucleotide states of the microtubule lattice influence MCAK binding is novel. This information will be interesting to researchers working on the mechanism of motors and microtubules. However, there are some issues with some experiments. In the paper, the authors say they measure MCAK residency of growing end microtubules, but in the kymographs, the microtubules don't appear dynamic- in addition, in Figure 1A, MCAK is at microtubule ends and does not cause depolymerization. I would have expected to see depolymerization of the microtubule after MCAK targeting. The MCAK mutants are not well characterized. Do they still have ATPase activity? Are they folded? Can the authors also highlight T537 and discuss this?

Finally, a few experiments are done with MCAK and XMAP215, after the authors say they have demonstrated the binding sites overlap. The data supporting this statement were not obvious and the conclusions that the effect of the two molecules are additive would argue against competing binding sites. Overall, while there are some interesting quantitative measurements of MCAK on microtubules - in particular in relation to the nucleotide state of the microtubule lattice - the insights into end-recognition are modest and do not address or discuss how it might happen in cells. Often the number of events is not recorded. Histograms with large SEM bars are presented, so it is hard to get a good idea of data distribution and robustness. Figures lack annotations. This compromises therefore their quantifications and conclusions. The discussion was hard to follow and needs streamlining, as well as putting their work in the context of what is known from other groups who produced work on this in the past few years.

eLife assessment

This study by Sheng and colleagues provides valuable insights into the mechanism of competitive inhibitors of P2X receptors. The structural and functional evidence supporting the subtype specificity of pyridoxal-5'-phosphate derivatives is solid and provides information for designing drugs that selectively target different subtypes of P2X receptor channels. The written presentation could be improved for clarity. The work will be of interest to biochemists, structural biologists, and pharmacologists.

Reviewer #1 (Public Review):

This work provides new mechanistic insights into the competitive inhibition in the mammalian P2X7 receptors using structural and functional approaches. The authors solved the structure of panda (pd) P2X7 in the presence of the classical competitive antagonists PPNDS and PPADS. They find that both drugs bind to the orthosteric site employed by the physiological agonist ATP. However, owing to the presence of a single phosphate group, they prevent movements in the flipper domain required for channel opening. The authors performed structure-based mutational analysis together with electrophysiological characterization to understand the subtype-specific binding of these drugs. It is known from previous studies that P2X1 and P2X3 are more sensitive to these drugs as compared to P2X7, hence, the residues adjacent to the ATP binding site in pdP2X7 were mutated to those present in P2X1. They observed that mutations of Q143, I214, and Q248 into lysine (hP2X1) increased the P2X7 sensitivity to PPNDS, whereas in P2X1, mutations of these lysines to alanine reduced sensitivity to PPNDS, suggesting that these key residues contribute to the subunit-specific sensitivity to these drugs. Similar experiments were done in hP2X3 to demonstrate its higher sensitivity to PPNDS. This preprint provides a useful framework for developing subtype-specific drugs for the family of P2X receptor channels, an area that is currently relatively unexplored.

The conclusions of the paper are mostly well supported, but need some clarification for the following:

1) Why was the crystallization construct of panda P2X7 used for structural studies instead of rat P2X7 with the cytoplasmic ballast which is a more complete receptor that is closely related to the human receptor? Can the authors provide a justification for this choice?

2) Was there a good reason why hP2X1 and hP2X3 currents were recorded in perforated patches, whereas pdP2X7 currents were recorded using the whole-cell configuration? It seems that the extent of rundown is less of a problem with perforated patch recordings. Can the authors comment and perhaps provide a justification? It would also be good to present data for repeated applications of ATP alone using protocols similar to those for testing antagonists so the reader can better appreciate the extent of run down with different recording configurations for the different receptors.

3) The data in Fig. S1, panel A shows multiple examples where the currents activated by ATP after removal of the antagonist are considerably smaller than the initial ATP application. Is this due to rundown or incomplete antagonist unbinding? It is interesting that this wasn't observed with hP2X1 and hP2X3 even though they have a higher affinity for the antagonist. Showing examples of rundown without antagonist application would help to distinguish these distinct phenomena and it would be good for the authors to comment on this in the text. It is also curious why a previous study on pdP2X7 did not seem to have problems with rundown (see Karasawa and Kawate. eLife, 2016).

4) The written presentation could be improved as there are many instances where the writing lacks clarity and the reader has to guess what the authors wish to communicate.

Reviewer #2 (Public Review):

Summary:<br /> P2X receptors play pivotal roles in physiological processes such as neurotransmission and inflammation, making them promising drug targets. This study, through cryo-EM and functional experiments, reveals the structural basis of the competitive inhibition of the PPNDS and PPADS on mammalian P2X7 receptors. Key findings include the identification of the orthosteric site for these antagonists, the revelation of how PPADS/PPNDS binding impedes channel-activating conformational changes, and the pinpointing of specific residues in P2X1 and P2X3 subtypes that determine their heightened sensitivity to these antagonists. These insights present a comprehensive understanding that could guide the development of improved drugs targeting P2X receptors. This work will be a valuable addition to the field.

Strengths and weaknesses:<br /> The combination of structural experiments and mutagenesis analyses offers a deeper understanding of the mechanism. While the inclusion of MD simulation is appreciated, providing more insights from the simulation might further strengthen this already compelling story.

eLife assessment

This important manuscript details the characterization of ClpL from L. monocytogenes as an effective and autonomous AAA+ disaggregase that provides enhanced heat resistance to this food-borne pathogen. The authors convincingly demonstrate that ClpL has DnaK-independent disaggregase activity towards a variety of aggregated model substrates, which is more potent than that observed with the endogenous canonical DnaK/ClpB bi-chaperone system. The work will be of broad interest to microbiologists and biochemists.

Reviewer #1 (Public Review):

Summary:<br /> This work describes the mechanism of protein disaggregation by the ClpL AAA+ protein of Listeria monocytogenes. Using several model subtrate proteins the authors first show that ClpL possesses a robust disaggregase activity that does not further require the endogenous DnaK chaperone in vitro. In addition, they found that ClpL is more thermostable than the endogenous L. monocytogenes DnaK and has the capacity to unfold tightly folded protein domains. The mechanistic basis for the robust disaggregase activity of ClpL was also dissected in vitro and in some cases, supported by in vivo data performed in chaperone-deficient E. coli strains. The data presented show that the two AAA domains, the pore-2 site and the N-terminal domain (NTD) of ClpL are critical for its disaggregase activity. Remarkably, grafting the NTD of ClpL to ClpB converted ClpB into an autonomous disaggregase, highlighting the importance of such a domain in the DnaK-independent disaggregation of proteins. The role of the ClpL NTD domain was further dissected, identifying key residues and positions necessary for aggregate recognition and disaggregation. Finally, using sets of SEC and negative staining EM experiments combined with conditional covalent linkages and disaggregation assays the authors found that ClpL shows significant structural plasticity, forming dynamic hexameric and heptameric active single rings that can further form higher assembly states via their middle domains.

Strengths:<br /> The manuscript is well-written and the experimental work is well executed. It contains a robust and complete set of in vitro data that push further our knowledge of such important disaggregases. It shows the importance of the atypical ClpL N-terminal domain in the disaggregation process as well as the structural malleability of such AAA+ proteins. More generally, this work expands our knowledge of heat resistance in bacterial pathogens.

Weaknesses:<br /> There is no specific weakness in this work, although it would have helped to have a drawing model showing how ClpL performs protein disaggregation based on their new findings. The function of the higher assembly states of ClpL remains unresolved and will need further extensive research. Similarly, it will be interesting in the future to see whether the sole function of the plasmid-encoded ClpL is to cope with general protein aggregates under heat stress.

Reviewer #3 (Public Review):

Summary:<br /> This manuscript details the characterization of ClpL from L. monocytogenes as a potent and autonomous AAA+ disaggregase. The authors demonstrate that ClpL has potent and DnaK-independent disaggregase activity towards a variety of aggregated model substrates and that this disaggregase activity appears to be greater than that observed with the canonical DnaK/ClpB co-chaperone. Furthermore, Lm ClpL appears to have greater thermostability as compared to Lm DnaK, suggesting that ClpL-expressing cells may be able to withstand more severe heat stress conditions. Interestingly, Lm ClpP can provide thermotolerance to E. coli that have been genetically depleted of either ClpB or in cells expressing a mutant DnaK103. The authors further characterized the mechanisms by which ClpL interacts with protein aggregates, identifying that the N-terminal domain of ClpL is essential for disaggregase function. Lastly, by EM and mutagenesis analysis, the authors report that ClpL can exist in a variety of larger macromolecular complexes, including dimer or trimers of hexamers/heptamers, and they provide evidence that the N-terminal domains of ClpL prevent dimer ring formation, thus promoting an active and substrate-binding ClpL complex. Throughout this manuscript the authors compare Lm ClpL to ClpG, another potent and autonomous disaggregase found in gram-negative bacteria that have been reported on previously, demonstrating that these two enzymes share homologous activity and qualities. Taken together this report clearly establishes ClpL as a novel and autonomous disaggregase.

Strengths:<br /> The work presented in this report amounts to a significant body of novel and significant work that will be of interest to the protein chaperone community. Furthermore, by providing examples of how ClpL can provide in vivo thermotolerance to both E. coli and L. gasseri the authors have expanded the significance of this work and provided novel insight into potential mechanisms responsible for thermotolerance in food-borne pathogens.

Weaknesses:<br /> The figures are clearly depicted and easy to understand, though some of the axis labeling is a bit misleading or confusing and may warrant revision. While I do feel that the results and discussion as presented support the authors' hypothesis and overall goal of demonstrating ClpL as a novel disaggregase, interpretation of the data is hindered as no statistical tests are provided throughout the manuscript. Because of this only qualitative analysis can be made, and as such many of the concluding statements involving pairwise comparisons need to be revisited or quantitative data with stats needs to be provided. The addition of statistical analysis is critical and should not be difficult, nor do I anticipate that it will change the conclusions of this report.

eLife assessment

This study is valuable and contains results that are supported by convincing evidence. In the future, the observations could be further strengthened by independent validation, and by looking at larger numbers of patients, as well as by determining whether patient heterogeneity is either contributing to or obscuring certain patterns. The work will be of interest to a broad audience in the oncology and immunology fields as it is on a cancer type that does not respond well to immune checkpoint therapeutics.

Reviewer #1 (Public Review):

Summary:<br /> In this manuscript, Schmassmann et al. present a study on the immune microenvironment of grade 4 gliomas using single-cell RNA-seq data from the tumor center, periphery, and peripheral blood of patients. This manuscript is overall well written and reads easily. The approach to studying the TME at various spatial locations is innovative and interesting, and the dataset presented has the potential to become a useful resource for the community. However, the size of the dataset, notably in the context of the important inter-patient variability on key clinical information, hinders the generalizability of the results. The analysis presented by the authors seems at times somewhat shallow as compared to other studies in the literature, being almost solely based on the analysis of a single dataset with extremely limited biological validation of the observations, and some claims made by the authors do not seem appropriately backed by the data they present. While I appreciate the vast analysis effort undertaken by the authors, it seems more work is required to make the most of this interesting dataset and substantiate the conclusions.

Strengths:<br /> The authors have provided useful insights into diverse GBMs (IDH mutant and IDH wild-type) that provide a deep assessment of individual tumors with spatial information.

Weaknesses:<br /> A larger set of tumors will need to be explored before general principles of immune biology and GBM immune evasion can be uncovered. This is a descriptive study that provides some interesting new hypotheses - but these will need deeper functional exploration.

Reviewer #2 (Public Review):

Summary:<br /> Most of this paper concerns scRNA-seq data generated from glioblastoma patients, from three regions: tumor center, tumor periphery, and peripheral blood mononuclear cells. They focus on immune cells, and especially microglia and T-cells, where they look at the presence/absence/changes in different types of immune signatures. The data and analysis are sound and supportive of the conclusions they draw, though future studies with more patients and/or low-throughput validation would strengthen their evidence. This study adds to our knowledge of the immune cell environment in glioblastoma patients and its regional variation.

Strengths:<br /> A key strength of the paper lies in the novelty of the data, which simultaneously examines, at single-cell resolution, gene expression in two different tumor regions (center and periphery) along with peripheral blood. The authors provide numerous detailed and state-of-the-art analyses of this data, including gene differential expression, differential abundance of cell types, gene ontology analyses, tSNE visualizations, etc.

They focus in particular on differences in immune cell types. There are some suggestive differences in immune cell composition of center versus periphery, although the number of patients (5, one of whom is missing center data) does not allow one to draw a definitive conclusion.

They identified more definitive gene expression differences in center versus peripheral microglia -- differences that were not reflected in other cell types, and which included downregulation of a number of immune response functions. They also identified gene expression differences between two subsets of microglia, although those may partly reflect regional differences (the subsets are differentially enriched in the center versus periphery) or differential representation of different patients.

Finally, they identify differences in CD8+ T cells and NK cells in the center versus the periphery, where the latter were less activated/proliferative/cytotoxic.

Data analysis is performed to a high standard, using best-available methods and in some cases backed up with alternative approaches showing similar results.

Weaknesses:<br /> While the nature of the dataset is novel, the relatively low patient numbers (five) and patient diversity (e.g. with regard to IHD1 status) may be obscuring differences in cell type abundances or cell state between regions.

Most discoveries based on the scRNA-seq discussed in the paper remain to be validated by low-throughput methods in either the same patient samples, if material remains, or in other patients.

eLife assessment

The intrinsic chirality of actin filaments (F-actin) is implicated in the chiral arrangement and movement of cellular structures, but it was unknown how opposite chiralities can arise when the chirality of F-actin is invariant. Kwong et al. present evidence that two actin filament-based cytoskeletal structures, transverse actin arcs and radial stress fibers, drive clockwise and anti-clockwise rotation, respectively. This fundamental work, which has broad implications for cell biology, is supported by solid data, although the effect of the perturbations should be interpreted with caution.

Reviewer #1 (Public Review):

Summary:<br /> Kwong et al. present evidence that two actin-filament based cytoskeletal structures regulate the clockwise and anticlockwise rotation of the cytoplasm. These claims are based on experiments using cells plated on micropatterned substrates (circles). Previous reports have shown that the actomyosin network that forms on the dorsal surface of a cell plated on a circle drives a rotational or swirling pattern of movement in the cytoplasm. This actin network is composed of a combination of non-contractile radial stress fibers (AKA dorsal stress fibers) which are mechanically coupled to contractile transverse actin arcs (AKA actin arcs). The authors claim that directionality of the rotation of the cytoplasm (i.e., clockwise or anticlockwise) depends on either the actin arcs or radial fibers, respectively. While this would interesting, the authors are not able to remove either actin-based network without effecting the other. This is not surprising, as it is likely that the radial fibers require the arcs to elongate them, and the arcs require the radial fibers to stop them from collapsing. As such, it is difficult to make simple interpretations such as the clockwise bias is driven by the arcs and anticlockwise bias is driven by the radial fibers.

Weaknesses:<br /> There are also multiple problems with how the data is displayed and interpreted. First, it is difficult to compare the experimental data with the controls as the authors do not include control images in several of the figures. For example, Figure 6 has images showing myosin IIA distribution, but Figure 5 has the control image. Each figure needs to show controls. Otherwise, it will be difficult for the reader to understand the differences in localization of the proteins shown. This could be accomplished by either adding different control examples or by combining figures.

It is important that the authors should label the range of gray values of the heat maps shown. It is difficult to know how these maps were created. I could not find a description in the methods, nor have previous papers laid out a standardized way of doing it. As such, the reader needs some indication as to whether the maps showing different cells were created the same and show the same range of gray levels. In general, heat maps showing the same protein should have identical gray levels. The authors already show color bars next to the heat maps indicating the range of colors used. It should be a simple fix to label the minimum (blue on the color bar) and the maximum (red on the color bar) gray levels on these color bars. The profiles of actin shown in Figure 3 and Figure 3- figure supplement 3 were useful for interpretating the distribution of actin filaments. Why did not the authors show the same for the myosin IIa distributions?<br /> Line 189 "This absence of radial fibers is unexpected". The authors should clarify what they mean by this statement. The claim that the cell in Figure 3B has reduced radial stress fiber is not supported by the data shown. Every actin structure in this cell is reduced compared to the cell on the larger micropattern in Figure 3A. It is unclear if the radial stress fibers are reduced more than the arcs. Are the authors referring to radial fiber elongation?<br /> The choice of the small molecule inhibitors used in this study is difficult to understand, and their results are also confusing. For example, sequestering G actin with Latrunculin A is a complicated experiment. The authors use a relatively low concentration (50 nM) and show that actin filament-based structures are reduced and there are more in the center of the cell than in controls (Figure 3E). What was the logic of choosing this concentration? Using a small molecule that binds the barbed end (e.g., cytochalasin) could conceivably be used to selectively remove longer actin filaments, which the radial fibers have compared to the lamellipodia and the transverse arcs. The authors should articulate how the actin cytoskeleton is being changed by latruculin treatment and the impact on chirality. Is it just that the radial stress fibers are not elongating? There seems to be more radial stress fibers than in controls, rather than an absence of radial stress fibers. Similar problems arise from the other small molecules as well. LPA has more effects than simply activating RhoA. Additionally, many of the quantifiable effects of LPA treatment are apparent only after the cells are serum starved, which does not seem to be the case here. Furthermore, inhibiting ROCK with, Y-27632, effects myosin light chain phosphorylation and is not specific to myosin IIA. Are the two other myosin II paralogs expressed in these cells (myosin IIB and myosin IIC)? If so, the authors' statements about this experiment should refer to myosin II not myosin IIa. None of the uses of the small molecules above have supporting data using a different experimental method. For example, backing up the LPA experiment by perturbing RhoA tho.<br /> The use of SMIFH2 as a "formin inhibitor" is also problematic. SMIFH2 also inhibits myosin II contractility, making interpreting its effects on cells difficult to impossible. The authors present data of mDia2 knockdown, which would be a good control for this SMIFH2. However, the authors claim that mDia2 "typically nucleates tropomyosin-decorated actin filaments, which recruit myosin II and anneal endwise with α-actinin- crosslinked actin filaments." There is no reference to this statement and the authors own data shows that both arcs and radial fibers are reduced by mDia2 knockdown. Overall, the formin data does not support the conclusions the authors report.<br /> The data in Figure 7 does not support the conclusion that myosin IIa is exclusively on top of the cell. There are clear ventral stress fibers in A (actin) that have myosin IIa localization. The authors simply chose to not draw a line over them to create a height profile.

Reviewer #2 (Public Review):

Summary:<br /> Chirality of cells, organs, and organisms can stem from the chiral asymmetry of proteins and polymers at a much smaller lengthscale. The intrinsic chirality of actin filaments (F-actin) is implicated in the chiral arrangement and movement of cellular structures including F-actin-based bundles and the nucleus. It is unknown how opposite chiralities can be observed when the chirality of F-actin is invariant. Kwong, Chen, and co-authors explored this problem by studying chiral cell-scale structures in adherent mammalian cultured cells. They controlled the size of adhesive patches, and examined chirality at different timepoints. They made various molecular perturbations and used several quantitative assays. They showed that forces exerted by antiparallel actomyosin bundles on parallel radial bundles are responsible for the chirality of the actomyosin network at the cell scale.

Strengths:<br /> Whereas previously, most effort has been put into understanding radial bundles, this study makes an important distinction that transverse or circumferential bundles are made of antiparallel actomyosin arrays. A minor point that was nice for the paper to make is that between the co-existing chirality of nuclear rotation and radial bundle tilt, it is the F-actin driving nuclear rotation and not the other way around. The paper is clearly written.

Weaknesses:<br /> The paper could benefit from grammatical editing.

eLife assessment

This important, well-conducted study in a large data set - the UK BioBank population - reports that both circulating omega-6 and omega-3 PUFAs as well as the ratio of omega-6 to omega-3 PUFAs are associated with lower all-cause, cancer and cardiovascular mortality. The study is convincing and these findings will be of broad interest to epidemiologists, nutritionists, medical practitioners and the general population.

Reviewer #2 (Public Review):

Summary: This study utilized a large sample from the UK Biobank which enhanced statistical robustness, employed a prospective design to establish clear temporal relationships, used objective biomarkers for assessing plasma omega-6/omega-3 ratio, and investigated various mortality causes including CVD and cancer for a holistic health understanding.

Strengths: The authors used a large sample size, employed a prospective design, and investigated various mortality.

Weaknesses: Analyzing n-3 and n-6 PUFAs separately might be less instructive. It might not be methodologically sound to treat TG, HDL, LDL, and apolipoproteins as mediators. It's imperative to exercise caution when drawing causal conclusions from the observed correlations. The manuscript might propose potential research trajectories.

Reviewer #3 (Public Review):

Summary: The authors are trying to find out whether the levels of omega-6 and omega-3 fatty acids in the blood are linked to the likelihood of dying from anything, of dying from cancer and of dying from cardiovascular disease. They use a large dataset called UK Biobank where fatty acid levels were measured in blood at the start of the study and what happened to the participants over the following years (average of 12.7 years) was followed. They find that both omega-6 AND omega-3 fatty acids were linked with less likelihood of dying from anything, from cancer and from cardiovascular disease. The effects of omega-3s were stronger. They then made a ratio of omega-6 to omega-3 fatty acids and found that as that ratio increased risk of dying also increased,. This supports the idea that omega-3s have stronger effects than omega-6s.

Strengths: This is a large study (over 85,000 participants) with a good follow up period (average 12.7 years). Using blood levels of fatty acids is superior to using estimated dietary intakes. The authors take account of many variables that could interfere with the findings (confounding variables) - they do this using statistical methods.

Weaknesses: There are several omega-6 and omega-3 fatty acids - it is not clear which ones were actually measured in this study.

eLife assessment

This study presents a valuable finding on the optimal prioritization in different malaria transmission settings for the distribution of insecticide-treated nets to reduce the malaria burden. The evidence supporting the claims of the authors is solid. The work will be of interest from a global funder perspective, though somewhat less relevant for individual countries.

Reviewer #1 (Public Review):

Schmit et al. analyze and compare different strategies for the allocation of funding for insecticide-treated nets (ITNs) to reduce the global burden of malaria. They use previously published models of Plasmodium falciparum and Plasmodium vivax malaria transmission to quantify the effect of ITN distribution on clinical malaria numbers and the population at risk. The impact of different resource allocation strategies on the reduction of malaria cases or a combination of malaria cases and achieving pre-elimination is considered to determine the optimal strategy to allocate global resources to achieve malaria eradication.

Strengths:

Schmit et al. use previously published models and optimization for a rigorous analysis and comparison of the global impact of different funding allocation strategies for ITN distribution. This provides evidence of the effect of three different approaches: the prioritization of high-transmission settings to reduce the disease burden, the prioritization of low-transmission settings to "shrink the malaria map", and a resource allocation proportional to the disease burden.

Weaknesses:

The analysis and optimization which provide the evidence for the conclusions and are thus the central part of this manuscript necessitate some simplifying assumptions which may have important practical implications for the allocation of resources to reduce the malaria burden. For example, seasonality, mosquito species-specific properties, stochasticity in low transmission settings, and changing population sizes were not included. Other challenges to the reduction or elimination of malaria such as resistance of parasites and mosquitoes or the spread of different mosquito species as well as other beneficial interventions such as indoor residual spraying, seasonal malaria chemoprevention, vaccinations, combinations of different interventions, or setting-specific interventions were also not included. Schmit et al. clearly state these limitations throughout their manuscript.

This work considers different ITN distribution strategies, other interventions are not considered. It also provides a global perspective but an analysis of the specific local setting (as also noted by Schmit et al.) and different interventions as well as combinations of interventions should also be taken into account for any decisions. Nonetheless, the rigorous analysis supports the authors' conclusions and provides evidence that supports the prioritization of funding of ITNs for settings with high Plasmodium falciparum transmission. Overall, this work may contribute to making evidence-based decisions regarding the optimal prioritization of funding and resources to achieve a reduction in the malaria burden.

Reviewer #2 (Public Review):

Summary:

In this article, the authors discuss an optimal resource allocation strategy to best allocate funding in maximising malaria eradication efforts. Though achieving elimination by only using insecticide-treated bed nets (ITNs) is not the best practice, and countries utilise different interventions simultaneously, this analysis could be relevant in allocating funding for the global malaria elimination effort. To analyse and compare the impact of ITNs on P. falciparum and P. vivax cases and the total populations at risk, the authors use two previously published models (for P. falciparum and P. vivax).

Strengths:

The authors use models for both P. falciparum and P. vivax to analyse the impact of different strategies for allocating ITNs and provide the best strategies for funding to minimise malaria burden across different transmission settings. Using previously published models that account for various malaria aspects, including demography, heterogeneity in bite exposure, immunity, variation in hypnozoite across bites (P. vivax), mosquito larval dynamics, etc., gives a solid foundation for the analysis performed here.

Weaknesses:

Though the objective of the study is to identify the best setting to allocate funding to eradicate malaria, the authors use prevalence estimates (P. falciparum and P. vivax) based on the year 2000 as the baseline. Given their reasoning behind this choice, the analysis would be more relevant or useful if the proposed strategy were compared to the current Global Technical Strategy for Malaria (GTS 2016-2030). That is, using estimates based on around the year 2016.

In settings where both P. falciparum and P. vivax are co-endemic, using models that do not account for the interplay between the species, especially regarding immunity, somewhat underplays the overall disease dynamics. Furthermore, assuming the transmission within each setting (very low, low, moderate, high) is homogenous is also a weakness as there is heterogeneity in transmission intensity, bite exposure, etc, within each setting.

eLife assessment

This valuable study underscores the significance of PfMORC in shaping chromatin and guiding transitions in the malaria parasite, Plasmodium falciparum, that are essential for its survival. Solid evidence reveals PfMORC's influence on genes related to antigenic variation and the parasite's lifecycle, marking PfMORC as a key regulator of parasite heterochromatin.

Reviewer #1 (Public Review):

Summary: The authors investigated the function of Microrchidia (MORC) proteins in the human malaria parasite Plasmodium falciparum. Recognizing MORC's implication in DNA compaction and gene silencing across diverse species, the study aimed to explore the influence of PfMORC on transcriptional regulation, life cycle progression and survival of the malaria parasite. Depletion of PfMORC leads to the collapse of heterochromatin and thus to the killing of the parasite. The potential regulatory role of PfMORC in the survival of the parasite suggests that it may be central to the development of new antimalarial strategies.

Strengths: The application of the cutting-edge CRISPR/Cas9 genome editing tool, combined with other molecular and genomic approaches, provides a robust methodology. Comprehensive ChIP-seq experiments indicate PfMORC's interaction with sub-telomeric areas and genes tied to antigenic variation, suggesting its pivotal role in stage transition. The incorporation of Hi-C studies is noteworthy, enabling the visualization of changes in chromatin conformation in response to PfMORC knockdown.

Weaknesses: Although disruption of PfMORC affects chromatin architecture and stage-specific gene expression, determining a direct cause-effect relationship requires further investigation. Furthermore, while numerous interacting partners have been identified, their validation is critical and understanding their role in directing MORC to its targets or in influencing the chromatin compaction activities of MORC is essential for further clarification. In addition, the authors should adjust their conclusions in the manuscript to more accurately represent the multifaceted functions of MORC in the parasite.

Reviewer #2 (Public Review):

Summary: This paper, titled "Regulation of Chromatin Accessibility and Transcriptional Repression by PfMORC Protein in Plasmodium falciparum," delves into the PfMORC protein's role during the intra-erythrocytic cycle of the malaria parasite, P. falciparum. Le Roch et al. examined PfMORC's interactions with proteins, its genomic distribution in different parasite life stages (rings, trophozoites, schizonts), and the transcriptome's response to PfMORC depletion. They conducted a chromatin conformation capture on PfMORC-depleted parasites and observed significant alterations. Furthermore, they demonstrated that PfMORC depletion is lethal to the parasite.

Strengths: This study significantly advances our understanding of PfMORC's role in establishing heterochromatin. The direct consequences of the PfMORC depletion are addressed using chromatin conformation capture.

Weaknesses: The study only partially addressed the direct effects of PfMORC depletion on other heterochromatin markers.

eLife assessment

Using a variety of methods including mutant analyses, the authors study chromatin structure during spermatogenesis in Drosophila and transcriptional profiling in single cells/nuclei. This description of the dramatic changes in chromatin structure during spermatogenesis leads to some new observations, with convincing evidence, and it is useful for the field.

Reviewer #1 (Public Review):

Anderson, Henikoff and Ahmad et al. performed a series of genomics assays to study Drosophila spermatogenesis. Their main approaches include (1) Using two different genetic mutants that arrest male germ cell differentiation at distinct stages, bam and aly mutant, they performed CUT&TAG using H3K4me2, a histone modification for active promoters and enhancers; (2) Using FACS sorted pure spermatocytes, they performed CUT&TAG using antibodies against RNA PolII phosphorylated Ser 2, H4K16ac, H3K9me2, H3K27me3, and ubH2AK118. They also compare these chromatin profiling results with the published single-cell and single-nucleus RNA-seq data. Their analyses are across the genome but the major conclusions are about the chromatin features of the sex chromosomes. For example, the X chromosome is lack of dosage compensation as well as inactivation in spermatocytes, while Y chromosome is activated but enriched with ubH2A in spermatocytes. Overall, this work provides high quality epigenome data in testes and in purified germ cells. The analyses are very informative to understand and appreciate the dramatic chromatin structure change during spermatogenesis in Drosophila.

Reviewer #2 (Public Review):

Anderson et al profiled chromatin features, including active chromatin marks, RNA polymerase II distribution, and histone modifications in the sex chromosomes of spermatogenic cells in Drosophila. The experiments and analyses were well done, by a combination of the latest and appropriate methods. They include appropriate numbers of replicates. Results were parsed by comparing them among wildtype and two mutant with different arrest stages in spermatogenesis, as well as in FACS-sorted spermatocytes. The authors profiled larval wing discs as reference-somatic cells, allowing focus on features associated with germ cells; comparisons to testis somatic cells provided further specificity. Results were further refined by categorizing genes of interest based on available single nucleus RNA seq expression profiles. The authors acknowledge that the paper's interpretations are based on subtractive logic using the mutants, but comment that more precise ways of staging would not have yielded sufficient sample for their methods.

The authors documented differences in the distribution of RNAPIIS2p on some genes in germ cells vs somatic cells, the presence of a uH2A body beginning in early spermatocytes, and high levels of uH2A on the Y chromosome with little or none on the X, which is intriguing because uH2A is usually associated with silencing, yet the Y chromosome is active in spermatogenic cells. All of these are new, interesting, and important. Also importantly, the authors' data provide molecular details consistent with lack of MSCI, and lack of dosage compensation of the X chromosome in Drosophila spermatocytes.

eLife assessment

This study makes a valuable contribution by characterizing the role of the exocyst in secretory granule exocytosis in the Drosophila larval salivary gland. The results lead to the novel interpretation that the exocyst participates not only in exocytosis, but also in earlier steps of secretory granule biogenesis and maturation. Although these ideas are potentially of interest to a wide range of membrane traffic researchers, the evidence is incomplete, and the authors are urged to consider the possibility that inactivation of an essential exocytosis component might have indirect effects on other parts of the secretory pathway.

Reviewer #1 (Public Review):

Suarez-Freire et al. analyzed here the function of the exocyst complex in the secretion of the glue proteins by the salivary glands of the Drosophila larva. This is a widely used, genetically accessible system in which the formation, maturation and precisely timed exocytosis of the glue secretory granules can be beautifully imaged. Using RNAi, the authors show that all units of the exocyst complex are required for exocytosis. They show that not just granule fusion with the plasma membrane is affected (canonical role), but also, with different penetrance, that glue protein is retained in the ER, secretory granules fail to fuse homo-typically or fail to acquire maturation features. The authors document these phenotypes and postulate specific roles for the exocyst in these additional processes to explain them: exocyst as an ER-Golgi tether and exocyst as a granule-granule tether. However, the evidence for these highly novel, potentially interesting roles would need to be more compelling to support direct involvement. For instance, the localization of exocyst to Golgi or to granule-granule contact sites does not seem substantial. Instead, it is possible that defects in Golgi traffic and granule homotypic fusion are not due to direct involvement of the exocyst in these processes, but secondary to a defect in canonical exocyst roles at the plasma membrane. A block in the last step of glue exocytosis could perhaps propagate backward in the secretory pathway to disrupt Golgi complexes or cause poor cellular health due to loss of cell polarity or autophagy. In the absence of stronger evidence for these other exocyst roles, I would suggest focusing the study on the canonical role (interesting, as it was previously reported that Drosophila exocyst had no function in the salivary gland and limited function elsewhere [DOI: 10.1034/j.1600-0854.2002.31206.x]), and leave the alternative roles for discussion and deeper study in the future.

Reviewer #2 (Public Review):

The manuscript from Wappner and Melani labs claims a novel for the exocyst subunits in multiple aspects of secretory granule exocytosis. This an intriguing paper that suggests multiple roles of the exocyst in granule maturation and fusion with roles at the ER/Golgi interface, TGN, and granule homotypic fusion.

A key strength is the breadth of the assays and study of all 8 exocyst subunits in a powerful model system (fly larvae). Many of the assays are quantitated and roles of the exocyst in early phases of granule biogenesis have not been ascribed.

However there are several weaknesses, both in terms of experimental controls, concrete statements about the granules (better resolution), and making a clear conceptual framework.

Namely, why do KD of different exocysts have different effects on presumed granule formation? Why does just overexpression of a single subunit (Sec15) induce granule fusion? While the paper is fascinating, the major comments need to be addressed to really be able to make better sense of this work, which at present is hard to disentangle direct vs. secondary effects, especially as much of the TGN seems to be altered in the KDs. The authors conveniently ascribe many of the results to the holocomplex, but their own data (Fig. 4 and Fig. 6) are at odds with this.

Major Comments:

1. Resolution not sufficient. Identification of "mature secretory granules" (MSG) in Fig. 3 is based on low-resolution images in which the MSG are not clearly seen (see control in Fig. 3A) and rather appear as a diffuse haze, and not as clear granules. There may be granules here, but as shown it is not clear. Thus it would be helpful to acquire images at higher resolution (at the diffraction limit, or higher) to see and count the MSG. (Note: the authors are not clear on which objective was used. The 20x/0.8 NA or 63x/1.4 NA? Maybe the air objective as the resolution appears poor). They need to prove that the diffuse Sgs3-GFP haze is indeed due to MSG. Related it is unclear what are the granule structures that correspond to Immature secretory granules (ISG) and cells with mesh-like structures (MLS)? Similarly, Sgs3 images of KD of 8 exocyst subunits were interpreted to be identical, in Fig. 4, but the resolution is poor.

2. Explanation of variability of exocyst KD on the appearance of MSG. What is remarkable is a highly variable effect of different subunit KD on the percentage of cells with MLS (Fig. 4C). Controls = 100 %, Exo70=~75% (at 19 deg), Sec3 = ~30%, Sec10 = 0%, Exo84 = 100% ... This is interesting for the functional exocyst is an octameric holocomples, thus why the huge subunit variability in the phenotypes? The trivial explanation is either: i) variable exocyst subunit KD (not shown) or ii) variability between experiments (no error bars are shown). Both should be addressed by quantification of the KD of different proteins and secondly by replicating the experiments. If their data holds up then the underlying mechanism here needs to be considered. (Note: there is some precedent from the autophagy field of differential exocyst effects).

3. In the salivary glands the authors state that the exocyst is needed for Sgs3-GFP exit from the ER. First, Pearson's coefficient should be shown so as to quantitate the degree of ER localizations of all KDs. Second, there should be some rescue performed (if possible) to support specificity. Third, importantly other proteins that should traffic to the PM need to be shown to traffic normally so as to rule out a non-specific effect.

4. Golgi: It is unclear from their model (Fig. 5) why after exocyst KD of Sec15 the cis-Golgi is more preserved than the TGN, which appears as large vacuoles. This is not quantitated and not shown for the 8 subunits.

5. Acute/Chronic control: It would be nice to acutely block the exocyst so as to better distinguish if the effects observed are primary or secondary effects (e.g. on a recycling pathway).

6. Higher Resolution imaging (EM or super-resolution) - this would be nice to better understand the morphological interpretations.

7. Granule homotypic fusion. Strangely over-expression of just one subunit, Sec15-GFP, made giant secretory granules (SG) that were over 8 microns big! Why is that, especially if normally the exocyst is normally a holocomplex. Was this an effect that was specific to Sec15 or all exocyst subunits? Is the Sec15 level rate limiting in these cells? It may be that a subcomplex of Sec15/10 plays earlier roles, but in any case this needs to be addressed across all (or many) of the exocyst subcomplex members.

In summary, there are clearly striking effects on secretory granule biogenesis by dysfunction of the exocyst, however right now it is hard to disentangle effects on ERGolgi traffic, loss of the TGN, and a problem in maturation or fusion of granules. It is also confusing if the entire exocyst holocomplex or subcomplex plays a key role.

Reviewer #3 (Public Review):

Freire and co-authors examine the role of the exocyst complex during the formation and secretion of mucins from secretory granules in the larval salivary gland of Drosophila melanogaster. Using transgenic lines with a tagged Sgs3 mucin the authors KD expression of exocyst subunit members and observe a defect in secretory granules with a heterogeneity of phenotypes. By carefully controlling RNAi expression using a Gal4-based system the authors can KD exocyst subunit expression to varying degrees. The authors find that the stronger the inhibition of expression of exocyst the earlier in the secretory pathway the defect. The manuscript is well written, the model system is physiological, and the techniques are innovative.

My major concern is that the evidence underlying the fundamental claim of the manuscript that "the exocyst complex participates" in multiple secretory processes lacks direct evidence. It is clear from multiple lines of evidence, which are discussed by the authors, that exocyst is essential for an array of exocytic events. The fundamental concern is that loss of homeostasis on the plasma membrane proteome and lipidome might have severe pleiotropic effects on the cell. Indeed exocyst is essential, even in tissue culture conditions, and loss is lethal. Therefore, is an alternative explanation not that they are observing varying degrees of pleiotropic defect on the secretory pathway? Perhaps the authors have more evidence that exocyst is important for homeotypic fusion of the SGs, as supported by the localisation of Sec15 on the fusion sites.

The second question that I think is important to address is, what exactly do the varying RNAi levels correspond to in terms of experiments, and have these been validated? Due to the fundamental claim being that the severity of the phenotype being correlated with the level of KD, I think validation of this model is absolutely essential.

eLife assessment

Connelly and colleagues provide convincing genetic evidence that importation from mainland Tanzania is a major source of Plasmodium falciparum lineages currently circulating in Zanzibar. This study also reveals ongoing local malaria transmission and occasional near-clonal outbreaks in Zanzibar. Overall, this research highlights the role of human movements in maintaining residual malaria transmission in an area targeted for intensive control interventions over the past decades and provides valuable information for epidemiologists and public health professionals.

eLife assessment

This important study combines laboratory and field studies to quantify the force of human malaria parasite transmission. The methods of malaria parasite quantification in the mosquito midgut and salivary glands are compelling, but the statistical analyses seem to be biased by high infection loads of laboratory infections and would benefit from a more granular assessment of low parasite loads observed in the field. The study establishes a correlation between the sporozoite loads in the mosquito and the number of expelled parasites and would be of interest to vector biologists, parasitologists, immunologists, and mathematical modellers.

Reviewer #1 (Public Review):

Summary:

There is a long-believed dogma in the malaria field; a mosquito infected with a single oocyst is equally infectious to humans as another mosquito with many oocysts. This belief has been used for goal setting (and modeling) of malaria transmission-blocking interventions. While recent studies using rodent malaria suggest that the dogma may not be true, there was no such study with human P. falciparum parasites. In this study, the numbers of oocysts and sporozoite in the mosquitoes and the number of expelled sporozoites into artificial skin from the infected mosquito was quantified individually. There was a significant correlation between sporozoite burden in the mosquitoes and expelled sporozoites. In addition, this study showed that highly infected mosquitoes expelled sporozoites sooner.

Strengths:

• The study was conducted using two different parasite-mosquito combinations; one was lab-adapted parasites with Anopheles stephensi and the other was parasites, which were circulated in infected patients, with An. coluzzii. Both combinations showed statistically significant correlations between sporozoite burden in mosquitoes and the number of expelled sporozoites.

• Usually, this type of study has been done in group bases (e.g., count oocysts and sporozoites at different time points using different mosquitoes from the same group). However, this study determined the numbers in individual bases after multiple optimization and validation of the approach. This individual approach significantly increases the power of correlation analysis.

Weaknesses:

• In a natural setting, most mosquitoes have less than 5 oocysts. Thus, the conclusion is more convincing if the authors perform additional analysis for the key correlations (Fig 3C and 4D) excluding mosquitoes with very high total sporozoite load (e.g., more than 5-oocyst equivalent load).

• As written as the second limitation of the study, this study did not investigate whether all expelled sporozoites were equally infectious. For example, Day 9 expelled sporozoites may be less infectious than Day 11 sporozoites, or expelled sporozoites from high-burden mosquitoes may be less infectious because they experience low nutrient conditions in a mosquito. Ideally, it is nice to test the infectivity by ex vivo assays, such as hepatocyte invasion assay, and gliding assay at least for salivary sporozoites. But are there any preceding studies where the infectivity of sporozoites from different conditions was evaluated? Citing such studies would strengthen the argument.

• Since correlation analyses are the main points of this paper, it is important to show 95%CI of Spearman rank coefficient (not only p-value). By doing so, readers will understand the strengths/weaknesses of the correlations. The p-value only shows whether the observed correlation is significantly different from no correlation or not. In other words, if there are many data points, the p-value could be very small even if the correlation is weak.

Reviewer #2 (Public Review):

Summary:

The malaria parasite Plasmodium develops into oocysts and sporozoites inside Anopheles mosquitoes, in a process called sporogony. Sporozoites invade the insect salivary glands in order to be transmitted during a blood meal. An important question regarding malaria transmission is whether all mosquitoes harboring Plasmodium parasites are equally infectious. In this paper, the authors investigated the progression of P. falciparum sporozoite development in Anopheles mosquitoes, using a sensitive qPCR method to quantify sporozoites and an artificial skin system to probe for parasite expelling. They assessed the association between oocyst burden, salivary gland infection intensity, and sporozoites expelled.

The data show that higher sporozoite loads are associated with earlier colonization of salivary glands and a higher prevalence of sporozoite-positive salivary glands and that higher salivary gland sporozoite burdens are associated with higher numbers of expelled sporozoites. Intriguingly, there is no clear association between salivary gland burdens and the prevalence of expelling, suggesting that most infections reach a sufficient threshold to allow parasite expelling during a mosquito bite. This important observation suggests that low-density gametocyte carriers, although less likely to infect mosquitoes, could nevertheless contribute to malaria transmission.

Strengths:

The paper is well written and the work is well conducted. The authors used two experimental models, one using cultured P. falciparum gametocytes and An. stephensi mosquitoes, and the other one using natural gametocyte infections in a field setup with An. coluzzii mosquitoes. Both studies gave similar results, reinforcing the validity of the observations. Parasite quantification relies on a robust and sensitive qPCR method, and parasite expelling was assessed using an innovative experimental setup based on artificial skin.

Weaknesses:

There is no clear association between the prevalence of sporozoite expelling and the parasite burden. However, high total sporozoite burdens are associated with earlier and more efficient colonization of the salivary glands, and higher salivary gland burdens are associated with higher numbers of expelled sporozoites. While these observations suggest that highly infected mosquitoes could transmit/expel parasites earlier, this is not directly addressed in the study. In addition, whether all expelled sporozoites are equally infectious is unknown. The central question, i.e. whether all infected mosquitoes are equally infectious, therefore remains open.

Reviewer #3 (Public Review):

Summary:

This study uses a state-of-the-art artificial skin assay to determine the quantity of P. falciparum sporozoites expelled during feeding using mosquito infection (by standardised membrane feeding assay SMFA) using both cultured gametocytes and natural infection. Sporozoite densities in salivary glands and expelled into the skin are quantified using a well-validated molecular assay. These studies show clear positive correlations between mosquito infection levels (as determined by oocyst numbers), sporozoite numbers in salivary glands, and sporozoites expelled during feeding. This indicates potentially significant heterogeneity in infectiousness between mosquitoes with different infection loads and thus challenges the often-made assumption that all infected mosquitoes are equally infectious.

Strengths:

Very rigorously designed studies using very well validated, state-of-the-art methods for studying malaria infections in the mosquito and quantifying load of expelled sporozoites. This resulted in very high-quality data that was well-analyzed and presented. Both sources of gametocytes (cultures vs. natural infection) show consistent results further strengthening the quality of the results obtained.

Weaknesses:

As is generally the case when using SMFAs, the mosquito infections levels are often relatively high compared to wild-caught mosquitoes (e.g. Bombard et al 2020 IJP: median 3-4 ), and the strength of the observed correlations between oocyst sheet and salivary gland sporozoite load even more so between salivary gland sporozoite load and expelled sporozoite number may be dominated by results from mosquitoes with infection levels rarely observed in wild-caught mosquitoes. This could result in an overestimation of the importance of these well-observed positive relationships under natural transmission conditions.

The results obtained from these excellently designed and executed studies very well supported their conclusion - with a slight caveat regarding their application to natural transmission scenarios

This work very convincingly highlights the potential for significant heterogeneity in the infectiousness between individual P. falciparum-infected mosquitoes. Such heterogeneity needs to be further investigated and if again confirmed taken into account both when modelling malaria transmission and when evaluating the importance of low-density infections in sustaining malaria transmission.

Reviewer #4 (Public Review):

The study compares the number of sporozoites expelled by mosquitoes with different Plasmodium infection burden. To my knowledge this is the first report comparing the number of expelled P. falciparum sporozoites and their relation to oocyst burden (intact and ruptured) and residual sporozoites in salivary glands. The study provides important evidence on malaria transmission biology although conclusions cannot be drawn on direct impact on transmission.

Although there is some evidence from malaria challenge studies that the burden of sporozoites injected into a host is directly correlated with the likelihood of infection, this has been done using experimental infection models which administer sporozoites intravenously. It is unclear whether the same correlation occurs with natural infections and what the actual threshold for infection may be. Host immunity and other host related factors also play a critical role in transmission and need to be taken into consideration; these have not been mentioned by the authors. This is of particular importance as host immunity is decreasing with reduction in transmission intensity.

The natural infections reported in the study were not natural as the authors described. Gametocyte enrichment was done to attain high oocyst infection numbers. Studying natural infections would have been better without the enrichment step. The infected mosquitoes have much larger infection burden than what occurs in the wild.<br /> Nevertheless, the findings support the same results as in the experiments conducted in the Netherlands and therefore are of interest. I suggest the authors change the wording. Rather than calling these "natural" infections, they could be called, for example, "experimental infections with wild parasite strains".

I do not believe the study results generate sufficient evidence to conclude that lower infection burden in mosquitoes is likely to result in changes to transmission potential in the field. In study limitations section, the authors say "In addition, our quantification of sporozoite inoculum size is informative for comparisons between groups of high and low-infected mosquitoes but does not provide conclusive evidence on the likelihood of achieving secondary infections. Given striking differences in sporozoite burden between different Plasmodium species - low sporozoite densities appear considerably more common in mosquitoes infected with P. yoelli and P. Berghei the association between sporozoite inoculum and the likelihood of achieving secondary infections may be best examined in controlled human infection studies. However, in the abstract conclusion the authors state "Whilst sporozoite expelling was regularly observed from mosquitoes with low infection burdens, our findings indicate that mosquito infection burden is associated with the number of expelled sporozoites and may need to be considered in estimations of transmission potential." Kindly consider ending the sentence at "expelled sporozoites." Future studies on CHMI can be recommended as a conclusion if authors feel fit.

eLife assessment

In this important paper, Al-Hasani and colleagues provide the scientific community with a method to measure peptide concentrations in the brains of freely behaving animals. This support for this method is solid and expands upon previous methodological advances by this group by uncovering the role of these molecules during ongoing behavior. This solid contribution will be of broad interest to the neuroscientific community.

Reviewer #1 (Public Review):

Summary:

The present study by Mikati et al demonstrates an improved method for in-vivo detection of enkephalin release and studies the impact of stress on the activation of enkephalin neurons and enkephalin release in the nucleus accumbens (NAc). The authors refine their pipeline to measure met and leu enkephalin using liquid chromatography and mass spectrometry. The authors subsequently measured met and leu enkephalin in the NAc during stress induced by handling, and fox urine, in addition to calcium activity of enkephalinergic cells using fiber photometry. The authors conclude that this improved tool for measuring enkephalin reveals experimenter handling stress-induced enkephalin release in the NAc that habituates and is dissociable from the calcium activity of these cells, whose activity doesn't habituate. The authors subsequently show that NAc enkephalin neuron calcium activity does habituate to fox urine exposure, is activated by a novel weigh boat, and that fox urine acutely causes increases in met-enk levels, in some animals, as assessed by microdialysis.

Strengths:

A new approach to monitoring two distinct enkephalins and a more robust analytical approach for more sensitive detection of neuropeptides. A pipeline that potentially could help for the detection of other neuropeptides.

Weaknesses:

Some of the interpretations are not fully supported by the existing data or would require further testing to draw those conclusions. This can be addressed by appropriately tampering down interpretations and acknowledging other limitations the authors did not cover brought by procedural differences between experiments.

Reviewer #2 (Public Review):

Summary:

The authors aimed to improve the detection of enkephalins, opioid peptides involved in pain modulation, reward, and stress. They used optogenetics, microdialysis, and mass spectrometry to measure enkephalin release during acute stress in freely moving rodents. Their study provided better detection of enkephalins due to the implementation of previously reported derivatization reaction combined with improved sample collection and offered insights into the dynamics and relationship between Met- and Leu-Enkephalin in the Nucleus Accumbens shell during stress.

Strengths:

A strength of this work is the enhanced opioid peptide detection resulting from an improved microdialysis technique coupled with an established derivatization approach and sensitive and quantitative nLC-MS measurements. These improvements allowed basal and stimulated peptide release with higher temporal resolution, lower detection thresholds, and native-state endogenous peptide measurement.

Weaknesses:

The draft incorrectly credits itself for the development of an oxidation method for the stabilization of Met- and Leu-Enk peptides. The use of hydrogen peroxide reaction for the oxidation of Met-Enk in various biological samples, including brain regions, has been reported previously, although the protocols may slightly vary. Specifically, the manuscript writes about "a critical discovery in the stabilization of enkephalin detection" and that they have "developed a method of methionine stabilization." Those statements are incorrect and the preceding papers that relied on hydrogen peroxide reaction for oxidation of Met-Enk and HPLC for quantification of oxidized Enk forms should be cited. One suggested example is Finn A, Agren G, Bjellerup P, Vedin I, Lundeberg T. Production and characterization of antibodies for the specific determination of the opioid peptide Met5-Enkephalin-Arg6-Phe7. Scand J Clin Lab Invest. 2004;64(1):49-56. doi: 10.1080/00365510410004119. PMID: 15025428.

Another suggestion for this draft is to make the method section more comprehensive by adding information on specific tools and parameters used for statistical analysis:

1) Need to define "proteomics data" and explain whether calculations were performed on EIC for each m/z corresponding to specific peptides or as a batch processing for all detected peptides, from which only select findings are reported here. What type of data normalization was used, and other relevant details of data handling? Explain how Met- and Leu-Enk were identified from DIA data, and what tools were used.

2) Simple Linear Regression Analysis: The text mentions that simple linear regression analysis was performed on forward and reverse curves, and line equations were reported, but it lacks details such as the specific variables being regressed (although figures have labels) and any associated statistical parameters (e.g., R-squared values).

3) Violin Plots: The proteomics data is represented as violin plots with quartiles and median lines. This visual representation is mentioned, but there is no detail regarding the software/tools used for creating these plots.

4) Log Transformation: The text states that the data was log-transformed to reduce skewness, which is a common data preprocessing step. However, it does not specify the base of the logarithm used or any information about the distribution before and after transformation.

5) Two-Way ANOVA: Two-way ANOVA was conducted with peptide and treatment as independent variables. This analysis is described, but there is no information regarding the software or statistical tests used, p-values, post-hoc tests, or any results of this analysis.

6) Paired T-Test: A paired t-test was performed on predator odor proteomic data before and after treatment. This step is mentioned, but specific details like sample sizes, and the hypothesis being tested are not provided.

7) Correlation Analysis: The text mentions a simple linear regression analysis to correlate the levels of Met-Enk and Leu-Enk and reports the slopes. However, details such as correlation coefficients, and p-values are missing.

8) Fiber Photometry Data: Z-scores were calculated for fiber photometry data, and a reference to a cited source is provided. This section lacks details about the calculation of z-scores, and their use in the analysis.

9) Averaged Plots: Z-scores from individual animals were averaged and represented with SEM. It is briefly described, but more details about the number of animals, the purpose of averaging, and the significance of SEM are needed.

A more comprehensive and objective interpretation of results could enhance the overall quality of the paper.

Reviewer #3 (Public Review):

Summary:

This important paper describes improvements to the measurement of enkephalins in vivo using microdialysis and LC-MS. The key improvement is the oxidation of met- to prevent having a mix of reduced and oxidized methionine in the sample which makes quantification more difficult. It then shows measurements of enkephalins in the nucleus accumbens in two different stress situations - handling and exposure to predator odor. It also reports the ratio of released met- and leu-enkephalin matching what is expected from the digestion of proenkephalin. Measurements are also made by photometry of Ca2+ changes for the fox odor stressor. Some key takeaways are the reliable measurement of met-enkephalin, the significance of directly measuring peptides as opposed to proxy measurements, and the opening of a new avenue into the research of enkephalins due to stress based on these direct measurements.

Strengths:

-Improved methods for measurement of enkephalins in vivo.

-Compelling examples of using this method.

-Opening a new area of looking at stress responses through the lens of enkephalin concentrations.

Weaknesses:

1) It is not clear if oxidized met-enk is endogenous or not and this method eliminates being able to discern that.

2) It is not clear if the spatial resolution is really better as claimed since other probes of similar dimensions have been used.

3) Claims of having the first concentration measurement are not quite accurate.

4) Without a report of technical replicates, the reliability of the method is not as well-evaluated as might be expected.

Reviewer #1 (Public Review):

Summary:<br /> Mitochondria is the power plant of the cells including neurons. Thomas et al. characterized the distribution of mitochondria in dendrites and spines of L2/3 neurons from the ferret visual cortex, for which visually driven calcium responses of individual dendritic spines were examined. The authors analyzed the relationship between the position of mitochondria and the morphology or orientation selectivity of nearby dendrite spines. They found no correlation between mitochondrion location and spine morphological parameters associated with the strength of synapses, but correlation with the spine-somatic difference in orientation preference and local heterogeneity in preferred orientation of nearby spines. Moreover, they reported that the spines that have a mitochondrion in the head or neck are larger in size and have stronger orientation selectivity. Therefore, they proposed that "mitochondria are not necessarily positioned to support the energy needs of strong spines, but rather support the structurally and functionally diverse inputs."

Strengths:<br /> This paper attempted to address a fundamental question: whether the distribution of the mitochondria along the dendrites of visual cortical neurons is associated with the functions of the spines, postsynaptic sites of excitatory synapses. Two state of the art techniques (2 photon Ca imaging of somata and spines and EM reconstructions of cortical pyramidal neurons) had been used on the same neurons, which provides a great opportunity to examine and correlate the functional properties of spine ultrastructure and spatial distribution of dendritic mitochondria. The conclusion that dendritic mitochondria support functional diversity of spines, but not synaptic strength is surprising and will inspire rethinking the role of mitochondria in synaptic functions.

Weaknesses:<br /> Overall, the findings are intriguing. However, the interpretations of these findings need extra cautions due to the limitations of experimental designs and tools in this study. Neurons in L2/3 of visual cortex are highly diverse in functional properties, which is represented by not only orientation selectivity, but also direction selectivity and spatial/temporal frequency selectivity, etc. The orientation tuning with fixed spatial and temporal frequency may not be the optimal way of stimulating individual synaptic inputs to evaluate synaptic strengths. And the correlation between mitochondria distribution and spine activity evoked by other visual stimulation parameters is worth exploration. Moreover, GCaMP6s measures only spine Ca signals mediated by NMDA and voltage-gated Ca channels, but not sodium currents mediated by ligand-gated or voltage-gated channels. Thus, it reports only some aspects of synaptic properties. Future studies with new tools might help resolve those issues.

Reviewer #2 (Public Review):

Summary:<br /> Mitochondria in synapses are important to support functional needs, such as local protein translation and calcium buffering. Thus, they may be strategically localized to maximize functional efficiency. In this study, the authors examine whether a correlation exists between the positioning of mitochondria and the structure or function of dendritic spines in the visual cortex of a ferret. Unexpectedly, the authors found no correlation between structural measures of synaptic strength to mitochondria positioning, which may indicate that they are not localized only because of the local energy needs. Instead, the authors discover that mitochondria are positioned preferably in spines that display heterogeneous responses, showing that they are localized to support specific functional needs probably distinct from ATP output.

Strengths:<br /> The thorough analysis provides a yet unprecedented insight into the correlation between synaptic tuning and mitochondrial positioning in the visual cortex in vivo.

Weaknesses:<br /> Analysis of this study suggested that mitochondrial volume does not correlate with structural measures of synaptic strength (e.g. spine volume and post-synaptic density (PSD) area), but it remains to be determined if mitochondria localization is also co-related to the frequency of synaptic activity, and what causes the correlation (driven by mitochondrial positioning, or by synaptic activity).

Reviewer #3 (Public Review):

Summary: This is a careful examination of the distribution of mitochondria in the basal dendrites of ferret visual cortex in a previously published volume electron microscopy dataset. The authors report that mitochondria are sparsely, as opposed to continuously distributed in the dendritic shafts, and that they tend to cluster near dendritic spines with heterogeneous orientation selectivity.

Strengths: Volume EM is the gold standard for quantification of organelle morphology. An unusual strength of this particular dataset is that the orientation selectivity of the dendritic spines was measured by calcium imaging prior to EM reconstruction. This allowed the authors to assess how spines with varying selectivity are organized relative to mitochondria, leading to an intriguing observation that they localize to heterogeneous spine clusters. The analysis is carefully performed. An additional strength is the use of a carnivore with a sophisticated visual system.

Weaknesses: Using threshold distances between mitochondria and synapses as opposed to absolute distances may overlook important relationships in the data.

eLife assessment

This study presents an important discovery that the RNA synthesis protein of SARS-CoV-2, the virus that is responsible for COVID 19, has fewer mutations and causes limited conformational changes. The evidence supporting the claims is convincing, with robust sequence alignment studies, state-of-the-art protein-protein interaction analysis, and molecular conformational analysis. This work has implications for drug design and will be of broad interest to the general biophysics and structural biology community.

Joint Public Review:

The work is of fundamental importance and is a useful structural resource to the SARS-CoV2 proteome. The work relies on large-scale SARS-CoV2 genomes and extracts frequent mutations in two key proteins NSP16 and NSP10. The impact of these mutations was studied using x-ray crystallogrpahy, biophysical assays and simulations to propose structural changes. The evidence is, therefore, convincing to suggest NSP10 conformational changes are limited. More analysis on functional implications would be useful to understand the underlying reasons of limited structural variability. The questions raised during the review of the original submission have been addressed by the authors.

eLife assessment

Through a combination of careful experimental designs and computational modelling, this study provides solid evidence highlighting the role of attention in shaping temporal binding. Overall, the findings of this paper are important, supporting the cognitive link between time perception, temporal binding, and spatial attention.

Reviewer #1 (Public Review):

Summary:<br /> This study addressed an alternative hypothesis to temporal binding phenomena. In temporal binding, two events that are separated in time are "pulled" towards one another, such that they appear more coincidental. Previous research has shown evidence of temporal binding events in the context of actions and multisensory events. In this context, the author revisits the well-known Libet clock paradigm, in which subjects view a moving clock face, press a button at a time of their choosing to stop the clock, a tone is played (after some delay), and then subjects move the clock dial to the point where the one occurred (or when the action occurred). Classically, the reported clock time is a combination of the action and sound times. The author here suggests that attention can explain this by a mechanism in which the clock dial leads to a roving window of spatiotemporal attention (that is, it extends in both space and time around the dial). To test this, the author conducted a number of experiments where subjects performed the Libet clock experiment, but with a variety of different stimulus combinations. Crucially, a visual detection task was introduced by flashing a disc at different positions along the clock face. The results showed that detection performance was also "pulled" towards the action event or sensory event, depending on the condition. A model of roving spatiotemporal attention replicated these effects, providing further evidence of the attentional window.

Strengths:<br /> The study provides a novel explanation for temporal binding phenomena, with clear and cleverly designed experiments. The results provide a nice fit to the proposed model, and the model itself is able to recapitulate the observed effects.

Weaknesses:<br /> Despite the above, the paper could be clearer on why these effects are occurring. In particular, the control experiment introduced in Experiment 3 is not well justified. Why should a tactile stimulus not lead to a similar effect? There are possibilities here, but the author could do well to lay them out. Further, from a perspective related to the attentional explanation, other alternatives are not explored. The author cites and considers work suggesting that temporal binding relies on a Bayesian cue combination mechanism, in which the estimate is pulled towards the stimulus with the lowest variance, but this is not discussed. None of this necessarily detracts from the findings, but otherwise makes the case for attention less clear.

Reviewer #2 (Public Review):

Summary:<br /> Temporal binding, generally considered a timing illusion, results from actions triggering outcomes after a brief delay, distorting perceived timing. The present study investigates the relationship between attention and the perception of timing by employing a series of tasks involving auditory and visual stimuli. The results highlight the role of attention in event timing and the functional relevance of attention in outcome binding.

Strengths:<br /> - Experimental Design: The manuscript details a well-structured sequence of experiments investigating the attention effect in outcome binding. Thoughtful variations in manipulation conditions and stimuli contribute to a thorough and meaningful investigation of the phenomenon.<br /> - Statistical Analysis: The manuscript employs a diverse set of statistical tests, demonstrating careful selection and execution. This statistical approach enhances the reliability of the reported findings.<br /> - Narrative Clarity: Both in-text descriptions and figures provide clear insights into the experiments and their results, facilitating readers in following the logic of the study.

Weaknesses:<br /> - Conceptual Clarity: The manuscript aims to integrate key concepts in human cognitive functions, including attention, timing perception, and sensorimotor processes. However, before introducing experiments, there's a need for clearer definitions and explanations of these concepts and their known and unknown interrelationships. Given the complexity of attention, a more detailed discussion, including specific types and properties, would enhance reader comprehension.

- Computational Modeling: The manuscript lacks clarity in explaining the model architecture and setup, and it's unclear if control comparisons were conducted. These details are critical for readers to properly interpret attention-related findings in the modeling section. Providing a clearer overview of these aspects will improve the overall understanding of the computational models used.

eLife assessment

In this useful study, Millard et al. assessed the effects of nicotine on pain sensitivity and peak alpha frequency (PAF). The evidence shown is incomplete to support the key claim that nicotine modulates PAF or pain sensitivity, considering the effect sizes observed. This raises the question of whether the chosen experimental intervention was the most suitable approach for investigating their research question. Nonetheless, the work can be incorporated into the literature investigating the relationship between nicotine and pain, and could be of broad interest to pain researchers.

Reviewer #1 (Public Review):

Summary:<br /> In this study, Millard and colleagues investigated if the analgesic effect of nicotine on pain sensitivity, assessed with two pain models, is mediated by Peak Alpha Frequency (PAF) recorded with resting state EEG. The authors found indeed that nicotine (4 mg, gum) reduced pain ratings during phasic heat pain but not cuff pressor algometry compared to placebo conditions. Nicotine also increased PAF (globally). However, mediation analysis revealed that the reduction in pain ratings elicited by the phasic heat pain after taking nicotine was not mediated by the changes in PAF. Also, the authors only partially replicated the correlation between PAF and pain sensitivity at baseline (before nicotine treatment). At the group-level no correlation was found, but an exploratory analysis showed that the negative correlation (lower PAF, higher pain sensitivity) was present in males but not in females. The authors discuss the lack of correlation.<br /> In general, the study is rigorous, methodology is sound and the paper is well-written. Results are compelling and sufficiently discussed.

Strengths:<br /> Strengths of this study are the pre-registration, proper sample size calculation, and data analysis. But also the presence of the analgesic effect of nicotine and the change in PAF.

Weaknesses:<br /> It would even be more convincing if they had manipulated PAF directly.

Reviewer #2 (Public Review):

Summary: The study by Millard et al. investigates the effect of nicotine on alpha peak frequency and pain in a very elaborate experimental design. According to the statistical analysis, the authors found a factor-corrected significant effect for prolonged heat pain but not for alpha peak frequency in response to the nicotine treatment.

Strengths: I very much like the study design and that the authors followed their research line by aiming to provide a complete picture of the pain-related cortical impact of alpha peak frequency. This is very important work, even in the absence of any statistical significance. I also appreciate the preregistration of the study and the well-written and balanced introduction. However, it is important to give access to the preregistration beforehand.

Weaknesses: The weakness of the study revolves around three aspects:

(1) I am not entirely convinced that the authors' analysis strategy provides a sufficient signal-to-noise ratio to estimate the peak alpha frequency in each participant reliably. A source separation (ICA or similar) would have been better suited than electrode ROIs to extract the alpha signal. By using a source separation approach, different sources of alpha (mu, occipital alpha, laterality) could be disentangled.

(2) Also, there's a hint in the literature (reference 49 in the manuscript) that the nicotine treatment may not work as intended. Instead, the authors' decision to use nicotine to modulate the peak alpha frequency and pain relied on other, not suitable work on chronic pain and permanent smokers. In the present study, the authors use nicotine treatment and transient painful stimulation on non-smokers.

In my view, the discussion could be more critical for some aspects and the authors speculate towards directions their findings can not provide any evidence. Speculations are indeed very important to generate new ideas but should be restricted to the context of the study (experimental pain, acute interventions). The unfortunate decision to use nicotine severely hampered the authors' aim of the study.

Impact: The impact of the study could be to show what has not worked to answer the research questions of the authors. The authors claim that their approach could be used to define a biomarker of pain. This is highly desirable but requires refined methods and, in order to make the tool really applicable, more accurate approaches at subject level.

Reviewer #3 (Public Review):

In this manuscript, Millard et al. investigate the effects of nicotine on pain sensitivity and peak alpha frequency (PAF) in resting state EEG. To this end, they ran a pre-registered, randomized, double-blind, placebo-controlled experiment involving 62 healthy adults who received either 4 mg nicotine gum (n=29) or placebo (n=33). Prolonged heat and pressure were used as pain models. Resting state EEG and pain intensity (assessed with a visual analog scale) were measured before and after the intervention. Additionally, several covariates (sex at birth, depression and anxiety symptoms, stress, sleep quality, among others) were recorded. Data was analyzed using ANCOVA-equivalent two-wave latent change score models, as well as repeated measures analysis of variance. Results do not show *experimentally relevant* changes of PAF or pain intensity scores for either of the prolonged pain models due to nicotine intake.

The main strengths of the manuscript are its solid conceptual framework and the thorough experimental design. The researchers make a good case in the introduction and discussion for the need to further investigate the association of PAF and pain sensitivity. Furthermore, they proceed to carefully describe every aspect of the experiment in great detail, which is excellent for reproducibility purposes. Finally, they analyze the data from almost every possible angle and provide an extensive report of their results.<br /> The main weakness of the manuscript is the interpretation of these results. Even though some of the differences are statistically significant (e.g., global PAF, pain intensity ratings during heat pain), these differences are far from being experimentally or clinically relevant. The effect sizes observed are not sufficiently large to consider that pain sensitivity was modulated by the nicotine intake, which puts into question all the answers to the research questions posed in the study.

eLife assessment

The ThermoMaze represents a valuable tool to control the rest/exploration states of an animal. The data, collected and analyzed using solid and validated methodology, demonstrate its use in addressing previously elusive questions. More in-depth analysis of place cell activity would provide better support for some of the claims.

Reviewer #1 (Public Review):

Summary:<br /> This manuscript introduced a new behavioral apparatus to regulate the animal's behavioral state naturally. It is a thermal maze where different sectors of the maze can be set to different temperatures; once the rest area of the animal is cooled down, it will start searching for a warmer alternative region to settle down again. They recorded with silicon probes from the hippocampus in the maze and found that the incidence of SWRs was higher at the rest areas and place cells representing a rest area were preferentially active during rest-SWRs as well but not during non-REM sleep.

Strengths:<br /> The maze can have many future applications, e.g., see how the duration of waking immobility can influence learning, future memory recall, or sleep reactivation. It represents an out-of-the-box thinking to study and control less-studies aspects of the animals' behavior.

Weaknesses:<br /> The impact is only within behavioral research and hippocampal electrophysiology.

Reviewer #2 (Public Review):

Summary:<br /> In this manuscript, Vöröslakos and colleagues describe a new behavioural testing apparatus called ThermoMaze, which should facilitate controlling when a mouse is exploring the environment vs. remaining immobile. The floor of the apparatus is tiled with 25 plates, which can be individually heated, whereas the rest of the environment is cooled. The mouse avoids cooled areas and stays immobile on a heated tile. The authors systematically changed the location of the heated tile to trigger the mouse's exploratory behaviours. The authors showed that if the same plate stays heated longer, the mouse falls into an NREM sleep state. The authors conclude their apparatus allows easy control of triggering behaviours such as running/exploration, immobility and NREM sleep. The authors also carried out single-unit recordings of CA1 hippocampal cells using various silicone probes. They show that the location of a mouse can be decoded with above-chance accuracy from cell activity during sharp wave ripples, which tend to occur when the mouse is immobile or asleep. The authors suggest that consistent with some previous results, SPW-Rs encode the mouse's current location and any other information they may encode (such as past and future locations, usually associated with them).

Strengths:<br /> Overall, the apparatus may open fruitful avenues for future research to uncover the physiology of transitions from different behavioural states such as locomotion, immobility, and sleep. The setup is compatible with neural recordings. No training is required.

Weaknesses:<br /> I have a few concerns related to the authors' methodology and some limitations of the apparatus's current form. Although the authors suggest that switching between the plates forces animal behaviour into an exploratory mode, leading to a better sampling of the enclosure, their example position heat maps and trajectories suggest that the behaviour is still very stereotypical, restricted mostly to the trajectories along the walls or the diagonal ones (between two opposite corners). This may not be ideal for studying spatial responses known to be affected by the stereotypicity of the animal's trajectories. Moreover, given such stereotypicity of the trajectories mice take before and after reaching a specific plate, it may be that the stable activity of SWR-P ripples used for decoding different quadrants may be representing future and/or past trajectories rather than the current locations suggested by the authors. If this is the case, it may be confusing/misleading to call such activity ' place-selective firing', since they don't necessarily encode a given place per se (line 281).

Another main study limitation is the reported instability of the location cells in the Thermomaze. This may be related to the heating procedure, differences in stereotypical sampling of the enclosure, or the enclosure size (too small to properly reveal the place code). It would be helpful if the authors separate pyramidal cells into place and non-place cells to better understand how stable place cell activity is. This information may also help to disambiguate the SPW-R-related limitations outlined above and may help to solve the poor decoding problem reported by the authors (lines 218-221).

eLife assessment

Building on previous toolboxes to distinguish 1/f noise from oscillatory activity, this study introduces an important advancement in neural signal analysis to identify oscillatory activity in electrophysiological data that refines the accuracy of identifying non-sinusoidal neural oscillations. Extensive validation, using synthetic and various empirical data, provides convincing evidence for the accuracy of the method and outlines practical implications for relevant scientific problems in the field.

Reviewer #1 (Public Review):

Summary: The study introduces and validates the Cyclic Homogeneous Oscillation (CHO) detection method to precisely determine the duration, location, and fundamental frequency of non-sinusoidal neural oscillations. Traditional spectral analysis methods face challenges in distinguishing the fundamental frequency of non-sinusoidal oscillations from their harmonics, leading to potential inaccuracies. The authors implement an underexplored approach, using the auto-correlation structure to identify the characteristic frequency of an oscillation. By combining this strategy with existing time-frequency tools to identify when oscillations occur, the authors strive to solve outstanding challenges involving spurious harmonic peaks detected in time-frequency representations. Empirical tests using electrocorticographic (ECoG) and electroencephalographic (EEG) signals further support the efficacy of CHO in detecting neural oscillations.

Strengths:

1. The paper puts an important emphasis on the 'identity' question of oscillatory identification. The field primarily identifies oscillations through frequency, space (brain region), and time (length, and relative to task or rest). However, more tools that claim to further characterize oscillations by their defining/identifying traits are needed, in addition to data-driven studies about what the identifiable traits of neural oscillations are beyond frequency, location, and time. Such tools are useful for potentially distinguishing between circuit mechanistic generators underlying signals that may not otherwise be distinguished. This paper states this problem well and puts forth a new type of objective for neural signal processing methods.

2. The paper uses synthetic data and multimodal recordings at multiple scales to validate the tool, suggesting CHO's robustness and applicability in various real-data scenarios. The figures illustratively demonstrate how CHO works on such synthetic and real examples, depicting in both time and frequency domains. The synthetic data are well-designed, and capable of producing transient oscillatory bursts with non-sinusoidal characteristics within 1/f noise. Using both non-invasive and invasive signals exposes CHO to conditions which may differ in extent and quality of the harmonic signal structure. An interesting followup question is whether the utility demonstrated here holds for MEG signals, as well as source-reconstructed signals from non-invasive recordings.

3. This study is accompanied by open-source code and data for use by the community.

Weaknesses:

1. Due to the proliferation of neural signal processing techniques that have been designed to tackle issues such as harmonic activity, transient and event-like oscillations, and non-sinusoidal waveforms, it is naturally difficult for every introduction of a new tool to include exhaustive comparisons of all others. Here, some additional comparisons may be considered for the sake of context, a selection of which follows, biased by the previous exposure of this reviewer. One emerging approach that may be considered is known as state-space models with oscillatory and autoregressive components (Matsuda 2017, Beck 2022). State-space models such as autoregressive models have long been used to estimate the auto-correlation structure of a signal. State-space oscillators have recently been applied to transient oscillations such as sleep spindles (He 2023). Therefore, state-space oscillators extended with auto-regressive components may be able to perform the functions of the present tool through different means by circumventing the need to identify them in time-frequency. Another tool that should be mentioned is called PAPTO (Brady 2022). Although PAPTO does not address harmonics, it detects oscillatory events in the presence of 1/f background activity. Lastly, empirical mode decomposition (EMD) approaches have been studied in the context of neural harmonics and non-sinusoidal activity (Quinn 2021, Fabus 2022). EMD has an intrinsic relationship with extrema finding, in contrast with the present technique. In summary, the existence of methods such as PAPTO shows that researchers are converging on similar approaches to tackle similar problems. The existence of time-domain approaches such as state-space oscillators and EMD indicates that the field of time-series analysis may yield even more approaches that are conceptually distinct and may theoretically circumvent the methodology of this tool.

2. The criteria that the authors use for neural oscillations embody some operating assumptions underlying their characteristics, perhaps informed by immediate use cases intended by the authors (e.g., hippocampal bursts). The extent to which these assumptions hold in all circumstances should be investigated. For instance, the notion of consistent auto-correlation breaks down in scenarios where instantaneous frequency fluctuates significantly at the scale of a few cycles. Imagine an alpha-beta complex without harmonics (Jones 2009). If oscillations change phase position within a timeframe of a few cycles, it would be difficult for a single peak in the auto-correlation structure to elucidate the complex time-varying peak frequency in a dynamic fashion. Likewise, it is unclear whether bounding boxes with a pre-specified overlap can capture complexes that maneuver across peak frequencies.

3. Related to the last item, this method appears to lack implementation of statistical inferential techniques for estimating and interpreting auto-correlation and spectral structure. In standard practice, auto-correlation functions and spectral measures can be subjected to statistical inference to establish confidence intervals, often helping to determine the significance of the estimates. Doing so would be useful for expressing the likelihood that an oscillation and its harmonic has the same auto-correlation structure and fundamental frequency, or more robustly identifying harmonic peaks in the presence of spectral noise. Here, the authors appear to use auto-correlation and time-frequency decomposition more as a deterministic tool rather than an inferential one. Overall, an inferential approach would help differentiate between true effects and those that might spuriously occur due to the nature of the data. Ultimately, a more statistically principled approach might estimate harmonic structure in the presence of noise in a unified manner transmitted throughout the methodological steps.

4. As with any signal processing method, hyperparameters and their ability to be tuned by the user need to be clearly acknowledged, as they impact the robustness and reproducibility of the method. Here, some of the hyperparameters appear to be: a) number of cycles around which to construct bounding boxes and b) overlap percentage of bounding boxes for grouping. Any others should be highlighted by the authors and clearly explained during the course of tool dissemination to the community, ideally in tutorial format through the Github repository.

5. Most of the validation demonstrations in this paper depict the detection capabilities of CHO. For example, the authors demonstrate how to use this tool to reduce false detection of oscillations made up of harmonic activity and show in simulated examples how CHO performs compared to other methods in detection specificity, sensitivity, and accuracy. However, the detection problem is not the same as the 'identity' problem that the paper originally introduced CHO to solve. That is, detecting a non-sinusoidal oscillation well does not help define or characterize its non-sinusoidal 'fingerprint'. An example problem to set up this question is: if there are multiple oscillations at the same base frequency in a dataset, how can their differing harmonic structure be used to distinguish them from each other? To address this at a minimum, Figure 4 (or a followup to it) should simulate signals at similar levels of detectability with different 'identities' (i.e. different levels and/or manifestations of harmonic structure), and evaluate CHO's potential ability to distinguish or cluster them from each other. Then, does a real-world dataset or neuroscientific problem exist in which a similar sort of exercise can be conducted and validated in some way? If the "what" question is to be sufficiently addressed by this tool, then this type of task should be within the scope of its capabilities, and validation within this scenario should be demonstrated in the paper. This is the most fundamental limitation at the paper's current state.

References:

Beck AM, He M, Gutierrez R, Purdon PL. An iterative search algorithm to identify oscillatory dynamics in neurophysiological time series. bioRxiv. 2022. p. 2022.10.30.514422. doi:10.1101/2022.10.30.514422

Brady B, Bardouille T. Periodic/Aperiodic parameterization of transient oscillations (PAPTO)-Implications for healthy ageing. Neuroimage. 2022;251: 118974.

Fabus MS, Woolrich MW, Warnaby CW, Quinn AJ. Understanding Harmonic Structures Through Instantaneous Frequency. IEEE Open J Signal Process. 2022;3: 320-334.

Jones SR, Pritchett DL, Sikora MA, Stufflebeam SM, Hämäläinen M, Moore CI. Quantitative analysis and biophysically realistic neural modeling of the MEG mu rhythm: rhythmogenesis and modulation of sensory-evoked responses. J Neurophysiol. 2009;102: 3554-3572.

He M, Das P, Hotan G, Purdon PL. Switching state-space modeling of neural signal dynamics. PLoS Comput Biol. 2023;19: e1011395.

Matsuda T, Komaki F. Time Series Decomposition into Oscillation Components and Phase Estimation. Neural Comput. 2017;29: 332-367.

Quinn AJ, Lopes-Dos-Santos V, Huang N, Liang W-K, Juan C-H, Yeh J-R, et al. Within-cycle instantaneous frequency profiles report oscillatory waveform dynamics. J Neurophysiol. 2021;126: 1190-1208.

Reviewer #2 (Public Review):

Summary: A new toolbox is presented that builds on previous toolboxes to distinguish between real and spurious oscillatory activity, which can be induced by non-sinusoidal waveshapes. Whilst there are many toolboxes that help to distinguish between 1/f noise and oscillations, not many tools are available that help to distinguish true oscillatory activity from spurious oscillatory activity induced in harmonics of the fundamental frequency by non-sinusoidal waveshapes. The authors present a new algorithm which is based on autocorrelation to separate real from spurious oscillatory activity. The algorithm is extensively validated using synthetic (simulated) data, and various empirical datasets from EEG, intracranial EEG in various locations and domains (i.e. auditory cortex, hippocampus, etc.).

Strengths: Distinguishing real from spurious oscillatory activity due to non-sinusoidal waveshapes is an issue that has plagued the field for quite a long time. The presented toolbox addresses this fundamental problem which will be of great use for the community. The paper is written in a very accessible and clear way so that readers less familiar with the intricacies of Fourier transform and signal processing will also be able to follow it. A particular strength is the broad validation of the toolbox, using synthetic, scalp EEG, EcoG, and stereotactic EEG in various locations and paradigms.

Weaknesses: At many parts in the results section critical statistical comparisons are missing (e.g. FOOOF vs CHO). Another weakness concerns the methods part which only superficially describes the algorithm. Finally, a weakness is that the algorithm seems to be quite conservative in identifying oscillatory activity which may render it only useful for analysing very strong oscillatory signals (i.e. alpha), but less suitable for weaker oscillatory signals (i.e. gamma).

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Summary: Hansen et al. dissect the molecular mechanisms of bacterial ice nucleating proteins mutating the protein systematically. They assay the ice nucleating ability for variants changing the R-coils as well as the coil capping motifs. The ice nucleation mechanism depends on the integrity of the R-coils, without which the multimerization and formation of fibrils are disrupted.

Strengths: The effects of mutations are really dramatic, so there is no doubt about the effect. The variants tested are logical and progressively advance the story. The authors identify an underlying mechanism involving multimerization, which is plausible and compatible with EM data. The model is further shown to work in cells by tomography.

Weaknesses: The theoretical model presented for how the proteins assemble into fibrils is simple, but not supported by much data.

Agreed. This theoretical INP multimer model was introduced to promote discussion and elicit ideas on how to prove or disprove it. The length and width of the fibres are defined by cryo-ET results, in which the narrow width is just sufficient to accommodate a dimer of the INPs, and the long length requires that several INPs are joined end to end. Their antiparallel arrangement produces identical ends to the dimer and avoids steric clash of the C-terminal cap structures as well as the C-terminal GFP tag. This model can accommodate the wide range of INPs lengths seen in nature (due to different numbers of water-organizing coils) and introduced in mutagenesis experiments (Forbes et al. 2022). It defines a critical role for the R-coil subdomain in joining the dimers together and explains why this region cannot be shortened by more than a few coils either in nature or by experimentation.

In response to specific criticisms of the model (Fig. 9), we have redesigned this to be less schematic and to incorporate several copies of the AlphaFold-predicted structure.

Reviewer #2 (Public Review):

Summary:

This paper further investigates the role of self-assembly of ice-binding bacterial proteins in promoting ice-nucleation. For the P. borealis Ice Nucleating Protein (PbINP) studied here, earlier work had already determined clearly distinct roles for different subdomains of the protein in determining activity. Key players are the water-organizing loops (WO-loops) of the central beta-solenoid structure and a set of non-water-organizing C-terminal loops, called the R-loops in view of characteristically located arginines. Previous mutation studies (using nucleation activity as a read-out) had already suggested the R-loops interact with the WO loops, to cause self-assembly of PbINP, which in turn was thought to lead to enhanced ice-nucleating activity. In this paper, the activities of additional mutants are studied, and a bioinformatics analysis on the statistics of the number of WO- and R-loops is presented for a wide range of bacterial ice-nucleating proteins, and additional electron-microscopy results are presented on fibrils formed by the non-mutated PbINP in E coli lysates.

Strengths:

-A very complete set of additional mutants is investigated to further strengthen the earlier hypothesis.

-A nice bioinformatics analysis that underscores that the hypothesis should apply not only to PbINP but to a wide range of (related) bacterial ice-nucleating proteins.

-Convincing data that PbINP overexpressed in E coli forms fibrils (electron microscopy on E coli lysates).

Weaknesses:

-The new data is interesting and further strengthens the hypotheses put forward in the earlier work. However, just as in the earlier work, the proof for the link between self-assembly and ice-nucleation remains indirect. Assembly into fibrils is shown for E coli lysates expressing non-mutated pbINP, hence it is indeed clear that pbINP self-associates. It is not shown however that the mutations that lead to loss of ice-nucleating activity also lead to loss of self-assembly. A more quantitative or additional self-assembly assay could shine light on this, either in the present or in future studies.

The control cryo-ET experiment where the R-coils were deleted and INP fibres were not seen is consistent with a link between the loss of ice-nucleating activity and the loss of self-assembly. However, we agree that a more direct measurement of the physical state of INP molecules is needed to prove the link.

-Also the "working model" for the self-assembly of the fibers remains not more than that, just as in the earlier papers, since the mutation-activity relationship does not contain enough information to build a good structural model. Again, a better model would require different kinds of experiments, that yield more detailed structural data on the fibrils.

Reviewer #1 also raised these criticisms of the model, which we have responded to (above). Testing the model is a focus of our continuing experiments on INPs.

Reviewer #3 (Public Review):

Summary: in this manuscript, Hansen and co-authors investigated the role of R-coils in the multimerization and ice nucleation activity of PbINP, an ice nucleation protein identified in Pseudomonas borealis. The results of this work suggest that the length, localization, and amino acid composition of R-coils are crucial for the formation of PbINP multimers.

Strengths: The authors use a rational mutagenesis approach to identify the role of the length, localisation, and amino acid composition of R-coils in ice nucleation activity. Based on these results, the authors hypothesize a multimerization model. Overall, this is a multidisciplinary work that provides new insights into the molecular mechanisms underlying ice nucleation activity.

Weaknesses: Several parts of the work appear cryptic and unsuitable for non-expert readers. The results of this work should be better described and presented.

In revising the manuscript for reposting we have rewritten sections to make it more accessible to the non-expert. Incorporating the detailed recommendations of the reviewers has been helpful in this effort.

Recommendations for the authors: please note that you control which revisions to undertake from the public reviews and recommendations for the authors

Reviewer #1 (Recommendations For The Authors):

Introduction: Curiously, there is no mention at all in the introduction of what the biological function of these ice-nucleating proteins is.

We added the following text to the first paragraph of the Introduction: ”INP-producing bacteria are widespread in the environment where they are responsible for initiating frost (4) and atmospheric precipitation (5). As such, these bacteria play a significant role in the Earth’s hydrological cycle and in agricultural productivity.”

Line 70: TXT, SLT, and Y motifs are mentioned, but only the first is described. Also, TXT name alternates between TXT and TxT in the manuscript. (I think the latter is more correct).

These putative water-organizing motifs are introduced in the preceding paper (new ref 8). We now use TxT consistently throughout the manuscript and have converted SLT to SxT because L is an inward-pointing residue that is not directly involved in water organization.

Line 236: A construct with repeats deleted is tested for thermostability, but it is not really explained what hypothesis this experiment is supposed to test.

This is an observation that adds information about the stability of the INP multimers and will need to be explained by the structure.

Line 267: The authors test a mutant where the N-terminal coil is disrupted and find a big effect. Nevertheless, no conclusion is drawn. What does this result mean?

On the contrary, INP activity is not appreciably affected by N-terminal deletion.

Line 269: The CryoEM begins rather abruptly with technical details. Consider introducing the paragraph with a brief statement about what you want to investigate. Also, the analysis seems a little half-hearted.

Given that the authors describe other EM studies of fibrils of the same protein it would be nice with a clear statement about what is new in their study and how it compares to previous studies.

We have added this statement about why we used Cryo-EM: “The idea that INPs must assemble into larger structures to be effective at ice nucleation has persisted since their discovery (6). In the interim the resolving power of cryo-EM has immensely improved. Here we elected to use cryo-electron tomography to view the INP multimers in situ and avoid any perturbation of their superstructure during isolation.”

Fig. 7B: Single-letter amino acid codes are always capitalized.

We have revised this figure to use capital letters for the amino acids.

Fig. 9: This figure is really hard to read even though it is very simplistic. I would consider making a figure with several copies of the AlphaFold model instead. Especially panel D, I do not know what is supposed to show.

We have followed this advice and have completely revised the figure using copies of the AlphaFold model. Panel D (now C) shows two cross-sections through the AlphaFold model.

Line 355 onwards: The model of the INP is the weakest part of the manuscript. This reviewer considers that the model is crude and it is unclear what information the model is supported by. The authors might want to consider running an AlphaFold multimer to get a better model of at least the dimer.

Our objective now is to validate or disprove the model by experimentation using protein-protein cross-linking in conjunction with mass spectrometry, and higher resolution cryo-EM methods.

Reviewer #2 (Recommendations For The Authors):

I would suggest more frankly discussing the weaknesses mentioned in my public review, as well as approaches that could be used in the future to address these.

In the cryo-ET analysis, INP mutations of the R-coils that lead to loss of ice-nucleating activity fail to show fibres in the bacteria (Fig. S4), which is consistent with the loss of self-assembly. We are working on physical methods that can assess the degree of assembly of the different INP constructs and mutations. We are working to validate and improve the working model of INP multimers.

Reviewer #3 (Recommendations For The Authors):

Abstract

Line 18. Below 0 Celsius should be < 0 {degree sign}C.

Done

Line 25. E. coli should be Escherichia coli

Done

Line 29. E. coli should be in italics.

Done

Introduction

The introduction is weak and not suitable for non-expert readers. Moreover, in some parts it is cryptic and it is not clear whether the authors are describing INP in general or PbINP. The introduction should be reorganized to highlight the novelty of this paper compared to Forbes et al. 2022.

The changes we have made to the Introduction can be seen in the ‘documents compared’ version where the changes are tracked.

Line 45. It is unclear whether this paragraph is a result reported in the literature or the result of this work. Please clarify.

These are results reported in the literature as indicated by the references cited in the paragraph.

Line 54. It is not clear whether this paragraph describes PbINP or INP in general.

This paragraph begins with INPs in general and then focuses on PbINP.

Results

Line 109. This section would benefit from a paragraph in which the authors describe the rationale for this bioinformatic analysis.

We added the following Statement: “A bioinformatic analysis of bacterial INPs was undertaken to identify their variations in size and sequence to understand what is common to all that could guide experiments to probe higher order structure and help develop a collective model of the INP multimer.”

Some information is needed on the selected sequences such as sequence identity, what do the authors mean by nr database?

The abbreviation nr has been replaced by ‘non-redundant’. As explained in that same paragraph the sequences selected were those from long-read sequences that could be relied on to accurately count the number of solenoid coils.

Line 144. The standard deviation is necessary to understand whether these differences are statistically significant.

These have been added as p values.

Figure 2. I noticed that the authors used GFP-tagged PbINP. Why? In addition, panel C is never mentioned in the manuscript.

The GFP tag was used to confirm expression of the PbINP in E. coli. We have added this sentence: “As previously described these constructs were tagged with GFP as an internal control for INP production, and its addition had no measured effect on ice nucleation activity (8).”The GFP tag was also useful as in internal control for the heat denaturation experiments featured in Fig. 6, where it lost its fluorescence between 65 and 75 °C.

Fig. 2C is now cited alongside Fig. 2B.

Figure 3. In my opinion, the results of the R-coil deletion should also be shown in Figure 2. Line 171. This section is cryptic. A logo sequence or an alignment of WO-coils and R-coils of PbINP could be helpful for the reader. Instead of the architecture of the whole protein, it would be useful to have the sequence of the R-coils with the residues that the authors mutagenised.

The logo sequences are available in Fig. 1.

Line 202. Here, the authors describe a new experimental setup. As the Materials and Methods section follows the Discussion, the authors should state in the first paragraph of the Results section that IN activity was measured on whole cells.

We have now modified the introductory sentence to read: “Ice nucleation assays were performed on intact E. coli expressing PbINP to assess the activity of the incremental replacement mutants.”

Line 202. The authors investigated the effects of pH and temperature (Line 223) on the IN activity. The authors should better introduce the rationale for these experiments and how they fit within the work.

We have now modified the following sentence to provide the rationale: “To see how important electrostatic interactions were in the multimerization of PbINP as reflected by its ice nucleation activity, it was necessary to lyse the E. coli to change the pH surrounding the INP multimers.”

Line 245. This work is supported by a model provided by Alphafold. I wonder how reliable this model is; the authors should indicate the quality of the model and provide the accuracy values of the residuals.

This information is now provided in Figure S1.

Line 259. Typically in mutagenesis studies, a key residue is substituted with alanine to create a loss of function variant. In this case, the authors have made the following substitutions F1204D, D1208L, and Y1230D, it is not clear to me why the authors have replaced an aromatic residue with one of aspartic acid that is negatively charged.

We have justified these more extreme changes as follows: “For an enhanced effect of the mutations hydrophobic residues were replaced with charged ones and vice versa.”

Line 269. This paragraph seems completely unrelated to the section entitled: The β-solenoid of INPs is stabilized by a capping structure at the C terminus, but not at the N terminus.

We had omitted the sub-heading “Cryo-electron tomography reveals INPs multimers form bundled fibres in recombinant cells”, which is now in place.

Discussion

Overall, the discussion is too long and some parts appear cryptic, this section should be reorganized.

The changes we have made to the Discussion can be seen in the ‘documents compared’ version where the changes are tracked.

Line 354. It is not clear what experimental evidence supports this model. In the results, this model is never mentioned and it is not clear whether it was obtained by computational analysis or not.

The model is presented in the Discussion because it was not arrived at by experimentation but is an attempt to integrate the observations made in the Results section. The experimental evidence that supports this model is reviewed in the Discussion section: “Working model of the INP multimer is consistent with the properties of INPs and their multimers.”

Line 354. The authors used GFP-tagged PbINP. The Authors should discuss the role of GFP in this model and IN activity. A measurement of IN activity on PbINP without GFP would be useful.

We have previously shown in Ref 8 that the GFP tag has no detrimental effect on ice nucleation activity. Our model for the INP multimer can accommodate this C-terminal tag without any steric hindrance.

Line 364. The Authors hypothesize that electrostatic interactions stabilize end-to-end dimer associations. To test this hypothesis, the authors should measure the activity of IN at increasing concentrations of NaCl. It is known that high salt concentrations shield charges by preventing the formation of electrostatic intermolecular interactions.

We have added this sentence to the Discussion: “Another useful test of the electrostatic component to the multimer model would be to study the effects of increasing salt concentration on ice nucleation activity of the E. coli extracts.”

Line 439. Conclusions should be useful for the reader.

Material and Methods

In several sections, the authors refer to what has already been published in Forbes et al. However, the minimum information should also be described in this work. In addition, the Authors should indicate the number of replicates.

The ice nucleation assays on whole cells were done on the WISDOM apparatus, which integrates 100’s of individual measurements to obtain a T50 value. These T50 values were confirmed by assays on the nanoliter osmometer apparatus. The numbers of replicates used on the nanoliter osmometer apparatus are indicated by box and whisker plots in Figs. 5 & 6 with boxes and bars showing quartiles, with medians indicated by a centre line.

Line 500. This paragraph should be removed as the results are not described in the manuscript.

This is a Methods section that describes how that INPs were expression in E. coli. It has details that are important for researchers who want to repeat our findings, such as the use of the Arctic Express strain for producing INP.

Reviewer #2 (Public Review):

Summary:<br /> This paper further investigates the role of self-assembly of ice-binding bacterial proteins in promoting ice-nucleation. For the P. borealis Ice Nucleating Protein (PbINP) studied here, earlier work had already determined clearly distinct roles for different subdomains of the protein in determining activity. Key players are the water-organizing loops (WO-loops) of the central beta-solenoid structure and a set of non-water-organizing C-terminal loops, called the R-loops in view of characteristically located arginines. Previous mutation studies (using nucleation activity as a read-out) had already suggested the R-loops interact with the WO loops, to cause self-assembly of PbINP, which in turn was thought to lead to enhanced ice-nucleating activity. In this paper, the activities of additional mutants are studied, and a bioinformatics analysis on the statistics of the number of WO- and R-loops is presented for a wide range of bacterial ice-nucleating proteins, and additional electron-microscopy results are presented on fibrils formed by the non-mutated PbINP in E coli lysates.

Strengths:<br /> -A very complete set of additional mutants is investigated to further strengthen the earlier hypothesis.<br /> -A nice bioinformatics analysis that underscores that the hypothesis should apply not only to PbINP but to a wide range of (related) bacterial ice-nucleating proteins.<br /> -Convincing data that PbINP overexpressed in E coli forms fibrils (electron microscopy on E coli lysates).

Weaknesses:<br /> -The new data is interesting and further strengthens the hypotheses put forward in the earlier work. However, just as in the earlier work, the proof for the link between self-assembly and ice-nucleation remains indirect. Assembly into fibrils is shown for E coli lysates expressing non-mutated pbINP, hence it is indeed clear that pbINP self-associates. It is not shown however that the mutations that lead to loss of ice-nucleating activity also lead to loss of self-assembly. A more quantitative or additional self-assembly assay could shine light on this, either in the present or in future studies.

-Also the "working model" for the self-assembly of the fibers remains not more than that, just as in the earlier papers, since the mutation-activity relationship does not contain enough information to build a good structural model. Again, a better model would require different kinds of experiments, that yield more detailed structural data on the fibrils.

Reviewer #1 (Public Review):

Summary: Hansen et al. dissect the molecular mechanisms of bacterial ice nucleating proteins mutating the protein systematically. They assay the ice nucleating ability for variants changing the R-coils as well as the coil capping motifs. The ice nucleation mechanism depends on the integrity of the R-coils, without which the multimerization and formation of fibrils are disrupted.

Strengths: The effects of mutations are really dramatic, so there is no doubt about the effect. The variants tested are logical and progressively advance the story. The authors identify an underlying mechanism involving multimerization, which is plausible and compatible with EM data. The model is further shown to work in cells by tomography.

Weaknesses: The theoretical model presented for how the proteins assemble into fibrils is simple, but not supported by much data.

Reviewer #3 (Public Review):

Summary: in this manuscript, Hansen and co-authors investigated the role of R-coils in the multimerization and ice nucleation activity of PbINP, an ice nucleation protein identified in Pseudomonas borealis. The results of this work suggest that the length, localization, and amino acid composition of R-coils are crucial for the formation of PbINP multimers.

Strengths: The authors use a rational mutagenesis approach to identify the role of the length, localisation, and amino acid composition of R-coils in ice nucleation activity. Based on these results, the authors hypothesize a multimerization model. Overall, this is a multidisciplinary work that provides new insights into the molecular mechanisms underlying ice nucleation activity.

Weaknesses: Several parts of the work appear cryptic and unsuitable for non-expert readers. The results of this work should be better described and presented.

Author Response

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

Thank you for the e-mail of 27th September that includes the eLife assessment and reviewers comments on manuscript eLife-RP-RA-2023-91861. We have considered these, added additional data and made various changes to the text as detailed below. We now submit a modified version that we would be happy to view as the ‘Version of Record’.

We are very pleased to note the highly positive reports from the reviewers. The major change we have made is to alter the Introduction to include further consideration of the development of the ‘bar-code’ hypothesis. As highlighted by reviewer 2 the Lefkowitz/Duke University Group have been major proponents of this concept. However, as with many topics their views did not emerge in isolation. Indeed we (specifically Tobin) were developing similar ideas in the same period (see Tobin et al., (2008) Trends Pharmacol Sci 29, 413-420). Moreover, other groups, particularly that of Clark and collaborators at University of Texas, were developing similar ideas using the beta2-adrenoceptor as a model at least as early as this (e.g. Tran et al., (2004) Mol Pharmacol 65, 196-206). As such we have re-written parts of the Introduction to reflect these early studies whilst retaining information on more recent studies that have greatly expanded such early work. This has resulted in the addition of extra references and re-numbering of the Reference section. We have also provided statistical analysis of agonist-induced arrestin interactions with the receptor as requested by a reviewer and performed additional studies to assess the effect of the GRK2/3 inhibitor in agonist-regulation of phosphorylation of the hFFA2-DREADD receptor. This has led to an additional author (Aisha M. Abdelmalik) being added to the paper.

To address first the ‘public reviews’

Reviewer 1

1. We agree that we do not at this point explore the implications of the tissue specific barcoding we observe and report. However, as noted by the reviewer these will be studies for the future.

2. The question of why these are only 2 widely expressed arrestins and very many GPCRs is not one we attempt to address here and groups using various arrestin ‘conformation’ sensors are probably much better placed to do so than we are.

Reviewer 2

1. It is difficult to address the potential low level of ‘background’ staining in some of the immunocytochemical images versus the ‘cleaner’ background in some of the immunoblotting images. The methods and techniques used are very distinct. However, it should be apparent that the immunoblotting studies are performed (both using cell lines and tissues) post-immunoprecipitation and this is likely to reduce such background to a minimum. This is obviously not the case in the immunocytochemical studies. It is also likely, even though the antisera are immune-selected against the peptide target, there may be some level of immune-recognition this is not limited to the phosphorylated residues.

2. Whilst this reviewer has commented in detail in the ‘recommendations’ section on the use of English, the other reviewers have not, and we do not find the manuscript challenging to follow or read.

Reviewer 3

1. We agree that the mass-spectrometry presented is not quantitative. The intention was for the mass spec to be a guide for the development of the antisera used in the study. We have re-written the initial part of the Results section (page 7) to state that phosphorylation of Ser297 was evident in the basal and agonist-stimulated receptor whilst phosphorylation of Ser296 was only evident following agonist addition.

2. Immunoblotting is intrinsically variable as parameters of antiserum titre in re-used samples is not assessed and although we are aware that FFA2 displays a degree of constitutive activity (see for example Hudson et al., (2012) J Biol Chem. 287(49):41195-209) we did not make any specific effort to supress this by, for example, including an inverse agonist ligand. Agonist-regulation of phosphorylation of the receptor, as detected in cell lines by the anti- pThr306/pThr310antiserum, is exceptionally clear cut in all the images displayed, and as we note for the pSer296/pSer297 antiserum this was always, in part, agonist-independent.

The point about compound 101 not being tested directly in the immunoblotting studies performed on the cell line-expressed receptor is a good one. We have now performed such studies which are shown as Figure 2E. These illustrate that the GRK2/3 inhibitor compound 101 does not reduce substantially agonist-induced phosphorylation of the receptor at least as detected by the pThr306/pThr310antiserum or by the pSer296/pSer297 antiserum. Equally this compound had little effect on recognition of the receptor. As the PD2 mutations which correspond to the targets for the pThr306/pThr310antiserum have no significant effect on recruitment of arrestin 3 in response to MOMBA (please see additional statistical analysis in modified Figure 2C) this is perhaps not surprising. Moreover, the PD1 mutations that correspond to the pSer296/pSer297antiserum also, in isolation, only have a partial effect of MOMBA-induced interactions with arrestin 3.

1. The use of phosphatase inhibitors is an integral part of these studies. As noted in Materials we used PhosSTOP (Roche, 4906837001). However, we failed to make it sufficiently clear that this reagent was present throughput sample preparation for both cell lines and tissue studies. This had been specified previously by two of us (SS, FN, see Fritzwanker S, Nagel F, Kliewer A, Stammer V, Schulz S. In situ visualization of opioid and cannabinoid drug effects using phosphosite-specific GPCR antibodies. Commun Biol. 6, 419 (2023)) but we agree this was insufficient and we now correct this oversight by making this explicit in Results.

Recommendations

Reviewer 1

Competing interest: We apologise for this typographic error. It is now corrected.

Figures: We have upgraded the figure images to 300dpi and this markedly improves readability

Reviewer 2

Revisiting writing: We thank the reviewer for their assessment of the text. However, we do not feel that ‘every sentence in the entire manuscript could be clarified’ is a reasonable statement. Neither of the other reviewers commented on this. Each of the authors read and approved the manuscript.

Figures: see response to Reviewer 1. We have greatly enhanced image quality at this part of the process.

Statistics on Figure 2: We apologise for this oversight. Although there were no significant differences in potency for MOMBA to promote interactions with arrestin-3 to each of the PD mutants versus wild type receptor, there were in terms of maximal effect. Statistical analysis was performed via one-way ANOVA followed by Dunnett’s multiple comparisons test. This is now detailed directly in Figure 2C and its associated legend. As noted by the reviewer there was indeed a highly significant effect of the GRK2/3 inhibitor compound 101 and this is now also noted in Figure 2D and its associated legend.

Units on page 9: pEC50 is considered as Molar by default but we have now specified this. PD1-4: It would be cumbersome to write out (and to read) 8 mutations that make up PD1-4 and hence we think this is specified appropriately in the Figure.

Reviewer 3

1. Mass spec: Please see comment point 1 to reviewer 3.

2. Immunoblotting and compound 101: We have done so.

3. Phosphatase inhibition: see public comments, reviewer 3.

Reviewer #2 (Public Review):

The strengths of this paper begin with the topic. Specifically, this approaches the question of how GPCR signals are directed to different outcomes under different conditions. There is rich complexity within this question; there are potentially billions of molecules that could interact with >800 human GPCRs and thousands of molecular effectors that may be activated. However, these outcomes are filtered through a small number of GPCR-interacting proteins that direct the signal.

Experimentally, strengths include the initial experimental controls employed in characterizing their ever-important antisera, on which their conclusions hinge. In showing strong agonist-dependent and phosphosite-dependent recognition, as well as the addition of GRK inhibitors and eventually an antagonist and phosphatase treatment, the authors substantiate the role of the antiserum in recognizing their intended motifs. When employed, those antisera overall give clear indications of differences across variables in immunoblots, and while the immunocytochemical studies are qualitative and at times not visually significantly different across all variables, they are in large part congruent with the results of the immunoblots and provide secondary supporting evidence for the author's major claims. One confounding aspect of the immunocytochemical images is the presence of background pThr306/pThr310, like in Figures 4C and 6A and B. In 4A and C, while the immunoblot shows a complete absence of pThr306/pThr310, Figure 4C's immuno image does not. In 6A and B, a similar presence of pThr306/pThr310 is seen in the vehicle image, which is not strikingly over-shown by the MOMBA-treated image. In addition, only Ser/Thr residues of the C-terminus were investigated, while residues of ICL3 have long been known to direct signaling in many GPCRs. Because of the presentations of sequences, it was not clear whether there were residues of ICL3 that have the possibility of being involved.

It may be possible and further testable to show whether the residues that maintain basal phosphorylation could also be tissue-specific, especially considering the presence of pThr306/pThr310 detection in both the Figure 6A immunoblot's vehicle lane (but not MOMBA lane). The aforementioned detection in the immunocytochemical vehicle image could support differential basal phosphorylation in the enteroendocrine cells. Should this be the case, it could have confounded the initial mass-spec screen wherein the Ser residues were basally active in that cell type, while in a distinct cell type that may not be the case. Lastly, should normalized quantification of these images be possible, it may help in clearing up these hard-to-compare visual images.

It is noted that aspects of the writing and presentation may lead to confusion for some readers, but this does not affect the overall significance of the work.

Nevertheless, in terms of the global goal of the authors, the indication of differences in phosphorylation states between tissues is still evident across the experiments. Accordingly, the paper is overall strongly well-researched, well-controlled, and the conclusions made by the authors are data-grounded and not overly extrapolated. Providing direct evidence for the tissue-based branch of the barcode hypothesis is both novel and significant for the field, and the paper leaves room for much more exciting research to be done in the area, opening the door for new questions and hypotheses.

eLife assessment

In this study, the authors present important tools for monitoring distinct tissue-specific patterns of agonist-induced Free Fatty Acid receptor 2 phosphorylation. The work includes several validation experiments, which provide convincing evidence that will be beneficial for the scientific community.

Reviewer #1 (Public Review):

Summary:

Very systematic generation of phosphosite-specific antisera to monitor FFA2 phosphorylation in native cells and tissues. Provides evidence that FFA2 phosphorylation is tissue-specific.

Strengths:

Technical tour de force, rigorous experimental approaches taking advantage of wt and DREADD versions of FFA2 to make sure that ligand-and receptor-dependent phosphorylations are indeed specific to FFA2.

Weaknesses:

In this reviewer's opinion, the only shortcoming is that the implications of tissue-selective phosphorylation barcoding remain unexplored. However, I understand that tool development is required before tools are used to provide insight into the functional outcomes of receptor regulation by phosphorylation. The study is a technical tour de force to generate highly valuable tools. I have no major criticisms but suggest adding an additional aspect to the discussion as specified below.

Arrestins are highly flexible and dynamic phosphate sensors. If two arrestins have to recognize 800 different phosphorylated GPCRs, is it possible that any barcode serves the same purpose: arrestin recognition followed by signal arrest and internalization? Because phosphorylation barcoding is linked to G protein-independent signaling, which is claimed by some but is experimentally unsupported, and because arrestins don't transduce receptor signals on their own (they only scaffold signaling components and shuttle receptors within cellular compartments), I would also include this option in the discussion, i.e. that the different barcodes are a way nature may have chosen to regulate the location of 800 GPCRs by only 2 arrestins.

Reviewer #3 (Public Review):

Summary:

The authors generate and characterize two phosphospecific antisera for FFA2 receptor and claim a "bar code" difference between white fat and Peyers patches.

Strengths:

The question is interesting and the antibody characterization is convincing.

Weaknesses:

The mass spectrometry analysis is not convincing because the method is not quantitative (no SILAC, TMT, internal standards etc). Figure 1 shows single tryptic peptides with one and two phosphorylation fragmentations as claimed, but there is no data testing the abundance of these so the differences claimed between cell treatment conditions are not established.

The blot analysis cannot distinguish 296/7 but it does convincingly show an agonist increase. Can the authors clarify why the amount of constitutive phosphorylation is much higher in the example blot in Figure 2 than in Figure 3? It would be helpful to quantify this across more than one example, like in Figures 4 and 5 for tissue.

Compound 101 is shown in Figure 2 to block barrestin recruitment. I agree this suggests phosphorylation mediated by GRK2/3 but this is not tested. The new antibodies should be good for this so I don't understand why the indirect approach.

The conditions used to inhibit dephosphorylation are not specified, the method only says "phosphatase inhibitors". How do the authors know that low P at 306/7 in white fat is not a result of dephosphorylation during sample preparation? If these sites are GRK2/3 dependent (see above) then does adipose tissue lack this GRK?

Author Response

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

We are grateful to all the reviewers for their thoughtful comments and the efforts they put into reviewing our manuscript. These are highly positive and constructive reviews. Thank you! We have updated our manuscript to include further discussion of several important points (as suggested by reviewers) and addressed reviewer suggestions individually below.

Reviewer #1 (Public Review):

This remarkable and creative study from the Asbury lab examines the extent to which mechanical coupling can coordinate the growth of two microtubules attached to isolated kinetochores. The concept of mechanical coupling in kinetochores was proposed in the mid-1990s and makes sense intuitively (as shown in Fig. 1B). But intuitive concepts still need experimental validation, which this study at long last provides. The experiments described in this paper will serve as a foundation for the transition of an intuitive concept into a robust, quantitative, and validated model.

The introduction cites at least 5 papers that proposed mechanical coupling in kinetochores, as well as 5 theoretical studies on mechanical coupling within microtubule bundles, so it's clear that this manuscript will be of considerable interest to the field. The intro is very well written (as is the manuscript in general), but I recommend that the authors include a brief review of the variable size of k-fibers across species, to help the reader contextualize the problem.

We agree with the reviewer’s suggestion and have added a brief review of variable k-fiber sizes to the Introduction section (lines 30-35).

For example, budding yeast kinetochores are built around a single microtubule (Winey 1995), so mechanical coupling is not relevant for this species.

Indeed, the use of yeast kinetochores to study mechanical coupling is an odd fit, because these structures did not evolve to support such coupling. There is no doubt that yeast kinetochores are useful for demonstrating mechanical coupling and for measuring the stiffnesses necessary to achieve coupling, but I recommend that the authors include a caveat somewhere in the manuscript, perhaps in the place where they discuss their use of simple elastic coupling as compared to viscoelastic coupling or strain-stiffening. It's easy to imagine that kinetochores with large k-fibers might require complex coupling mechanisms, for example.

Even though yeast kinetochores are built around single microtubules, mechanical coupling has still been proposed to help coordinate the dynamics of sister kinetochores in yeast (Gardner et al. 2005, see main text for full reference). We have added this important point to the Introduction section of the manuscript (lines 33-35). The microtubules attached to sister kinetochores are oriented oppositely to one another, in an anti-parallel arrangement that differs from the parallel arrangement we studied here. Nevertheless, it seems likely to us that coordination of anti-parallel microtubule growth between the single microtubules attached to sister kinetochores in yeast relies at least partly on mechanical coupling. One of the many ways we foresee our dual-trap assay being useful in the future is to test how anti-parallel microtubule growth and shortening can be coordinated via mechanical coupling. Of course, since kinetochores can change the dynamics of their attached microtubules (Umbreit et al., 2012, “The Ndc80 kinetochore complex directly modulates microtubule dynamics”), the kinetochores from different species may have also evolved unique mechanisms of modifying microtubule tension-dependent dynamics to achieve coordination of their attached microtubules. Thus far, in vitro reconstitutions using kinetochore assemblies from metazoans have not yet achieved the coupling stability that we routinely achieve with isolated yeast kinetochores. As reconstitutions with kinetochores from other species improve, it will be very interesting to test for species-specific differences in how the kinetochores influence microtubule dynamics and in how effectively they can coordinate microtubules via mechanical coupling.

We note that the (visco)elastic properties of yeast kinetochores, and their relative simplicity compared to other kinetochores, shouldn’t significantly affect our primary experimental results. Yeast kinetochores are relatively small and the force on each bead changes very slowly in our experiments (see Figure S3-1 for examples), so the kinetochore’s change in length over time is very slow and very small. We have added this point to the Methods section of the manuscript (lines 479-484). We agree that mechanical coupling in species with large k-fibers might rely on more complex material properties, such as viscoelasticity or strain-stiffening. In principle, that type of complexity could be incorporated into our dual-trap experiments by altering the simulated linker. We view this as an interesting area for future study.

And is mechanical coupling relevant for holocentric kinetochores like those found in C. elegans?

This is a very interesting question. While holocentric kinetochores do not form k-fiber bundles (O’Toole et al., 2003, “Morphologically distinct microtubule ends in the mitotic centrosome of Caenorhabditis elegans” and Redemann et al., 2017, C. elegans chromosomes connect to centrosomes by anchoring into the spindle network), mechanical coupling could be even more important for them compared to monocentric kinetochores because tip-attached microtubules both near each other AND at opposite ends of the same chromosome must grow at similar enough rates to stay attached to the same chromosome. In C. elegans prometaphase, opposite chromosome ends move towards the same pole as the chromosome itself oscillates, suggesting that microtubule plus ends attached to the same chromosome are growing in the same direction at the same time (Maddox et al., 2004, ““Holo”er than thou: Chromosome segregation and kinetochore function in C. elegans”). Microtubules appear to stop growing or shortening after chromosome alignment is complete (Redemann et al., 2017), at which time the plus ends of kinetochore microtubules are in close proximity to the chromosome surface (O’Toole et al., 2003, Redemann et al., 2017). The tight clustering of kinetochore microtubule tips near the chromosome at metaphase, as well as the coordinated movement of chromosome arms preceding metaphase, suggests a high level of inter-microtubule coordination in the congression leading up to metaphase. We propose this coordination could be achieved by mechanical coupling through the kinetochore proteins on the surface of holocentric chromosomes and through the underlying chromosome itself.

The paper shows considerable rigour in terms of experimental design, statistical analysis, and presentation of results. My only comment on this topic relates to the bandwidth of the dual-trap assay, which I recommend describing in the main text in addition to the methods. For example, the authors note that the stage position is updated at 50 Hz. The authors should clearly explain that this bandwidth is sufficiently fast relative to microtubule growth speeds.

Thank you for this suggestion. We have added to the Results section (lines 131-133) that updating the stage position at 50 Hz is sufficient to maintain the desired force. We also modified the Methods section (lines 488-491) to clarify that the stage position is sampled at 200 Hz, which is more than sufficient to accurately show the growth variability present in dual-trap experiments.

After describing their measurements, the authors use Monte Carlo simulations to show that pauses are essential to a quantitative explanation of their coupling data. Apparently, there is a history of theoretical approaches to coupling, as the introduction cites 5 theoretical studies. I would have appreciated it if the authors had engaged with this literature in the Results section, e.g. by describing which previous study most closely resembles their own and/or comparing and contrasting their approach with the previous work.

Thank you for this excellent suggestion. We have added a brief comparison of our work to previous theoretical studies examining the role of mechanical coupling in k-fiber coordination to the Results section (lines 179-185).

Overall, this paper is rigorous, creative, and thought-provoking. The unique experimental approach developed by the Asbury lab shows great promise, and I very much look forward to future iterations.

Reviewer #2 (Public Review):

Leeds et al. investigated the role of mechanical coupling in coordinating the growth kinetics of microtubules in kinetochore-fibers (k-fibers). The authors developed a dual optical-trap system to explore how constant load redistributed between a pair of microtubules depending on their growth state coordinates their growth.

The main finding of the paper is that the duration and frequency of pausing events during individual microtubule growth are decreased when tension is applied at their tips via kinetochore particles coupled to optically trapped beads. However, the study does not offer any insight into the possible mechanism behind this dependency. For example, it is not clear whether this is a specific property of the kinetochore particles that were used in this experiment, whether it could be attributed to specific proteins in these particles, or if this could potentially be an inherent property of the microtubules themselves.

We agree that the experiments described in our work do not distinguish between tension-dependence inherent to the microtubule itself and tension-dependence conferred by the kinetochore. We speculate about reasons why tension might disfavor pausing in paragraph 5 of the discussion (lines 356-366). Given that microtubule growth is suppressed by compression without the presence of kinetochores or other microtubule-associated proteins (Dogterom & Yurke, 1997, Janson et al., 2003, see main text for full reference), it seems plausible to us that tension-dependent dynamics, including pausing behaviors, might be inherent to microtubules. However, experiments with different tension-bearing plus-end couplers will be required to test this idea rigorously. We view this as an interesting area for future study.

The authors simulate the coordination between two microtubules and show that by using the parameters of pausing and variability in growth rates both measured experimentally they can explain coordination between two microtubules measured in their experiments. This is a convincing result, but k-fibers typically have many more microtubules, and it seems important to understand how the ability to coordinate growth by this mechanism scales with the number of microtubules. It is not obvious whether this mechanism could explain the coordination of more than two microtubules.

We wholeheartedly agree, it is of vital importance to understand how the coordination of growth via mechanical coupling scales with the number of microtubules. Indeed, we have already begun studying simulations of bundles of ten to twenty microtubules based on the pausing model developed in this paper. Simulated microtubule tips appear significantly limited when linked by mechanical couplers of similar stiffnesses to those used in the dual-trap assay, supporting the idea that mechanical coupling may be able to explain much of the coordination between microtubules in growing k-fiber bundles. We hope to use these simulations to continue exploring the degree to which mechanical coupling can coordinate k-fiber microtubules in future publications.

The range of stiffnesses chosen to simulate the microtubule coupling allows linkers to stretch hundreds of nanometers linearly. However, most proteins including those at kinetochore must have finite size and therefore should behave more like worm-like chains rather than linear springs. This means they may appear soft for small elongations, but the force would increase rapidly once the length gets close to the contour length. How this more realistic description of mechanics might affect the conclusions of the work is not clear.

While the worm-like chain is likely a better model for individual linker molecules, deformation of the underlying centromeric chromatin is also likely to be important, with viscoelastic properties that are still poorly understood. Rather than using a complicated (viscoelastic or worm-like-chain-based) model with many unconstrained parameters, we felt a simple model with a single stiffness parameter to characterize the coupling material was a better starting point, allowing a straightforward comparison between stiffer and softer coupling. In future work, simulations could be used to study the effects of strain-stiffening and viscoelasticity and ask if these effects might further improve (or degrade) the efficacy of mechanical coupling for coordinating kinetochore microtubules.

The novel dual-bead assay is interesting. However, it only provides virtual coupling between two otherwise independently growing microtubules. Since the growth of one affects the growth of the other only via software, it is unclear whether the same insight can be gained from the single-bead setup, for example, by moving the bead at a constant speed and monitoring how microtubule growth adjusts to the fixed speed. The advantages of the double-bead setup could have been demonstrated better.

Thank you for your suggestion to clarify the advantages of our dual-trap approach compared to single-trap experiments. We have added a paragraph to the Discussion section (lines 315-327) to explain the following points: In a real k-fiber bundle, each microtubule can dynamically adjust its growth speed to the current force being applied. In the same way, the dual-trap assay allows us to examine how both leading and lagging tips dynamically adjust to the other’s growth speed simultaneously. In addition, in our dual-trap assay each microtubule in the pair is grown at the same time relative to preparing the slide and comes from an identical batch of kinetochore-bead and tubulin-containing growth buffer. Any differences in growth speeds between paired microtubules can be attributed to intrinsic microtubule variability, rather than prep-to-prep or sample-to-sample differences in microtubule dynamics.

Reviewer #3 (Public Review):

Leeds et al. employ elegant in vitro experiments and sophisticated numerical modeling to investigate the ability of mechanical coupling to coordinate the growth of individual microtubules within microtubule bundles, specifically k-fibers. While individual microtubules naturally polymerize at varying rates, their growth must be tightly regulated to function as a cohesive unit during chromosome segregation. Although this coordination could potentially be achieved biochemically through selective binding of polymerases and depolymerases, the authors demonstrate, using a novel dual laser trap assay, that mechanical coupling alone can also coordinate the growth of in vitro microtubule pairs.

By reanalyzing recordings of single microtubules growing under constant force (data from their own previous work), the authors investigate the stochastic kinetics of pausing and show that pausing is suppressed by tension. Using a constant shared load, the authors then show that filament growth is tightly coordinated when pairs of microtubules are mechanically coupled by a material with sufficient stiffness. In addition, the authors develop a theoretical model to describe both the natural variability and force dependence of growth, using no freely adjustable parameters. Simulations based on this model, which accounts for stochastic force-dependent pausing and intrinsic variability in microtubule growth rate, fit the dual-trap data well.

Overall, this study illuminates the potential of mechanical coupling in coordinating microtubule growth and offers a framework for modeling k-fibers under shared loads. The research exhibits meticulous technical rigor and is presented with exceptional clarity. It provides compelling evidence that a minimal, reconstituted biological system can exhibit complex behavior. As it currently stands, the paper is highly informative and valuable to the field.

To provide further clarity regarding the implications of their study, the authors may wish to address the following points in more detail:

• Considering the authors' understanding of the quantitative relationship between forces, microtubule growth, and coordination, is the dual trap assay necessary to demonstrate this coordination? What advantages does the (semi)experimental system offer compared to a purely in silico treatment?

Thank you for your suggestion to explain the advantages of our dual-trap approach compared to simulations based on previous recordings of individual microtubules growing under tension. We have added a paragraph about this to the Discussion section (lines 315-327). Previously we knew that a shared load should theoretically tend to coordinate a growing microtubule pair, but we did not know how well, nor did we know the degree of variability that would need to be overcome to achieve coordination. Moreover, there are myriad ways one could model the variability and force dependence in microtubule growth, but not all of them can successfully explain the tip separations we now measure between real microtubule pairs. For instance, our non-pausing model, although entirely derived from force-clamp data, had too much variability and too little coordination between microtubule pairs when we compared simulation results to our dual-trap measurements. Thus, the dual-trap assay allows us to test our assumptions about how variability in microtubule growth arises and how mechanical coupling affects it using real microtubules. Reviewer 2 likewise asked about the advantages of the dual-trap approach relative to single-trap experiments, and we suggest also examining our response to their comment above.

• What are the limitations of studying a system comprising only two individual microtubules? How might the presence of crosslinkers, which are typically present in vivo between microtubules, influence their behavior in this context?

This is a very interesting question. K-fiber microtubules in many organisms are subject to forces along their lattices from crosslinkers that attach them to each other and to other microtubules outside the k-fiber. Bridging fibers, for example, are pushed apart at the spindle equator by kinesin motors like Eg5, and are thought to maintain tension on k-fiber microtubule tips by sliding them towards the pole (Vukusic et al., 2017, “Microtubule Sliding within the Bridging Fiber Pushes Kinetochore Fibers Apart to Segregate Chromosomes"). Passive crosslinkers can also produce diffusion-like forces that drive microtubules to move relative to one another (although to our knowledge this has only been demonstrated with antiparallel microtubules—see Braun et al., 2017, “Changes in microtubule overlap length regulate kinesin-14-driven microtubule sliding”). Testing how these various lattice-based forces might affect k-fiber coordination is of great interest to us, but it is not easy to envision how it could be done in our dual-trap setup, where the two coupled microtubules only interact through mechanical forces and are biochemically isolated from one another (in separate assay chambers). Perhaps a clever new assay could be devised in the future to study the role of crosslinkers in combination with mechanical coupling on the coordination of growing microtubules in parallel.

• How dependent are the results on the chosen segmentation algorithm? Can the distributions of pause and run durations truly be fitted by "simple" Gaussians, as indicated in Figure S5-2? Given the inherent limitations in accurately measuring short durations and the application of threshold durations, it is likely that the first bins in the histograms underestimate events. Cumulative plots could potentially address this issue.

The qualitative trends of tension suppressing pause entrance and promoting pause exit seemed to be insensitive to the choices we made in our segmentation algorithm. We have added a paragraph to the Methods section (lines 558-569) to explain how other choices we tried (a smoothing window of 5 s compared to 2 s and a minimum event duration of 0.01 s compared to 1 s) had only mild effects on the measured force sensitivities but did not affect their signs. This suggests that while imposing a threshold duration almost certainly underestimates the number of shorter events, it does not substantially affect our overall conclusion that tension reduces the rate of pause entry, accelerates pause exit, and speeds assembly during the ‘runs’ between pauses.

For segmenting each position-vs-time record into pause and run intervals, we fit the velocity distribution for each individual recording with a mixture of Gaussians. The distributions from some recordings fit quite well to a sum of Gaussians, while others did not fit as well. However, we found that the exact threshold used to separate runs from pauses (typically between 2 and 4 nm/s) had a surprisingly small effect on what the algorithm differentiated as a pause or a run. The segmentation algorithm and its performance on every record we analyzed can be directly viewed by downloading and running our force-clamp viewer, publicly available at https://doi.org/10.5061/dryad.6djh9w16v.

Reviewer #2 (Recommendations For The Authors):

In Figure 3a it would be helpful to see the traces of forces applied to individual microtubules. This would help to understand both, how the force is distributed between individual microtubules depending on their dynamic state and also to see the fluctuations of individual forces.

We completely agree that understanding how force is distributed between microtubules in our dual-trap assay is both interesting and of great value. Although we decided not to include force vs time traces in the main figures, please refer to Figure S3-1, which shows the force-vs-time curves corresponding to the example position-vs-time traces displayed in Figure 3a, plus examples from two additional microtubule pairs.

eLife assessment

In this technically advanced and important piece of work, the authors study the coordination of microtubule growth in kinetochore fibers using force spectroscopy and numerical simulations. With compelling evidence the authors address the question of how microtubules, which naturally exhibit variable growth rates, can coordinate their behavior by mechanical coupling so as to function as a single unit in generating forces during chromosome segregation.

Reviewer #1 (Public Review):

This remarkable and creative study from the Asbury lab examines the extent to which mechanical coupling can coordinate the growth of two microtubules attached to isolated kinetochores. The concept of mechanical coupling in kinetochores was proposed in the mid-1990s and makes sense intuitively (as shown in Fig. 1B). But intuitive concepts still need experimental validation, which this study at long last provides. The experiments described in this paper will serve as a foundation for the transition of an intuitive concept into a robust, quantitative, and validated model.

The introduction cites at least 5 papers that proposed mechanical coupling in kinetochores, as well as 5 theoretical studies on mechanical coupling within microtubule bundles, so it's clear that this manuscript will be of considerable interest to the field. The intro is very well written (as is the manuscript in general), but I recommend that the authors include a brief review of the variable size of k-fibers across species, to help the reader contextualize the problem. For example, budding yeast kinetochores are built around a single microtubule (Winey 1995), so mechanical coupling is not relevant for this species.

Indeed, the use of yeast kinetochores to study mechanical coupling is an odd fit, because these structures did not evolve to support such coupling. There is no doubt that yeast kinetochores are useful for demonstrating mechanical coupling and for measuring the stiffnesses necessary to achieve coupling, but I recommend that the authors include a caveat somewhere in the manuscript, perhaps in the place where they discuss their use of simple elastic coupling as compared to viscoelastic coupling or strain-stiffening. It's easy to imagine that kinetochores with large k-fibers might require complex coupling mechanisms, for example. And is mechanical coupling relevant for holocentric kinetochores like those found in C. elegans?

The paper shows considerable rigour in terms of experimental design, statistical analysis, and presentation of results. My only comment on this topic relates to the bandwidth of the dual-trap assay, which I recommend describing in the main text in addition to the methods. For example, the authors note that the stage position is updated at 50 Hz. The authors should clearly explain that this bandwidth is sufficiently fast relative to microtubule growth speeds.

After describing their measurements, the authors use Monte Carlo simulations to show that pauses are essential to a quantitative explanation of their coupling data. Apparently, there is a history of theoretical approaches to coupling, as the introduction cites 5 theoretical studies.

Overall, this paper is rigorous, creative, and thought-provoking. The unique experimental approach developed by the Asbury lab shows great promise, and I very much look forward to future iterations.

Reviewer #2 (Public Review):

Leeds et al. employ elegant in vitro experiments and sophisticated numerical modeling to investigate the ability of mechanical coupling to coordinate the growth of individual microtubules within microtubule bundles, specifically k-fibers. While individual microtubules naturally polymerize at varying rates, their growth must be tightly regulated to function as a cohesive unit during chromosome segregation. Although this coordination could potentially be achieved biochemically through selective binding of polymerases and depolymerases, the authors demonstrate, using a novel dual laser trap assay, that mechanical coupling alone can also coordinate the growth of in vitro microtubule pairs.

By reanalyzing recordings of single microtubules growing under constant force (data from their own previous work), the authors investigate the stochastic kinetics of pausing and show that pausing is suppressed by tension. Using a constant shared load, the authors then show that filament growth is tightly coordinated when pairs of microtubules are mechanically coupled by a material with sufficient stiffness. In addition, the authors develop a theoretical model to describe both the natural variability and force dependence of growth, using no freely adjustable parameters. Simulations based on this model, which accounts for stochastic force-dependent pausing and intrinsic variability in microtubule growth rate, fit the dual-trap data well.

Overall, this study illuminates the potential of mechanical coupling in coordinating microtubule growth and offers a framework for modeling k-fibers under shared loads. The research exhibits meticulous technical rigor and is presented with exceptional clarity. It provides compelling evidence that a minimal, reconstituted biological system can exhibit complex behavior. As it currently stands, the paper is highly informative and valuable to the field.

Author Response

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

Reviewer #1 (Public Review):

The paper offers some potentially interesting insight into the allosteric communication pathways of the CTFR protein. A mutation to this protein can cause cystic fibrosis and both synthetic and endogenous ligands exert allosteric control of the function of this pivotal enzyme. The current study utilizes Gaussian Network Models (GNMs) of various substrate and mutational states of CFTR to quantify and characterize the role of individual residues in contributing to two main quantities that the authors deem important for allostery: transfer entropy (TE) and cross correlation. I found the TE of the Apo system and the corresponding statistical analysis particularly compelling. I found it difficult, however, to assess the limitations of the chosen model (GNM) and thus the degree of confidence I should have in the results. This mainly stems from a lack of a proposed mechanism by which allostery is achieved in the protein. Proposing a mechanism and presenting logical alternatives in the introduction would greatly benefit this manuscript. It would also allow the authors to place the allosteric mechanism of this protein in the broader context of protein allostery.

As detailed below, we went to great lengths to address these concerns, with an emphasis on the limitations of the model and a proposed mechanism. These revisions should hopefully warrant a re-evaluation of our manuscript.

Reviewer #1 (Recommendations For The Authors):

1. It would greatly benefit the paper to state a proposed mechanism by which allostery is achieved in this protein. Is this through ensemble selection, ensemble induction, or a purely dynamic mechanism? What is the rationale for choosing the proposed mechanism and what are reasonable alternative mechanisms? How does this mechanism fit in the broader context of protein allostery?

Following this comment, we added a VERY extensive description of the proposed mechanism by which allostery is achieved in CFTR and present the rationale for choosing this mechanism (lines 445-97 and Figure 7). Briefly, based on previous experimental results and our results we propose that no single model can explain allostery in CFTR, and that its allosteric mechanism is a combination of induced fit, ensemble selection, and a dynamic mechanism.

1. With a proposed mechanism in place, the choice of a GNM to investigate the mechanism and eliminate alternative mechanisms should be rationalized.

The rational for choosing GNM (and ANM-LD) to study the proposed mechanism is now given in lines 498-510. Please note however, that as mentioned in the response to point 1 (and detailed in lines 445-97), the choice of allosteric mechanism, and ruling out other alternatives was not based solely on GNM and ANM-LD, but also on previous experimental results.

1. A discussion of the strengths and limitations of the GNM are pivotal to understanding the limitations of the results shown. How sensitive are the results to specific details of the model(s)?

a. A discussion of the strengths and limitations of the GNM have been added to the introduction. Please see lines 107-122.

b. Sensitivity of the results to the specific details of GNM:

GNM uses two parameters: the force constant of harmonic interactions and the cutoff distance within which the existence of the interactions is considered. The force constant is uniform for all interactions and is taken as unity. Its value affects only the absolute values of the fluctuations (i.e., their scale) but not their distribution. As we are only looking at fluctuations in relative terms our results are insensitive to its value. GNM uses a cutoff distance of 7-10 Å in which interactions are considered (10 Å used in this study). To test the sensitivity of the results to the cutoff distance we repeated the calculations using 7 Å. As now discussed in lines 170-73 and shown in Figure S2 the results remained largely unchanged.

c. Sensitivity of the results to the specific details of TE: To identify cause-and-effect relations TE introduces a time delay (τ) between the movement of residues. The choice of τ is important: when τ is too small, only local cause-and-effect relations (between adjacent amino acids) will be revealed. if τ is too big, few (if any) cause-and-effect relations will manifest. This is analogous to the effects of a stone throne into a lake: look too soon, before the stone hits the water, and you’ll see no ripples. Look too late, the ripples will have already subsided. In a previous work (PMID 32320672), we studied in detail the effects of choosing different τ values and found that an optimal value of τ which maximizes the degree of collectivities of net TE values is in most cases 3× τopt (τopt is the time window in which the total TE of residues is maximized). Details of how τ was chosen were added to the methods section.

In general, the limitations of the chosen model(s) is difficult to determine from the current manuscript because it is devoid of details of the model. While I understand that GNMs have been widely used to study protein systems, the specifics of the model are central to the current work and thus should be provided somewhere in the manuscript.

a. As mentioned in our response above, the limitations of GNM are now presented (lines 107-122).

b. The specifics of the model are now given in more detail in the methods section.

c. In addition, as mentioned above, the results are largely independent of the values of the model’s parameters.

b. Would changing the force constants to a more anisotropic model qualitatively change the results?

a. GNM assumes isotropic fluctuations, and the calculations are based on this assumption. Therefore, GNM is inherently an isotropic model.

b. Importantly, we complement the GNM-TE calculations with ANM-LD simulations, which predict the normal modes in 3D using an anisotropic network model.

1. How repeatable is the difference between no ATP bound and ATP bound CFTR? I worry that the differences in TE in Figures 1 and 3A are mainly due to two different crystallization conditions. Is there evidence that two different structures of the same protein in the same ligand state lead to small changes in TE?

To address this concern, we repeated the calculations using the structures of the ATP-free and bound forms of zebrafish CFTR. As now explained in text (lines 298-303) and shown in Figure S8 the effects of ATP are highly repeatable.

1. Collective modes - why should we expect allostery to be in the most collective modes? Let alone the 10 most? Why not do a mode by mode analysis? Why, for example, were two modes removed page 9 first full paragraph?

a. Collective modes: We have erroneously referred to the slow modes as collective modes. This has now been corrected throughout the manuscript.

b. Let alone the 10 most?

c. why should we expect allostery to be in the most collective modes? Residues that are allosterically coupled are expected to display correlated motions. The slow modes (formerly referred to as “collective modes”) are generally the most collective ones, i.e., display the greatest degree of concerted motions. We therefore expect these modes to contain the allosteric information.

d. Furthermore, as now explained in the text (lines 163-69) and in Figure S1 the Eigenvalue decays of ATP-free and -bound CFTR demonstrate that the 10 slowest GNM modes sufficiently represent the entire dynamic spectrum (the distribution converges after the 10th slow mode).

e. Why not do a mode by mode analysis? It is entirely possible to do a mode-by-mode analysis. However, our view is that the allosteric dynamics of a protein is best represented by an ensemble of modes, rather than by individual ones. We found (as detailed here PMID 32320672) that it is more informative to first use the complete set of modes that encompasses the dynamics (the 10 slowest modes in our case) and then gradually remove the dominant modes.

f. As explained in text (lines 254-7) and more elaborately in our previous work (PMID 35644497), the large amplitude of the slowest modes may hide the presence of “faster” modes that may nevertheless be of functional importance. Removal of the 1-2 slowest modes often helps reveal such modes.

g. Why, for example, were two modes removed page 9 first full paragraph? As explained for the ATP-free form (lines 257-60), removal of these two slowest modes allowed the “surfacing” of dynamic features which were hidden before. We propose that these dynamic features are functionally relevant (see lines 304-19). Removal of other modes did not provide additional insight.

Minor issues:<br /> 1. Statements like "see shortly below" should be made more specific (or removed completely).

Corrected as suggested

1. "interfered" should be "inferred" page 10 middle of the first full paragraph

Corrected as suggested

1. End parenthesis after "(for an excellent explanation about the correlation between TE and allostery see (41)." Page 4 middle of first full paragraph

Corrected as suggested

Reviewer #2 (Public Review):

In this study, the authors used ANM-LD and GNM-based Transfer Entropy to investigate the allosteric communications network of CFTR. The modeling results are validated with experimental observations. Key residues were identified as pivotal allosteric sources and transducers and may account for disease mutations.

The paper is well written and the results are significant for understanding CFTR biology.

Reviewer #2 (Recommendations For The Authors):

Technical comments:

p4 Please explain how is the time delay parameter tau chosen (ie. three times the optimum tau value...)? It seems this unknown time should depend on the separation between i and j. Is the TE result sensitive to the choice of tau? How does the choice of cutoff distance of GNM affect the TE result?

a. The choice of τ is important: when τ is too small, only local cause-and-effect relations (between adjacent amino acids) will be revealed. if τ is too big, few (if any) cause-and-effect relations will manifest. This is analogous to the effects of a stone throne into a lake: look too soon, before the stone hits the water, and you’ll see no ripples. Look too late, the ripples will have already subsided. In a previous work (PMID 32320672), we studied in detail the effects of choosing different τ values and found that an optimal value of τ which maximizes the degree of collectivities of net TE values is in most cases 3× τopt (τopt is the time window in which the total TE of residues is maximized). Details of how τ was chosen were added to the methods section.

b. To test the sensitivity of the results to the cutoff distance we repeated the calculations using 7 Å. As now discussed in lines 170-173 and shown in Figure S2 the results remained largely unchanged.

It would be nice to directly validate the causal prediction by GNM-based TE. For example, is it in agreement with direct causal observation of MD simulation? If the dimer is too big for MD, perhaps MD is more feasible for the monomer (NBD1+TMD1).

a. The causality we determined using GNM-based TE is in good agreement with conclusions drawn from single channel electrophysiological recordings and rate-equilibrium free-energy relationship analysis (Sorum et al; Cell 2015, and see lines 8691, and 364-70).

b. To the best of our knowledge, causality relations in CFTR are yet to be determined by MD simulations (This is likely because the protein is too big and the conformational changes are very slow). We cannot therefore compare the causality.

c. Conducting MD simulations on half of CFTR (NBD1+TMD1) is not likely to be very informative: the ATP binding sites are formed at the interface of NBD1 and NBD2, and the ion translocation pathway at the interface of the TMDs.

p5 How are the TE peak positions different from other key positions as predicted by GNM, such as the hinge positions with minimal mobility of the dominant GNM modes?

Following this comment, we compared the positions of the GNM-TE peaks and the hinge positions as determined by GNM. As now discussed in lines 173-178 and shown in Figure S3 we observed partial overlap which was nevertheless statistically significant (Figure S3).

p7 How to select the 10 most collective GNM modes? Why not use the 10 slowest GNM modes?

We have actually used the 10 slowest GNM modes, but in an attempt to cater for the non-specialist reader, we referred to them as the most collective ones. This has now been corrected throughout the manuscript and the terminology that is now used is “10 slowest modes”

p9 There exist other ANM-based methods for conformational transition modeling. So it would be nice to discuss their similarity and differences from ANM-LD, and compare their predictions.

Alternative ANM (and other elastic network models) -based methods are now mentioned and referenced in lines 144-50. These methods are different from ANM-LD in the details of the all atom simulations and in their integration with the elastic network model. It is not trivial to reanalyze CFTR’s allostery using these methods and is beyond the scope of this work.

Regarding the prediction of order of residue motions, can one directly observe such order by superimposing some intermediate conformation of ANM-LD with the initial and end structure?

This would indeed be very attractive approach to visualize the order of events and following this comment we have tried to do just so. Unfortunately, we failed: Superimposing pairs of frames provided little insight, and we therefore compiled a video comprising all frames, or videos based on averages of several time delayed frames. We found that it is next to impossible to discern (using the naked eye) the directionality of the fluctuations and follow the order of conformational changes. Therefore, at this point, we have abandoned this endeavor.

Reviewer #3 (Public Review):

This study of CFTR, its mutants, dynamics, and effects of ATP binding, and drug binding is well written and highly informative. They have employed coarse-grained dynamics that help to interpret the dynamics in useful and highly informative ways. Overall the paper is highly informative and a pleasure to read.

The investigation of the effects of drugs is particularly interesting, but perhaps not fully formed.

This is a remarkably thorough computational investigation of the mechanics of CFTR, its mutants, and ATP binding and drug binding. It applies some novel appropriate methods to learn much about structure's allostery and the effects of drug bindings. It is, overall, an interesting and well written paper.

There are only two main questions I would like to ask about this quite thorough study.

Reviewer #3 (Recommendations For The Authors):

1. Is it possible that the relatively large exothermic ATP hydrolysis itself exerts a force that causes the observed transitions? Jernigan and others have explored this effect for GroEL and some other structures. The effects of ATP binding and hydrolysis are likely often confused, and both are likely to be important.

It is well established by many studies that ATP hydrolysis is not required to drive the conformational changes or to open the channel, and that ATP binding per-se is sufficient (e.g., We have clarified this point in lines 521-30.

1. For the case of ivacaftor, would a comparison of the motion's directions show that ivacaftor might be compensating simply by its mass being located in a site to compensate for the mass changes from the mutations (ENMs with masses needed to address this). We have observed such cases on opposite sides of a hinge.

We do not think that this is the case, from the following reasons:

a. Ivacaftor corrects many gating mutations (e.g., G551D, G178R, S549N, S549R, G551S, G970R, G1244E, S1251N, S1255P, G1349D) which are spread all over the protein. Ivacaftor binds to a single site in CFTR, and it is therefore unlikely that its mass contribution corrects all these diverse mass changes.

b. The residues that comprise the Ivacaftor binding were identified as allosteric “hotspots” in both the ATP-free and -bound forms (Figures 2B, 3B, and 6A), also in the absence of the drug. This indicates that the dynamic traits of this site is intrinsic to it, and that once bound, the drug acts by modulating these dynamics

The Abstract does not repeat some of the more interesting points made in the paper and would benefit from a substantial revision.

Corrected as suggested

There are just a few minor points (just words):

P 3 line 2 of first full ¶: "effects" should be "affects"

Corrected as suggested

P 6 first lilne "per-se" should be "per se"

Corrected as suggested

Further down that page "two set" should be "two sets"

Corrected as suggested

Even further down that same page "testimony" should be "support"

Corrected as suggested

P 10, 5 lines from the bottom "impose that" is awkward

Changed to “define”

Reviewer #1 (Public Review):

The paper offers interesting insight into the allosteric communication pathways of the CTFR protein. A mutation to this protein can cause cystic fibrosis and both synthetic and endogenous ligands exert allosteric control of the function of this pivotal enzyme. The current study utilizes Gaussian Network Models (GNMs) of various substrate and mutational states of CFTR to quantify and characterize the role of individual residues in contributing to two main quantities that the authors deem important for allostery: transfer entropy (TE) and cross correlation. I found the TE of the Apo system and the corresponding statistical analysis particularly compelling. The authors updated the manuscript nicely to include the limitations of the chosen model (GNM) and thus allow the reader to assess the limitations of the results. I appreciated the comprehensive discussion of a proposed mechanism by which allostery is achieved in the protein (though I would have put that in the introduction and had it motivate the choice of methods). This discussion allows the reader to place the allosteric mechanism of this protein in the broader context of protein allostery.

Reviewer #2 (Public Review):

In this study, the authors used ANM-LD and GNM-based Transfer Entropy to investigate the allosteric communications network of CFTR. The modeling results are validated with experimental observations. Key residues were identified as pivotal allosteric sources and transducers and may account for disease mutations.

The paper is well written and the results are significant for understanding CFTR biology.

eLife assessment

This important study advances our knowledge of Drosophila Bonus, the sole ortholog of the mammalian transcriptional regulator Tif1. Solid evidence, both in vivo and in vitro, shows how SUMOylation controls the function of the Bonus protein and what the impact of SUMOylation on the function of Bonus protein in the ovary is.

Reviewer #1 (Public Review):

Summary:<br /> This important study from Godneeva et al. establishes a Drosophila model system for understanding how the activity of Tif1 proteins is modified by SUMO. The authors convincingly show that Bonus, like homologous mammalian Tif1 proteins, is a repressor, and that it interacts with other co-repressors Mi-2/NuRD and SetDB1 in Drosophia ovaries and S2 cells. They also show that Bonus is SUMOylated by Su(var)2-10 on one lysine at its N-terminus to promote its interaction with SetDB1. By combining biochemistry with an elegant reporter gene approach, they show that SUMOylation is important for Bonus interaction with SetDB1, and that this SUMO-dependent interaction triggers high levels of H3K9me3 deposition and gene silencing. While there are still major questions of how SUMO molecularly promotes this process, the authors conducted several experiments that will guide future work. For example, they showed that SUMOylation likely indirectly promotes Bon interaction with SetDB1 because mostly unSUMOylated Bon copurifies with SetDB1. They also show that SUMOylated and unSUMOylated Bon differentially localize within the cell, and preventing Bon SUMOylation alters its subcellular localization. These important experiments disfavor a simple model where SUMO bridges the Bon/SetDB1 interaction and hint at a more complex multi-step assembly process that regulates Bon-dependent transcriptional silencing.

Reviewer #2 (Public Review):

Summary:<br /> The authors analyze the functions and regulation of Bon, the sole Drosophila ortholog of the TIF1 family of mammalian transcriptional regulators. Bon has been implicated in several developmental programs, however the molecular details of its regulation have not been well understood. Here, the authors reveal the requirement of Bon in oogenesis, thus establishing a previously unknown biological function for this protein. Furthermore, careful molecular analysis convincingly established the role of Bon in transcriptional repression. This repressor function requires interactions with the NuRD complex and histone methyltransferase SetDB1, as well as sumoylation of Bon by the E3 SUMO ligase Su(var)2-10. Overall, this work represents a significant advance in our understanding of the functions and regulation of Bon and, more generally, the TIF1 family. Since Bon is the only TIF1 family member in Drosophila, the regulatory mechanisms delineated in this study may represent the prototypical and important modes of regulation of this protein family. The presented data are rigorous and convincing. As discussed below, this study can be strengthened by a demonstration of a direct association of Bon with its target genes, and by analysis of the biological consequences of the K20R mutation.

Strengths:<br /> 1. This study identified the requirement for Bon in oogenesis, a previously unknown function for this protein.<br /> 2. Identified Bon target genes that are normally repressed in the ovary, and showed that the repression mechanism involves the repressive histone modification mark H3K9me3 deposition on at least some targets.<br /> 3. Showed that Bon physically interacts with the components of the NuRD complex and SetDB1. These protein complexes are likely mediating Bon-dependent repression.<br /> 4. Identified Bon sumoylation site (K20) that is conserved in insects. This site is required for repression in a tethering transcriptional reporter assay, and SUMO itself is required for repression and interaction with SetDB1. Interestingly, the K20-mutant Bon is mislocalized in the nucleus in distinct puncta.<br /> 5. Showed that Su(var)2-10 is a SUMO E3 ligase for Bon and that Su(var)2-10 is required for Bon-mediated repression.

Weaknesses:<br /> The study would be strengthened by demonstrating a direct recruitment of Bon to the target genes identified by RNA-seq. - It appears that the authors have attempted such an experiment, but it was not successful due to the current technical limitations, as the authors describe in their rebuttal.

The second area where the manuscript can be improved is to analyze the biological function of the K20R mutant Bonus protein. The molecular data suggest that this residue is important for function, and it would be important to confirm this in vivo. - Fig. 5G indeed shows that the 3KR mutant is deficient in inducing repression, which partially addresses this concern. In the future, it would be interesting to test if the single K20R is similarly deficient, and to analyze any resulting phenotypes.

Reviewer #1 (Public Review):

In their study, Zhou et al. unveil the pivotal role of ULK4 in conjunction with STK36, shedding light on their collective impact on GLI2 phosphorylation and the subsequent activation of the SHH pathway. The research delves deep into the intricate interactions between ULK4 and various components of the SHH pathway within the primary cilium.

The main strength of the study lies in the careful and systematic sequence of logical methods. The authors apply the expression of a range of different deletion and mutation constructs and carry out a comprehensive biochemical study of the consequences of depletion and reintroduction of various components in the context of STK36 and ULK4.

Their findings reveal that ULK4 forms dynamic interactions with a complex composed of STK36 and GLI2. It is proposed that ULK4 acts as a scaffold, facilitating the essential interaction between STK36 and GLI2, thereby driving GLI2 phosphorylation by STK36. Notably, the research reveals that the N-terminal pseudokinase domain of ULK4 binds to Stk36, while the C-terminal regulatory domain of ULK4 interacts with Gli2. Moreover, the study presents compelling evidence for co-localization of ULK4 and STK36 with GLI2 at the ciliary tip within NIH 3T3 cells. Importantly, ULK4 and STK36 mutually rely on each other for their accumulation at this ciliary tip.

This intricate mechanism, orchestrated by ULK4, brings to light the nuanced modulation of the SHH pathway. The research is substantiated by rigorous Co-IP experiments, kinase assays, and confocal imaging localization studies. To unravel the fine details of GLI2 phosphorylation at the primary cilium tip, the authors meticulously employ a diverse array of mutated and wild-type constructs of STK36 and ULK4.

In summary, the studiy provide compelling insights into the intricate regulation of signaling pathways. Zhou et al.'s work on ULK4 and STK36 in the SHH pathway deepen our understanding of these complex processes, offering potential avenues for drug development, particularly in the context of cancer therapeutics.

eLife assessment

This fundamental study substantially advances our understanding of how the pseudokinase ULK4 interacts with an active member of the same kinase subfamily (STK36) to promote GLI phosphorylation and Hedgehog pathway activation. The evidence supporting the proposed mechanism is compelling, with rigorous biochemical assays and state-of-the-art cell based imaging techniques. The work will be of broad interest to cell biologists and biochemists.

Reviewer #2 (Public Review):

The authors provide solid molecular and cellular evidence that ULK4 and STK36 not only interact, but that STK36 is targeted (transported?) to the cilium by ULK4. Their data helps generate a model for ULK4 acting as a scaffold for both STK36 and its substrate, Gli2, which appear to co-localise through mutual binding to ULK4. This makes sense, given the proposed role of most pseuodkinases as non-catalytic signaling hubs. There is also an important mechanistic analysis performed, in which ULK4 phosphorylation in an acidic consensus by STK36 is demonstrated using IP'd STK36 or an inactive 'AA' mutant, which suggests this phosphorylation is direct.

The major strength of the study is the well-executed combination of logical approaches taken, including expression of various deletion and mutation constructs and the careful (but not always quantified in immunoblot) effects of depleting and adding back various components in the context of both STK36 and ULK3, which broadens the potential impact of the work. The biochemical analysis of ULK4 phosphorylation appears to be solid, and the mutational study at a particular pair of phosphorylation sites upstream of an acidic residue (notably T2023) is further strong evidence of a functional interaction between ULK4/STK36. The possibility that ULK4 requires ATP binding for these mechanisms is not approached, though would provide significant insight: for example it would be useful to ask if Lys39 in ULK4 is involved in any of these processes, because this residue is likely important for shaping the ULK4 substrate-binding site as a consequence of ATP binding; this was originally shown in PMID 24107129 and discussed more recently in PMID: 33147475 in the context of the large amount of ULK4 proteomics data released.

The discussion is excellent, and raises numerous important future work in terms of potential transportation mechanisms of this complex. It also explains why the ULK4 pseudokinase domain is linked to an extended C-terminal region. Does AF2 predict any structural motifs in this region that might support binding to Gli2?

A weakness in the study, which is most evident in Figure 1, where Ulk4 siRNA is performed in the NIH3T3 model (and effects on Shh targets and Gli2 phosphorylation assessed), is that we do not know if ULK4 protein is originally present in these cells in order to actually be depleted. Also, we are not informed if the ULK4 siRNA has an effect on the 'rescue' by HA-ULK4; perhaps the HA-ULK4 plasmid is RNAi resistant, or if not, this explains why phosphorylation of Gli2 never reaches zero? Given the important findings of this study, it would be useful for the authors to comment on this, and perhaps discuss if they have tried to evaluate endogenous levels of ULK4 (and Stk36) in these cells using antibody-based approaches, ideally in the presence and absence of Shh. The authors note early on the large number of binding partners identified for ULK4, and siRNA may unwittingly deplete some other proteins that could also be involved in ULK4 transport/stability in their cellular model.

The sequence of ULK4 siRNAs is not included in the materials and methods as far as I can see, though this is corrected in the next version of the manuscript.

Reviewer #3 (Public Review):

In this manuscript, Zhou et al. demonstrate that the pseudokinase ULK4 has an important role in Hedgehog signaling by scaffolding the active kinase Stk36 and the transcription factor Gli2, enabling Gli2 to be phosphorylated and activated.<br /> Through nice biochemistry experiments, they show convincingly that the N-terminal pseudokinase domain of ULK4 binds Stk36 and the C-terminal Heat repeats bind Gli2.

Lastly, they show that upon Sonic Hedgehog signaling, ULK4 localizes to the cilia and is needed to localize Stk36 and Gli2 for proper activation.

This manuscript is very solid and methodically shows the role of ULK4 and STK36 throughout the whole paper, with well controlled experiments. The phosphomimetic and incapable mutations are very convincing as well.<br /> I think this manuscript is strong and stands as is, and there is no need for additional experiments.

Overall, the strengths are the rigor of the methods, and the convincing case they bring for the formation of the ULK4-Gli2-Stk36 complex. There are no weaknesses noted. I think a little additional context for what is being observed in the immunofluorescence might benefit readers who are not familiar with these cell types and structures.

The revised manuscript has improved some of the unclear areas.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

The authors have previously employed micrococcal nuclease tethered to various Mcm subunits to the cut DNA to which the Mcm2-7 double hexamers (DH) bind. Using this assay, they found that Mcm2-7 DH are located on many more sites in the S. cerevisiae genome than previously shown. They then demonstrated that these sites have characteristics consistent with origins of DNA replication, including the presence of ARS consensus sequences, location of very inefficient sites of initiation of DNA replication in vivo, are free of nucleosomes, they contain a G-C skew and they locate to intergenic regions of the genome. The authors suggest, consistent with published single molecule results, that there are many more potential origins in the S. cerevisiae genome than previously annotated.

The results are convincing and are consistent with prior observations. The analysis of the origin associated features is informative.

Reviewer #2 (Public Review):

By mapping the sites of the Mcm2-7 replicative helicase loading across the budding yeast genome using high-resolution chromatin endogenous cleavage or ChEC, Bedalov and colleagues find that these markers for origins of DNA replication are much more broadly distributed than previously appreciated. Interestingly, this is consistent with early reconstituted biochemical studies that showed that the ACS was not essential for helicase loading in vitro (e.g. Remus et al., 2009, PMID: 19896182). To accomplish this, they combined the results of 12 independent assays to gain exceptionally deep coverage of Mcm2-7 binding sites. By comparing these sites to previous studies mapping ssDNA generated during replication initiation, they provide evidence that at least a fraction of the 1600 most robustly Mcm2-7-bound sequences act as origins. A weakness of the paper is that the group-based (as opposed to analyzing individual Mcm2-7 binding sites) nature of the analysis prevents the authors from concluding that all of the 1,600 sites mentioned in the title act as origins. The authors also show that the location of Mcm2-7 location after loading are highly similar in the top 500 binding sites, although the mobile nature of loaded Mcm2-7 double hexamers prevents any conclusions about the location of initial loading. Interestingly, by comparing subsets of the Mcm2-7 binding sites, they find that there is a propensity of at least a subset of these sites to be nucleosome depleted, to overlap with at least a partial match to the ACS sequence (found at all of the most well-characterized budding yeast origins), and a GC-skew. Each of which is a characteristic of previously characterized origins of replication.

Overall, this manuscript greatly broadens the number of sites that are capable of loading Mcm2-7 in budding yeast cells and shows that a subset of these additional sites act as replication origins. Although these sites do have a propensity to include a match to the ACS, these studies suggest that the mechanism of helicase loading in yeast and multicellular organisms is more similar than previously thought.

Reviewer #1 (Recommendations For The Authors):

Specific Comments:

1. The proposal, based on this study, that replication in S. cerevisiae is similar to that in Human cells (mentioned in the abstract, introduction and end of discussion) is not supported by the evidence, either in this paper or elsewhere. The authors suggest that even these inefficient origins are directed by specific sequences that load Mcm2-7 DH, but there is no evidence that this occurs outside a limited clade of budding yeasts and certainly no in human cells. Furthermore, the distribution and efficiency of origins of replication Human cells has not been shown to parallel the findings in this paper. Thus, the conclusion should be removed since it makes a statement that S. cerevisiae and Human cells have similar mechanisms for origin location. This might confuse non-specialists who do not appreciate the subtleties.

The reviewer's concern that we could confuse non-specialists is well-founded. We have made the following changes to emphasize the point that, while a wider distribution of origins makes S phase in yeast more like that in humans, the genome replication programs in the two organisms remain distinctly different:

1) The last sentence of the abstract was changed as follows:

a. These results shed light on recent reports that as many as 15% of replication events initiate outside of known origins, and they reveal S phase in yeast to be surprisingly similar to that in humans.

b. These results shed light on recent reports that as many as 15% of replication events initiate outside of known origins, and this broader distribu5on of replica5on origins suggest that S phase in yeast may be less dis5nct from that in humans than is widely assumed.

1. A sentence in the results was changed as follows:

a. Another characteris5c of known origins that we could use as a criterion to assess the nature of Mcm binding sites is the presence of an ACS.

b. Another characteris5c of known origins in S. cerevisiae (although not in most other organisms) that we could use as a criterion to assess the nature of Mcm binding sites is the presence of an ACS.

1. We changed the last sentence of the Discussion as follows:

a. On the other hand, the sharply focused nature of its replication origins made S phase in yeast appear distinct from that in other organisms. Our discovery that sites of replica5on ini5a5on in yeast are much more widely dispersed than previously believed, with at least 1600 and possibly as many as 5500 origins, emphasizes its continued relevance to understanding genome duplication in humans.

b. On the other hand, the sharply focused nature of its replication origins made S phase in yeast appear dis?nct from that in other organisms. Although by no means elimina5ng this dis5nc5on, our discovery that sites of replication ini5a5on in yeast are much more widely dispersed than previously believed, with at least 1600 and possibly as many as 5500 origins, emphasizes yeast's continued relevance to understanding S phase in humans.

1. The authors discuss in the introduction that origins in S. cerevisiae are equivalent to ARS sequences. Why didn't they ask if the inefficient origins also confer ARS activity? This would be a valuable addition and a very simple experiment.

The inefficient origins are not expected to confer ARS activity, because origins that are not licensed in essentially every G1 will be diluted out by cell division. We confirmed the absence of our inefficiently licensed origins in a data set generated by high throughput sequencing of a genomic library that was selected for origin activity (PMID: 23241746), but we did not note the results of this analysis in our manuscript, because the low complexity of the library used made this negative result uninformative. To clarify this point, we added the bolded clauses to the following sentences in the Introduction and Discussion:

1. Origins vary widely in their efficiency, with some being used in almost every cell cycle while others may be used in only one in one thousand S phases (Boos and Ferreira, 2019), with only the former being capable of supporting plasmid replication in the traditional ARS assay.
2. "Thus, we can detect Mcm complexes that are loaded in as few as 1 in 500 cells (Foss et al., 2021), even though such low affinity Mcm binding sites are not expected to be capable of supporting autonomous replication of a plasmid."

1. While the authors have shown that Mcm2-7 is loaded adjacent to the principal ARS consensus sequence, consistent with biochemical studies on pre-RC assembly, two reports have shown that the Mcm2-7 ChIP is dependent on the B2 element of ARS1, but the ORC ChIP is not, suggesting that Mcm2-7 is loaded there (See Lipford and Bell, Mol. Cell 2007 and Zou and Stillman, Mol. Cell. Biol. 2000).

We have added the following two sentences in the Results section to note these reports:

"Furthermore, in the case of ARS1, two reports have demonstrated a requirement for the B2 element for Mcm loading, though not for Orc binding, suggesting that Orc may bind to the ACS but then load Mcm at the B2 element (Zou and Stillman 2000; Lipford and Bell 2001). This would still leave Mcm loaded downstream of the ACS, but we note this result to emphasize that not all details of Mcm loading in vitro have been definitively established."

**Reviewer #2 (Recommendations For The Authors):>>

Specific points:

1. The authors state "It is notable that the Mcm-ChEC panel of Figure 3A shows no obvious change in Mcm stoichiometry across the entire range, from low abundance, at the bottom, to high abundance, at the top." The ChEC method does not intrinsically measure stoichiometry so this conclusion needs more explanation. The authors appear to be referring to the distribution of Mcm2-7 reads being similar across all origins, but this does not measure how many double hexamers are present at an origin. If the stoichiometry argument is based on a finding that each origin has only a single 60 bp region that is protected by Mcm2-7 (rather than a distribution of 60 bp regions spread across the origin), then the authors should provide more compelling evidence than what is shown in Fig. 3A.

We agree with the reviewer that our conclusion needs more explanation, and we have therefore made the following change, which we believe clarifies the point that we were trying to convey:

We agree with the reviewer that our conclusion needs more explanation, and we have therefore made the following change, which we believe clarifies the point that we were trying to convey:

1. Original version: It is notable that the Mcm-ChEC panel of Figure 3A shows no obvious change in Mcm stoichiometry across the entire range, from low abundance, at the bottom, to high abundance, at the top. This argues against models in which higher replication activity at more active origins reflect the loading of more Mcm double-hexamers at those origins within a single cell.

2. Updated version: It is notable that, when Mcm is present, it is present predominantly as a single double-hexamer (right panel of Figure 3A), and that this remains true across the entire range of abundance shown in Figure 3A. This argues against models in which higher replication activity at more active origins is caused by the loading of more Mcm double-hexamers at those origins within a single cell, since such models predict that multiple Mcm footprints should be more prevalent at the top (high abundance) of the Mcm-ChEC heat map in Figure 3A than at the bottom.

1. The authors state "we estimate that ~1-2 % cells have an Mcm complex loaded at the Mcm binding sites in the eighth cohort (ranks 1401-1600)" but it is not clear how this estimate is calculated. An explanation would help the reader to understand this statement.

We have expanded on our earlier statement to clarify how we arrived at the estimate:

1. Original version: Based on our previous analysis of MCM occupancy (Foss et al., 2021), which showed that approximately 90% cells have an MCM complex loaded at one of the most active known replication origins, we estimate that ~1-2 % cells have an Mcm complex loaded at the Mcm binding sites in the eighth cohort (ranks 1401-1600).

2. Updated version: We have previously used Southern blodng to demonstrate that approximately 90% of the DNA at one of the most active known origins (ARS1103) is cut by Mcm-MNase (Foss et al., 2021), and to thereby infer that 90% of cells have a doublehelicase loaded at this origin. Using this as a benchmark, we estimate that ~1-2 % cells have an Mcm complex loaded at the Mcm binding sites in the eighth cohort (ranks 14011600).

1. Although there is evidence that some subset of the CMBS sites exhibit nucleosome depletion, an ACS, and a GCskew, the authors should do a better job of making the reader aware that it is likely that a decreasing percentage of the individual origins in a group include these characteristic and that this is a likely factor explaining the increasingly rare use of these sites as Mcm2-7 loading sites and origins of replication.

We have added the following text to the Discussion to draw the reader's attention to this possibility, while also noting that we do not believe it to be a major factor in the increasingly rare use of sites within the first 5,500 CMBSs as replication origins:

Furthermore, it is possible that, as one moves to lower abundance groups of CMBSs within the most abundant 5500 sites, a smaller fraction of sites within those groups have any origin function at all. If one takes this model to the extreme, it would suggest that the continuous decline in replication activity seen in Figure 2B between the group comprised of ranks 1-200 and that comprised of ranks 1401-1600 reflects an ever increasing fraction of CMBSs with zero origin activity. At the other extreme, the decline in replication activity could be interpreted within a framework in which 100% of CMBSs in each group function as replication origins, but that their replication activity declines with rank, perhaps because continuously decreasing fractions of cells in the population contain a single double-hexamer. While the truth presumably lies between these two extremes, we favor a model that tilts toward the latter view, because of the abruptness of the transition that appears around rank 5,000 in (1) nucleosomal architecture (Figures 3A, 3B and S3); (2) intergenic versus genic localization and transcription levels (Figure 4A); (3) EACS position weight matrix scores (Figure 5B); and (4) GC skew (Figure 6B). By these criteria, the CMBSs below rank 5000 appear relatively homogeneous, while still showing a gradual decline in replication activity with MCM abundance within the range of detection (11600). Our assumption is that the qualitative homogeneity is more consistent with a quantitative, but not qualitative, change in CMBSs with declining MCM abundance among the top 5000 CMBSs.

1. The argument that there are as many as 5,500 origins is not well justified. Similarly, the evidence that there are even 1,600 origins is not compelling. As the authors state, to see the peaks observed in the various analyses (ssDNA association, nucleosome depletion, etc.) of the increasingly less populated CMBSs (e.g. those with fewer ChEC reads), only a small subset of the CMBS are likely to have a given characteristic. Given that the loading of a Mcm2-7 double hexamer makes any site a potential origin, it would be more appropriate to say that there could be as many as 5,500 potential origins but many if not most are unlikely to ever direct initiation.

The reviewer is correct that, because many of our analyses rely on group averages rather than individual measurements, we are oien unable to make statements that can be applied to every member of a group. We had tried to emphasize this point in our original manuscript with the following two sentences (in bold), which were in the Results and Discussion, respectively:

1. First, clear peaks of ssDNA signal extend down to the eighth cohort (brown line), which corresponds to CMBSs ranked 1401-1600. Of course, this does not imply that all of these sites function as replication origins, and nor does it imply that no sites below that rank do so, since we have reached the limits of detection of this ssDNA-based assay. Nonetheless, it suggests that replication activity is common among sites extending at least down to rank 1600.

2. Of course, we do not conclude that all CMBSs with ranks lower than 5500 function as replication origins, nor that none with ranks above 5500 do so, but only that the number of replication origins is likely to be approximately an order of magnitude higher than widely believed.

We have now added a third sentence to further underline this point (in bold):

Second, by averaging signals of replication from multiple Mcm binding sites, we were able to extract weak signals of replication. This is due to the fact that noise, which is randomly distributed, will tend to cancel itself out, while signals of replication will consistently augment the signal at the midpoint of the origin (Figure 2). An inevitable shortcoming to this approach is that it precludes analysis of specific sites; in other words, not every member of the group will share the average characteristic of that group.

A separate issue that this touches on is the distinction between a replication origin and a site at which Mcm2-7 has been loaded. While it strikes us as unlikely that a loaded Mcm complex would be completely incalcitrant to activation, it is a formal possibility. To alert the reader to this issue, we have added the following clause, in bold, to the Abstract, and we have also added the sentence below that to the Discussion:

We conclude that, if sites at which Mcm double-hexamers are loaded can function as replication origins, then DNA replication origins are at least 3-fold more abundant than previously assumed, and we suggest that replication may occasionally initiate in essentially every intergenic region.

Finally, it is important to note that, in equating Mcm binding sites with potential replication origins, we are assuming that if an Mcm double-hexamer is loaded onto the DNA, then it is conceivable that that complex can be activated.

1. The author's discussion of the relationship between Mcm2-7 location relative to the ACS and the mechanism of of Mcm2-7 loading does not consider that Mcm2-7 double hexamers can slide on DNA after loading (for example, Remus et al., 2009 PMID: 19896182). Thus, the authors are not looking at sites of loading only the distribution of Mcm2-7 molecules after loading. In addition, biochemical experiments do not predict a particular Mcm2-7 position relative to the ACS. Indeed, at ARS1, one would predict that the close proximity of the second weak match to the ACS (the B2 element) to the primary ACS would lead the Mcm2-7 double hexamer being initially formed at a site overlapping the ARS1 ACS. It is much more likely that the explanation for the distribution of Mcm2-7 locations relative to the ACS is that the ORC-bound ACS and the nucleosomes immediately flanking the origin prevents Mcm2-7 from occupying the right-side of the origin as illustrated in Fig. 5D.

We have tried to emphasize this point more clearly. In our original manuscript, we had brought up the possibility of Mcms sliding after being loaded in the following context (see bolded clause):

Specifically, in 112 out of 146 instances in which a peak of Mcm signal was within 100 base pairs of a known ACS, that peak was downstream of the ACS. The 34 exceptions may reflect (1) incorrect identification of the ACS; (2) incorrect inference of the directionality of the site; or (3) sliding of the Mcm complex after it has been loaded.

We have now added the following to further emphasize the point:

In interpreting the results above, it is important to remember that the locations at which we are detecting Mcm complexes by ChEC do not necessarily reflect the locations at which those complexes were loaded, since Mcm double-hexamers can slide along the DNA after loading (Remus et al. 2009; Gros et al. 2015; Foss et al. 2019).

We have also softened the following conclusion by changing "confirmation of" to "support for":

"...our results...provide in vivo support for in vitro predictions of the directionality of Mcm loading by Orc..."

There are missing references in several places:

1. "For example, 15 of the 56 genes that contained a high abundance site have been implicated in meiosis and sporulation and are not expressed during vegetative growth (~5 out of 56 expected from random sampling), consistent with previous observations (Mori and Shirahige, 2007)." Should include Blitzblau et al., 2012 (PMC3355065) which showed that Mcm2-7 loading was impacted by differences in meiotic and mitotic transcription.

2. "In contrast to the low abundance sites, the most abundant 500 sites showed a preference for convergent over divergent transcription (left of vertical dotted line in Figure 4B), in agreement with a previous report (Li et al., 2014)." This preference was first pointed out in MacAlpine and Bell, 2005 (PMID: 15868424).

3. "This sequence is recognized by the Origin Recognition Complex (Orc), a 6-protein complex that loads MCM (Broach et al., 1983; Deshpande and Newlon, 1992; Eaton et al., 2010; Kearsey, 1984; Newlon and Theis, 1993; Singh and Krishnamachari, 2016; Srienc et al., 1985)." This list should include a reference to Bell and Stillman, 1992 (PMID: 1579162), which first described ORC and showed that it recognized the ACS. It would also be more helpful to the reviewer to distinguish the references that identified that ACS from those concerning ORC binding to it.

We thank the reviewer for pointing out these missing references, and we have added them. We have also separated the references that note the identification of the ACS sequence from those that demonstrate Orc binding to that sequence.

eLife assessment

This study represents a valuable addition to the understanding of the DNA replication origin selection process in the budding yeast. The authors provide convincing evidence that the number of possible origins of replication is much higher than previously appreciated, although many of the newly identified origins are likely to only direct replication initiation rarely. This work will be of interest to those studying DNA replication and investigating protein-DNA interactions across the genome.

Reviewer #1 (Public Review):

The authors have previously employed micrococcal nuclease tethered to various Mcm subunits to the cut DNA to which the Mcm2-7 double hexamers (DH) bind. Using this assay, they found that Mcm2-7 DH are located on many more sites in the S. cerevisiae genome than previously shown. They then demonstrated that these sites have characteristics consistent with origins of DNA replication, including the presence of ARS consensus sequences, the location of very inefficient sites of initiation of DNA replication in vivo, and for the most part are free of nucleosomes. They contain a G-C skew and they locate to intergenic regions of the genome. The authors suggest, consistent with published single molecule results, that there are many more potential origins in the S. cerevisiae genome than previously annotated, but also conclude that many of the newly discovered Mcm2-7 DH are very infrequently used as active origins of DNA replication.

The results are convincing and are consistent with prior observations. The analysis of the origin associated features is informative.

Specific Comments:

1. Page 8. The addition of an estimate of the most active origins using Southern blotting is fine for highly active origins, but how was Southern blotting used to calculate that 1-2% of cells in the eight cohort have an Mcm complex loaded.

Reviewer #2 (Public Review):

By mapping the sites of the Mcm2-7 replicative helicase loading across the budding yeast genome using high-resolution chromatin endogenous cleavage or ChEC, Bedalov and colleagues find that these markers for origins of DNA replication are much more broadly distributed than previously appreciated. Interestingly, this is consistent with early reconstituted biochemical studies that showed that the ACS was not essential for helicase loading in vitro (e.g. Remus et al., 2009, PMID: 19896182). To accomplish this, they combined the results of 12 independent assays to gain exceptionally deep coverage of Mcm2-7 binding sites. By comparing these sites to previous studies mapping ssDNA generated during replication initiation, they provide evidence that at least a fraction of the 1600 most robustly Mcm2-7-bound sequences act as origins. A weakness of the paper is that the group-based (as opposed to analyzing individual Mcm2-7 binding sites) nature of the analysis prevents the authors from concluding that all of the 1,600 sites mentioned in the title act as origins. The authors also show that the location of Mcm2-7 location after loading are highly similar in the top 500 binding sites, although the mobile nature of loaded Mcm2-7 double hexamers prevents any conclusions about the location of initial loading. Interestingly, by comparing subsets of the Mcm2-7 binding sites, they find that there is a propensity of at least a subset of these sites to be nucleosome depleted, to overlap with at least a partial match to the ACS sequence (found at all of the most well-characterized budding yeast origins), and a GC-skew centered around the site of Mcm loading. Each of these characteristics is related to previously characterized S. cerevisiae origins of replication.

Overall, this manuscript greatly broadens the number of sites that are capable of loading Mcm2-7 in budding yeast cells and shows that a subset of these additional sites act as replication origins. Although these studies show that the sequence specificity of S. cerevisiae replication origins still sets it apart from metazoan origins, the ability to license and initiate replication from sites with increasing sequence divergence suggests a previously unappreciated versatility.

Specific points:

1. The authors need to come up with a consistent name for loaded Mcms at an origin. In the manuscript they variously use 'MCM'(page 3), 'Mcm complexes' (page 4), 'MCM double hexamer' (page 6), and 'double-helicase' (page 8) to describe the Mcm2-7 complexes detected in their ChEC experiments. They should pick one name (Mcm2-7 double hexamer or MCM double hexamer would be the most accurate and clear) and stick with it throughout the manuscript.

2. The authors state that "It is notable that, when Mcm is present, it is present predominantly as a single double-hexamer (right panel of Figure 3A), and that this remains true across the entire range of abundance shown in Figure 3A." This statement would be improved by prefacing it with "Based on the size of the protected regions" or some other clarifying statement that lets the reader know what they should be looking for in the data in 3A.


3. The revised statements that "We have previously used Southern blotting to demonstrate that approximately 90% of the DNA at one of the most acive known origins (ARS1103) is cut by Mcm-MNase (Foss et al., 2021), and to thereby infer that 90% of cells have a double- helicase loaded at this origin. Using this as a benchmark, we estimate that ~1-2 % cells have an Mcm complex loaded at the Mcm binding sites in the eighth cohort (ranks 1401- 1600)." partially clarifies how the authors came to the 1-2% number, however, the calculation is still unclear. Based on Figure 1A, there are at least three logs (1,00 fold) difference in the number of CBMSs between the best origins (which is what they state the 90% comes from) to anywhere close to the 1400-1600 rank. Seems like the number should be at best 0.1% and probably less. Either way, the authors need to explain this calculation either in the text or in the text. This sort of number tends to get thrown around later and without a clear explanation readers cannot evaluate its credibility. 


4. The authors make the point in the introduction and discussion that recent single-molecule studies of replication origins indicate that as many as 20% of the origins identified are outside of known origins. This is very interesting but there seems to be a missed opportunity of comparing the location of these origins with the CBMSs. It would improve the manuscript to include some sort of comparison rather than using only the much older and less accurate ssDNA analysis.

5. The authors state at the end of the first paragraph on page 6 that the ChEC data is "very reproducible" which does seem to be the case but it is a little confusing for the knowledgeable reader since one would expect quite different results for an HU arrested strain versus a asynchronous or G1 arrested strain. This is hidden in the analysis in Figure S1 since 13 experiments are compared against one in each plot, however, if one x one comparisons were done there would certainly be substantial differences (or if there are not, there is a problem with the data - e.g. HU arrested cells should lack licensing at early firing origins).

6. On page 8 the authors state, "First, clear peaks of ssDNA extend down to the eighth cohort..." This seems to be stretching the data. There are clear peaks for the first five cohorts and then there is a notable change with any peak being much broader, extending over at least 10,000 bp. The authors should reconsider their statement here as it is not well supported by the data.

7. There is one last missing reference. Wherever Eaton et al, 2010 is referenced Berbenetz, et al, 2010 (full ref below) should also be referenced as they come to very similar conclusions.

Berbenetz, N. M., Nislow, C. & Brown, G. W. Diversity of eukaryotic DNA replication origins revealed by genome-wide analysis of chromatin structure. PLoS Genet 6, (2010).

Author Response

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

Reviewer #1 (Public Review):

"MAGIC" was introduced by the Rong Li lab in a Nature letters article in 2017. This manuscript is an extension of this original work and uses a genome wide screen the Baker's yeast to decipher which cellular pathways influence MAGIC. Overall, this manuscript is a logical extension of the 2017 study, however the manuscript is challenging to follow, complicated by the data often being discussed out of sequence. Although the manuscripts make claims of a mechanism being pinpointed, there are many gaps and the true mechanisms of how the factors identified in the screen influence MAGIC is not clear. A key issue is that there are many assumptions drawn on previous literature, but central aspects of the mechanisms being proposed are not adequately shown.

Key comments:

1. Reasoning and pipelines presented in the first two sections of the results are disordered and do not follow figure order. In some instances, the background to experimental analyses such as detailing the generation of spGFP constructs in the YKO mutant library, or validation of Snf1 activation are mentioned after respective results are discussed. This needs to be fixed.

We thank the reviewer for pointing out potential confusion to readers. We have revised the first two sections according to reviewer’s suggestion. (Page 4-6)

1. In general there is a lack of data to support microscopy data and supporting quantification analysis. The validity of this data could be significantly strengthened with accompanying western blots showing accumulation of a given constructs in mitochondrial sub compartments (as was the case in the lab’s original paper in 2017).

We appreciate the reviewer’s suggestion on biochemical validations. However, the validity of this imaging-based assay for detecting import of cytosolic misfolded proteins into mitochondria, including the use of FlucSM as a model misfolding-prone protein, was carefully established in our previous study by using appropriate controls, super resolution imaging, APEX-based proximity labeling, and classical biochemical fractionation and protease protection assay (Ruan et al., 2017 Nature, ref. 10). We have reminded readers of these validation experiments in the previous study on Page 4, line 14-17.

In recent years, advancements in imaging-based tools have allowed many protein interactions and dynamic processes, which were previously examined by using biochemical assays in lysates of populations of cells, to be observed with various level of quantitation in live cells with intact cellular compartments. Many of these assays, e.g., the RUSH assay for ER to Golgi transport, FRAP-based analysis for nuclear/cytoplasmic shuttling of proteins, or FRET-based assays for protein-protein interactions, have been well accepted and even embraced by the respective fields of study once validated with genetic and biochemical approaches. The advantages for live-cell imaging-based assays are often their unique ability to report dynamic processes or unstable molecular species with spatiotemporal sensitivity. Respectfully, it is our view, based on our own experience, that the traditional protease protection assay is not adequate or sufficiently quantitative for examining the presence of unstable misfolded proteins in mitochondrial sub-compartments, given the obligatorily lengthy in vitro cell lysis and mitochondrial isolation process, during which the unstable proteins are continuously being degraded. This likely explains our previous biochemical fractionation result that only weak protein signals were detected in the matrix fraction (Ruan et al., 2017 Nature, ref. 10). In addition, unlike stably folded, native mitochondrial matrix proteins, misfolded/unfolded proteins such as Lsg1 or FlucSM are highly susceptible to protease treatment. This sensitivity makes the assay unreliable for detecting such proteins if trace amount of the protease penetrates mitochondrial membranes during cell lysis even without detergent treatment.

While we agree that protease protection assay is highly valuable for qualitative detection of the presence of a protein in certain mitochondrial compartments or determining its topology on membranes, this assay (regrettably in our hands) does not allow quantitative comparisons that were necessary for this study, because of inherent sample to sample variation, yet the laborious and low throughput nature of this assay makes it difficult for adequate statistical analysis. Furthermore, the level of protein detection in various fractions is highly sensitive to how the sample is treated with protease and detergent. Our imaging-based quantification, on the other hand, allows us to compare increased or decreased presence of GFP11-tagged proteins in mitochondria under different metabolic conditions or in different mutant or wild-type strains. Data from hundreds of cells and at least three independent biological replicates allowed us to apply adequate statistical analysis to aid our conclusion.

1. Much of the mechanisms proposed relies on the Snf1 activation. This is however not shown but assumed to be taking place. Given that this activation is central to the mechanism proposed, this should be explicitly shown here - for example survey the phosphorylation status of the protein.

Both REG1 deletion and low glucose conditions have been demonstrated extensively for Snf1 phosphorylation and activation in yeast (e.g., many seminal papers from Marian Carlson’s and other lab, such as ref. 24-28). In our study, we have indeed corroborated this by showing that Mig1 was exported from the nucleus in Δreg1 mutant and in low glucose conditions (Figure 1—figure supplement 2H and I. The mechanism of Snf1-mediated nuclear export of Mig1 has been characterized in detail as well (e.g., ref. 29-31).

Recommendations for the authors: please note that you control which, if any, revisions, to undertake

Reviewer #1 (Recommendations For The Authors):

SPECIFIC COMMENTS

Genetic Screen o Line 20 - the narrative moves to SNF1, but the reasoning for the selection of this Class I substrate is not defined. What was the basis for this selection - what happened to the other Class I substrates. It is stated in the text that the other Class I proteins show the same increase in spGFP signal. The data showing this should be included in the Supp Figure 1 for transparency.

We have moved the narratives of Snf1 function to the second section and clarified that we were interested in this gene due to its central role in metabolism and mitochondrial functions that may influence MAGIC (Page 5: line 16-20). Other genes in class 1 were shown in Table S1. Detailed discussion of other genes in this category is beyond the scope of this study.

Snf1/AMPK prevents MP accumulation in mitochondria:

The FlucDM data in human RPE-1 mitochondria seems to be added to only increase the significance of the work. The mechanisms suggested here with Hap4 would not be possible in human cells as there is no homologue of this protein in human cells. Making generalisations that these pathways are conserved based on this one experiment is not appropriate.

We appreciate this feedback. Although the focus of this study is the regulation of MAGIC by the yeast AMPK Snf1, we would like to share our initial observation that suggests a similar role of AMPK in human RPE-1 cells. We acknowledge that the underlying mechanisms regarding the downstream transcription factors and pathway for misfolded protein import could be different in mammalian cells, but the overall effect of AMPK in mitochondrial biogenesis is well known to resemble that of Snf1. To avoid making over-generalization, we changed our statement of conclusion to: ‘These results suggest that AMPK in human cells regulates MP accumulation in mitochondria following a similar trend as in yeast, although the underlying mechanisms might differ between these organisms.’ (Page 7: line 2-4)

Mechanisms of MAGIC regulation by Snf1:

While the lysosome is ruled out here the authors have not considered the proteasomes. Is there a reason for this? Given accumulation of aggregates outside of mitochondria, and previous connections of the proteasome to mitochondrial quality control this would be an obvious thing to check. We examined the role of lysosomal degradation here because it is known to be activated under Snf1active condition (ref. 37). We appreciate this feedback and have included a new analysis on MG132treated FlucSM spGFP strains in which PDR5 gene was deleted to avoid drug efflux.

This result suggests that the proteosome inhibitor did not ablate the difference in FlucSM accumulation between these conditions. That MG132 promoted mitochondrial accumulation of FlucSM in both high glucose and low glucose conditions was not surprising, as FlucSM is also degraded by proteasome in the cytosol (Ruan et al., 2017 Nature, ref. 10), and preventing this pathway could divert more of such protein molecules toward MAGIC. (Page 7: line 26-29).

Line 13 "we hypothesized that elevated expression of mitochondrial preproteins induced by the activation of Snf1-Hap4 axis (REF) may outcompete MPs for import channels". This statement has some assumptions. The authors have not shown that Snf1 is activated in thier models and more importantly that they have an accumulation of mitochondrial preproteins. The data that follows using the cytosolic domains of the receptors is hard to rationalise without seeing evidence that there is in fact pre-protein accumulation or impacts on the mitochondrial proteome in this system.

As stated in our response to main point [3], Snf1 activation in reg1 mutant or in low glucose is evidenced by our data showing Mig1 export from nucleus to cytoplasm and had also been shown in many previous publications. A recent study (Tsuboi et al., 2020 eLife) also showed a dramatic increase in mitochondrial volume fraction in Δreg1 cells and wild-type cells in respiratory conditions, further supporting the role of Snf1 in mitochondrial biogenesis. We have provided relevant references in the manuscript (ref. 24-28).

The ability of Tom70 cytosolic domain (Tom70cd), which can bind mitochondrial preproteins but not localize to mitochondria due to lack of N-terminal targeting sequence, to compete with endogenous Tom70 for mitochondrial preproteins has been well documented (ref. 47-49). However, we agree with the reviewer that a future quantitative proteomics study to measure changes in mitochondrial proteome under Tom70cd over-expression could allow more accurate interpretation of our experimental result.

AMPK protects cellular fitness during proteotoxic stress:

The inhibition of preprotein import by overexpressing the cytosolic domains of receptors is not supported with some proof of principle data. If this was working as the authors assume, it is not clear why only an effect with Tom70 is observed. The majority of the mitochondrial proteome is imported via Tom20/Tom22 so this does not align with what the authors are suggesting. Is the Tom70CD and any associated Hsp proteins facilitating the observed changes to the MPs?

We thank the reviewer for raising this point. We expressed different TOM receptor cytosolic domains but found that Tom70cd had the strongest rescue on MAGIC under AMPK activation conditions. It is possible that certain Tom70 substrates or Tom70-assoicated heat shock proteins inhibit the import of MAGIC substrates. We admit that a clear explanation of this unexpected observation necessitates a better understanding of how native and MAGIC substrates are selected and imported by the outer-membrane channel. We can only offer our best interpretation based on the current state of the understanding, and we feel that we have been careful to acknowledge such in the manuscript.

While the effect of AMPK inactivation reducing FUS accumulation was striking, this was all in the context of overexpression and may not be physiologically relevant - or may occur very transiently under basal conditions. Is GST an appropriate control here, why not use WT FUS? Likewise, one representative image is shown in Figure 5 - can the authors show western blotting that mitochondrial accumulation of FUS can be reduced with AMPK activation?

We thank the reviewer for this suggestion, however, overexpressed FUS WT is also aggregation prone (Zhihui Sun et al., 2011, PloS Biology; Shulin Ju, 2011, PloS Biology; Jacqueline C. Mitchell et., 2013, Acta Neuro). We believe that GST, as a well-folded protein, is an appropriate control (Ruan et al., 2017 Nature, ref. 10). As we discussed in response to main point [1], the in vitro assay involving protease protection and western blots do not allow reliable quantitative comparison in our hands.

In text changes.

The analysis pipeline of the YKO mutant library should be introduced at the very start of the first paragraph, not the end.

Addressed on Page 4, second paragraph

"Fluc" should be introduced as "Firefly luciferase" within the first paragraph of the first section, also need to define SM and DM in FlucSM/FlucDM - these appear to be missing.

Addressed in both Introduction (Page 2: line 29; Page 3: line 8-9) and re-clarified in Result (Page 5: line 27-29)

The role of Reg1 should be explicitly stated in the text, not just in the figure.

Addressed on Page 6: line 3-6

Figure 1H legend states Reg1 (WT) is Snf1-inactive and Reg1 KO is Snf1-active. This wording is confusing and is not supported by data, but by assumption. If the authors want to use this wording then evidence needs to be provided - as suggested above.

We have changed this and other legends to only show genotypes and medium conditions.

"Tom70cd overexpression also exacerbated growth rate reduction due to FlucSM expression in HG medium (Figure 4A; Figure 4 - figure supplement 1A)" should be figure supplement 1B.

Fixed on Page 10: line 10

"These results suggest that glucose limitation protects mitochondria and cellular fitness during FlucSM induced proteotoxic stress through Snf1-dependent inhibition of MP import into mitochondria". The phrase "Snf1-dependent inhibition of MP import into mitochondria" may be misleading, as Snf1 isn't modulating import directly but is acting on transcriptional regulators to modulate mitochondrial import under stress.

We restated the conclusion as follows: ‘These results suggest that Snf1 activation under glucose limitation protects mitochondrial and cellular fitness under FlucSM-associated proteotoxic stress.’ (Page 10: line 20- 21)

"... Significantly increased the fraction of spGFP-positive and MMP-low cells in both HG and LG medium (Figure 4G-K)" should be (Figure 4J-K).

Fixed on Page 11: line 3

Reviewer #2 (Public Review):

Work of Rong Li´s lab, published in Nature 2017 (Ruan et al, 2017), led the authors to suggest that the mitochondrial protein import machinery removes misfolded/aggregated proteins from the cytosol and transports them to the mitochondrial matrix, where they are degraded by Pim1, the yeast Lon protease. The process was named mitochondria as guardian in cytosol (MAGIC).

The mechanism by which MAGIC selects proteins lacking mitochondrial targeting information, and the mechanism which allows misfolded proteins to cross the mitochondrial membranes remained, however, enigmatic. Up to my knowledge, additional support of MAGIC has not been published. Due to that, MAGIC is briefly mentioned in relevant reviews (it is a very interesting possibility!), however, the process is mentioned as a "proposal" (Andreasson et al, 2019) or is referred to require "further investigation to define its relevance for cellular protein homeostasis (proteostasis)" (Pfanner et al, 2019).

Rong Li´s lab now presents a follow-up story. As in the original Nature paper, the major findings are based on in vivo localization studies in yeast. The authors employ an aggregation prone, artificial luciferase construct (FlucSM), in a classical split-GFP assay: GFP1-10 is targeted to the matrix of mitochondria by fusion with the mitochondrial protein Grx5, while GFP11 is fused to FlucSM, lacking mitochondrial targeting information. In addition the authors perform a genetic screen, based on a similar assay, however, using the cytosolic misfolding-prone protein Lsg1 as a read-out.

My major concern about the manuscript is that it does not provide additional information which helps to understand how specifically aggregated cytosolic proteins, lacking a mitochondrial targeting signal could be imported into mitochondria. As it stands, I am not convinced that the observed FlucSM-/Lsg1-GFP signals presented in this study originate from FlucSM-/Lsg1-GFP localized inside of the mitochondrial matrix. The conclusions drawn by the authors in the current manuscript, however, rely on this single approach.

In the 2017 paper the authors state: "... we speculate that protein aggregates engaged with mitochondria via interaction with import receptors such as Tom70, leading to import of aggregate proteins followed by degradation by mitochondrial proteases such as Pim1." Based on the new data shown in this manuscript the authors now conclude "that MP (misfolded protein) import does not use Tom70/Tom71 as obligatory receptors." The new data presented do not provide a conclusive alternative. More experiments are required to draw a conclusion.

In my view: to confirm that MAGIC does indeed result in import of aggregated cytosolic proteins into the mitochondrial matrix, a second, independent approach is needed. My suggestion is to isolate mitochondria from a strain expressing FlucSM-GFP and perform protease protection assays, which are well established to demonstrate matrix localization of mitochondrial proteins. In case the authors are not equipped to do these experiments I feel that a collaboration with one of the excellent mitochondrial labs in the US might help the MAGIC pathway to become established.

We thank Reviewer 2 for these suggestions, but we would like to respectfully offer our difference in opinion:

a. Regarding the suggestion “to isolate mitochondria from a strain expressing FlucSM-GFP and perform protease protection assays”, in our previous study (Ruan et al., 2017 Nature, ref. 10), we have indeed applied two independent biochemical approaches: APEX-mitochondrial matrix proximity labeling and classic protease protection assay using non-spGFP strains, both consistently confirmed the entry of misfolded proteins into mitochondria under proteotoxic stress. Our super-resolution imaging further confirmed the import of the split GFP-labeled proteins to be inside mitochondria. Moreover, as we discussed in response to Reviewer 1’s main point [2], while the suggested biochemical assay is useful for validating topology within mitochondria, it is not quantitative and may not reliably report the in vivo accumulation of misfolded proteins in mitochondria due to the isolation process that takes hours, during which the unstable proteins could be continuously degraded within mitochondria.

While we agree with the reviewer that we do not yet understand how misfolded proteins are imported into mitochondria, it would be unfair to state “as it stands, I am not convinced..” simply because the underlying mechanism remains to be elucidated. We would like to point out that targeting sequences for many well-established mitochondrial proteins are still not well defined. It is well known that mitochondrial targeting sequences are not as uniformly predictable as, for example, nuclear targeting sequences. Our finding that deletion of TOM6 enhances the import of misfolded proteins suggest that their import may involve the TOM channel in a more promiscuous conformation, which may reduce the requirement for a specific sequence-based targeting signal associated with the substrate.

b. Regarding the role of Tom70, in our 2017 study, using proteomics and subsequently immunoprecipitation we validated the binding, albeit not necessarily direct, between misfolded protein FlucSM and Tom70. Therefore, “we speculate that protein aggregates engaged with mitochondria via interaction with import receptors such as Tom70”. Recent studies from different labs confirmed the interactions between Tom70 and aggregation prone proteins (Backes et al., 2021, Cell Reports; Liu et al., 2023, PNAS). In the current study, surprisingly, knockout of TOM70 did not block MAGIC, suggesting redundant components of mitochondria import system may facilitate the recruitment of misfolded proteins in the absence of Tom70, and this does not contradict the notion that Tom70 helps tether protein aggregates to mitochondria.

c. Regarding other studies also showing the import of misfolding or aggregation-prone cytosolic proteins into mitochondria, there have been at least several recent studies in the literature for mammalian cells involving either model substrates or disease proteins (e.g., ref. 12-15; 56-58; Vicario, M. et al. 2019 Cell Death Dis.). The studies are briefly mentioned in Introduction (Page 3, paragraph 2). The present manuscript documents a major effort from our group using whole genome screen in yeast to understand the mechanism and regulation of MAGIC. Many of the screen hits have yet to be studied in detail. We full agree that much remains to be understood about whether and how this pathway affects proteostasis and what might be the evolutionary origin for such a mechanism.

Additional comments:

The genetic screen:

The genetic screen identified five class 1 deletion strains, which lead to enhanced accumulation of Lsg1GFP and a larger set of class 2 mutants, which lead to reduced accumulation. Please note, in my opinion it is not clear that accumulation of the reporters occurs inside the mitochondria. In any case, the authors selected one single protein for further analysis: Snf1, the catalytic subunit of the yeast SNF complex, which is required for respiratory growth of yeast.

The results of the screen are not discussed in any detail. The authors mention that ribosome biogenesis factors are abundant among class 2 mutants. Noteworthy, Lsg1 is involved in 60S ribosomal subunit biogenesis. As Lsg1-GFP11 is overexpressed in the screen this should be discussed. Class 2 mutants also .include several 40S ribosomal subunit proteins (only one of the 60S subunit). What does this imply for the MAGIC model? Also, it should be discussed that the screen did not identify reg1 and hap4, which I had expected as hits based on the data shown in later parts of the manuscript.

We apologize for the confusion, but the GFP11 tag was in fact knocked into the C-terminus of Lsg1 in the endogenous LSG1 locus, and so Lsg1 was not overexpressed in the screen. We have made sure that this information is clearly conveyed in the revised manuscript (Page 4: line 20-22). How the ribosome small subunit affects MAGIC is beyond the focus of the current study and will be pursued in the future.

Regarding why certain mutants did not come out of our initial screen, this is not unexpected as the YKO collection, although extremely valuable to the community, is known to be potentially affected by false knockouts, suppressor accumulation and cross contamination (for references, e.g., Puddu et al., 2019 Nature). Additionally, high-through screens can also miss real hits. In our experience using this collection in several studies, we often found additional hits from analysis of genes implicated by known genetic or biochemical interactions.

Mutant yeast strains and growth assays:

The Δreg1 strain grows poorly in all growth conditions and frequently accumulates extragenic suppressor mutations (Barrett et al, 2012). It would be good to make sure that this is not the case in the strains employed in this study. My suggestion is to do (and show) standard yeast plating assays with the relevant mutant strains including Δreg1, snf1, hap4, Δreg1Δhap4 without the split GFP constructs and also with them (i.e. the strains that were used in the assays).

We thank the reviewer for the suggestion. We were indeed aware of potential accumulation of suppressor mutations from the YKO library. Therefore, deletion mutants like Δreg1 and loss of TFs downstream of Snf1 that we used in the study after the initial screen were all freshly made and validated. At least 3 independent colonies were analyzed for each mutant (mentioned in Methods & Materials; Page 33, line 57). Moreover, the plating assay suggested here may not reveal additional information other than growth, which was taken into consideration during our experiments.

Activation of Snf1 in the relevant strains should be tested with the commercially available antibody recognizing active Snf1, which is phosphorylated at Snf1-T210.

Snf1 activation was validated by the Mig1 exporting from the nucleus. We also noted above that many studies have clearly demonstrated Snf1 activation in reg1 mutant and under low glucose growth (e.g., ref. 24-28).

Effects of Snf1, Reg1, Hap4 and respiratory growth conditions:

The authors show that split GFP reporters show enhanced accumulation during fermentative growth, in Δsnf1, and Δreg1Δhap4 and fail to accumulate during respiratory growth, in Δreg1 and upon overexpression of HAP4. Analysis of Δhap4 should be included in Fig. 2. The suggestion that upon activation of Snf1 enhanced Hap4-dependent expression "outcompetes" misfolded protein import seems unlikely as only a fraction of mitochondrial genes is under control of Hap4. Without further experimental evidence I do not find that a valid assumption. More likely, the membrane potential plays a role: it is low during fermentative growth, in Δsnf1 and Δreg1Δhap4, and high during respiratory growth and in Δreg1 (Hübscher et al, 2016). Such an effect of the membrane potential seems to contradict the findings in the 2017 paper and the issue should be clarified and discussed. In any case, these data do not reveal that GFP reporters accumulate inside of the mitochondria. Based on the currently available evidence they may accumulate in close proximity/attached to the mitochondria. This has to be tested directly (see above).

We have included our analysis of Δhap4 in Page 8: line 14-15 and Figure 2—figure supplement 1H. Consistent with our result for Δreg1Δhap4 in glucose-rich medium, HAP4 deletion also resulted in a significant increase in mitochondrial accumulation of FlucSM in low glucose medium compared to WT. It did not have effect in high glucose condition in which Snf1 is largely inactive.

It is our view that the importance of Hap4 should not be judged by the number of nuclear encoded mitochondrial proteins they regulate. Still, this sub-group comprises a considerable number of proteins (at least 55 genes upregulated by Hap4 overexpression, ref. 43), and certain substrates may be more competitive with misfolded cytosolic proteins for import. Our genetic data strongly suggest that the inhibitory effect of active Snf1 on MAGIC is through Hap4, although we agree with the reviewer that detailed mechanism on how Hap4 substrates may compete with misfolded proteins need to be addressed in future studies.

Membrane potential is important for mitochondrial import. During respiratory growth and in Δreg1, membrane potential is well known to be elevated comparing to fermentative condition (e.g., Figure 4C). Our observation that the import of misfolded proteins into mitochondria is reduced under these conditions simply suggests that this reduction is not due to a lack of membrane potential. This is not in any way contradictory to our 2017 finding that misfolded protein import requires membrane potential (ref. 10).

Again, the accumulation of misfolded proteins in mitochondria, especially the model protein FlucSM, has been validated by using super resolution imaging (Figure 1—figure supplement 1A) in addition to the protease protection assay in our 2017 study.

Introduction and Discussion:

Both are really short, too short in my view. Please provide some background of the general principals of mitochondrial protein import and information of how exactly translocation of cytosolic, aggregated proteins (lacking targeting information) is supposed to work. I do not understand exactly how the authors actually envisage the process.

We thank the reviewer for the suggestion. In the revised manuscript, we have extended both Introduction (Page 2-3) and Discussion section (Page 11-13)

The results from the 2022 eLife paper (Liu et al, 2022), which suggests that Tom70 may "regulate both the transcription/biogenesis and import of mitochondrial proteins so the nascent mitochondrial proteins do not compromise cytosolic proteostasis or cause cytosolic protein aggregation" should be discussed with regard to the data obtained with overexpression of the Tom70 soluble domain.

We thank the reviewer for pointing out that study and we have included a brief comment in Discussion section (Page 12: line 13-16). As the function of Tom70 appears to be complex, we cannot exclude the possibility that overexpression of the cytosolic domain has additional or indirect effects in addition to that due to preprotein binding.

Andreasson, C., Ott, M., and Buttner, S. (2019). Mitochondria orchestrate proteostatic and metabolic stress responses. EMBO Rep 20, e47865.

Barrett, L., Orlova, M., Maziarz, M., and Kuchin, S. (2012). Protein kinase A contributes to the negative control of Snf1 protein kinase in Saccharomyces cerevisiae. Eukaryot Cell 11, 119-128.

Hubscher, V., Mudholkar, K., Chiabudini, M., Fitzke, E., Wolfle, T., Pfeifer, D., Drepper, F., Warscheid, B., and Rospert, S. (2016). The Hsp70 homolog Ssb and the 14-3-3 protein Bmh1 jointly regulate transcription of glucose repressed genes in Saccharomyces cerevisiae. Nucleic Acids Res. 44, 5629-5645.

Liu, Q., Chang, C.E., Wooldredge, A.C., Fong, B., Kennedy, B.K., and Zhou, C. (2022). Tom70-based transcriptional regulation of mitochondrial biogenesis and aging. Elife 11

Pfanner, N., Warscheid, B., and Wiedemann, N. (2019). Mitochondrial proteins: from biogenesis to functional networks. Nat Rev Mol Cell Biol 20, 267-284.

Ruan, L., Zhou, C., Jin, E., Kucharavy, A., Zhang, Y., Wen, Z., Florens, L., and Li, R. (2017). Cytosolic proteostasis through importing of misfolded proteins into mitochondria. Nature 543, 443-446.

I prefer to have "all in one", also due to time limitation.

It would be great to be able to upload the review file as otherwise formatting and symbols get lost.

Reviewer #3 (Public Review):

In this study, Wang et al extend on their previous finding of a novel quality control pathway, the MAGIC pathway. This pathway allows misfolded cytosolic proteins to become imported into mitochondria and there they are degraded by the LON protease. Using a screen, they identify Snf1 as a player that regulates MAGIC. Snf1 inhibits mitochondrial protein import via the transcription factor Hap4 via an unknown pathway. This allows cells to adapt to metabolic changes, upon high glucose levels, misfolded proteins an become imported and degraded, while during low glucose growth conditions, import of these proteins is prevented, and instead import of mitochondrial proteins is preferred.

This is a nice and well-structured manuscript reporting on important findings about a regulatory mechanism of a quality control pathway. The findings are obtained by a combination of mostly fluorescent protein-based assays. Findings from these assays support the claims well.

While this study convincingly describes the mechanisms of a mitochondria-associated import pathway using mainly model substrates, my major concern is that the physiological relevance of this pathway remains unclear: what are endogenous substrates of the pathway, to which extend are they imported and degraded, i.e. how much does MAGIC contribute to overall misfolded protein removal (none of the experiments reports quantitative "flux" information). Lastly, it remains unclear by which mechanism Snf1 impacts on MAGIC or whether it is "only" about being outcompeted by mitochondrial precursors.

We thank Reviewer 3 for the positive and encouraging comments on our manuscript. We agree with the reviewer that identifying MAGIC endogenous substrates and understanding what percentage of them are degraded in mitochondria are very important issues to be addressed. We are indeed carrying out projects to address these questions. We also agree with Reviewer 3 that the effect of Snf1 on MAGIC may have additional mechanisms in addition to precursors competition, such as Tom6 mediated conformational changes of TOM pores. In the revised manuscript, we had added a discussion to address these comments (Page 12: line 21-28).

Reviewer #3 (Recommendations For The Authors):

1. In their screen, the authors utilize differences in GFP intensity as a measure for import efficiency. However, reconstitution of the GFP from GFP1-10 and GFP11 in the matrix might also be affected (folding factors, differential degradation).

Upon Snf1 activation, the protein abundance of mitochondrial chaperones such as Hsp10, Hsp60, and Mdj1, and mitochondrial proteases such as Pim1 are not significantly changed (ref. 35). Therefore, it is unlikely that the folding and degradation capacity of mitochondrial matrix is drastically affected by Snf1 activation.

To examine the effect of Snf1 activation on spGFP reconstitution, Grx5 spGFP strain was constructed in which the endogenous mitochondrial matrix protein Grx5 was C-terminally tagged with GFP11 at its genomic locus, and GFP1-10 was targeted to mitochondria through cleavable Su9 MTS (MTS-mCherryGFP1-10) (ref. 10). Only modest reduction in Grx5 spGFP intensity was observed in LG compared to HG, and no significant difference after adjusting the GFP1-10 abundance (spGFP/mCherry ratio) (Figure 1— figure supplement 3A-D). These data suggest that any effect on spGFP reconstitution is insufficient to explain the drastic reduction of MP accumulation in mitochondria under Snf1 activation. Overall, our results demonstrate that Snf1 activation primarily prevents mitochondrial accumulation of MPs, but not that of normal mitochondrial proteins. (Page 6: line 17-25).

We admit, however, that to fully rule out these factors, specific intra-mitochondrial folding or degradation reporter assays would be needed.

1. Scoring of protein import always takes place using fluorescence-based assays. These always require folding of the "sensors" in the matrix. An additional convincing approach that would not rely on matrix folding could be pulse chase approaches coupled to fractionation assays and immunoprecipitation.

We thank reviewer 3 for this suggestion. In our previous study, we applied two different biochemical assays: APEX proximity labeling, and mitochondrial fractionation followed by protease protection. Both confirmed the entry of misfolded proteins into mitochondria as observed by using split GFP. As we discussed in response to Reviewer 1’s main point [3], the fractionation assays are not quantitative enough for the comparisons made in our study. In particular, during the over 2-hour assay, misfolded proteins continue to be degraded within mitochondria. By using proper controls, our spGFP system provides quantitative comparisons for mitochondrial accumulation of misfolded proteins in non-disturbed physiological conditions.

1. Could the pathway be reconstituted in vitro with isolated mitochondria to test for the "competition hypothesis"

This is an excellent suggestion, but setting up such a reconstituted system is a project on its own. The study documented in this manuscript already encompasses a large amount of work that we feel should be published timely.

1. Fluorescence figures are not colour blind friendly (red-green). This should be improved by changing the color scheme.

We thank reviewer 3 for pointing this out and sincerely apologize for any inconvenience. However, we are unfortunately unable to change all images within a limited time. We will adopt another color scheme in future work.

1. spGFP in human cells appears to form "spot-like" structures. What are these granules?

We indeed observed granule-like structures by spGFP labeled FUS in mitochondria, which is interesting, but we did not investigate this further because it is a not a focus of this study.

Joint Public Review:

In this study, Wang et al extend on their previous finding of a novel quality control pathway, the MAGIC pathway. This pathway allows misfolded cytosolic proteins to become imported into mitochondria and there they are degraded by the LON protease. Using a screen, they identify Snf1 as a player that regulates MAGIC. Snf1 inhibits mitochondrial protein import via the transcription factor Hap4 via an unknown pathway. This allows cells to adapt to metabolic changes, upon high glucose levels, misfolded proteins become imported and degraded, while during low glucose growth conditions, import of these proteins is prevented, and instead import of mitochondrial proteins is preferred.

This is a nice and well-structured manuscript reporting on important findings about a regulatory mechanism of a quality control pathway. The findings are obtained by a combination of mostly fluorescent protein-based assays. Findings from these assays support the claims well.

While this study convincingly describes the mechanisms of a mitochondria-associated import pathway using mainly model substrates, my major concern is that the physiological relevance of this pathway remains unclear: what are endogenous substrates of the pathway, to which extent are they imported and degraded, i.e. how much does MAGIC contribute to overall misfolded protein removal (none of the experiments reports quantitative "flux" information). Lastly, it remains unclear by which mechanism Snf1 impacts on MAGIC or whether it is "only" about being outcompeted by mitochondrial precursors.

Author Response

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

Response to Reviewers

To whom it may concern, Thank you for your constructive feedback on our manuscript. I appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback. We are grateful to the reviewers for their insightful comments and suggestions for our paper. I have been able to incorporate changes to reflect the majority of these suggestions provided. I have updated the analysis scripts (at https://github.com/neurogenomics/reanalysis_Mathys_2019) and have listed these changes in blue below:

eLife assessment:

This work is useful as it highlights the importance of data analysis strategies in influencing outcomes during differential gene expression testing. While the manuscript has the potential to enhance awareness regarding data analysis choices in the community, its value could be further enhanced by providing a more comprehensive comparison of alternative methods and discussing the potential differences in preprocessing, such as scFLOW. The current analysis, although insightful, appears incomplete in addressing these aspects.

We thank the reviewing editors for this note. We agree that the differences in preprocessing will affect the results and conceal which step in our reanalysis resulted in the discrepancies we noted. To address this, we have split out our reanalysis into two separate parts - In the main body of the text we discuss the differences resulting from just changing the differential expression approach where we use the same processed data as the authors to enable a fair comparison. Secondly, we still provide the reprocessed data and perform differential expression analysis on it and discuss the cause and impact the differences in the processing steps made to the results.

Reviewer 1:

I think readers would be interested to learn more about the genes that were found "significant" by the original paper but sorted out by the authors. Did they just fall short of the cutoffs? If so, how many more samples would have been required to ascertain significance? This would yield a recommendation for future studies and an overall more positive/productive spirit to the manuscript. On the other hand, I suspect a fraction of DEGs were false positives due to differences in the proportions of cells from different individuals compared to the original analysis. Which percentage of DEGs does this apply to? Again, this would raise awareness of the issue and support the use of pseudobulk approaches.

To investigate the relationship between the genes and how they differ across our analysis we have added a correlation analysis between our different DE approaches (using the same processed data), see paragraph 5 in the manuscript and supplementary table 3. In short, we find that there is a high correlation in the genes’ fold change values across our pseudobulk analysis and the author’s pseudoreplication analysis on the same dataset (pearson R of 0.87 for an adjusted p-value of 0.05) which is somewhat expected given the DE approaches are applied to the same dataset. However, the p-values, which pertain to the likelihood that a gene’s expressional changes is related to the case/control differences in AD, and resulting DEGs vary considerably due to the artificially inflated confidence of the author’s approach (Fig. 1c-e). Despite there being a correlation between the pseudoreplciation and pseudobulk approaches here, we do not think it makes sense to consider how many more samples would have been required to ascertain significance. The differences in results between the two approaches is not negatable with sample size as many DEGs identified by pseudoreplication will be false positives as highlighted in previous work1,2,3,4. However, perhaps we are misinterpreting the reviewer, who may have meant a power analysis which we have not conducted. Such an undertaking would require analysing a multitude of snRNA-Seq of large sample sizes to garner a confident estimate for power calculations based on pseudobulk approaches. Although we agree with the reviewer that this would be beneficial to the field, we do not believe it is in scope for this work. On the reviewer’s note regarding a fraction of DEGs being false positives due to differences in the proportions of cells from different individuals compared to the original analysis - We have analysed the same processed data the authors used to negate the differences caused by the differing processing steps. We thank the reviewer for this suggestion. We also give more insight into the cause of these differences, namely on filtering our nuclei with large proportions of mitochondrial reads and discuss their effect in paragraph 3 (also see Supplementary Figure 2).

Given there are only a few DEGs, it would be good to show more data about these genes to allow better assessment of the robustness of the results, i.e., boxplots of the pseudobulk counts in the compared groups and perhaps heatmaps of the raw counts prior to aggregation. This could rule out concerns about outliers affecting the results.

In Supplementary Figure 3, we have added boxplots of the sum pseudobulked, trimmed mean of M-values (TMM) normalised counts for three of our identified DEGs (b) and three of the authors’ DEGs which they discuss in their manuscript (a) to show the differences in counts across AD pathology and controls for these genes. We hope this gives some insight into the transcriptional changes highlighted by the differing approaches. In our opinion, there is a clear difference in the transcriptional signal in the genes identified from pseudobulk which is not present for the genes identified from the authors approach.

Overall, I believe the paper would deliver a clearer message by mainlining the QC from the original study and only changing the DE analysis. However, if keeping the part about QC/batch correction:

• Assess to which degree changes in cell type proportion are indeed due to batch correction (as suggested in the text) and not filtering by looking at the annotated cell types in the original publication and those in your analysis.

• Also perform the analysis without changing QC and state the # of DEGs in both cases, to at least allow some disentanglement of the effect of different steps of the analysis.

• Please state the number of cells removed by each QC step in the supplementary note.

We thank the reviewer for this suggestion. We agree with performing the DE analysis on the same processed data as the original authors and have split out our reanalysis into two separate parts, primarily focussing on the discrepancies caused by the choice of differential expression (DE) approach. By splitting our analysis in this manner, we can identify the substantial differences in results caused by differing the DE approach in the study. Secondly, we can see how differences in preprocessing affects the DE results in isolation too – see paragraph 8 but in short, the fold change correlation between pseudobulk DE analyses on the reprocessed data vs authors processed data only had a moderate correlation (Pearson R of 0.57).

In regards to the number of cells removed by each QC step, we have added an aggregated view for all samples in supplementary table 3 and also give the full statistics per sample in our Github repository: https://github.com/neurogenomics/reanalysis_Mathys_2019. Moreover, we investigated the root cause in the differences in nuclei numbers, uncovering filtering down to mitochondrial read proportions as the main culprit (Supplementary Figure 2).

I recommend the authors read the following papers, assess whether their methodology agrees with them, and add citations as appropriate to support statements made in the manuscript.

We thank the reviewer for this comprehensive list. We have updated our manuscript and supplementary file and main text throughout to cite many of these where appropriate. We believe this helps add context to our decisions for the differing tools and approaches used as part of the processing pipeline with scFlow and the differential expression approach.

I believe the authors' intention was to show the results of their reanalysis not as a criticism of the original paper (which can hardly be faulted for their strategy which was state-of-the-art at the time and indeed they took extra measures attempting to ensure the reliability of their results), but primarily to raise awareness and provide recommendations for rigorous analysis of sc/snRNA-seq data for future studies.

We thank the reviewer for this note, this was exactly our intent. Furthermore, we are based in a dementia research institute and our aim is to ensure that ensure that the Alzheimer’s disease research field does not focus on spuriously identified genes.We have updated the text of the manuscript (start paragraph 2) to explicitly state this so our message is not misconstrued.

In my opinion, the purpose of the paper might be better served by focusing on the DE strategy without changing QC and instead detailing where/how DEGs were gained/lost and supporting whether these were false positives.

We agree that the differences in preprocessing will affect the results and conceal which step in our reanalysis resulted in the discrepancies we noted. To address this, we have split out our reanalysis into two separate parts - In the main body of the text we discuss the differences resulting from just changing the differential expression approach where we use the same processed data as the authors to enable a fair comparison. Secondly, we still provide the reprocessed data and perform differential expression analysis on it and discuss the impact the differences in the processing steps made to the results. As previously mentioned, we have also added further investigation into the DEGs identified, looking at the correlation across the differing approaches and plotting the counts for selected genes.

For instance, removal with a mitochondrial count of <5% seems harsh and might account for a large proportion of additional cells filtered out in comparison to the original analysis. There is no blanket "correct cutoff" for this percentage. For instance, the "classic" Seurat tutorial https://satijalab.org/seurat/articles/pbmc3k_tutorial.html uses the 5% threshold chosen by the authors, an MAD-based selection of cutoff arrived at 8% here https://www.sc-best-practices.org/preprocessing_visualization/quality_control.html, another "best practices" guide choses by default 10% https://bioconductor.org/books/3.17/OSCA.basic/quality-control.html#quality-control-discarded, etc. Generally, the % of mitochondrial reads varies a lot between datasets.

Apologies, the 5% cut-off was a misprint – the actual cut-off used was 10% which, as the reviewer notes, is on the higher side of what is recommended. We have updated our manuscript to rectify this mistake and discuss the differences in the number of cells caused by the two approaches to mitochondrial filtering in the manuscript (paragraph 3). We found that over 16,000 nuclei that were removed in our QC pipeline were kept by the author’s (Supplementary Fig. 2), explaining the discrepancy in the number of nuclei after QC. Based on Supplementary Fig. 2, it is clear the author’s approach was ineffective at removing nuclei with high proportions of mitochondrial reads which is indicative of cell death5,6. We hope this alleviates the reviewer’s concerns around our alternative processing approach. Moreover, as mentioned, we swapped to compare the differences by DE approaches on the same data to avoid any effect by this.

Reviewer 2:

The paper would be better if the authors merged this work with the scFLOW paper so that they can justify their analysis pipeline and show it in an influential dataset.

We thank the reviewer for this note. We would like to clarify that the purpose of our work was not to show the scFlow analysis pipeline on an influential dataset but rather to raise awareness and provide recommendations for rigorous analysis of single-cell and single-nucleus RNA-Seq data (sc/snRNA-Seq) for future studies and to help redirect the focus of the Alzheimer’s disease research field away from possible spuriously identified genes. We have updated our manuscript text to highlight this (see start paragraph 2). Furthermore, we are aware our original approach reprocessing the data with scFlow will affect the results and conceal which step in our reanalysis resulted in the discrepancies we noted. Thus, we have split out our reanalysis into two separate parts - In the main body of the text we discuss the differences resulting from just changing the differential expression approach where we use the same processed data as the authors to enable a fair comparison. Secondly, we still provide the reprocessed data so that the community can benefit from it and perform differential expression analysis on it and discuss the impact the differences in the processing steps made to the results. We have also added further references supporting the choice of steps and tools used in scFlow in the supplementary text which should address the reviewer’s concerns about justifying the analysis pipeline. Moreover, we identified the cause of the nuclei count differences caused by the two processing approaches, namely on filtering our nuclei with large proportions of mitochondrial reads and discuss their effect in paragraph 3 (also see Supplementary Figure 2).

A major contribution is the use of the authors' own inhouse pipeline for data preparation (scFLOW), but this software is unpublished since 2021 and consequently not yet refereed. It isn't reasonable to take this pipeline as being validated in the field.

We believe our answer to the previous point addresses these concerns - We have added references supporting the choice of steps and tools used in scFlow in the supplementary text which should address the reviewer’s concerns about justifying the analysis pipeline. Moreover, as a result of the pipeline we identified that 16,000 of the nuclei kept by the authors are likely of low quality and indicative of cell death with high mitochondrial read proportions5,6.

They also worry that the significant findings in Mathys' paper are influenced by the number of cells of each type. I'm sure it is since power is a function of sample size, but is this a bad thing? It seems odd that their approach is not influenced by sample size.

We thank the reviewer for highlighting this point. As they noted, we conclude that the original authors number of DEGs is just a product of the number of cells. However, the reviewer states that ‘It seems odd that their approach is not influenced by sample size’. An increase in the number of cells is not an increase in sample size since these cells are not independent from one another - they come from the same sample. Therefore, an increase in the number of cells should not result in an increase in the number of DEGs whereas an increase in the number of samples would. This point is the major issue with pseudoreplication approaches which over-estimate the confidence when performing differential expression due to the statistical dependence between cells from the same patient not being considered. See these references for more information on this point1,2,7,8. We have added a discussion of this point to our manuscript in paragraph 6.

Moreover, recent work has established that the genetic risk for Alzheimer’s disease acts primarily via microglia9,10. Thus, it would be reasonable to expect that the majority of large effect size DEGs identified would be found in this cell type. This is what we found with our pseudobulk differential expression approach – 96% of all DEGs were in microglia. We have updated the text of our manuscript (paragraph 5) to highlight this last point.

References 1. Murphy, A. E. & Skene, N. G. A balanced measure shows superior performance of pseudobulk methods in single-cell RNA-sequencing analysis. Nat. Commun. 13, 7851 (2022).

1. Squair, J. W. et al. Confronting false discoveries in single-cell differential expression. Nat. Commun. 12, 5692 (2021).

2. Crowell, H. L. et al. muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data. Nat. Commun. 11, 6077 (2020).

3. Soneson, C. & Robinson, M. D. Bias, robustness and scalability in single-cell differential expression analysis. Nat. Methods 15, 255–261 (2018).

4. Ilicic, T. et al. Classification of low quality cells from single-cell RNA-seq data. Genome Biol. 17, 29 (2016).

5. Heumos, L. et al. Best practices for single-cell analysis across modalities. Nat. Rev. Genet. 24, 550–572 (2023).

6. Zimmerman, K. D., Espeland, M. A. & Langefeld, C. D. A practical solution to pseudoreplication bias in single-cell studies. Nat. Commun. 12, 738 (2021).

7. Lazic, S. E. The problem of pseudoreplication in neuroscientific studies: is it affecting your analysis? BMC Neurosci. 11, 5 (2010).

8. Skene, N. G. & Grant, S. G. N. Identification of Vulnerable Cell Types in Major Brain Disorders Using Single Cell Transcriptomes and Expression Weighted Cell Type Enrichment. Front. Neurosci. 0, (2016).

9. McQuade, A. & Blurton-Jones, M. Microglia in Alzheimer’s disease: Exploring how genetics and phenotype influence risk. J. Mol. Biol. 431, 1805–1817 (2019).

eLife assessment

This paper reports a useful finding on the impact of choices of quality control and differential analysis methods on the discovery of disease-associated gene expression signatures. The study provides a solid comparison of the data process by re-analysis of a large-scale snRNA-seq dataset for Alzheimer's Disease. This paper would be of interest to the community as to rigorous analyses for large-scale single cell datasets.

Joint Public Review:

Murphy, Fancy and Skene performed a reanalysis of snRNA-seq data from Alzheimer Disease (AD) patients and healthy controls published previously by Mathys et al. (2019), arriving at the conclusion that many of the transcriptional differences described in the original publication were false positives. This was achieved by revising the strategy for both quality control and differential expression analysis. With this re-analysis, the authors aim to raise awareness of the impact of data analysis choices for scRNA-seq data and to caution focus on putatively wrongly identified genes in the AD research community. The revised manuscript has been improved by separating QC and DE analysis, which makes interpretation of both steps more straightforward.

STRENGTHS:

The authors demonstrate that the choice of data analysis strategy can have a vast impact on the results of a study, which in itself may not be obvious to many researchers.

The authors apply a pseudobulk-based differential expression analysis strategy (essentially, adding up counts from all cells per individual and comparing those counts with standard RNA-seq differential expression tests), which is (a) in line with latest community recommendations, (b) different from the "default options" in most popular scRNA-seq analysis suites, and (c) explains the vastly different number of DEGs identified by the authors and the original publication. The recommendation of this approach together with a detailed assessment of the DEGs found by both methodologies could potentially be a useful finding for the research community. Unfortunately, it is currently not sufficiently substantiated.

All code and data used in this study are publicly available to the readers.

WEAKNESSES:

The authors interpret the fact that they found fewer DEGs with their method than the original paper as a good thing by making the assumption that all genes that were not found were false positives. However, they do not prove this, and it is likely that at least some genes were not found due to a lack of statistical power and not because they were actually "incorrect". The original paper also had performed independent validations of some genes that were not found here. I had raised this weakness in my first review, but it was not explicitly addressed and still pertains to the revised manuscript. The authors have added an analysis that shows that "pseudoreplication" is prone to false positive (FP) discoveries for high cell numbers (Fig. 1f), but this does not prove that all of Mathys' DEGs were wrong.

I am concerned that almost all DEGs found by the authors are in the rare cell types, foremost the rare microglia (see Fig. 1e). Indeed, there is a weak negative correlation between cell counts and numbers of DEGs (Fig. 1e), if the correlation analysis is to be believed (see next point). It is unclear to me how many cells the pseudo-bulk counts were based on for these cell types, but it seems that (a) there were few and (b) there were quite few reads per cells. If both are the case, the pseudobulk counts for these cell populations might be rather noisy and the DEG results liable to outliers with extreme fold changes. Supp. Fig. 3b now shows three examples of DEGs, of which one (EGR1) looks like the DE call is indeed largely driven by four outliers, while Supp. Fig 3a shows at least one gene (BEX1) that could be FP of the pseudobulk approach due to insufficient statistical power. The authors go on to cite two papers (one is their own, published in a journal with suspected lack of appropriate quality assurance measures https://predatoryreports.org/the-predatory-journals-1), to support that the finding of DEGs in microglia "makes more sense" (l. 127). In summary, neither the presented examples nor the supporting literature are convincing. Lastly, the authors even show themselves that their approach is liable to FPs if applied with very low cell numbers in the range of those for microglia and OPCs (Fig. 1g).

The correlation analysis between cell counts and number of DEGs found is weak. In all three cases (Fig. 1c, d, e) the correlation is largely driven by a single outlier data point.

The authors claim they improved the quality control of the dataset but offer no objective metric to assess this putative improvement. The authors' QC procedure removes some 20k cells that had not been filtered out by Mathys' et al. As the authors state themselves, this difference is mostly due to the removal of cells with a high mitochondrial read content. Murphy et al use a fixed threshold for the mitochondrial percentage of reads, while the original paper had removed cell clusters with an "abnormally high" mitochondrial read fraction. That also seems reasonable, given that some cells might have a higher mitochondrial read content for reasons other than being "low quality". Simply stating that Mathys' approach was ineffective at removing cells with high mitochondrial read content is a self-fulfilling prophecy given the difference in approach, and itself not proof that the original QC procedure was inferior.

Batch correction: "Dataset integration has become a common step in single-cell RNA-Seq protocols and is recommended to remove confounding sources of variation" (l. 38). While it is true that many authors now choose to perform an integration step as part of their analysis workflow, this is by no means uncontroversial as there is a risk of "over-integration" and loss of true biological differences. I had raised this point previously, but the authors chose not to address it (quoted text and line numbers updated). Given that there is controversy in the literature and "community opinion" on the topic of data integration, this is another example of the authors claiming superiority in analysis without showing proof.

Due to a lack of comparison with other methods and due to the fact that the author's methodology was only applied to a single dataset, the paper presents merely a case study, which could be useful but falls short of providing a general recommendation for a best practice workflow.

APPRAISAL:

The manuscript could help to increase awareness of data analysis choices in the community, but only if the superiority of the methodology was clearly demonstrated. However, the authors only show that there are differences but have no convincing (orthogonal) evidence that their methodology was indeed better. This applies to both QC and DE analysis.

Reviewer #1 (Public Review):

This work provides a new dataset of 71,688 images of different ape species across a variety of environmental and behavioral conditions, along with pose annotations per image. The authors demonstrate the value of their dataset by training pose estimation networks (HRNet-W48) on both their own dataset and other primate datasets (OpenMonkeyPose for monkeys, COCO for humans), ultimately showing that the model trained on their dataset had the best performance (performance measured by PCK and AUC). In addition to their ablation studies where they train pose estimation models with either specific species removed or a certain percentage of the images removed, they provide solid evidence that their large, specialized dataset is uniquely positioned to aid in the task of pose estimation for ape species.

The diversity and size of the dataset make it particularly useful, as it covers a wide range of ape species and poses, making it particularly suitable for training off the shelf pose estimation networks or for contributing to the training of a large foundational pose estimation model. In conjunction with new tools focused on extracting behavioral dynamics from pose, this dataset can be especially useful in understanding the basis of ape behaviors using pose.

Overall this work is a terrific contribution to the field, and is likely to have a significant impact on both computer vision and animal behavior.

Strengths:<br /> - Open source dataset with excellent annotations on the format, as well as example code provided for working with it<br /> - Properties of the dataset are mostly well described<br /> - Comparison to pose estimation models trained on humans vs monkeys, finding that models trained on human data generalized better to apes than the ones trained on monkeys, in accordance with phylogenetic similarity. This provides evidence for an important consideration in the field: how well can we expect pose estimation models to generalize to new species when using data from closely or distantly related ones.<br /> - Sample efficiency experiments reflect an important property of pose estimation systems, which indicates how much data would be necessary to generate similar datasets in other species, as well as how much data may be required for fine tuning these types of models (also characterized via ablation experiments where some species are left out)<br /> - The sample efficiency experiments also reveal important insights about scaling properties of different model architectures, finding that HRNet saturates in performance improvements as a function of dataset size sooner than other architectures like CPMs (even though HRNets still perform better overall).

Reviewer #2 (Public Review):

The authors have used transcranial magnetic stimulation (TMS) and motor evoked potentials (MEPs) and TMS-electroencephalography (EEG) evoked potentials (TEPs) to determine how experimental heat pain could induce alterations these metrics.
In Experiment 1 thermal stimuli were administered over the forearm, with the first, second and third block of stimuli consisting of warm but non painful (pre-pain block), painful heat (pain block) and warm but non-painful (post-pain block) temperatures respectively. Painful stimuli led to an increase in the amplitude of the fronto-central N45, with a larger increase associated with higher pain ratings. Experiments 2 and 3 studied the correlation between the increase in the N45 in pain and the effects of a sham stimulation protocol/higher stimulation intensity. They found that the centro-frontal N45 TEP was decreased in acute pain. While their results are in line with reductions seen in motor evoked responses during pain and effort was made to address possible confounding factors (study 2 and 3). This study opens the way for the use exploration of cortical excitability outside M1 in acute pain, and potentially in chronic pain instances. While there is still open discussion on the best strategy to handle auditory and mechanical tactile noise, technological and methodological improvements seen in the last years have greatly improved the signal to noise ratio of TMS-EEG.

Reviewer #1 (Public Review):

Summary: The study provides valuable insights into the role of PfMORC in Plasmodium's epigenetic regulation, backed by a comprehensive methodological approach. The overarching goal was to understand the role of PfMORC in epigenetic regulation during asexual blood stage development, particularly its interactions with ApiAP2 TFs and its potential involvement in the regulation of genes vital for Plasmodium virulence. To achieve this, they conducted various analyses. These include a proteomic analysis to identify nuclear proteins interacting with PfMORC, a study to determine the genome-wide localization of PfMORC at multiple developmental stages, and a transcriptomic analysis in PfMORCHA-glmS knockdown parasites. Taken together, this study suggests that PfMORC is involved in chromatin assemblies that contribute to the epigenetic modulation of transcription during the asexual blood stage development.

Strengths: The study employed a multi-faceted approach, combining proteomic, genomic, and transcriptomic analyses, providing a holistic view of PfMORC's role. The proteomic analysis successfully identified several nuclear proteins that may interact with PfMORC. The genome-wide localization offered valuable insights into PfMORC's function, especially its predominant recruitment to subtelomeric regions. The results align with previous findings on PfMORC's interaction with ApiAP2 TFs. Notably, the authors meticulously contextualized their findings with prior research, including pre-prints, adding credibility to their work.

Weaknesses: While the study identifies potential interacting partners and loci of binding, direct functional outcomes of these interactions remain an inference. The authors heavily rely on past research for some of their claims. While it strengthens some assertions, it might indicate a lack of direct evidence in the current study for particular aspects. The declaration that PfMORC may serve as an attractive drug target is substantial. While the data suggests its involvement in essential processes, further studies are required to validate its feasibility as a drug target.

Reviewer #2 (Public Review):

Summary:<br /> This is a paper entitled "Plasmodium falciparum MORC protein modulates gene expression through interaction with heterochromatin" describes the role of PfMORC during the intra-erythrocytic cycle of Plasmodium falciparum. Garcia et al. investigated the PfMORC-interacting proteins and PfMORC genomic distribution in trophozoites and schizonts. They also examined the transcriptome of the parasites after partial knockdown of the transcript.

Strengths:<br /> This study is a significant advance in the knowledge of the role of PfMORC in heterochromatin assembly. It provides an in-depth analysis of the PfMORC genomic localisation and its correlation with other chromatin marks and ApiAP2 transcription factor binding.

Weaknesses:<br /> However, most of the conclusions are based on the function of interacting proteins and the genomic localisation of the protein. The authors did not investigate the direct effects of PfMORC depletion on heterochromatin marks. Furthermore, the results of the transcriptomic analysis are puzzling as 50% of the transcripts are downregulated, a phenotype not expected for a heterochromatin marker.

eLife assessment

This important study investigates the structural organization of a series of diblock elastin-like polypeptide condensates. The methodology is highly compelling, as it combines multiscale simulations and fluorescence lifetime imaging microscopy experiments. The results increase our understanding of model biomolecular condensates. The manuscript would benefit from more details for the concepts and terminology introduced, as well as the analysis of the simulations.

Reviewer #1 (Public Review):

This is an interesting, informative, and well-designed study that combines theoretical and experimental methodologies to tackle the phenomenon of higher-resolution structures/substructures in model biomolecular condensates. The results should be published. However, there is significant room for improvement in the presentation and interpretation of the results. As it stands, the precise definition of "frustration," which is a main theme of this manuscript (as emphasized in the title), is not sufficiently well articulated. This situation should be rectified to avoid "frustration" becoming a "catch-all" term without a clear perimeter of applicability rather than a precise, informative description of the physical state of affairs. There are also a few other concerns, e.g., regarding interpretation of correlation of phase-separation critical temperature and transfer free energy of amino acid residues as well as the difference between critical temperature and onset temperature, and the way the simulated configurations are similar to that of gyroids. Accordingly, the manuscript should be revised to address the following:

1. It is accurately pointed out on p.4 that elastin-like polypeptides (ELPs) undergo heat-induced phase separation and therefore exhibit lower critical solution temperatures (LCSTs). But it is not entirely clear how this feature is reproduced by the authors' simulation. A relationship between simulated surface tension and "transition temperature" is provided in Fig.1C; but is the "transition temperature" (authors cited ref.41 by Urry) the same as critical temperature? Apparently, Urry's Tt is "critical onset temperature", the temperature when phase separation happens at a given polymer concentration. This is different from the (global) critical temperature LCST - though the two may be correlated-or not-depending on the shape of the phase boundary. Moreover, is the MOFF coarse-grained forcefield (first step in the multi-scale simulation), by itself, capable of reproducing heat-induced phase separation in a way similar to the forcefield of Dignon et al., ACS Cent Sci 5, 821-230 (2019)? Or is this temperature-dependent effect appearing only subsequently, after the implementation of the MARTINI and/or all-atom steps? Clarification is needed. To afford a more informative context for the authors' introductory discussion, the aforementioned Dignon et al. work and the review by Cinar et al. [Chem Eur J 25, 13049-13069 (2019)], both touching upon the physical underpinning of the LCST feature of elastin, should also be cited along with refs.41-43.

2. "Frustration" and "frustrated" are used prominently in the manuscript to characterize certain observed molecular configurations (11 times total, in both the title and in the abstract). Apparently, it is the most significant conceptual pronouncement of this work, hence its precise meaning is of central importance to the authors' thesis. Whereas one should recognize that the theoretical and experimental observations are striking without invocation of the "frustration" terminology, usage of the term can be useful if it offers a unifying conceptual framework. However, as it stands, a clear definition of the term "frustration" is lacking, leaving readers to wonder what molecular configurations are considered "frustrated" and what are not (i.e., is the claim of observation of frustration falsifiable?). For instance, "frustrated microphase separation" appears in both the title and abstract. A logical question one may ask is: "Are all microphase separations frustrated"? If the answer is in the affirmative, does invocation of the term "frustration" add anything to our physical insight? If the answer is not in the affirmative, then how does one distinguish between microphase separations that are frustrated from those that are not frustrated? Presumably all simulated and experimental molecular configurations in the present study are those of lowest free energy for the given temperature. In other words, they are what they are. In the discussion about frustrated phase separation on p.13, for example, the authors appear to refer to the fact that chain connectivity is preventing hydrophobic residues to come together in a way to achieve the most favorable interactions as if there were no chain connectivity (one may imagine in that case all the hydrophobic residues will form a large cluster without microphase separation). Is this what the authors mean by "frustration"? If that's true, isn't that merely stating the obvious, at least for the observed microphase separation? In general, does "frustration" always mean deviation of actual, physical molecular configurations from certain imagined/hypothetical/reference molecular configurations, and therefore dependent upon the choice of the imagined reference configuration? If this is how the authors apply the term "frustration" in the present work, what is the zero-frustration reference state/configuration for microphase separation? And, similarly, what is the zero-frustration reference state/configuration when frustrated EPS-water interactions are discussed (~p.14-p.15, Fig.5)? How do non-frustrated water-protein interactions look like? Is the classic clathrate-like organization of water hydrogen bonds around small nonpolar solute "frustrated"?

3. In the discussion about the correlation of various transfer free energy scales for amino acids and Urry's critical onset temperature (ref.41) on p.11 and Fig.4, is there any theoretical relationship to be expected between the interactions among amino acids of ELPs and their critical onset temperatures? While a certain correlation may be intuitively expected if the free energy scale "is working", is there any theoretical insight into the mathematical form of this relationship? A clarifying discussion is needed because it bears logically on whether the observed correlation or lack thereof for different transfer energy scales is a good indication of the adequacy of the energy scales in describing the actual physical interactions at play. This question requires some prior knowledge of the expected mathematical relationship between interaction parameters and onset temperature.

4. To provide a more comprehensive context for the present study, it is useful to compare the microphase separation seen in the authors' simulation with the micelle-like structures observed in recent simulated condensed/aggregated states of hydrophobic-polar (HP) model sequences in Statt et al., J Chem Phys 152, 075101 (2020) [see esp. Fig.6] and Wessén et al., J Phys Chem B 126, 9222-9245 (2022) [see, e.g., Fig.10].

5. "Gyroid-like morphology" is mentioned several times in the manuscript (p.4, p.8, p.17, Fig.S3). This is apparently an interesting observation, but a clear explanation is lacking. A more detailed and specific discussion, perhaps with additional graphical presentations, should be provided to demonstrate why the simulated condensed-phase ELP configurations are similar to the classical description of gyroid as in, e.g., Terrones & Mackay, Chem Phys Lett 207, 45-50 (1993) and Lambert et al., Phil Trans R Soc A 354, 2009-2023 (1996).

Reviewer #2 (Public Review):

Summary:<br /> Latham A.P. et al. apply simulations and FLIM to analyse several di-block elastin-like polypetides and connect their sequence to the micro-structure of coacervates resulting from their phase-separation.

Strengths:<br /> Understanding the molecular grammar of phase separating proteins and the connection with mesoscale properties of the coacervates is highly relevant. This work provides insights into micro-structures of coacervates resulting from di-block polypetides.

Weaknesses:<br /> The results apply to a very specific architecture (di-block polypetides) with specific sequences.

eLife assessment

This valuable study reports on the structure and function of capsid size-determining external scaffolding protein encoded by a Vibrio phage satellite. The structural work is of high quality and the presented reconstructions are compelling. The paper offers a substantial advance in the field of phage and virus structure and assembly, with implications for understanding the evolution of phage satellites.

Reviewer #1 (Public Review):

This paper describes the discovery, functional analysis and structure of TcaP, a protein encoded by the Vibrio phage satellite PLE, that forms a size-determining scaffold around PLE procapsids made from helper phage ICP1 structural proteins.

The system displays a fascinating similarity to the P2/P4 system, which had previously been unique in its use of a dominant, size-determining external scaffolding protein (Sid). An interesting observation is that PLE appears to be dependent on small capsids for efficient transduction, a phenomenon not previously seen in headful packaging phage/satellite pairs. It is not clear why this is the case.

The work is interesting, comprehensive and of high quality. The reconstruction and modeling statistics are good; unfortunately, although the map has clear alpha-helical density around the threefold axes, the TcaP model does not include this critical region. The comparison to Sid provides an illustration of probable convergent evolution.

The paper constitutes an important contribution to the field of phage and virus structure and assembly, with implications for understanding the evolution of phage satellites and for macromolecular assembly processes in general.

Reviewer #2 (Public Review):

Phage satellites are fascinating elements that have evolved to hijack phages for induction, packaging, and transfer, promoting their widespread dissemination in nature. It is remarkable how different satellites use conserved strategies of parasitism, utilising unrelated proteins that perform similar roles in their cognate elements. In the current manuscript, Dr. Seed and coworkers elucidated the mechanism used by one family of satellites, the PLEs, to produce small capsids, a process that inhibits phage reproduction while increasing PLE transmission. The work is presented beautifully, and the results are astonishing. The authors identified the gene responsible for generating the small capsids, characterised its role in the PLE transfer and phage inhibition, and determined the structure of the PLE-sized small capsids. It is a truly impressive piece of work.

Reviewer #3 (Public Review):

The manuscript by Boyd and co-authors "A Vibrio cholerae viral satellite maximizes its spread and inhibits phage by remodelling hijacked phage coat proteins into small capsids" reports important results related to self-defending mechanisms that bacteria are used against phages that infect them. It has been shown previously that bacteria produce phage-inducible chromosomal island-like elements (PLE) that encode proteins that are integrated into bacterial genome. These proteins are used by bacteria to amend the phage capsids and to create phage-like particles (satellites) that move between cells and transfer the genetic material of PLE to another bacteria. That study highlights the interactions between a PLE-encoded protein, TcaP, and capsid proteins of the phage ICP1.

The manuscript is well written, provides a lot of new information and the results are supported by biochemical analysis.

eLife assessment

This manuscript is a valuable study of the responses of GPi neurons to DBS stimulation in human PD and dystonia patients and it finds evidence for altered short-term and long-term plasticity in response to DBS between the two patient populations. This data set is of interest to both basic and clinical researchers working in the field of DBS and movement disorders. While there was enthusiasm for the potential significance of these findings, support for their conclusions was incomplete. Their data may be indicative of more interesting and complex interpretations than currently considered in the article.

Reviewer #1 (Public Review):

Summary:<br /> Sumarac et al investigate differences in globus pallidus internus (GPi) spike activity and short- and long-term plasticity of direct pathway projections in patients with Parkinson's disease (PD) and dystonia. Their main claims are that GPi neurons exhibit distinct characteristics in these two disorders, with PD associated with specific power-frequency oscillations and dystonia showing lower firing rates, increased burstiness, and less regular activity. Additionally, long-term plasticity and synaptic depression appear to differ between the two conditions. The authors suggest that these findings support the concept of hyperfunctional GPi output in PD and hypofunctional output in dystonia, possibly driven by variations in the plasticity of striato-pallidal synapses. Overall enthusiasm is relatively high, but I think the discussion omits discussing findings that don't align well with standard models.

Strengths:<br /> These types of studies are valuable as the data arise from patients who have dystonia or PD. This could provide unique insights into disease pathophysiology that might not be recapitulated in animal systems work.

Weaknesses:<br /> - The rate model and indirect/direct pathway ideas lack explanatory power; too much of the hypothesis generation and discussion in this manuscript is set in the context of these old ideas. Their data in my view emphasize this somewhat emphatically. Most patients with the 'hypokinetic' movement disorder PD have dystonia as a part of their motor features. Dystonia is a form of excessive muscle activation that on the one hand is 'hyperkinetic' but on the other usually decreases the speed of motor tasks, even in patients with primary dystonia. Similarly, PD patients display a bewildering variety of hyperkinetic manifestations as well (rest tremor, dystonia, dyskinesia). If these are truly independent classifications, i.e. hyper- versus hypo-kinetic, the authors must acknowledge that there is considerable overlap in the spike activity across groups - numerous dystonia patients display higher discharge rates than the majority of the PD sample. Based on the firing rate alone, it would not be possible to distinguish these groups.

- If beta power is pathognomonic of parkinsonism, the authors found no differences in beta-related spike discharges across the groups. One would have predicted greater beta power in PD than in primary dystonia. This should be discussed explicitly and an interpretation should be provided.

- The study lacks a healthy control group, making it challenging to differentiate disease-specific findings from normal variations in GPi activity and plasticity. Although this is acknowledged in the discussion, this complicates the interpretation of the results. The sample sizes for PD and dystonia patients are relatively small, and the study combines various forms of dystonia, potentially masking subtype-specific differences. A larger and more homogenous sample could enhance the study's reliability.

- While they mention that data are available on request, sharing data openly would increase transparency and allow for independent validation of the results. It is unclear how sharing deidentified data would compromise patient privacy or present ethical issues of any kind, as claimed by the authors.

- They appropriately acknowledge several limitations, such as the inability to use pharmacological interventions and the need for further research in the chronic setting.

- The manuscript highlights differences in GPi activity and plasticity between PD and dystonia but could provide more context on the clinical implications of these findings, particularly regarding what the implications would be novel paradigms for deep brain stimulation.

- While statistical tests are mentioned, the manuscript could benefit from a more detailed presentation of statistical methods, including correction for multiple comparisons and effect sizes. Did the authors consider different recording sites within each patient as independent observations? I think this is not appropriate if that was the case.

- The manuscript could elaborate on the potential mechanisms underlying the observed differences in GPi activity and plasticity and their relevance to the pathophysiology of PD and dystonia.

Reviewer #2 (Public Review):

Summary:<br /> The authors investigated how neuronal activity and metrics of plasticity using local electrical stimulation in the GPi were different between Parkinson's disease and dystonia patients.

Strengths:<br /> The introduction highlights the importance of the work and the fundamental background needed to understand the rest of the paper. It also clearly lays out the novelty (i.e., that the dynamics of plastic effects in GPi between dystonia and PD have not been directly compared).

The methods are clearly described and the results are well organized in the figures.

The results are strong with measurements from a large population of patients for each disease group and with distinct findings for each group.

Weaknesses:<br /> The discussion was hard to follow in several places, making it difficult to fully appreciate how well the authors' claims and conclusions are justified by their data, mostly in relation to the plasticity results. It may help to summarize the relevant findings for each section first and then further expand on the interpretation, comparison with prior work, and broader significance. Currently, it is hard to follow each section without knowing which results are being discussed until the very end of the section. With the current wording in the "Neuronal correlates.." section, it is not always clear which results are from the current manuscript, and where the authors are referring to past work.

Also, I felt that more discussion could be used to highlight the significance of the current results by comparing and/or contrasting them to prior relevant work and mechanisms. The novelty or impact is not very clear as written. Could this be further substantiated in the Discussion?

Some specific comments and questions about the Discussion:<br /> Lines 209-211 - This sentence was hard to understand, could it be clarified?<br /> Lines 211-213 - What do phasic and tonic components mean exactly? Could this be specifically defined? Are there specific timescales (as referred to in Intro)?<br /> Lines 215-217 - It's not clear what was delayed in dystonia, and how the authors are trying to contrast this with the faster time course in PD. I think some of this is explained in the introduction, but could also be re-summarized here as relevant to the results discussed.<br /> Lines 223-224 - I'm not sure I follow the implication that network reorganization leads to delayed functional benefits. Could this be further elaborated?

Could the absence of a relationship between FR and disease in PD be discussed?

It wasn't very clear how the direct pathway can be attributed to plasticity changes if the GPi makes up both the direct and indirect pathways. Could this be further clarified?

The mechanism of short- and long-term plasticity as applied in the protocols used in this work are outlined in reference to previous citations [15, 16, 18]. Because this is a central aspect of the current work and interpreting the results, it was difficult to appreciate how these protocols provide distinct metrics of short and long-term plasticity in GPi without some explanation of how it applies to the current work and the specific mechanisms. It would also help to be able to better link how the results fit with the broader conclusions.

In the Conclusion, it was difficult to understand the sentence about microcircuit interaction (line 232) and how it selectively modulates the efficacy of target synapses. Some further explanation here would be helpful. Also, it was not clear how these investigations (line 237) provide cellular-level support for closed-loop targeting. Could the reference to closed-loop targeting also be further explained?

How is the burst index calculated (Methods)?

Figures and figure captions are missing some details:

Fig. 1 - What does shading represent?

Fig. 2 - Can the stimulation artifact be labeled so as not to be confused with the physiological signal? Is A representing the average of all patients or just one example? Are there confidence intervals for this data as it's not clear if the curves are significantly different or not (may not be important to show if just one example)? Same for D. What is being plotted in E? Is this the exponential fitted on data? Can this be stated in the figure citation directly so readers don't have to find it in the text, where it may not be directly obvious which figure the analyses are being applied towards?

What does shading here represent?

eLife assessment

The article has important scientific merit in the field of cardiovascular research and other fields where the design and rigor of scientific experiments is key for translation of preclinical research to clinical studies. This study holds convincing evidence that sheds light on the lack of progress in this area over the past decade, despite a substantial body of existing research. Although there is a need to re-evaluate the statistical test used, the descriptive paper outcomes serves as a compelling call to action for the wider scientific community.

Reviewer #1 (Public Review):

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

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

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

Reviewer #2 (Public Review):

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

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

Weaknesses<br /> The authors did not use the exact same journals as they did in the 2017 study. This makes comparing the results complicated. Also, they pulled papers from 2011 to 2021, and hence cannot assess the impact of their own prior paper.<br /> The authors write "the proportion of studies including animals of both biological sexes generally increased between 2011 and 2021, though not significantly (R2= 0.0762, F(1,9)= 0.742, p= 0.411 (corrected p=8.2". This statement is not rigorous because the regression result is not statistically significant. Their data supports neither a claim of an increase nor a decrease over time. A similar problem repeats several times in the remainder of their results presentation.<br /> I think the Introduction and the Discussion are somewhat repetitive and the wording could be reduced.

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

eLife assessment

This study presents a valuable finding on the identification of tumor-reactive T lymphocytes (TRLs) using paired single-cell sequencing and PDX models for cell therapy and marker selection in uveal melanoma treatment. The evidence supporting the claims of the authors is convincing, although the inclusion of detailed explanations of the results for a broader audience would have strengthened the study. The work will be of interest to clinicians and medical biologists working on uveal melanoma (UM).

Reviewer #1 (Public Review):

Summary:<br /> This work successfully identified and validated TRLs in hepatic metastatic uveal melanoma, providing new horizons for enhanced immunotherapy. Uveal melanoma is a highly metastatic cancer that, unlike cutaneous melanoma, has a limited effect on immune checkpoint responses, and thus there is a lack of formal clinical treatment for metastatic UM. In this manuscript, the authors described the immune microenvironmental profile of hepatic metastatic uveal melanoma by sc-RNAseq, TCR-seq, and PDX models. Firstly, they identified and defined the phenotypes of tumor-reactive T lymphocytes (TRLs). Moreover, they validated the activity of TILs by in vivo PDX modeling as well as in vitro co-culture of 3D tumorsphere cultures and autologous TILs. Additionally, the authors found that TRLs are mainly derived from depleted and late-activated T cells, which recognize melanoma antigens and tumor-specific antigens. Most importantly, they identified TRLs-associated phenotypes, which provide new avenues for targeting expanded T cells to improve cellular and immune checkpoint immunotherapy.

Strengths:<br /> Jonas A. Nilsson, et al. has been working on new therapies for melanoma. The team has also previously performed the most comprehensive genome-wide analysis of uveal melanoma available, presenting the latest insights into metastatic disease. In this work, the authors performed paired sc-RNAseq and TCR-seq on 14 patients with metastatic UM, which is the largest single-cell map of metastatic UM available. This provides huge data support for other studies of metastatic UM.

Weaknesses:<br /> Although the paper does have strengths in principle, the weaknesses of the paper are that these strengths are not directly demonstrated. That is, insufficient analyses are performed to fully support the key claims in the manuscript by the data presented. In particular：

The author's description of the overall results of the article should be logical, not just a description of the observed phenomena. For example, the presentation related to the results of TRLs lacked logic. In addition, the title of the article emphasizes the three subtypes of hepatic metastatic UM TRLs, but these three subtypes are not specifically discussed in the results as well as the discussion section. The title of the article is not a very comprehensive generalization and should be carefully considered by the authors.

The authors' claim that they are the first to use autologous TILs and sc-RNAseq to study immunotherapy needs to be supported by the corresponding literature to be more convincing. This can help the reader to understand the innovation and importance of the methodology. In addition, the authors argue that TILs from metastatic UM can kill tumor cells. This is the key and bridging point to the main conclusion of the article. Therefore, the credibility of this conclusion should be considered. Metastatic UM1 and UM9 remain responsive to autologous tumors under in vitro conditions with their autologous TILs. In contrast, UM22, also as a metastatic UM, did not respond to TIL treatment. In particular, the presence of MART1-responsive TILs. The reliability of the results obtained by the authors in the model of only one case of UM22 liver metastasis should be considered. The authors should likewise consider whether such a specific cellular taxon might also exist in other patients with metastatic UM, producing an immune response to tumor cells. The results would be more comprehensive if supported by relevant data.

In addition, the authors in that study used previously frozen biopsy samples for TCR-seq, which may be associated with low-quality sequencing data, high risk of outcome indicators, and unfriendly access to immune cell information. The existence of these problems and the reliability of the results should be considered. If special processing of TCR-seq data from frozen samples was performed, this should also be accounted for.

Reviewer #2 (Public Review):

Summary:<br /> The study's goal is to characterize and validate tumor-reactive T cells in liver metastases of uveal melanoma (UM), which could contribute to enhancing immunotherapy for these patients. The authors used single-cell RNA and TCR sequencing to find potential tumor-reactive T cells and then used patient-derived xenograft (PDX) models and tumor sphere cultures for functional analysis. They discovered that tumor-reactive T cells exist in activated/exhausted T cell subsets and in cytotoxic effector cells. Functional experiments with isolated TILs show that they are capable of killing UM cells in vivo and ex vivo.

Strengths:<br /> The study highlights the potential of using single-cell sequencing and functional analysis to identify T cells that can be useful for cell therapy and marker selection in UM treatment. This is important and novel as conventional immune checkpoint therapies are not highly effective in treating UM. Additionally, the study's strength lies in its validation of findings through functional assays, which underscores the clinical relevance of the research.

Weaknesses:<br /> The manuscript may pose challenges for individuals with limited knowledge of single-cell analysis and immunology markers, making it less accessible to a broader audience.

Author Response

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

eLife assessment

The findings of this article provide valuable information on the changes of cell clusters induced by chronic periodontitis. The observation of a new fibroblast subpopulation, named AG fibroblasts, is interesting, and the strength of evidence presented is solid.

We thank the Reviewing Editor and the Senior Editor for the positive assessment and strong support for our study.

Public Reviews:

Reviewer #1 (Public Review):

In this article, the authors found a distinct fibroblast subpopulation named AG fibroblasts, which are capable of regulating myeloid cells, T cells and ILCs, and proposed that AG fibroblasts function as a previously unrecognized surveillant to orchestrate chronic gingival inflammation in periodontitis. Generally speaking, this article is innovative and interesting.

We truly appreciate this public review.

Reviewer #2 (Public Review):

This study proposed the AG fibroblast-neutrophil-ILC3 axis as a mechanism contributing to pathological inflammation in periodontitis. In this study single-cell transcriptomic analysis was performed. But the signal mechanism behind them was not evaluated.

The authors achieved their aims, and the results partially support their conclusions.

We agree that we must conduct future studies to evaluate our hypothesis.

The mouse ligatured periodontitis models differ from clinical periodontitis in human, this study supplies the basis for future research in human.

This is an important subject. We have previously expressed a concern on the mouse ligature model that the microbial composition of the mouse ligature did not mirror the human oral microbial composition. Therefore, we developed the maxillary topical application (MTA) model, in which human oral biofilm was directly applied to the maxillary gingiva. In this study, the newly developed MTA model was further dissected by single cell RNA seq, which revealed that the extracellular substances of human oral biofilm might be an important trigger of gingival inflammation. RESULT has been revised.

Recommendations for the authors:

Reviewer #1 (Recommendations For The Authors):

I appreciate the authors' efforts. I think it would be much better to simplify INTRODUCTION.

INTRODUCTION has been simplified as suggested.

Reviewer #2 (Recommendations For The Authors):

1. Many host cells participate in immune responses, such as gingival epithelial cells. AG fibroblast is not the only cell involved in the immune response, and the weight of its role needs to be clarified. So the expression in the conclusion should be appropriate.

RESPONSE: We agree with this comment. Our study identified the AG fibroblast–neutrophil–ILC3 axis as a previously unrecognized mechanism which could play an additional role in the complex interplay between oral barrier immune cells.

1. The main results should be included in the Abstract.

Abstract has been revised.

---

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

We thank all reviewers for constructive critiques. We plan to perform new experiments and revise our manuscript accordingly. The text and Figures are currently undergoing the revision process. Below highlights our revision plan.

eLife assessment

The findings of this article provide valuable information on the changes of cell clusters induced by chronic periodontitis. The observation of a new fibroblast subpopulation, which was named as AG fibroblasts, was quite interesting, but needs further evidence. The strength of evidence presented is incomplete.

We discovered a new subpopulation of gingival fibroblasts, named AG fibroblasts, using non-biased single cell RNA sequencing (scRNA-seq) of mouse gingival samples undergoing the development of ligature-induced periodontitis. AG fibroblasts exhibited a unique gene expression profile: [1] constitutive expression of type XIV collagen; and [2] ligatureinduced upregulation of Toll-Like Receptors and their downstream signals as well as chemokines such as CXCL12. Thus, we have hypothesized that AG fibroblasts initially sense the pathological stress including oral microbial stimuli and secrete inflammatory signals through chemokine expression.

The current manuscript examined the relationship between AG fibroblasts and oral barrier immune cells focusing on the chemokines and other ligands derived from AG fibroblasts and their putative receptors in those immune cells. Using scRNA-seq data mining programs, our data demonstrated the compelling evidence that AG fibroblasts should play a critical role in orchestrating the oral barrier immunity, at least at the early stages of periodontal inflammation.

We agree that it is important to explore the functional/pathological role of AG fibroblasts. In this revision, we further investigated the role of TLRs in the pathogen sensing mechanism of AG fibroblasts. To accomplish this goal, we applied a newly developed mouse model in which mice were exposed to the maxillary topical application (MTA) of oral microbial pathogens without the ligature placement. With 1 hr exposure with human oral biofilm, not with planktonic microbiota, the mice maxillary tissue exhibited measurable degradation as evidenced by the activation of cathepsin K. To dissect the role of TLRs, we applied the putative stimulants of TLR9 and TLR2/4 using the discrete MTA model. The scRNA-seq from the MTA model revealed that the application of unmethylated CpG oligonucleotide and P. gingivalis lipopolysaccharide (LPS), respectively, induced the activation of chemokines by AG fibroblast.

The revised manuscript reported this critical data with the detailed information. As such the additional figures and corresponding results, discussion and materials & methods were included.

Public Reviews:

Reviewer #1 (Public Review):

In this article, the authors found a distinct fibroblast subpopulation named AG fibroblasts, which are capable of regulating myeloid cells, T cells and ILCs, and proposed that AG fibroblasts function as a previously unrecognized surveillant to orchestrate chronic gingival inflammation in periodontitis. Generally speaking, this article is innovative and interesting, however, there are some problems that need to be addressed to improve the quality of the manuscript.

We appreciate this comment. As suggested, we further investigated the surveillant function of AG fibroblasts by reanalyzing the scRNA-seq data for stress sensing receptors such as Toll-Like Receptors (TLR). In the revision, we addressed the role of TLR in the activation of AG fibroblasts using a newly developed mouse model employing the maxillary topical application (MTA) of putative TLR stimulants. The new information clearly demonstrated that AG fibroblasts play a pivotal role as the surveillant and translating the pathogenic stimulants to oral barrier inflammation through chemokine expression.

Reviewer #2 (Public Review):

This study proposed the AG fibroblast-neutrophil-ILC3 axis as a mechanism contributing to pathological inflammation in periodontitis. However, the immune response in the vivo is very complex. It is difficult to determine which is the cause and which is the result. This study explores the relevant issue from one dimension, which is of great significance for a deeper understanding of the pathogenesis of periodontitis. It should be fully discussed.

We appreciate this comment. We expanded the current understanding of oral immune signal communication in Discussion and highlight how AG fibroblast may fit to it. To address this question, we expanded our investigation in the pathological signal detection by AG fibroblasts by employing the newly developed maxillary topical application (MTA) model. The revised manuscript contains the new information and expanded the discussion in the context of complex immune response.

Reviewer #1 (Recommendations For The Authors):

Detailed comments are listed below:

Abstract：<br /> I am confused about the expression of "human periodontitis-like phenotype". How does the authors define this concept? Periodontitis is a complex disease, despite that alveolar bone resorption is a typical manifestation of periodontitis, its characteristics remain to be further studied. I hope the authors can provide some detailed information about this concept or describe it in another way.

This is an important comment. Radiographically, human periodontitis is diagnosed by alveolar bone resorption from the cervical region, not from root apex. To highlight this, we present dental radiographs of human periodontitis as supplementary information. However, we agree with this comment, our statement should be limited to alveolar bone resorption pattern in Rag2KO and Rag2gcKO mice. Abstract be revised accordingly.

Introduction:<br /> It is recommended to simplify the first to third paragraphs, and briefly explain the functions of various types of cells in different stages of periodontitis, as well as the role of different cluster markers play across the time course of periodontal inflammation development.

Following this recommendation, INTRODUCTION has been simplified.

Results:<br /> 1. It is recommended to add HE staining and immunohistochemistry staining to observe the inflammation, tissue damage, and repair status from 0 to 7 days, so that readers can understand cell phenotype changes corresponding to the periodontitis stage. The observation index can include inflammation and vascular related indicators.

As recommended, representative histological figures were included. We further performed new immunohistochemistry experiment of mouse gingival tissue (D0, D1, D3, D7) highlighting the infiltration of CD45+ immune cells. We found that inflammatory vascular formation in the H&E histology, which was highlighted. To characterize the tissue damage, the histological sections were stained by picrosirius red to highlight the change in collagen connective tissue of PDL and gingiva.

1. Figure 1A-1D can be placed in the supplementary figure.

Combining the new data above, Figure 1 was revised as suggested.

1. I suggest the authors to put the detection of the existence of AG fibroblasts before exploring its relationship with other types of cells.

2. The layout of the picture should be closely related to the topic of the article. It is recommended to readjust the layout of the picture. Figure 1 should be the detection of AG cells and their proportion changes from 0 to 7 days. In other figures, the authors can separately describe the proportion changes of myeloid cells, T cells and ILCs, and explored the association between AG fibroblasts and these cell types.

As suggested, the presentation order of Figures and text was revised to bring the information about AG fibroblasts first. The chemokine-receptor analysis was moved below.

1. Please provide the complete form of "KT" in Line 162.

KT fibroblasts (fibroblasts keeping typical phenotype) was described in the text.

Methods:<br /> It is recommended to separately list the statistical methods section. The statistical method used in the article should be one-way ANOVA.

A separate statistical method section is created. As pointed out, we used one-way ANOVA with post-hoc Tukey test (when multiple groups were compared).

Discussion:<br /> I suggest the authors remove Figures 3-6 from the discussion section. For example, in Line 283, "(Figure 3 and 4)" should be removed.

Revised as suggested.

Reference:<br /> Some information for the references is missing. For example, "Lin P, et al. Application of Ligature-Induced Periodontitis in Mice to Explore the Molecular Mechanism of Periodontal Disease. Int J Mol Sci 22, (2021)" should be "Lin P, et al. Application of Ligature-Induced Periodontitis in Mice to Explore the Molecular Mechanism of Periodontal Disease. Int J Mol Sci 22, 8900 (2021)". It is necessary to recheck all references.

The reference has been checked for the accuracy and the omission pointed out was corrected. Although we used EndNote program, we found some more inaccuracy in the references that were manually corrected. We appreciate your suggestion.

Reviewer #2 (Recommendations For The Authors):

1. Many host cells participate in immune responses, such as gingival epithelial cells. AG fibroblast is not the only cell involved in the immune response, and the weight of its role needs to be clarified. So the expression in the conclusion should be appropriate.

Following this critique, we revised INTRODUCTION, DISCUSSION and CONCLUSION, to highlight how AG fibroblasts function within a comprehensive immune response network.

1. This study cannot directly answer the issue of the relationship between periodontitis and systemic diseases.

We agree with this critique. We either deleted or de-emphasized the relationship between periodontitis and systemic diseases throughout the text.

Author response

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

We thank the editor for the eLife assessment and reviewers for their remaining comments. We will address them in this response.

First, we thank eLife for the positive assessment. Regarding the point of visual acuity that is mentioned in this assessment, we understand that this comment is made. It is not an uncommon comment when rodent vision is discussed. However, we emphasize that we took the lower visual acuity of rats and the higher visual acuity of humans into account when designing the human study, by using a fast and eccentric stimulus presentation for humans. As a result, we do not expect a higher discriminability of stimuli in humans. We have described this in detail in our Methods section when describing the procedure in the human experiment:

“We used this fast and eccentric stimulus presentation with a mask to resemble the stimulus perception more closely to that of rats. Vermaercke & Op de Beeck (2012) have found that human visual acuity in these fast and eccentric presentations is not significantly better than the reported visual acuity of rats. By using this approach we avoid that differences in strategies between humans and rats would be explained by such a difference in acuity”

Second, regarding the remaining comment of Reviewer #2 about our use of AlexNet:

While it is indeed relevant to further look into different computational architectures, we chose to not do this within the current study. First, it is a central characteristic of the study procedure that the computational approach and chosen network is chosen early on as it is used to generate the experimental design that animals are tested with. We cannot decide after data collection to use a different network to select the stimuli with which these data were collected. Second, as mentioned in our first response, using AlexNet is not a random choice. It has been used in many previously published vision studies that were relatively positive about the correspondence with biological vision (Cadieu et al., 2014; Groen et al., 2018; Kalfas et al., 2018; Nayebi et al., 2023; Zeman et al., 2020). Third, our aim was not to find a best DNN model for rat vision, but instead examining the visual features that play a role in our complex discrimination task with a model that was hopefully a good enough starting point. The fact that the designs based upon AlexNet resulted in differential and interpretable effects in rats as well as in humans suggests that this computational model was a good start. Comparing the outcomes of different networks would be an interesting next step, and we expect that our approach could work even better when using a network that is more specifically tailored to mimic rat visual processing.

Finally, regarding the choice to specifically chose alignment and concavity as baseline properties, this choice is probably not crucial for the current study. We have no reason to expect rats to have an explicit notion about how a shape is built up in terms of a part-based structure, where alignment relates to the relative position of the parts and concavity is a property of the main base. For human vision it might be different, but we did not focus on such questions in this study.

---

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

We would like to thank you for giving us the opportunity to submit a revised draft our manuscript. We appreciate the time and effort that you dedicated to providing insightful feedback on our manuscript and are grateful for the valuable comments and improvements on our paper. It helped us to improve our manuscript. We have carefully considered the comments and tried our best to address every one of them. We have added clarifications in the Discussion concerning the type of neural network that we used, about which visual features might play a role in our results as well as clarified the experimental setup and protocol in the Methods section as these two sections were lacking key information points.

Below we provide a response to the public comments and concerns of the reviewers.

Several key points were addressed by at least two reviewers, and we will respond to them first.

A first point concerns the type of network we used. In our study, we used AlexNet to simulate the ventral visual stream and to further examine rat and human performance. While other, more complex neural networks might lead to other results, we chose to work with AlexNet because it has been used in many other vision studies that are published in high impact journals ((Cadieu et al., 2014; Groen et al., 2018; Kalfas et al., 2018; Nayebi et al., 2023; Zeman et al., 2020). We did not try to find a best DNN model for rat vision but instead, we were looking for an explanation of which visual features play a role in our complex discrimination task. We added a consideration to our Discussion addressing why we worked with AlexNet. Since our data will be published on OSF, we encourage to researchers to use our data with other, more complex neural networks and to further investigate this issue.

A second point that was addressed by multiple reviewers concerns the visual acuity of the animals and its impact on their performance. The position of the rat was not monitored in the setup. In a previous study in our lab (Crijns & Op de Beeck, 2019), we investigated the visual acuity of rats in the touchscreen setups by presenting gratings with different cycles per screen to see how it affects their performance in orientation discrimination. With the results from this study and general knowledge about rat visual acuity, we derived that the decision distance of rats lies around 12.5cm from the screen. We have added this paragraph to the Discussion.

A third key point that needs to be addressed as a general point involves which visual features could explain rat and human performance. We reported marked differences between rat and human data in how performance varied across image trials, and we concluded through our computationally informed tests and analyses that rat performance was explained better by lower levels of processing. Yet, we did not investigate which exact features might underlie rat performance. As a starter, we have focused on taking a closer look at pixel similarity and brightness and calculating the correlation between rat/human performance and these two visual features.

We calculated the correlation between the rat performances and image brightness of the transformations. We did this by calculating the difference in brightness of the base pair (brightness base target – brightness base distractor), and subtracting the difference in brightness of every test target-distractor pair for each test protocol (brightness test target – brightness test distractor for each test pair). We then correlated these 287 brightness values (1 for each test image pair) with the average rat performance for each test image pair. This resulted in a correlation of 0.39, suggesting that there is an influence of brightness in the test protocols. If we perform the same correlation with the human performances, we get a correlation of -0.12, suggesting a negative influence of brightness in the human study.

We calculated the correlation between pixel similarity of the test stimuli in relation to the base stimuli with the average performance of the animals on all nine test protocols. We did this by calculating the pixel similarity between the base target with every other testing distractor (A), the pixel similarity between the base target with every other testing target (B), the pixel similarity between the base distractor with every other testing distractor (C) and the pixel similarity between the base distractor with every other testing target (D). For each test image pair, we then calculated the average of (A) and (D), and subtracted the average of (C) and (B) from it. We correlated these 287 values (one for each image pair) with the average rat performance on all test image pairs, which resulted in a correlation of 0.34, suggesting an influence of pixel similarity in rat behaviour. Performing the same correlation analysis with the human performances results in a correlation of 0.12.

We have also addressed this in the Discussion of the revised manuscript. Note that the reliability of the rat data was 0.58, clearly higher than the correlations with brightness and pixel similarity, thus these features capture only part of the strategies used by rats.

We have also responded to all other insightful suggestions and comments of the reviewers, and a point-by-point response to the more major comments will follow now.  

Reviewer #1, general comments:

The authors should also discuss the potential reason for the human-rat differences too, and importantly discuss whether these differences are coming from the rather unusual approach of training used in rats (i.e. to identify one item among a single pair of images), or perhaps due to the visual differences in the stimuli used (what were the image sizes used in rats and humans?). Can they address whether rats trained on more generic visual tasks (e.g. same-different, or category matching tasks) would show similar performance as humans?

The task that we used is typically referred to as a two-alternative forced choice (2AFC). This is a simple task to learn. A same-different task is cognitively much more demanding, also for artificial neural networks (see e.g. Puebla & Bowers, 2022, J. Vision). A one-stimulus choice task (probably what the reviewer refers to with category matching) is known to be more difficult compared to 2AFC, with a sensitivity that is predicted to be Sqrt(2) lower according to signal detection theory (MacMillan & Creelman, 1991). We confirmed this prediction empirically in our lab (unpublished observations). Thus, we predict that rats perform less good in the suggested alternatives, potentially even (in case of same-different) resulting in a wider performance gap with humans.

I also found that a lot of essential information is not conveyed clearly in the manuscript. Perhaps it is there in earlier studies but it is very tedious for a reader to go back to some other studies to understand this one. For instance, the exact number of image pairs used for training and testing for rats and humans was either missing or hard to find out. The task used on rats was also extremely difficult to understand. An image of the experimental setup or a timeline graphic showing the entire trial with screenshots would have helped greatly.

All the image pairs used for training and testing for rats and humans are depicted in Figure 1 (for rats) and Supplemental Figure 6 (for humans). For the first training protocol (Training), only one image pair was shown, with the target being the concave object with horizontal alignment of the spheres. For the second training protocol (Dimension learning), three image pairs were shown, consisting of the base pair, a pair which differs only in concavity, and a pair which differs only in alignment. For the third training protocol (Transformations) and all testing protocols, all combination of targets and distractors were presented. For example, in the Rotation X protocol, the stimuli consisted of 6 targets and 6 distractors, resulting in a total of 36 image pairs for this protocol. The task used on rats is exactly as shown in Figure 1. A trial started with two blank screens. Once the animal initiated a trial by sticking its head in the reward tray, one stimulus was presented on each screen. There was no time limit and so the stimuli remained on the screen until the animal made a decision. If the animal touched the target, it received a sugar pellet as reward and a ITI of 20s started. If the animal touched the distractor, it did not receive a sugar pellet and a time-out of 5s started in addition to the 20s ITI.

We have clarified this in the manuscript.

The authors state that the rats received random reward on 80% of the trials, but is that on 80% of the correctly responded trials or on 80% of trials regardless of the correctness of the response? If these are free choice experiments, then the task demands are quite different. This needs to be clarified. Similarly, the authors mention that 1/3 of the trials in a given test block contained the old base pair - are these included in the accuracy calculations?

The animals receive random reward on 80% on all testing trials with new stimuli, regardless of the correctness of the response. This was done to ensure that we can measure true generalization based upon learning in the training phase, and that the animals do not learn/are not trained in these testing stimuli. For the trials with the old stimuli (base pair), the animals always received real reward (reward when correct; no reward in case of error).

The 1/3rd trials with old stimuli are not included in the accuracy calculations but were used as a quality check/control to investigate which sessions have to be excluded and to assure that the rats were still doing the task properly. We have added this in the manuscript.

The authors were injecting noise with stimuli to cDNN to match its accuracy to rat. However, that noise potentially can interacted with the signal in cDNN and further influence the results. That could generate hidden confound in the results. Can they acknowledge/discuss this possibility?

Yes, adding noise can potentially interact with the signal and further influence the results. Without noise, the average training data of the network would lie around 100% which would be unrealistic, given the performances of the animals. To match the training performance of the neural networks with that of the rats, we added noise 100 times and averaged over these iterations (cfr. (Schnell et al., 2023; Vinken & Op de Beeck, 2021)).  

Reviewer #2, weaknesses:

1) There are a few inconsistencies in the number of subjects reported. Sometimes 45 humans are mentioned and sometimes 50. Probably they are just typos, but it's unclear.

Thank you for your feedback. We have doublechecked this and changed the number of subjects where necessary. We collected data from 50 human participants, but had to exclude 5 of them due to low performance during the quality check (Dimension learning) protocols. Similarly, we collected data from 12 rats but had to exclude one animal because of health issues. All these data exclusion steps were mentioned in the Methods section of the original version of the manuscript, but the subject numbers were not always properly adjusted in the description in the Results section. This is now corrected.

2) A few aspects mentioned in the introduction and results are only defined in the Methods thus making the manuscript a bit hard to follow (e.g. the alignment dimension), thus I had to jump often from the main text to the methods to get a sense of their meaning.

Thank you for your feedback. We have clarified some aspects in the Introduction, such as the alignment dimension.

4) Many important aspects of the task are not fully described in the Methods (e.g. size of the stimuli, reaction times and basic statistics on the responses).

We have added the size of the stimuli to the Methods section and clarified that the stimuli remained on the screen until the animals made a choice. Reaction time in our task would not be interpretable given that stimuli come on the screen when the animal initiates a trial with its back to the screen. Therefore we do not have this kind of information.

Reviewer #1

• Can the authors show all the high vs zero and zero vs high stimulus pairs either in the main or supplementary figures? It would be instructive to know if some other simple property covaried between these two sets.

In Figure 1, all images of all protocols are shown. For the High vs. Zero and Zero vs. High protocols, we used a deep neural network to select a total of 7 targets and 7 distractors. This results in 49 image pairs (every combination of target-distractor).

• Are there individual differences across animals? It would be useful for the authors to show individual accuracy for each animal where possible.

We now added individual rat data for all test protocols – 1 colour per rat, black circle = average. We have added this picture to the Supplementary material (Supplementary Figure 1).

• Figure 1 - it was not truly clear to me how many image pairs were used in the actual experiment. Also, it was very confusing to me what was the target for the test trials. Additionally, authors reported their task as a categorisation task, but it is a discrimination task.

Figure 1 shows all the images that were used in this study. Every combination of every target-distractor in each protocol (except for Dimension learning) was presented to the animals. For example in Rotation X, the test stimuli as shown in Fig. 1 consisted of 6 targets and 6 distractors, resulting in a total of 36 image pairs for this test protocol.

In each test protocol, the target corresponded to the concave object with horizontally attached spheres, or the object from the pair that in the stimulus space was closed to this object. We have added this clarification in the Introduction: “We started by training the animals in a base stimulus pair, with the target being the concave object with horizontally aligned spheres. Once the animals were trained in this base stimulus pair, we used the identity-preserving transformations to test for generalization.” as well as in the caption of Figure 1. We have changed the term “categorisation task” to “discrimination task” throughout the manuscript.

• Figure 2 - what are the red and black lines? How many new pairs are being tested here? Panel labels are missing (a/b/c etc)

We have changed this figure by adding panel labels, and clarifying the missing information in the caption. All images that were shown to the animals are presented on this figure. For Dimension Learning, only three image pairs were shown (base pair, concavity pair, alignment pair) and for the Transformations protocol, every combination of every target and distractor were shown, i.e. 25 image pairs in total.

• Figure 3 - last panel: the 1st and 2nd distractor look identical.

We understand your concern as these two distractors indeed look quite similar. They are different however in terms of how they are rotated along the x, y and z axes (see Author response image 1 for a bigger image of these two distractors). The similarity is due to the existence of near-symmetry in the object shape which causes high self-similarity for some large rotations.

Author response image 1.

• Line 542 – authors say they have ‘concatenated’ the performance of the animals, but do they mean they are taking the average across animals?

It is both. In this specific analysis we calculated the performance of the animals, which was indeed averaged across animals, per test protocol, per stimulus pair. This resulted in 9 arrays (one for each test protocol) of several performances (1 for each stimulus pair). These 9 arrays were concatenated by linking them together in one big array (i.e. placing them one after the other). We did the same concatenation with the distance to hyperplane of the network on all nine test protocols. These two concatenated arrays with 287 values each (one with the animal performance and one with the DNN performance) were correlated.

• Line 164 - What are these 287 image pairs - this is not clear.

The 287 image pairs correspond to all image pairs of all 9 test protocols: 36 (Rotation X) + 36 (Rotation Y) + 36 (Rotation Z) + 4 (Size) + 25 (Position) + 16 (Light location) + 36 (Combination Rotation) + 49 (Zero vs. high) + 49 (High vs. zero) = 287 image pairs in total. We have clarified this in the manuscript.

• Line 215 - Human rat correlation (0.18) was comparable to the best cDNN layer correlation. What does this mean?

The human rat correlation (0.18) was closest to the best cDNN layer - rat correlation (about 0.15). In the manuscript we emphasize that rat performance is not well captured by individual cDNN layers.  

Reviewer #2

Major comments

• In l.23 (and in the methods) the authors mention 50 humans, but in l.87 they are 45. Also, both in l.95 and in the Methods the authors mention "twelve animals" but they wrote 11 elsewhere (e.g. abstract and first paragraph of the results).

In our human study design, we introduced several Dimension learning protocols. These were later used as a quality check to indicate which participants were outliers, using outlier detection in R. This resulted in 5 outlying human participants, and thus we ended with a pool of 45 human participants that were included in the analyses. This information was given in the Methods section of the original manuscript, but we did not mention the correct numbers everywhere. We have corrected this in the manuscript. We also changed the number of participants (humans and rats) to the correct one throughout the entire manuscript.

• At l.95 when I first met the "4x4 stimulus grid" I had to guess its meaning. It would be really useful to see the stimulus grid as a panel in Figure 1 (in general Figures S1 and S4 could be integrated as panels of Figure 1). Also, even if the description of the stimulus generation in the Methods is probably clear enough, the authors might want to consider adding a simple schematic in Figure 1 as well (e.g. show the base, either concave or convex, and then how the 3 spheres are added to control alignment).

We have added the 4x4 stimulus grid in the main text.

• There is also another important point related to the choice of the network. As I wrote, I find the overall approach very interesting and powerful, but I'm actually worried that AlexNet might not be a good choice. I have experience trying to model neuronal responses from IT in monkeys, and there even the higher layers of AlexNet aren't that helpful. I need to use much deeper networks (e.g. ResNet or GoogleNet) to get decent fits. So I'm afraid that what is deemed as "high" in AlexNet might not be as high as the authors think. It would be helpful, as a sanity check, to see if the authors get the same sort of stimulus categories when using a different, deeper network.

We added a consideration to the manuscript about which network to use (see the Discussion): “We chose to work with Alexnet, as this is a network that has been used as a benchmark in many previous studies (e.g. (Cadieu et al., 2014; Groen et al., 2018; Kalfas et al., 2018; Nayebi et al., 2023; Zeman et al., 2020)), including studies that used more complex stimuli than the stimulus space in our current study. […] . It is in line with the literature that a typical deep neural network, AlexNet and also more complex ones, can explain human and animal behaviour to a certain extent but not fully. The explained variance might differ among DNNs, and there might be DNNs that can explain a higher proportion of rat or human behaviour. Most relevant for our current study is that DNNs tend to agree in terms of how representations change from lower to higher hierarchical layers, because this is the transformation that we have targeted in the Zero vs. high and High vs. zero testing protocols. (Pinto et al., 2008) already revealed that a simple V1-like model can sometimes result in surprisingly good object recognition performance. This aspect of our findings is also in line with the observation of Vinken & Op de Beeck (2021) that the performance of rats in many previous tasks might not be indicative of highly complex representations. Nevertheless, there is still a relative difference in complexity between lower and higher levels in the hierarchy. That is what we capitalize upon with the Zero vs. high and High vs. zero protocols. Thus, it might be more fruitful to explicitly contrast different levels of processing in a relative way rather than trying to pinpoint behaviour to specific levels of processing.”

• The task description needs way more detail. For how long were the stimuli presented? What was their size? Were the positions of the stimuli randomized? Was it a reaction time task? Was the time-out used as a negative feedback? In case, when (e.g. mistakes or slow responses)? Also, it is important to report some statistics about the basic responses. What was the average response time, what was the performance of individual animals (over days)? Did they show any bias for a particular dimension (either the 2 baseline dimensions or the identity preserving ones) or side of response? Was there a correlation within animals between performance on the baseline task and performance on the more complex tasks?

Thank you for your feedback. We have added more details to the task description in the manuscript.

The stimuli were presented on the screens until the animals reacted to one of the two screens. The size of the stimuli was 100 x 100 pixel. The position of the stimuli was always centred/full screen on the touchscreens. It was not a reaction time task and we also did not measure reaction time.

• Related to my previous comment, I wonder if the relative size/position of the stimulus with respect to the position of the animal in the setup might have had an impact on the performance, also given the impact of size shown in Figure 2. Was the position of the rat in the setup monitored (e.g. with DeepLabCut)? I guess that on average any effect of the animal position might be averaged away, but was this actually checked and/or controlled for?

The position of the rat was not monitored in the setup. In a previous study from our lab (Crijns & Op de Beeck, 2019), we investigated the visual acuity of rats in the touchscreen setups by presenting gratings with different cycles per screen to see how it affects their performance in orientation discrimination. With the results from this study and general knowledge about rat visual acuity, we derived that the decision distance of rats lies around 12.5cm from the screen. We have added this to the discussion.

Minor comments

• l.33 The sentence mentions humans, but the references are about monkeys. I believe that this concept is universal enough not to require any citation to support it.

Thank you for your feedback. We have removed the citations.

• This is very minor and totally negligible. The acronymous cDNN is not that common for convents (and it's kind of similar to cuDNN), it might help clarity to stick to a more popular acronymous, e.g. CNN or ANN. Also, given that the "high" layers used for stimulus selection where not convolutional layers after all (if I'm not mistaken).

Thank you for your feedback. We have changed the acronym to ‘CNN’ in the entire manuscript.

• In l.107-109 the authors identified a few potential biases in their stimuli, and they claim these biases cannot explain the results. However, the explanation is given only in the next pages. It might help to mention that before or to move that paragraph later, as I was just wondering about it until I finally got to the part on the brightness bias.

We expanded the analysis of these dimensions (e.g. brightness) throughout the manuscript.

• It would help a lot the readability to put also a label close to each dimension in Figures 2 and 3. I had to go and look at Figure S4 to figure that out.

Figures 2 and 3 have been updated, also including changes related to other comments.

• In Figure 2A, please specify what the red dashed line means.

We have edited the caption of Figure 2: “Figure 2 (a) Results of the Dimension learning training protocol. The black dashed horizontal line indicates chance level performance and the red dashed line represents the 80% performance threshold. The blue circles on top of each bar represent individual rat performances. The three bars represent the average performance of all animals on the old pair (Old), the pair that differs only in concavity (Conc) and on the pair that differs only in alignment (Align). (b) Results of the Transformations training protocol. Each cell of the matrix indicates the average performance per stimulus pair, pooled over all animals. The columns represent the distractors, whereas the rows separate the targets. The colour bar indicates the performance correct. ”

• Related to that, why performing a binomial test on 80%? It sounds arbitrary.

We performed the binomial test on 80% as 80% is our performance threshold for the animals

• The way the cDNN methods are introduced makes it sound like the authors actually fine-tuned the weights of AlexNet, while (if I'm not mistaken), they trained a classifier on the activations of a pre-trained AlexNet with frozen weights. It might be a bit confusing to readers. The rest of the paragraph instead is very clear and easy to follow.

We think the most confusing sentence was “ Figure 7 shows the performance of the network after training the network on our training stimuli for all test protocols. “ We changed this sentence to “ Figure 8 shows the performance of the network for each of the test protocols after training classifiers on the training stimuli using the different DNN layers.“

Reviewer #3

Main recommendations:

Although it may not fully explain the entire pattern of visual behavior, it is important to discuss rat visual acuity and its impact on the perception of visual features in the stimulus set.

We have added a paragraph to the Discussion that discusses the visual acuity of rats and its impact on perceiving the visual features of the stimuli.

The authors observed a potential influence of image brightness on behavior during the dimension learning protocol. Was there a correlation between image brightness and the subsequent image transformations?

We have added this to the Discussion: “To further investigate to which visual features the rat performance and human performance correlates best with, we calculated the correlation between rat performance and pixel similarity of the test image pairs, as well as the correlation between rat performance and brightness in the test image pairs. Here we found a correlation of 0.34 for pixel similarity and 0.39 for brightness, suggesting that these two visual features partly explain our results when compared to the full-set reliability of rat performance (0.58). If we perform the same correlation with the human performances, we get a correlation of 0.12 for pixel similarity and -0.12 for brightness. With the full-set reliability of 0.58 (rats) and 0.63 (humans) in mind, this suggests that even pixel similarity and brightness only partly explain the performances of rats and humans.”

Did the rats rely on consistent visual features to perform the tasks? I assume the split-half analysis was on data pooled across rats. What was the average correlation between rats? Were rats more internally consistent (split-half within rat) than consistent with other rats?

The split-half analysis was indeed performed on data pooled across rats. We checked whether rats are more internally consistent by comparing the split-half within correlations with the split-half between correlations. For the split-half within correlations, we split the data for each rat in two subsets and calculated the performance vectors (performance across all image pairs). We then calculated the correlation between these two vectors for each animal. To get the split-half between correlation, we calculated the correlation between the performance vector of every subset data of every rat with every other subset data from the other rats. Finally, we compared for each animal its split-half within correlation with the split-half between correlations involving that animal. The result of this paired t-test (p = 0.93, 95%CI [-0.09; 0.08]) suggests that rats were not internally more consistent.

Discussion of the cDNN performance and its relation to rat behavior could be expanded and clarified in several ways:

• The paper would benefit from further discussion regarding the low correlations between rat behavior and cDNN layers. Is the main message that cDNNs are not a suitable model for rat vision? Or can we conclude that the peak in mid layers indicates that rat behavior reflects mid-level visual processing? It would be valuable to explore what we currently know about the organization of the rat visual cortex and how applicable these models are to their visual system in terms of architecture and hierarchy.

We added a consideration to the manuscript about which network to use (see Discussion).

• The cDNN exhibited above chance performance in various early layers for several test protocols (e.g., rotations, light location, combination rotation). Does this limit the interpretation of the complexity of visual behavior required to perform these tasks?

This is not uncommon to find. Pinto et al. (2008) already revealed that a simple V1-like model can sometimes result in surprisingly good object recognition performance. This aspect of our findings is also in line with the observation of Vinken & Op de Beeck (2021) that the performance of rats in many previous tasks might not be indicative of highly complex representations. Nevertheless, there is still a relative difference in complexity between lower and higher levels in the hierarchy. That is what we capitalize upon with the High vs zero and the Zero vs high protocols. Thus, it might be more fruitful to explicitly contrast different levels of processing in a relative way rather than trying to pinpoint behavior to specific levels of processing. This argumentation is added to the Discussion section.

• How representative is the correlation profile between cDNN layers and behavior across protocols? Pooling stimuli across protocols may be necessary to obtain stable correlations due to relatively modest sample numbers. However, the authors could address how much each individual protocol influences the overall correlations in leave-one-out analyses. Are there protocols where rat behavior correlates more strongly with higher layers (e.g., when excluding zero vs. high)?

We prefer to base our conclusions mostly on the pooled analyses rather than individual protocols. As the reviewer also mentions, we can expect that the pooled analyses will provide the most stable results. For information, we included leave-one-out analyses in the supplemental material. Excluding the Zero vs. High protocol did not result in a stronger correlation with the higher layers. It was rare to see correlations with higher layers, and in the one case that we did (when excluding High versus zero) the correlations were still higher in several mid-level layers.

Author response image 2.

• The authors hypothesize that the cDNN results indicate that rats rely on visual features such as contrast. Can this link be established more firmly? e.g., what are the receptive fields in the layers that correlate with rat behavior sensitive to?

This hypothesis was made based on previous in-lab research ((Schnell et al., 2023) where we found rats indeed rely on contrast features. In this study, we performed a face categorization task, parameterized on contrast features, and we investigated to what extent rats use contrast features to perform in a face categorization task. Similarly as in the current study, we used a DNN that as trained and tested on the same stimuli as the animals to investigate the representations of the animals. There, we found that the animals use contrast features to some extent and that this correlated best with the lower layers of the network. Hence, we would say that the lower layers correlate best with rat behaviour that is sensitive to contrast. Earlier layers of the network include local filters that simulate V1-like receptive fields. Higher layers of the network, on the other hand, are used for object selectivity.

• There seems to be a disconnect between rat behavior and the selection of stimuli for the high (zero) vs. zero (high) protocols. Specifically, rat behavior correlated best with mid layers, whereas the image selection process relied on earlier layers. What is the interpretation when rat behavior correlates with higher layers than those used to select the stimuli?

We agree that it is difficult to pinpoint a particular level of processing, and it might be better to use relative terms: lower/higher than. This is addressed in the manuscript by the edit in response to three comments back.

• To what extent can we attribute the performance below the ceiling for many protocols to sensory/perceptual limitations as opposed to other factors such as task structure, motivation, or distractibility?

We agree that these factors play a role in the overall performance difference. In Figure 5, the most right bar shows the percentage of all animals (light blue) vs all humans (dark blue) on the old pair that was presented during the testing protocol. Even here, the performance of the animals was lower than humans, and this pattern extended to the testing protocols as well. This was most likely due to motivation and/or distractibility which we know can happen in both humans and rats but affects the rat results more with our methodology.

Minor recommendations:

• What was the trial-to-trial variability in the distance and position of the rat's head relative to the stimuli displayed on the screen? Can this variability be taken into account in the size and position protocols? How meaningful is the cDNN modelling of these protocols considering that the training and testing of the model does not incorporate this trial-to-trial variability?

We have no information on this trial-to-trial variability. We have information though on what rats typically do overall from an earlier paper that was mentioned in response to an earlier comment (Crijns et al.).

We have added a disclaimer in the Discussion on our lack of information on trial-to-trial variability.

• Several of the protocols varied a visual feature dimension (e.g., concavity & alignment) relative to the base pair. Did rat performance correlate with these manipulations? How did rat behavior relate to pixel dissimilarity, either between target and distractor or in relation to the trained base pair?

We have added this to the Discussion. See also our general comments in the Public responses.

• What could be the underlying factor(s) contributing to the difference in accuracy between the "small transformations" depicted in Figure 2 and some of the transformations displayed in Figure 3? In particular, it seems that the variability of targets and distractors is greater for the "small transformations" in Figure 2 compared to the rotation along the y-axis shown in Figure 3.

There are several differences between these protocols. Before considering the stimulus properties, we should take into account other factors. The Transformations protocol was a training protocol, meaning that the animals underwent several sessions in this protocol, always receiving real reward during the trials, and only stopping once a high enough performance was reached. For the protocols in Figure 3, the animals were also placed in these protocols for multiple sessions in order to obtain enough trials, however, the difference here is that they did not receive real reward and testing was also stopped if performance was still low.

• In Figure 3, it is unclear which pairwise transformation accuracies were above chance. It would be helpful if the authors could indicate significant cells with an asterisk. The scale for percentage correct is cut off at 50%. Were there any instances where the behaviors were below 50%? Specifically, did the rats consistently choose the wrong option for any of the pairs? It would be helpful to add "old pair", "concavity" and "alignment" to x-axis labels in Fig 2A .

We have added “old”, “conc” and “align” to the x-axis labels in Figure 2A.

• Considering the overall performance across protocols, it seems overstated to claim that the rats were able to "master the task."

When talking about “mastering the task”, we talk about the training protocols where we aimed that the animals would perform at 80% and not significantly less. We checked this throughout the testing protocols as well, where we also presented the old pair as quality control, and their performance was never significantly lower than our 80% performance threshold on this pair, suggesting that they mastered the task in which they were trained. To avoid discussion on semantics, we also rephrased “master the task” into “learn the task”.

• What are the criteria for the claim that the "animal model of choice for vision studies has become the rodent model"? It is likely that researchers in primate vision may hold a different viewpoint, and data such as yearly total publication counts might not align with this claim.

Primate vision is important for investigating complex visual aspects. With the advancements in experimental techniques for rodent vision, e.g. genetics and imaging techniques as well as behavioural tasks, the rodent model has become an important model as well. It is not necessarily an “either” or “or” question (primates or rodents), but more a complementary issue: using both primates and rodents to unravel the full picture of vision.

We have changed this part in the introduction to “Lately, the rodent model has become an important model in vision studies, motivated by the applicability of molecular and genetic tools rather than by the visual capabilities of rodents”.

• The correspondence between the list of layers in Supplementary Tables 8 and 9 and the layers shown in Figures 4 and 6 could be clarified.

We have clarified this in the caption of Figure 7

• The titles in Figures 4 and 6 could be updated from "DNN" to "cDNN" to ensure consistency with the rest of the manuscript.

Thank you for your feedback. We have changed the titles in Figures 4 and 6 such that they are consistent with the rest of the manuscript.

eLife assessment

This valuable manuscript attempts to identify the brain regions and cell types involved in habituation to dark flash stimuli in larval zebrafish. Habituation being a form of learning widespread in the animal kingdom, the investigation of neural mechanisms underlying it is a worthwhile endeavor. The authors use a combination of behavioral analysis, neural activity imaging, and pharmacological manipulation to investigate brain-wide mechanisms of habituation. While the data presented are solid, the authors conclude that there is no simple relationship between pharmacological intervention, neural activity patterns, and behavioral outcomes, and a robust causative link can therefore not be established.

Author Response

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

eLife assessment

This valuable manuscript attempts to identify the brain regions and cell types involved in habituation to dark flash stimuli in larval zebrafish. Habituation being a form of learning widespread in the animal kingdom, the investigation of neural mechanisms underlying it is an important endeavor. The authors use a combination of behavioral analysis, neural activity imaging, and pharmacological manipulation to investigate brain-wide mechanisms of habituation. However, the data presented are incomplete and do not show a convincing causative link between pharmacological manipulations, neural activity patterns, and behavioral outcomes.

We thank the reviewers and editors for their careful reading and reviews of our work. We are grateful that they appreciate the value in our experimental approach and results. We acknowledge what we interpret as the major criticism, that in our original manuscript we focused too heavily on the hypothesized role of GABAergic neurons in driving habituation. This hypothesis will remain only indirectly supported until we can identify a GABAergic population of neurons that drives habituation. Therefore, we have revised our manuscript, decreasing the focus on GABA, and rather emphasizing the following three points:

1) By performing the first Ca2+ imaging experiments during dark flash habituation, we identify multiple distinct functional classes of neurons which have different adaptation profiles, including non-adapting and potentiating classes. These neurons are spread throughout the brain, indicating that habituation is a complex and distributed process.

2) By performing a pharmacological screen for dark flash habituation modifiers, we confirm habituation behaviour manifests from multiple distinct molecular mechanisms that independently modulate different behavioural outputs. We also implicate multiple novel pathways in habituation plasticity, some of which we have validated through dose-response studies.

3) By combining pharmacology and Ca2+ imaging, we did not observe a simple relationship between the behavioural effects of a drug treatment and functional alterations in neurons. This observation further supports our model that habituation is a multidimensional process, for which a simple circuit model will be insufficient.

We would like to point out that, in our opinion, there appears to be a factual error in the final sentence of the eLife assessment:

“However, the data presented are incomplete and do not show a convincing causative link between pharmacological manipulations, neural activity patterns, and behavioral outcomes”.

We believe that a “convincing causative link” between pharmacological manipulations and behavioural outcomes has been clearly demonstrated for PTX, Melatonin, Estradiol and Hexestrol through our dose response experiments. Similarly a link between pharmacology and neural activity patterns has also been directly demonstrated. As mentioned in (3), we acknowledge that our data linking neural activity and behaviour is more tenuous, as will be more explicitly reflected in our revised manuscript.

Nevertheless, we maintain that one of the primary strengths of our study is our attempt to integrate analyses that span the behavioural, pharmacological, and neural activity-levels.

In our revised manuscript, we have substantially altered the Abstract and Discussion, removed the Model figure (previously Figure 8), and changed the title from :

“Inhibition drives habituation of a larval zebrafish visual response”

to:

“Functional and pharmacological analyses of visual habituation learning in larval zebrafish”

Text changes from the initial version are visible as track changes in the word document: “LamireEtAl_2022_eLifeRevisions.docx”

Reviewer #1 (Public Review):

This manuscript addresses the important and understudied issue of circuit-level mechanisms supporting habituation, particularly in pursuit of the possible role of increases in the activity of inhibitory neurons in suppressing behavioral output during long-term habituation. The authors make use of many of the striking advantages of the larval zebrafish to perform whole brain, single neuronal calcium imaging during repeated sensory exposure, and high throughput screening of pharmacological agents in freely moving, habituating larvae. Notably, several blockers/antagonists of GABAA(C) receptors completely suppress habituation of the O-bend escape response to dark flashes, suggesting a key role for GABAergic transmission in this form of habituation. Other substances are identified that strikingly enhance habituation, including melatonin, although here the suggested mechanistic insight is less specific. To add to these findings, a number of functional clusters of neurons are identified in the larval brain that has divergent activity through habituation, with many clusters exhibiting suppression of different degrees, in line with adaptive filtration during habituation, and a single cluster that potentiates during habituation. Further assessment reveals that all of these clusters include GABAergic inhibitory neurons and excitatory neurons, so we cannot take away the simple interpretation that the potentiating cluster of neurons is inhibitory and therefore exerts an influence on the other adapting (depressing) clusters to produce habituation. Rather, a variety of interpretations remain in play.

Overall, there is great potential in the approach that has been used here to gain insight into circuit-level mechanisms of habituation. There are many experiments performed by the authors that cannot be achieved currently in other vertebrate systems, so the manuscript serves as a potential methodological platform that can be used to support a rich array of future work. While there are several key observations that one can take away from this manuscript, a clear interpretation of the role of GABAergic inhibitory neurons in habituation has not been established. This potential feature of habituation is emphasized throughout, particularly in the introduction and discussion sections, meaning that one is obliged as a reader to interrogate whether the results as they currently stand really do demonstrate a role for GABAergic inhibition in habituation. Currently, the key piece of evidence that may support this conclusion is that picrotoxin, which acts to block some classes of GABA receptors, prevents habituation. However, there are interpretations of this finding that do not specifically require a role for modified GABAergic inhibition. For instance, by lowering GABAergic inhibition, an overall increase in neural activity will occur within the brain, in this case below a level that could cause a seizure. That increase in activity may simply prevent learning by massively increasing neural noise and therefore either preventing synaptic plasticity or, more likely, causing indiscriminate synaptic strengthening and weakening that occludes information storage. Sensory processing itself could also be disrupted, for instance by altering the selectivity of receptive fields. Alternatively, it could be that the increase in neural activity produced by the blockade of inhibition simply drives more behavioral output, meaning that more excitatory synaptic adaptation is required to suppress that output. The authors propose two specific working models of the ways in which GABAergic inhibition could be implemented in habituation. An alternative model, in which GABAergic neurons are not themselves modified but act as a key intermediary between Hebbian assemblies of excitatory neurons that are modified to support memory and output neurons, is not explored. As yet, these or other models in which inhibition is not required for habituation, have not been fully tested.

This manuscript describes a really substantial body of work that provides evidence of functional clusters of neurons with divergent responses to repeated sensory input and an array of pharmacological agents that can influence the rate of a fundamentally important form of learning.

We thank the reviewer for their careful consideration of our work, and we agree that multiple models of how habituation occurs remain plausible. As discussed above and below in more detail, we have revised our manuscript to better reflect this. We hope the reviewer will agree that this has improved the manuscript.

Reviewer #2 (Public Review):

In this study, Lamire et al. use a calcium imaging approach, behavioural tests, and pharmacological manipulations to identify the molecular mechanisms behind visual habituation. Overall, the manuscript is well-written but difficult to follow at times. They show a valuable new drug screen paradigm to assess the impact of pharmacological compounds on the behaviour of larval zebrafish, the results are convincing, but the description of the work is sometimes confusing and lacking details.

We thank the reviewer for identifying areas where our description lacked details. We apologize for these omissions and have attempted to add relevant details as described below. We note that all of the analysis code is available online, though we appreciate that navigating and extracting data from these files is not straightforward.

The volumetric calcium imaging of habituation to dark flashes is valuable, but the mix of responses to visual cues that are not relevant to the dark flash escape, such as the slow increase back to baseline luminosity, lowers the clarity of the results. The link between the calcium imaging results and free-swimming behaviour is not especially convincing, however, that is a common issue of head-restrained imaging with larval zebrafish.

We agree with the reviewer that the design of our stimulus, and specifically the slow increase back to baseline luminosity, is perhaps confusing for the interpretation of some of the response profiles of neurons. We originally chose this stimulus type (rather than a square wave of 1s of darkness, for example) in order to better highlight the responses of the larvae to the onset of darkness (rather than the response to abruptly returning to full brightness). We therefore believe that the slow return to baseline is an important feature of the stimulus,, which better separates activity related to the fast offset from activity related to light onset. And since all of the foundational behavioural data (Randlett et al., Current Biology 2019), and pharmacological data, used this stimulus type, we did not change it for the Ca2+ imaging experiments. Our use of relatively slow nuclear-targeted GCaMP indicators also means that the temporal resolution of our imaging experiments is relatively poor, and therefore we felt that using a stimulus that highlighted light offset might be best.

We also fully acknowledge in the Results section that the behaviour of the head embedded fish is not the same as that of free-swimming fish, and that therefore establishing a direct link between these types of experiments is complicated. This is an unavoidable caveat in the head-embedded style experiments. To further emphasize this, we have also added a paragraph to the discussion where this is acknowledged explicitly.

“We also found that the same pharmacological treatments that result in strong alterations to habituation behaviour in freely swimming larvae ([fig:5]), resulted in relatively subtle and complex functional alterations in the circuit ([fig:6]). Making direct comparisons between freely-swimming behaviour and head-fixed Ca2+ imaging is always challenging due to the differences in behaviour observed in the two contexts, and therefore our failure to identify a clear logic in these experiments may have technical explanations that will require approaches to measure neural activity from unrestrained and freely-behaving animals to resolve . Alternatively, these results are again consistent with the idea that habituation is a multidimensional and perhaps highly non-linear phenomenon in the circuit, which cannot be captured by a simple model.”

The strong focus on GABA seems unwarranted based on the pharmacological results, as only Picrotoxinin gives clear results, but the other antagonists do not give a consistent results. On the other hand, the melatonin receptor agonists, and oestrogen receptor agonists give more consistent results, including more convincing dose effects.

We agree that our manuscript focused too strongly on GABA and have toned this down. We are currently performing genetic experiments aimed at identifying the Melatonin, Estrogen and GABA receptors that function during habituation, which we think will be necessary to move beyond pharmacology and the necessary caveats that such experiments bring.

The pharmacological manipulation of the habituation circuits mapped in the first part does not arrive at any satisfying conclusion, which is acknowledged by the authors. These results do reinforce the disconnect between the calcium imaging and the behavioural experiments and undercut somewhat the proposed circuit-level model.

We agree with this criticism and have toned down the focus on GABA specifically in the circuit, and have removed the speculative model previously in Figure 8.

Overall, the authors did identify interesting new molecular pathways that may be involved in habituation to dark flashes. Their screening approach, while not novel, will be a powerful way to interrogate other behavioural profiles. The authors identified circuit loci apparently involved in habituation to dark flashes, and the potentiation and no adaptation clusters have not been previously observed as far as I know.

The data will be useful to guide follow-up experiments by the community on the new pathway candidates that this screen has uncovered, including behaviours beyond dark flash habituation.

We again thank the reviewer for both their support of our approach, and in pointing out where our conclusions were not well supported by our data.

Reviewer #3 (Public Review):

To analyze the circuit mechanisms leading to the habituation of the O-bed responses upon repeated dark flashes (DFs), the authors performed 2-photon Ca2+ imaging in larvae expressing nuclear-targeted GCaMP7f pan-neuronally panning the majority of the midbrain, hindbrain, pretectum, and thalamus. They found that while the majority of neurons across the brain depress their responsiveness during habituation, a smaller population of neurons in the dorsal regions of the brain, including the torus longitudinalis, cerebellum, and dorsal hindbrain, showed the opposite pattern, suggesting that motor-related brain regions contain non-depressed signals, and therefore likely contribute to habituation plasticity.

Further analysis using affinity propagation clustering identified 12 clusters that differed both in their adaptation to repeated DFs, as well as the shape of their response to the DF.

Next by the pharmacological screening of 1953 small molecule compounds with known targets in conjunction with the high-throughput assay, they found that 176 compounds significantly altered some aspects of measured behavior. Among them, they sought to identify the compounds that 1) have minimal effects on the naive response to DFs, but strong effects during the training and/or memory retention periods, 2) have minimal effects on other aspects of behaviors, 3) show similar behavioral effects to other compounds tested in the same molecular pathway, and identified the GABAA/C Receptor antagonists Bicuculline, Amoxapine, and Picrotoxinin (PTX). As partial antagonism of GABAAR and/or GABACR is sufficient to strongly suppress habituation but not generalized behavioral excitability, they concluded that GABA plays a very prominent role in habituation. They also identified multiple agonists of both Melatonin and Estrogen receptors, indicating that hormonal signaling may also play a prominent role in habituation response.

To integrate the results of the Ca2+ imaging experiments with the pharmacological screening results, the authors compared the Ca2+ activity patterns after treatment with vehicle, PTX, or Melatonin in the tethered larvae. The behavioral effects of PTX and Melatonin were much smaller compared with the very strong behavioral effects in freely-swimming animals, but the authors assumed that the difference was significant enough to continue further experiments. Based on the hypothesis that Melatonin and GABA cooperate during habituation, they expected PTX and Melatonin to have opposite effects. This was not the case in their results: for example, the size of the 12(Pot, M) neuron population was increased by both PTX and Melatonin, suggesting that pharmacological manipulations that affect habituation behavior manifest in complex functional alterations in the circuit, making capturing these effects by a simple difficult.

Since the 12(𝑃𝑜𝑡, 𝑀) neurons potentiate their responses and thus could act to progressively depress the responses of other neuronal classes, they examined the identity of these neurons with GABA neurons. However, GABAergic neurons in the habituating circuit are not characterized by their Adaptation Profile, suggesting that global manipulations of GABAergic signaling through PTX have complex manifestations in the functional properties of neurons.

Overall, the authors have performed an admirably large amount of work both in whole-brain neural activity imaging and pharmacological screening. However, they are not successful in integrating the results of both experiments into an acceptably consistent interpretation due to the incongruency of the results of different experiments. Although the authors present some models for interpretation, it is not easy for me to believe that this model would help the readers of this journal to deepen the understanding of the mechanisms for habituation in DF responses at the neural circuit level.

This reviewer would rather recommend the authors divide this manuscript into two and publish two papers by adding some more strengthening data for each part such as cellular manipulations, e.g. ablation to prove the critical involvement of 12(Pot, M) neurons in habituation.

We thank the reviewer for their careful consideration of our manuscript, and we agree that our emphasis on a particular model of DF habituation, namely the potentiation of GABAergic synapses, was overly speculative. We hope they will agree that our revised manuscript better reflect the results from our experiments, and we have tried to more specifically emphasize the incongruency in our behavioural and Ca2+ imaging data after pharmacological treatment, which we agree shows that a simple model is insufficient to capture both of these sets of observations.

We have opted not to split the paper into two, since we feel that the collective message of this paper and approach combining molecular and functional analysis will be of interest. Moreover, we feel that the molecular and functional analyses feed off of each other and provide a level of complementarity that would be lost if the manuscript would be split, even if the message in this particular case is rather complex

Reviewer #1 (Recommendations For The Authors):

There is much to commend about this manuscript. The advantages of studying habituation in the zebrafish larva are very clearly demonstrated, including the wonderful calcium imaging across the brain and the relatively high throughput screening of large numbers of different pharmacological agents. The habituation to dark flashes in freely moving larvae is also striking and the very large effect size serves the screening beautifully. Thus, if we take the really substantial amount of work of a very high standard that has been done here, there is clearly potential for an important new contribution to the literature. However, as you will see from my public review, I am of the opinion that a specific role for the modification of GABAergic inhibitory systems has not yet been established through this work. While the potential role for GABAergic inhibitory neurons in habituation, either as the key modifiable element or as an intermediary between memory and motor output, is an attractive theory with many strengths, your study as it currently stands does not categorically demonstrate that one of those two options holds. For instance, the more traditional view, that adaptive filtration is mediated by weakened synaptic connectivity between excitatory sensory systems and excitatory motor output or reduced intrinsic excitability in those same neurons, could still be in operation here. By lowering GABAergic influence over post-synaptic targets with picrotoxin, it is possible that motor output remains highly active, and even lower activity or synaptic drive from those excitatory sensory systems that feed into the output may still reliably produce behavioral output. Alternatively, it could be the formation of a memory of the familiar stimulus is disrupted by reduced inhibition that alters sensory coding either by introducing noise or reducing the selectivity of receptive fields. I believe that there are several options to address these concerns:

1) You could change the emphasis of the manuscript so that it is less focused on inhibition and instead emphasizes the categorization of clusters of neurons that have divergent responses during habituation, including either strong suppression to potentiation. To this, you add a high throughput screening system with a wide range of different agents being tested, several of which produce a significant effect on habituation in either direction. These observations in themselves provide powerful building blocks for future work.

2) If GABAergic neurons play a key role in habituation in this paradigm, then picrotoxin is having its effect by blocking receptors on excitatory neurons. Thus, it seems that selectively imaging GABAergic neurons before and after the application of these drugs is not likely to reveal the contribution of GABAergic synaptic influence on excitatory targets. More important is to get a stronger sense of how the GABAergic neurons change their activity throughout habituation and then influence the downstream target neurons of those GABAergic neurons (some of which may themselves be inhibitory and participating in disinhibition). For instance, you could interrogate whether anti-correlations in activity levels exist between presynaptic inhibitory neurons and putative post-synaptic targets. This analysis could be further bolstered by removing that relationship in the presence of Picrotoxin, thereby demonstrating a direct influence of inhibition from a GABAergic presynaptic partner on a postsynaptic target. While this would constitute a lot more work, it is likely to yield greater insight into a specific role for GABAergic neurons in habituation, and I suspect much of that information is in the existing datasets.

3) To really reveal causal roles for inhibition in this form of habituation, it seems to me that there needs to be some selective intervention in GABAergic neuronal activity, ideally bidirectionally, to transiently interrupt or enhance habituation. Optogenetic or chemogenetic stimulation/inactivation is one option in this regard, which I imagine would be challenging to implement and certainly involves a lot of further work, particularly if you are then going to target specific subpopulations of GABAergic neurons. I appreciate that this option seems way beyond the scope of a review process and would probably constitute a follow-up study.

We agree with the reviewer that we have not “categorically demonstrated” that GABAergic inhibitory neurons drive habituation by increasing their influence on the circuit, and appreciate the suggestions for how to reformulate our manuscript to better reflect this. We have opted to follow suggestion (1), and have considerably changed the focus of the manuscript.

The additional analysis suggested in (2) is very interesting, but since we can not identify which cells are inhibitory in our imaging experiments with picrotoxinin treatment, nor which are pre- or post-synaptic, we feel that this analysis will be very unconstrained. Also, if GABA is acting as an inhibitory neurotransmitter, it therefore is expected to act to drive anticorrelations among pre and postsynaptic neurons through inhibition. Therefore, blockage of GABA through PTX would be expected to result in increased correlations, regardless of our hypothesized role of neurons during habituation. Our current efforts are aimed at identifying critical neurons driving habituation plasticity, and we will perform such analysis once we have mechanisms for identifying these neurons.

Finally, we agree that (3) is the obvious and only way to demonstrate causation here, and this is where we are working towards. However, since we currently have no means of genetically targeting these neurons, we are not able to perform these suggested experiments today.

I have some additional concerns that I would really appreciate you addressing:

1) The behavioral habituation is striking in the freely moving larvae, but very hard to monitor in the larvae that are immobilized for calcium imaging. Are there steps that could be taken in the long run to improve direct observation of the habituation effect in these semi-stationary fish? For instance, is it possible to observe eye movements or some more subtle behavioral readout than the O-bend reflex? I apologize if this is a naïve question, but I am not entirely familiar with this specific experimental paradigm.

In the Dark Flash paradigm, we do not have readouts beyond the “O-bend” response itself, which is characterized by a large-angle bend of the tail and turning maneuver. We have not observed other, more subtle behavioural responses, such as eye or fin movements, for example. If we would be able to identify alternative behavioural outputs that were more robustly performed during head-embedded preparations, this would indeed be an advantage allowing us to more directly interpret the Ca2+ imaging results with respect to behaviour.

2) The dark flash as a stimulus to which the larvae habituate is obviously used as a powerful and ethologically relevant stimulus. However, it does leave an element of traditional habituation paradigms out, which is a novel stimulus that can be used to immediately re-instate the habituated response (otherwise known as dishabituation). Is there a way that you can imagine implementing that with zebrafish larvae, for instance through systematically altering a visual feature, such as spatial frequency or orientation? This would be a powerful development in my view as it would not only allow you to rule out motor or sensory fatigue as an underlying cause of reduced behavior but also it would provide an extra feature that strengthens your assessment of neuronal response profiles in candidate populations of inhibitory and excitatory neurons.

We agree that identifying a dishabituating stimulus would be very powerful for our experiments. For short-term habituation of the acoustic startle response, Wolman et al demonstrated that dishabituation occurs after a touch stimulus (Wolman et al., PNAS, 2011; https://doi.org/10.1073/pnas.1107156108). We attempted to dishabituate the O-Bend response with tap and touch stimuli, and this unfortunately did not occur. Our understanding of dishabituation is that this generally requires a second stimulus that elicits the same behaviour as the habituated stimulus (e.g. both acoustic and touch-stimuli elicit the Mauthner-dependent C-bend response). In zebrafish the only stimulus that has been identified that elicits the O-bend is a dark-flash. This lack of an appropriate alternative stimulus is perhaps why we have been unsuccessful in identifying a dishabituating stimulus.

3) You have written about the concept of 'short' and 'long' response shapes when using calcium imaging as a proxy for neural activity, surmising that the short response shape may reflect transient bursting. Although calcium imaging obviously has many advantages, this feature reveals one notable limitation of calcium imaging in contrast to electrophysiology, in that the time course of the signal is considerably longer and does not allow you with confidence to fully detect the response profile of neurons. Is there some kind of further deconvolution process that you could implement to improve the fidelity of your calcium imaging to the occurrence of action potentials? The burstiness of neurons is obviously important as it can indicate a particular type of neuron (for instance fast-spiking inhibitory neurons) or it might reveal a changing influence on post-synaptic neurons. For instance, bursting can be a response to inhibition due to the triggering of T-type calcium channels in response to hyperpolarization.

One of the major limitations to Ca2+ imaging is the lack of temporal resolution. In our particular approach, using nuclear-targeted H2B-GCaMP indicators, further reduces our temporal resolution. Deconvolution approaches can be used in some instances to approximate spike rate, since the rise-time of Ca2+ indicators can be relatively fast. However, in our imaging we chose to image larger volumes at the expense of scan rate, where our imaging is performed at only 2hz. Therefore, deconvolution and spike-rate estimation is not appropriate. Considering these limitations, we would argue that the fact that we can observe differences in kinetics of the 'short' and 'long' response shapes indicates that they likely show very different response kinetics, which we hope to confirm by electrophysiology once we have established ways of targeting these neurons for recordings.

4) I note that among the many substances you screened with is MK801. An obvious candidate mechanism in habituation is the NMDA receptor, given the importance of this receptor for so many forms of learning and bidirectional synaptic plasticity. If I am to understand correctly, this NMDA receptor blocker actually enhances habituation in the zebrafish larvae, similar to melatonin. That is a very surprising observation, which is worth looking into further or at least discussed in the manuscript. The finding would, at least, be consistent with the idea that plasticity is not occurring at excitatory synapses and could potentially bolster the argument that plasticity of inhibitory synapses is at play in this particular form of habituation.

This is a very important point. We were also particularly interested in MK801, which has been shown to inhibit other forms of habituation, like short-term acoustic habituation (Wolman et al., PNAS, 2011; https://doi.org/10.1073/pnas.1107156108). In our experiments we did see that fish become even less responsive to dark flashes when treated with MK-801 (SSMD fingerprint data: Prob-Train = -0.39, Prob-Test = -1.58) which would indicate that MK-801 promotes dark flash habituation, similar to Melatonin. However, we also observed that MK-801 caused a decrease in the performance in the other visual assay we tested: the optomotor response (OMR-Perf = -0.93), indicating that MK-801 causes a generalized decrease in visual responses, perhaps by acting on circuits within the retina. Therefore, based on these experiments with global drug applications, we cannot determine if MK-801 influences the plasticity process in dark-flash habituation, and this is why we did not pursue it further in this project.

Anyway, I hope that you take these suggestions as constructive and, in the spirit that they are intended, as possible routes for improving an already very interesting manuscript.

We are very grateful for your suggestions, which we feel has helped us to improve our manuscript substantially.

Reviewer #2 (Recommendations For The Authors):

Overall, the manuscript is well-written, but confusing at times. The results are not always presented in a consistent way, and I found myself having to dig in the raw data or code to find answers. There is a certain disconnect between the free-swimming results, and the calcium imaging, which is somewhat inevitable based on other published work. But I am unsure of what they each bring to the other, as the results from Fig.6 do not match at all the changes observed in the behavioural assays, it almost feels like two separate studies and the inconsistencies make the model appear unlikely.

We agree that there is a disconnect at the behavioural level in our free-swimming and head-embedded imaging experiments. However, this does not necessarily mean that the activity we observe during the imaging experiments cannot be informative about processes that are also occurring in freely-swimming fish. For example, it is possible that the dark-flash circuit is responding and habitating similarly in the head-embedded and freely-swimming preparations, but that in the latter context there is an additional blockade on motor output that massively decreases the propensity of the fish to initiate any movements. In such a case, the “disconnect between the free-swimming results, and the calcium imaging” would indicate that the relationship between neural activity and habituation behaviour is rather complex.

Without a method to record activity from freely swimming fish at our disposal, we can not determine this, one way or the other.

We hope that we now acknowledge these concerns appropriately in the discussion:

“We also found that the same pharmacological treatments that result in strong alterations to habituation behaviour in freely swimming larvae ([fig:5]), resulted in relatively subtle and complex functional alterations in the circuit ([fig:6]). Making direct comparisons between freely-swimming behaviour and head-fixed Ca2+ imaging is always challenging due to the differences in behaviour observed in the two contexts, and therefore our failure to identify a clear logic in these experiments may have technical explanations that will require approaches to measure neural activity from unrestrained and freely-behaving animals to resolve . Alternatively, these results are again consistent with the idea that habituation is a multidimensional and perhaps highly non-linear phenomenon in the circuit, which cannot be captured by a simple model. “

I am not convinced by the results surrounding GABA, from the inconsistent GABA receptor antagonist profile to the post hoc identification of GABAergic neurons as it is currently done in the manuscript. I think that the current focus on GABA does a disservice to the manuscript. However, the novel findings surrounding the potential role of Melatonin, and Estrogen, in habituation are quite interesting.

We agree that we focused too heavily on our hypothesized role for GABA in our original manuscript, and we hope that the reviewer agrees that our updated manuscript is an improvement. We also thank the reviewer for their interest in our Melatonin and Estrogen results, for which follow up studies are ongoing to characterize the effects of these hormones and their receptors on habituation.

There is an assumption that all the adaptation profiles are related to the DF (although that is somewhat alleviated in the discussions of the ON responses) and not to the luminosity changes. But there is no easy way to deconvolve those two in the current experiments. I would like the timing of the fluorescence rise to be quantified compared to the dark flash stimulus onset, potentially spike inference methods could help with giving a better idea of the timing of those responses. Based on the behavioural responses that were <500ms in Randlet O et al, eLife, 2019; we would expect only the fastest DF responses to be linked to the behaviour.

We agree that we are unable to disambiguate responses to the dark flash that initiate the O-bend response, and those that are related to only changes in luminosity. As discussed above, our Ca2+ imaging approach is severely limited in temporal resolution and therefore spike inference methods are not appropriate.

Major comments

Fig.1: There seems to be a very variable lag between the motor events and DF responses, furthermore, it does not seem that the motor responses follow a similar habituation rate as in 1Bi. Although this only shows the smoothed 'movement cluster' from the rastermap, it could hide individual variability. It would be important to know what the 'escape' rate was in the embedded experiment, as

Fig.1 sup.1 seems to indicate there was little to no habituation. It would also be needed to know which motor events are considered linked to the DF stimulus, and how that was decided. Was there a movement intensity threshold and lag limit in the response?

We interpret this concern as relating to the data presented in Figure 6A, where we quantify the habituation rate in the head-embedded experiments. As we have discussed, both above and in the manuscript, we saw very strongly muted responses to DFs in the head-embedded preparation, but we neglected to describe our method of quantifying the responses. We have added the following description to the methods:

“To quantify responses to the dark flash stimuli we used motion artifacts in the imaging data to identify frames associated with movements ([fig:1]-[fig:S1]). Motion artifact was quantified using the “corrXY” parameter from suite2p, which reflects the peak of phase correlation comparing each acquired frame and reference image used for motion correction. The “motion power” was quantified as the standard deviation of a 3-frame rolling window, which was smoothed in time using a Savitzky-Golay filter (window length = 15 frames, polyorder = 2). A response to a dark flash was defined as a “motion power” signal greater than 3 (z-score) occurring within 10-seconds of the dark-flash onset, and was used to quantify habituation in the head-embedded preparation ([fig:6]A).“

Line 94: This seems to be a strong claim based on the sparse presence of non-habituating, or potentiating, neurons in downstream regions. However, these neurons appear to be extremely rare, and as mentioned in my comment above, the behavioural habituation appears minimal. These neurons could encode the luminosity and be part of other responses, such as light-seeking in Karpenko S et al, eLife, 2020 or escape directionality in Heap et al, Neuron, 2018. Furthermore, dimming information has been shown to have parallel processing pathways in Robles E et al, JCN, 2020; so it would make sense that not all the observed responses in this manuscript would be involved in behavioural habituation to dark flashes.

We agree that without functional interventions, we do not know which of the neurons we have categorized are specifically involved in the dark flash response habituation. It is possible that the non-adapting and potentiating neurons are involved in other behaviours. We have therefore removed this statement.

Line 103: It appears that several of those responses are to the changes in luminosity and not the DF itself, especially the ON and sustained responses. Based on the previous DF habituation study from Randlet O et al, eLife, 2019; the latency of the response is below 0.5s. So the behaviour-relevant responses must only include the shortest latency one, as discussed above.

We appreciate the point that the reviewer is making here, but we are less clear about what the difference between “changes in luminosity” and a “dark flash” response are, since a dark flash consists of a change in luminosity. We take it that the reviewer means the difference between a luminance stimulus that elicits an O-bend, from one that does not. In order to disambiguate the two, one would likely need to use stimuli where the luminosity changes, but do not elicit O-bends.

Perhaps due to the limited temporal resolution of our Ca2+ imaging data, we do not see a clear difference in the onset of the stimulus response for any of the functional clusters that would help us to determine which neurons are more relevant to the acute DF response.

Fig.2B. It is very difficult to make out the actual average z-scored fluorescence, a supplementary figure would help by making these bigger. A plot to quantify the maximum response would also be useful to judge how it changes between the first few and few last DF. Another plot to give the time between the onset of the responses and the onset of the DF stimulus is also needed to judge which cluster may be relevant to the DF escapes observed in the free-swimming experiments.

We agree with the reviewer that interpreting these datasets are challenging. We did include the actual average z-scored fluorescence in Figure 6—figure supplement 1, panel D. This figure also includes a comparison between the predicted Ca2+ response to the dark flash (the stimulus convolved with the approximate GCaMP response kernel), which shows that all OFF-responding neuronal classes show very similar rise time response kinetics, and thus this analysis does not help to judge whether a cluster is more or less relevant to O-bend responses in the free-swimming experiments. We appreciate that there are differences in opinion about the best way to present the data, but we have opted to leave our original presentation.

Line 130: Is a correlation below 0.1 meaningful or significant? It does not seem like this cluster would be a motor or decision cluster.

Our goal with this correlational analysis to motor signals was to identify if certain clusters of DF responsive neurons were more associated with motor output, and therefore may be more downstream in the sensori-motor cascade. Cluster 4 showed the highest median correlation across the population of cells. Whether a median correlation of ~0.1 is “meaningful” is impossible for us to answer, but it is highly “significant” in the statistical sense, as is evident by the 99.99999% confidence intervals plotted. We note that these cells were not selected based on their correlation to the motor stimulus, but only to the dark flash stimulus. There are “motor” clusters that show much higher correlations to the motors signals, as is evident in Figure 1G.

Line 165: Did the changes observed for Pimozide fall below the significance threshold, were lethal, or were the results not repeated? It does not appear in source data 2.

Pimozide was lethal in our screen and therefore does not appear in the source data file. Indeed, in our previous experiments with Pimozide we had already established that a 10uM dose is lethal, and that the maximal effective dose we tried was 1uM as reported in (Randlett et al., Current Biology, 2019).

We have clarified this in the text:

“While the false negative rate is difficult to determine since so little is known about the pharmacology of the system, we note that of the three small molecules we previously established to alter dark flash habituation that were included in the screen, Clozapine, Haloperidol and Pimozide , the first two were identified among our hits while Pimozide was lethal at the 10\muM screening concentration.”

Fig.1B and Fig.3B are the same data, which is awkward and should be explicitly stated. But the legends do not match in terms of the rest period. Which is correct? It is also important to note the other behavioural assays in the 'rest' period.

We thank the reviewer for pointing out this discrepancy in the legend. We have corrected the typo in the figure legend of Figure 3B :

“Habituation results in a progressive decrease in responsiveness to dark flashes repeated at 1-minute intervals, delivered in 4 training blocks of 60 stimuli, separated by 1hr of rest (from 0:00-7:00).”

We have also added a statement that the data is the same as that in Figure 1B.

Figure 3-4: SSMD fingerprint, there is no description of the different behavioural parameters. What they represent is left to the reader's inference. There is no mention of SpontDisp in the GitHub for example, so it is hard to know how these different parameters were measured. Even referring to the previous manuscript on habituation (Randlet O et al, eLife, 2019) does not shed light on most of them, for example, I suppose TwoMvmt represents the 'double responses' from the previous manuscript. Furthermore, there are inconsistencies between 3C and 4B, some minor (SpontDisp becomes SpntDisp), but Curve-Tap has disappeared for example, and I suspect became BendAmp-Tap. A more thorough description of these measures, and making the naming scheme consistent, are essential for readers to know what they are looking at.

We again thank the reviewer for their careful assessment of our data, and we apologize for this sloppiness. We have gone through and made the naming of these parameters consistent in both figures, and have added another supplementary table that describes in more detail what each parameter is, and how it relates to the analysis code (Figure3_sourcedata3_SSMDFingerprintParameters.xls). This was an essential missing piece of information from our original manuscript.

Line 206: While this prioritization makes sense, how was it implemented, how was the threshold decided and which were they? A table, or supplementary figure, would help to clarify the reason behind the choices. Fig.4C being cropped only around the response probability makes it impossible to judge if the criteria were respected, as the main heatmap is too small. For example, the choice of GABA receptor antagonists is somewhat puzzling, as besides PTX it does not seem that the other compounds had strong effects, with Amoxapine for example having seemingly as much effect on Naive and Train, with little in Test. And Bicuculline gave negative SSMD for prob in the three cases. The dose-response for PTX does lend credence to its effect, but I would have liked the other compounds, especially bicuculline. The melatonin results, for example, are much more convincing and interesting in our opinion.

While in hindsight it may have been possible to do the hit prioritization in a systematic way using thresholding and ranking, we did this manually by inspecting the clustered fingerprints. We have clarified this in the text: “This manual prioritization led to the identification of the GABAA/C Receptor antagonists…”

While we agree that it is not possible to judge how well we performed this prioritization based on the images presented, we note that we do provide the full fingerprint data in the supplementary data, for which the reader is welcome to draw their own conclusions.

We have not performed further experiments with amoxapine, so we can not comment further on this. We did perform additional experiments with bicuculline, for which we did see effects similar to those of PTX, were habituation was inhibited. However, the effects are weaker and more variable than what we observe with PTX, and bicuculline also inhibits the initial responses of the larvae, causing their Naive response to be lower. Therefore we did not include it in our manuscript. We include these data here in Author response image 1 to reassure the Reviewer that picrotoxinin is not the only GABA Receptor antagonist for which we see inhibitory effects on habituation.

Author response image 1.

Fig.6: Why was the melatonin concentration used only 1um instead of 10um on the screen?

Based on dose response experiments (Figure 5B, and others not shown), we found that the effect of Melatonin on habituation saturates at about 1uM, and therefore we used this dose.

Line 277: As the correlation with motor output is marginal at best, and the authors recognize the lack of behaviour in tethered animals, I would be careful about such speculation. Especially since the other changes are complex and go in all directions.

While we appreciate the reviewer's caution, we feel that our statement is appropriately hedged using “might be”. We have also removed the statement “and thus is most closely associated with behavioural initiation”.

We now state:

“However, opposite effects of PTX and Melatonin were observed for 4_L^{strgD} neurons ([fig:6]C), which we found to be most strongly correlated with motor output ([fig:2]F). Therefore, this class might be most critical for habituation of response Probability.”

Fig.7: I am not sure how convincing these results are. 7F may have been more convincing, but to be thorough the authors would need to register the Gad1b identity to the calcium imaging and use their outline to extract the neuron's fluorescence. As it is, in the tectum, it is hard to be sure that all the identified neurons are indeed Gad1b positive, as that population is intermingled with other neuronal populations. The authors should consider the approach of Lovett-Barron M et al, Nat Neuro, 2020. Alternatively, the authors can tone down the language used in this section to match the confidence level of the association they propose.

Figure 7A-E are what can be considered “virtual colocalization” analyses, where we are comparing the localization of data acquired in different experiments using image registration to common atlas coordinates. We agree that these results alone will never be very strong evidence for the identification of individual cells. The MultiMAP approach of Lovett-Barron is a powerful approach, though it makes the assumption that registration accuracy will be subcellular, which in practice may often not be the case. We believe that a better approach is to label the cells of interest during the Ca2+ imaging experiment itself, as we did 7F and G. The challenge in this experiment is binarizing the ROIs and thus deciding what is and is not a Gad1b-positive cell. In our opinion, the fact that these two independent experiments came to the same conclusion regarding Cluster 10 and 11 is good evidence that these cell types are likely predominantly GABAergic.

As discussed above, we have re-written the manuscript to tone down our claims about the role of GABA and GABAergic neurons in habituation, which we hope the reviewer will agree better reflects the limitations of the data in Figure 6 and 7.

Line 317: Based on the somewhat inconsistent results of the other GABA antagonists, I would be careful. Picrotoxin has been reported to antagonize other receptors besides GABA, see Das P et al, Neuropharma, 2003. So the results may be explained by a complex set of effects on multiple pathways with PTX.

Off target effects are an important concern with any pharmacological experiment, and perhaps especially in zebrafish where receptors and targets can be quite divergent from those in mammals where most drug targets have been characterized. We have added this sentiment to the discussion:

“We cannot rule out the possibility that off-targets of PTX, or subtle non-specific changes in excitatory/inhibitory balance alter habituation behaviour.”

Line 400-403, 430: There are some conflicting statements regarding the potential role of clusters 1 and 2 in DF habituation. Do the authors think they play a role in the behaviour measured in this manuscript? Could they clarify what they mean?

We see how our original statement in line 429 about the presence of cluster 1 and 2 neurons in the TL implied a role in dark flash habituation. This was not our intent, and we have removed “which also contains high concentrations of on-responding neurons”.

Our thoughts on these neurons are now stated in the discussion as:

“We also observed classes exhibiting an On-response profile ( and ). These neurons fire at the ramping increase in luminance after the DF, making it unlikely that they play a role in aspects of acute DF behaviour we measured here. These neurons exist in both non-adapting and depressing forms suggesting a yet unidentified role in behavioural adaptation to repeated DFs.“

Minor comments

Line 73 (and elsewhere): Why use adaptation instead of habituation (also in the adaptation profile)? Do you suspect your observations do not reflect habituation, but a sensory adaptation mechanism?

We have used the convention that “habituation” refers to observations at the behavioural level, while “depression” and “potentiation” refer to observations at the neuronal level. We use the term “adaptation” to refer to neuronal adaptations of either sign (depression or potentiation), as in line 73.

We believe that our observations reflect neuronal adaptations that underlie habituation behaviour.

Line 71: It is debatable that the strongest learning happens in the first block, the difference between the first and last response seems to grow larger with each successive block. What do the authors mean by 'strongest'

We agree that “strongest” was ambiguous. We have changed this to “initial”:

“We focused on a single training block of 60 DFs to identify neuronal adaptations that occur during the initial phase of learning ”

Fig.1F: there is no rastermap call in the GitHub repository, was the embedding done in the GUI? If so, it should also be shared for reproducibility's sake.

Yes, Fig.1F was created using the suite2p GUI, as we have now clarified in the methods:

“The clustered heatmap image of neural activity (([fig:3]F) was generated using the suite2p GUI using the “Visualize selected cells” function, and sorting the neurons using the rastermap algorithm ”

The image is available in the “Figure1 - Ca2Imaging.svg” file available here: https://github.com/owenrandlett/lamire_2022/tree/main/LamireEtAl_2022

Line 101: while true that AffinityPropagation does not require input on the number of clusters, preference can influence the number of clusters. It seems that at least two values were tested in the search for the clusters, can the authors comment on how many clusters the other preference value converged (or failed to converge) on?

Indeed, as with any clustering approach, the resultant clusters are highly dependent on the input parameters, in this case the “preference”, as well as “damping” and the choice of affinity metric. By varying these parameters one can arrive at anywhere between 2 and hundreds of clusters.

It is for this reason that we feel that the anatomical analyses of these clusters is very important, making the assumption that neurons of differing functional types will have different localizations in the brain, as we explained in the Results:

“While these results indicate the presence of a dozen functionally distinct neuron types, such clustering analyses will force categories upon the data irrespective of if such categories actually exist. To determine if our cluster analyses identified genuine neuron types, we analyzed their anatomical localization ([fig:2]C-E). Since our clustering was based purely on functional responses, we reasoned that anatomical segregation of these clusters would be consistent with the presence of truly distinct types of neurons.”

We also acknowledge in the Results that the clustering approach has limitations:

“These results highlight a diversity of functional neuronal classes active during DF habituation. Whether there are indeed 12 classes of neurons, or if this is an over- or under-estimate, awaits a full molecular characterization. Independent of the precise number of neuronal classes, we proceed under the hypothesis that these clusters define neurons that play distinct roles in the DF response and/or its modulation during habituation learning“

Fig.2. My understanding is that the cluster numbers are arbitrary unless there is a meaning to them, which then should be explained. I would recommend grouping the clusters per functional category as in Fig.6 to make it easier for the reader.

Cluster number reflects the ordering in the hierarchical clustering tree shown in Figure 2B. We feel that this is the most logical representation of their functional similarity. We have clarified this in the Methods:

“ We then used the Affinity Propagation clustering from scikit-learn , with “affinity” computed as the Pearson product-moment correlation coefficients (corrcoef in NumPy ), preference=-9, and damping=0.9, and clustered using Hierarchical clustering (cluster.hierarchy in SciPy ). Cluster number was assigned based on the ordering of the hierarchical clustering tree. ”

Fig.3 SSMD fingerprint, it would be much easier for the readers if the list of parameters was clearer and rotated 90 degrees. Maybe in a supplementary figure to show what each represents.

We agree that the SSMD fingerprint is very difficult to interpret. As discussed above, we have now included a supplementary table (Figure3_sourcedata2_SSMDFingerprintParameters.xlsx) where we have clarified what each parameter represents.

Fig.4: The use of the same colours across the clustering methods is confusing, especially after the use of colours for the SSMD fingerprint in Fig.3. and at the bottom of 4A. Fig.4A for example could have been colour coded according to the most affected behaviour in the fingerprint at the bottom.

Fig.4B the coloured text is difficult to read, especially for the lighter colours.

We agree that our use of color is not perfect, but we have attempted to use them consistently: for example when referring to a functional cluster, or a drug manipulation. We don’t think that there is a sufficient number of distinguishable colors for us to never use the same color twice.

Fig.4C if the goal is to show similarity, the relevant drugs could be placed adjacent to each other. One could also report the Euclidean distance, or compute how correlated the different fingerprints are within one pharmacological target space.

The goal of Fig 4C is to highlight where Bicuculline, Amoxapine, Picrotoxinin, Melatonin, Ethinyl Estradiol and Hexestrol lie within the clustered heatmap of the behavioural fingerprints (Fig 4A), and<br /> demonstrate how the probability of response to dark flashes is modulated by these drugs. In our analyses, “similarity” is a function of the clustering distance.

Fig.6D 'Same data as M, ...' I assume should be 'Same data as C,...'

Indeed, thank you for pointing out this error that we have corrected.

Fig. 7 How many GCaMP6s double transgenic larvae were imaged?

6 fish were imaged, as is stated in the legend to Fig 7G

Line 407: all is repeated.

We apologize, but we do not see what is repeated at line 407. Can you please clarify?

Line 481: Would testing spontaneous activity after training for 7h be unbiased, could there be fatigue effects?

We tested for fatigue effects in our previous study, comparing larvae that received the training for 7hrs and those that did not, and we saw no deficits in spontaneous activity, tap response, or OMR performance (Figure S1, Randlett et al., Current Biology, 2019).

Line 610: There are some inconsistencies between the authors' contributions in the manuscript and the one provided to eLife.

Thank you, we will double check this in the resubmission forms. The authors' contributions in the manuscript are correct.

Reviewer #3 (Recommendations For The Authors):

I would rather recommend the authors divide this manuscript into two and publish two papers by adding some more strengthening data for each part such as cellular manipulations, e.g. ablation to prove the critical involvement of 12(Pot, M) neurons in habituation.

We thank the reviewer for their suggestion, but have opted not to split the paper into two. We feel that the collective message of this paper and approach combining molecular and functional analysis will be of interest, and we believe the incongruencies in our results reflects the complexity inherent within the system.

eLife assessment

This manuscript provides a useful reconstruction of the structure of the sirtuin-class histone deacetylase Sirt6 bound to a nucleosome based on cryo-EM observations, and additional characterization of the flexibility of the histone tails in the complex based on molecular dynamics simulations. Similar structures have recently been published, but this work provides solid support for the conclusions of those papers and also includes some novel insights into the potential dynamics of Sirt6 bound to a nucleosome that help explain its substrate specificity.

eLife assessment

Yang et al. investigate whether distinct sources of conflict are represented in a common cognitive space. The study uses an interesting task that mixes different sources of difficulty and reports that the brain appears to represent these sources as a mixture on a continuum in prefrontal areas. While the findings could be valuable to theory in this area, there are concerns with the analysis, design and results, that raise uncertainty regarding the main conclusion of a shared cognitive space. Thus, the evidence reported here remains incomplete.

eLife assessment

This useful study provides a systematic mutational analysis to elucidate mechanisms involved in transcriptional activation by the murine DUX protein, DUX is a master transcription factor regulating mammalian early embryonic gene activation and its human homolog DUX4 is also involved in a muscular disease, fascioscapulohumeral dystrophy (FSHD). The data are solid and the interpretations of the findings are reasonable. The work will be of interest to colleagues studying early embryonic development or FSHD.

eLife assessment

This important study examined the use of dantrolene, a Ryanodine Receptor stabilizer, in slowing pathological progression of pressure-overload heart failure in a guinea pig model and reducing arrhythmias. Convincing data were collected and analyzed using validated methodology and can be used as a starting point for future studies of dantrolene in Ca2+ handling in ROS production and further deterioration of cardiac function in chronic heart failure.

eLife assessment

This is valuable study on the mechanistic relationship between two prominent events in post-stimulus EEG: alpha desynchronization and P300 that are known for their slow/relatively late build up. The sample size is substantial. The data are compelling, showing that the P300 can be explained by desynchronization of a non-zero mean alpha oscillations over posterior sites through the baseline-shift model, at least partially. This makes a significant contribution to understanding and interpreting P300 generation (and possibly other ERP components) from concurrent changes in brain oscillations, with links to cognition.

Author Response

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

Thank you for the detailed and constructive reviews. We revised the paper accordingly, and a point-by-point reply appears below. The main changes are:

• An extended discussion section that places our work in context with other related developments in theory and modeling.

• A new results section that demonstrates a substantial improvement in performance from a non-linear activation function. This led to addition of a co-author.

• The mathematical proof that the resolvent of the adjacency matrix leads to the shortest path distances has been moved to a separate article, available as a preprint and attached to this resubmission. This allows us to present that work in the context of graph theory, and focus the present paper on neural modeling.

Reviewer #1 (Public Review):

This paper presents a highly compelling and novel hypothesis for how the brain could generate signals to guide navigation towards remembered goals. Under this hypothesis, which the authors call "Endotaxis", the brain co-opts its ancient ability to navigate up odor gradients (chemotaxis) by generating a "virtual odor" that grows stronger the closer the animal is to a goal location. This idea is compelling from an evolutionary perspective and a mechanistic perspective. The paper is well-written and delightful to read.

The authors develop a detailed model of how the brain may perform "Endotaxis", using a variety of interconnected cell types (point, map, and goal cells) to inform the chemotaxis system. They tested the ability of this model to navigate in several state spaces, representing both physical mazes and abstract cognitive tasks. The Endotaxis model performed reasonably well across different environments and different types of goals.

The authors further tested the model using parameter sweeps and discovered a critical level of network gain, beyond which task performance drops. This critical level approximately matched analytical derivations.

My main concern with this paper is that the analysis of the critical gain value (gamma_c) is incomplete, making the implications of these analyses unclear. There are several different reasonable ways in which the Endotaxis map cell representations might be normalized, which I suspect may lead to different results. Specifically, the recurrent connections between map cells may either be an adjacency matrix, or a normalized transition matrix. In the current submission, the recurrent connections are an unnormalized adjacency matrix. In a previous preprint version of the Endotaxis manuscript, the recurrent connections between the map cells were learned using Oja's rule, which results in a normalized state-transition matrix (see "Appendix 5: Endotaxis model and the successor representation" in "Neural learning rules for generating flexible predictions and computing the successor representation", your reference 17). The authors state "In summary, this sensitivity analysis shows that the optimal parameter set for endotaxis does depend on the environment". Is this statement, and the other conclusions of the sensitivity analysis, still true if the learned recurrent connections are a properly normalized state-transition matrix?

Yes, this is an interesting topic. In v.1 of our bioRxiv preprint we used Oja’s rule for learning, which will converge on a map connectivity that reflects the transition probabilities. The matrix M becomes a left-normalized or right-normalized stochastic matrix, depending on whether one uses the pre-synaptic or the post-synaptic version of Oja’s rule. This is explained well in Appendix 5 of Fang 2023.

In the present version of the model we use a rule that learns the adjacency matrix A, not the transition matrix T. The motivation is that we want to explain instances of oneshot learning, where an agent acquires a route after traversing it just once. For example, we had found experimentally that mice can execute a complex homing route on the first attempt.

An agent can establish whether two nodes are connected (adjacency) the very first time it travels from one node to the other. Whereas it can evaluate the transition probability for that link only after trying this and all the other available links on multiple occasions. Hence the normalization terms in Oja’s rule, or in the rule used by Fang 2023, all involve some time-averaging over multiple visits to the same node. This implements a gradual learning process over many experiences, rather than a one-shot acquisition on the first experience.

Still one may ask whether there are advantages to learning the transition matrix rather than the adjacency matrix. We looked into this with the following results:

• The result that (1/γ − A)−1 is monotonically related to the graph distances D in the limit of small γ (a proof now moved to the Meister 2023 preprint) , holds also for the transition matrix T. The proof follows the same steps. So in the small gain limit, the navigation model would work with T as well.

• If one uses the transition matrix to compute the network output (1/γ − T)-1 then the critical gain value is γc = 1. It is well known that the largest eigenvalue of any Markov transition matrix is 1, and the critical gain γc is the inverse of that. This result is independent of the graph. So this offers the promise that the network could use the same gain parameter γ regardless of the environment.

• In practice, however, the goal signal turned out to be less robust when based on T than when based on A. We illustrate this with the attached Author response image 1. This replicates the analysis in Figure 3 of the manuscript, using the transition matrix instead of the adjacency matrix. Some observations:

• Panel B: The goal signal follows an exponential dependence on graph distance much more robustly for the model with A than with T. This holds even for small gain values where the exponential decay is steep.

• Panel C: As one raises the gain closer to the critical value, the goal signal based on T scatters much more than when based on A.

• Panels D, E: Navigation based on A works better than based on T. For example, using the highest practical gain value, and a readout noise of ϵ = 0.01, navigation based on T has a range of only 8 steps on this graph, whereas navigation based on A ranges over 12 steps, the full size of this graph.

We have added a section “Choice of learning rule” to explain this. The Author response image 1 is part of the code notebook on Github.

Author response image 1.

Overall, this paper provides a very compelling model for how neural circuits may have evolved the ability to navigate towards remembered goals, using ancient chemotaxis circuits.

This framework will likely be very important for understanding how the hippocampus (and other memory/navigation-related circuits) interfaces with other processes in the brain, giving rise to memory-guided behavior.

Reviewer #2 (Public Review):

The manuscript presents a computational model of how an organism might learn a map of the structure of its environment and the location of valuable resources through synaptic plasticity, and how this map could subsequently be used for goal-directed navigation.

The model is composed of 'map cells', which learn the structure of the environment in their recurrent connections, and 'goal-cell' which stores the location of valued resources with respect to the map cell population. Each map cell corresponds to a particular location in the environment due to receiving external excitatory input at this location. The synaptic plasticity rule between map cells potentiates synapses when activity above a specified threshold at the pre-synaptic neuron is followed by above-threshold activity at the post-synaptic neuron. The threshold is set such that map neurons are only driven above this plasticity threshold by the external excitatory input, causing synapses to only be potentiated between a pair of map neurons when the organism moves directly between the locations they represent. This causes the weight matrix between the map neurons to learn the adjacency for the graph of locations in the environment, i.e. after learning the synaptic weight matrix matches the environment's adjacency matrix. Recurrent activity in the map neuron population then causes a bump of activity centred on the current location, which drops off exponentially with the diffusion distance on the graph. Each goal cell receives input from the map cells, and also from a 'resource cell' whose activity indicates the presence or absence of a given values resource at the current location. Synaptic plasticity potentiates map-cell to goal-cell synapses in proportion to the activity of the map cells at time points when the resource cell is active. This causes goal cell activity to increase when the activity of the map cell population is similar to the activity where the resource was obtained. The upshot of all this is that after learning the activity of goal cells decreases exponentially with the diffusion distance from the corresponding goal location. The organism can therefore navigate to a given goal by doing gradient ascent on the activity of the corresponding goal cell. The process of evaluating these gradients and using them to select actions is not modelled explicitly, but the authors point to the similarity of this mechanism to chemotaxis (ascending a gradient of odour concentration to reach the odour source), and the widespread capacity for chemotaxis in the animal kingdom, to argue for its biological plausibility.

The ideas are interesting and the presentation in the manuscript is generally clear. The two principle limitations of the manuscript are: i) Many of the ideas that the model implements have been explored in previous work. ii) The mapping of the circuit model onto real biological systems is pretty speculative, particularly with respect to the cerebellum.

Regarding the novelty of the work, the idea of flexibly navigating to goals by descending distance gradients dates back to at least Kaelbling (Learning to achieve goals, IJCAI, 1993), and is closely related to both the successor representation (cited in manuscript) and Linear Markov Decision Processes (LMDPs) (Piray and Daw, 2021, https://doi.org/ 10.1038/s41467-021-25123-3, Todorov, 2009 https://doi.org/10.1073/pnas.0710743106). The specific proposal of navigating to goals by doing gradient descent on diffusion distances, computed as powers of the adjacency matrix, is explored in Baram et al. 2018 (https://doi.org/10.1101/421461), and the idea that recurrent neural networks whose weights are the adjacency matrix can compute diffusion distances are explored in Fang et al. 2022 (https://doi.org/10.1101/2022.05.18.492543). Similar ideas about route planning using the spread of recurrent activity are also explored in Corneil and Gerstner (2015, cited in manuscript). Further exploration of this space of ideas is no bad thing, but it is important to be clear where prior literature has proposed closely related ideas.

We have added a discussion section on “Theories and models of spatial learning” with a survey of ideas in this domain and how they come together in the Endotaxis model.

Regarding whether the proposed circuit model might plausibly map onto a real biological system, I will focus on the mammalian brain as I don't know the relevant insect literature. It was not completely clear to me how the authors think their model corresponds to mammalian brain circuits. When they initially discuss brain circuits they point to the cerebellum as a plausible candidate structure (lines 520-546). Though the correspondence between cerebellar and model cell types is not very clearly outlined, my understanding is they propose that cerebellar granule cells are the 'map-cells' and Purkinje cells are the 'goal-cells'. I'm no cerebellum expert, but my understanding is that the granule cells do not have recurrent excitatory connections needed by the map cells. I am also not aware of reports of place-field-like firing in these cell populations that would be predicted by this correspondence. If the authors think the cerebellum is the substrate for the proposed mechanism they should clearly outline the proposed correspondence between cerebellar and model cell types and support the argument with reference to the circuit architecture, firing properties, lesion studies, etc.

On further thought we agree that the cerebellum-like circuits are not a plausible substrate for the endotaxis algorithm. The anatomy looks compelling, but plasticity at the synapse is anti-hebbian, and - as the reviewer points out - there is little evidence for recurrence among the inputs. We changed the discussion text accordingly.

The authors also discuss the possibility that the hippocampal formation might implement the proposed model, though confusingly they state 'we do not presume that endotaxis is localized to that structure' (line 564).

We have removed that confusing bit of text.

A correspondence with the hippocampus appears more plausible than the cerebellum, given the spatial tuning properties of hippocampal cells, and the profound effect of lesions on navigation behaviours. When discussing the possible relationship of the model to hippocampal circuits it would be useful to address internally generated sequential activity in the hippocampus. During active navigation, and when animals exhibit vicarious trial and error at decision points, internally generated sequential activity of hippocampal place cells appears to explore different possible routes ahead of the animal (Kay et al. 2020, https://doi.org/10.1016/j.cell.2020.01.014, Reddish 2016, https:// doi.org/10.1038/nrn.2015.30). Given the emphasis the model places on sampling possible future locations to evaluate goal-distance gradients, this seems highly relevant.

In our model, the possible future locations are sampled in real life, with the agent moving there or at least in that direction, e.g. via VTE movements. In this simple form the model has no provision for internal planning, and the animal never learns any specific route sequence. One can envision extending such a model with some form of sequence learning that would then support an internal planning mechanism. We mention this in the revised discussion section, along with citation of these relevant articles.

Also, given the strong emphasis the authors place on the relationship of their model to chemotaxis/odour-guided navigation, it would be useful to discuss brain circuits involved in chemotaxis, and whether/how these circuits relate to those involved in goal-directed navigation, and the proposed model.

The neural basis of goal-directed navigation is probably best understood in the insect brain. There the locomotor decisions seem to be initiated in the central complex, whose circuitry is getting revealed by the fly connectome projects. This area receives input from diverse sensory areas that deliver the signal on which the decisions are based. That includes the mushroom body, which we argue has the anatomical structure to implement the endotaxis algorithm. It remains a mystery how the insect chooses a particular goal for pursuit via its decisions. It could be revealing to force a change in goals (the mode switch in the endotaxis circuit) while recording from brain areas like the central complex. Our discussion now elaborates on this.

Finally, it would be useful to clarify two aspects of the behaviour of the proposed algorithm:

1) When discussing the relationship of the model to the successor representation (lines 620-627), the authors emphasise that learning in the model is independent of the policy followed by the agent during learning, while the successor representation is policy dependent. The policy independence of the model is achieved by making the synapses between map cells binary (0 or 1 weight) and setting them to 1 following a single transition between two locations. This makes the model unsuitable for learning the structure of graphs with probabilistic transitions, e.g. it would not behave adaptively in the widely used two-step task (Daw et al. 2011, https://doi.org/10.1016/ j.neuron.2011.02.027) as it would fail to differentiate between common and rare transitions. This limitation should be made clear and is particularly relevant to claims that the model can handle cognitive tasks in general. It is also worth noting that there are algorithms that are closely related to the successor representation, but which learn about the structure of the environment independent of the subjects policy, e.g. the work of Kaelbling which learns shortest path distances, and the default representation in the work of Piray and Daw (both referenced above). Both these approaches handle probabilistic transition structures.

Yes. Our problem statement assumes that the environment is a graph with fixed edge weights. The revised text mentions this and other assumptions in a new section “Choice of learning rule”.

2) As the model evaluates distances using powers of adjacency matrix, the resulting distances are diffusion distances not shortest path distances. Though diffusion and shortest path distances are usually closely correlated, they can differ systematically for some graphs (see Baram et al. ci:ted above).

The recurrent network of map cells implements a specific function of the adjacency matrix, namely the resolvent (Eqn 7). We have a mathematical proof that this function delivers the shortest graph distances exactly, in the limit of small gain (γ in Eqn 7), and that this holds true for all graphs. For practical navigation in the presence of noise, one needs to raise the gain to something finite. Figure 3 analyzes how this affects deviations from the shortest graph distance, and how nonetheless the model still supports effective navigation over a surprising range. The mathematical details of the proof and further exploration of the resolvent distance at finite gain have been moved to a separate article, which is cited from here, and attached to the submission. The preprint by Baram et al. is cited in that article.

Reviewer #3 (Public Review):

This paper argues that it has developed an algorithm conceptually related to chemotaxis that provides a general mechanism for goal-directed behaviour in a biologically plausible neural form.

The method depends on substantial simplifying assumptions. The simulated animal effectively moves through an environment consisting of discrete locations and can reliably detect when it is in each location. Whenever it moves from one location to an adjacent location, it perfectly learns the connectivity between these two locations (changes the value in an adjacency matrix to 1). This creates a graph of connections that reflects the explored environment. In this graph, the current location gets input activation and this spreads to all connected nodes multiplied by a constant decay (adjusted to the branching number of the graph) so that as the number of connection steps increases the activation decreases. Some locations will be marked as goals through experiencing a resource of a specific identity there, and subsequently will be activated by an amount proportional to their distance in the graph from the current location, i.e., their activation will increase if the agent moves a step closer and decrease if it moves a step further away. Hence by making such exploratory movements, the animal can decide which way to move to obtain a specified goal.

I note here that it was not clear what purpose, other than increasing the effective range of activation, is served by having the goal input weights set based on the activation levels when the goal is obtained. As demonstrated in the homing behaviour, it is sufficient to just have a goal connected to a single location for the mechanism to work (i.e., the activation at that location increases if the animal takes a step closer to it); and as demonstrated by adding a new graph connection, goal activation is immediately altered in an appropriate way to exploit a new shortcut, without the goal weights corresponding to this graph change needing to be relearnt.

As the reviewer states, allowing a graded strengthening of multiple synapses from the map cells increases the effective range of the goal signal. We have now confirmed this in simulations. For example, in the analysis of Fig 3E, a single goal synapse enables perfect navigation only over a range of 7 steps, whereas the distributed goal synapses allow perfect navigation over the full 12 steps. This analysis is included in the code notebook on Github.

Given the abstractions introduced, it is clear that the biological task here has been reduced to the general problem of calculating the shortest path in a graph. That is, no real-world complications such as how to reliably recognise the same location when deciding that a new node should be introduced for a new location, or how to reliably execute movements between locations are addressed. Noise is only introduced as a 1% variability in the goal signal. It is therefore surprising that the main text provides almost no discussion of the conceptual relationship of this work to decades of previous work in calculating the shortest path in graphs, including a wide range of neural- and hardwarebased algorithms, many of which have been presented in the context of brain circuits.

The connection to this work is briefly made in appendix A.1, where it is argued that the shortest path distance between two nodes in a directed graph can be calculated from equation 15, which depends only on the adjacency matrix and the decay parameter (provided the latter falls below a given value). It is not clear from the presentation whether this is a novel result. No direct reference is given for the derivation so I assume it is novel. But if this is a previously unknown solution to the general problem it deserves to be much more strongly featured and either way it needs to be appropriately set in the context of previous work.

As far as we know this proposal for computing all-pairs-shortest-path is novel. We could not find it in textbooks or an extended literature search. We have discussed it with two graph theorist colleagues, who could not recall seeing it before, although the proof of the relationship is elementary. Inspired by the present reviewer comment, we chose to publish the result in a separate article that can focus on the mathematics and place it in the appropriate context of prior work in graph theory. For related work in the area of neural modeling please see our revised discussion section.

Once this principle is grasped, the added value of the simulated results is somewhat limited. These show: 1) in practical terms, the spreading signal travels further for a smaller decay but becomes erratic as the decay parameter (map neuron gain) approaches its theoretical upper bound and decreases below noise levels beyond a certain distance. Both follow the theory. 2) that different graph structures can be acquired and used to approach goal locations (not surprising) .3) that simultaneous learning and exploitation of the graph only minimally affects the performance over starting with perfect knowledge of the graph. 4) that the parameters interact in expected ways. It might have been more impactful to explore whether the parameters could be dynamically tuned, based on the overall graph activity.

This is a good summary of our simulation results, but we differ in the assessment of their value. In our experience, simulations can easily demolish an idea that seemed wonderful before exposure to numerical reality. For example, it is well known that one can build a neural integrator from a recurrent network that has feedback gain of exactly 1. In practical simulations, though, these networks tend to be fickle and unstable, and require unrealistically accurate tuning of the feedback gain. In our case, the theory predicts that there is a limited range of gains that should work, below the critical value, but large enough to avoid excessive decay of the signal. Simulation was needed to test what this practical range was, and we were pleasantly surprised that it is not ridiculously small, with robust navigation over a 10-20% range. Similarly, we did not predict that the same parameters would allow for effective acquisition of a new graph, learning of targets within the graph, and shortest-route navigation to those targets, without requiring any change in the operation of the network.

Perhaps the most biologically interesting aspect of the work is to demonstrate the effectiveness, for flexible behaviour, of keeping separate the latent learning of environmental structure and the association of specific environmental states to goals or values. This contrasts (as the authors discuss) with the standard reinforcement learning approach, for example, that tries to learn the value of states that lead to reward. Examples of flexibility include the homing behaviour (a goal state is learned before any of the map is learned) and the patrolling behaviour (a goal cell that monitors all states for how recently they were visited). It is also interesting to link the mechanism of exploration of neighbouring states to observed scanning behaviours in navigating animals.

The mapping to brain circuits is less convincing. Specifically, for the analogy to the mushroom body, it is not clear what connectivity (in the MB) is supposed to underlie the graph structure which is crucial to the whole concept. Is it assumed that Kenyon cell connections perform the activation spreading function and that these connections are sufficiently adaptable to rapidly learn the adjacency matrix? Is there any evidence for this?

Yes, there is good evidence for recurrent synapses among Kenyon cells (map cells in the model), and for reward-gated synaptic plasticity at the synapses onto mushroom body output cells (goal cells in our model). We have expanded this material in the discussion section. Whether those functions are sufficient to learn the structure of a spatial environment has not been explored; we hope our paper might give an impetus, and are exploring behavioral experiments on flies with colleagues.

As discussed above, the possibility that an algorithm like 'endotaxis' could explain how the rodent place cell system could support trajectory planning has already been explored in previous work so it is not clear what additional insight is gained from the current model.

Please see our revised discussion section on “theories and models of spatial learning”. In short, some ingredients of the model have appeared in prior work, but we believe that the present formulation offers an unexpectedly simple end-to-end solution for all components of navigation: exploration, target learning, and goal seeking.

Reviewer #1 (Recommendations For The Authors):

Major concern:

See the public review. How do the results change depending on whether the recurrent connections between map cells are an adjacency matrix vs. a properly normalized statetransition matrix? I'm especially asking about results related to critical gain (gamma_c), and the dependence of the optimal parameter values on the environment.

Please see our response above including the attached reviewer figure.

Minor concerns:

It is not always clear when the learning rule is symmetric vs asymmetric (undirected vs directed graph), and it seems to switch back and forth. For example, line 127 refers to a directed graph; Fig 2B and the intro describe symmetric Hebbian learning. Most (all?) of the simulations use the symmetric rule. Please make sure it's clear.

For simplicity we now use a symmetric rule throughout, as is appropriate for undirected graphs. We mention that a directed learning rule could be used to learn directed graphs. See the section on “choice of learning rule”. M_ij is not defined when it's first introduced (eq 4). Consider labeling the M's and the G's in Fig 2.

Done.

The network gain factor (gamma, eq 4) is distributed over both external and recurrent inputs (v = gamma(u + Mv)), instead of local to the recurrent weights like in the Successor Representation. This notational choice is obviously up to the authors. I raise slight concern for two reasons -- first, distributing gamma may affect some of the parameter sweep results (see major concern), and second, it may be confusing in light of how gamma is used in the SR literature (see reviewer's paper for the derivation of how SR is computed by an RNN with gain gamma).

In our model, gamma represents the (linear) activation function of the map neuron, from synaptic input to firing output. Because the synaptic input comes from point cells and also from other map cells, the gain factor is applied to both. See for example the Dayan & Abbott book Eqn 7.11, which at steady state becomes our Eqn 4. In the formalism of Fang 2023 (Eqn 2), the factor γ is only applied to the recurrent synaptic input J ⋅ f, but somehow not to the place cell input ϕ. Biophysically, one could imagine applying the variable gain only to the recurrent synapses and not the feed-forward ones. Instead we prefer to think of it as modulating the gain of the neurons, rather than the synapses. The SR literature follows conventions from the early reinforcement learning papers, which were unconstrained by thinking about neurons and synapses. We have added a footnote pointing the reader to the uses of γ in different papers.

In eq 13, and simulations, noise is added to the output only, not to the activity of recurrently connected neurons. It is possible this underestimates the impact of noise since the same magnitude of noise in the recurrent network (map cells) could have a compounded effect on the output.

Certainly. The equivalent output noise represents the cumulative effect of noise everywhere in the network. We argue that a cumulative effect of 1% is reasonable given the overall ability of animals at stimulus discrimination, which is also limited by noise everywhere in the network. This has been clarified in the text.

Fig 3 E, F, it looks like the navigated distance may be capped. I ask because the error bars for graph distance = 12 are so small/nonexistent. If it's capped, this should be in the legend.

Correct. 12 is the largest distance on this graph. This has been added to the caption.

Fig 3D legend, what does "navigation failed" mean? These results are not shown.

On those occasions the agent gets trapped at a local maximum of the goal signal other than the intended goal. We have removed that line as it is not needed to interpret the data.

Line 446, typo (Lateron).

Fixed.

Line 475, I'm a bit confused by the discussion of birds and bats. Bird behavior in the real world does involve discrete paths between points. Even if they theoretically could fly between any points, there are costs to doing so, and in practice, they often choose discrete favorite paths. It is definitely plausible that animals that can fly could also employ Endotaxis, so it is confusing to suggest they don't have the right behavior for Endotaxis, especially given the focus on fruit flies later in the discussion.

Good points, we removed that remark. Regarding fruit flies, they handle much important business while walking, such as tracking a mate, fighting rivals over food, finding a good oviposition site.

Section 9.3, I'm a bit confused by the discussion of cerebellum-like structures, because I don't think they have as dense recurrent connections as needed for the map cells in Endotaxis. Are you suggesting they are analogous to the output part of Endotaxis only, not the whole thing?

Please see our reply in the public review. We have removed this discussion of cerebellar circuits.

Line 541, "After sufficient exploration...", clarify that this is describing learning of just the output synapses, not the recurrent connections between map cells?

We have revised this entire section on the arthropod mushroom body.

In lines 551-556, the discussion is confusing and possibly not consistent with current literature. How can a simulation prove that synapses in the hippocampus are only strengthened among immediately adjacent place fields? I'd suggest either removing this discussion or adding further clarification. More broadly, the connection between Endotaxis and the hippocampus is very compelling. This might also be a good point to bring up BTSP (though you do already bring it up later).

As suggested, we removed this section.

Line 621 "The successor representation (at least as currently discussed) is designed to improve learning under a particular policy" That's not actually accurate. Ref 17 (reviewer's manuscript, cited here) is not policy-specific, and instead just learns the transition statistics experienced by the animal, using a biologically plausible learning rule that is very similar to the Endotaxis map cell learning rule (see our Appendix 5, comparing to Endotaxis, though that was referencing the previous version of the Endotaxis preprint where Oja's rule was used).

We have edited this section in the discussion and removed the reference to policyspecific successor representations.

Line 636 "Endotaxis is always on" ... this was not clear earlier in the paper (e.g. line 268, and the separation of different algorithms, and "while learning do" in Algorithm 2).

The learning rules are suspended during some simulations so we can better measure the effects of different parts of endotaxis, in particular learning vs navigating. There is no interference between these two functions, and an agent benefits from having the learning rules on all the time. The text now clarifies this in the relevant sections.

Section 9.6, I like the idea of tracing different connected functions. But when you say "that could lead to the mode switch"... I'm a bit confused about what is meant here. A mode switch doesn't need to happen in a different brain area/network, because winnertake-all could be implemented by mutual inhibition between the different goal units.

This is an interesting suggestion for the high-level control algorithm. A Lorenzian view is that the animal’s choice of mode depends on internal states or drives, such as thirst vs hunger, that compete with each other. In that picture the goal cells represent options to be pursued, whereas the choice among the options occurs separately. But one could imagine that the arbitrage between drives happens through a competition at the level of goal cells: For example the consumption of water could lead to adaptation of the water cell, such that it loses out in the winner-take-all competition, the food cell takes over, and the mouse now navigates towards food. In this closed-loop picture, the animal doesn’t have to “know” what it wants at any given time, it just wants the right thing. This could eliminate the homunculus entirely! Of course this is all a bit speculative. We have edited the closing comments in a way that leaves open this possibility.

Line 697-704, I need more step-by-step explanation/derivation.

We now derive the properties of E step by step starting from Eqn (14). The proof that leads to Eqn 14 is now in a separate article (available as a preprint and attached to this submission).

Reviewer #3 (Recommendations For The Authors):

• Please include discussion and comparison to previous work of graph-based trajectory planning using spreading activation from the current node and/or the goal node. Here is a (far from comprehensive) list of papers that present similar algorithms:

Glasius, R., Komoda, A., & Gielen, S. C. (1996). A biologically inspired neural net for trajectory formation and obstacle avoidance. Biological Cybernetics, 74(6), 511-520.

Gaussier, P., Revel, A., Banquet, J. P., & Babeau, V. (2002). From view cells and place cells to cognitive map learning: processing stages of the hippocampal system. Biological cybernetics, 86(1), 15-28.

Gorchetchnikov A, Hasselmo ME. A biophysical implementation of a bidirectional graph search algorithm to solve multiple goal navigation tasks. Connection Science. 2005;17(1-2):145-166

Martinet, L. E., Sheynikhovich, D., Benchenane, K., & Arleo, A. (2011). Spatial learning and action planning in a prefrontal cortical network model. PLoS computational biology, 7(5), e1002045.

Ponulak, F., & Hopfield, J. J. (2013). Rapid, parallel path planning by propagating wavefronts of spiking neural activity. Frontiers in computational neuroscience, 7, 98.

Khajeh-Alijani, A., Urbanczik, R., & Senn, W. (2015). Scale-free navigational planning by neuronal traveling waves. PloS one, 10(7), e0127269.

Adamatzky, A. (2017). Physical maze solvers. All twelve prototypes implement 1961 Lee algorithm. In Emergent computation (pp. 489-504). Springer, Cham.

Please see our reply to the public review above, and the new discussion section on “Theories and models of spatial learning”, which cites most of these papers among others.

• Please explain, if it is the case, why the goal cell learning (other than a direct link between the goal and the corresponding map location) and calculation of the overlapping 'goal signal' is necessary, or at least advantageous.

Please see our reply in the public review above.

• Map cells are initially introduced (line 84) as getting input from "only one or a few point cells". The rest of the paper seems to assume only one. Does it work when this is 'a few'? Does it matter that 'a few' is an option?

We simplified the text here to “only one point cell”. A map cell with input from two distant locations creates problems. After learning the map synapses from adjacencies in the environment, the model now “believes” that those two locations are connected. This distorts the graph on which the graph distances are computed and introduces errors in the resulting goal signals. One can elaborate the present toy model with a much larger population of map cells that might convey more robustness, but that is beyond our current scope.

• (line 539 on) Please explain what feature in the mushroom body (or other cerebellumlike) circuits is proposed to correspond to the learning of connections in the adjacency matrix in the model.

Please see our response to this critique in the public review above. In the mushroom body, the Kenyon cells exhibit sparse responses and are recurrently connected. These would correspond to map cells in Endotaxis. For vertebrate cerebellum-like circuits, the correspondence is less compelling, and we have removed this topic from the discussion.

eLife assessment

This valuable work proposes a framework inspired by chemotaxis for understanding how the brain might implement behaviours related to navigating toward a goal. The evidence supporting the conceptual claim is convincing. The manuscript proposes a hypothesis that would be of interest to the broad systems neuroscience community, although it was noted the relationship to existing similar hypotheses could be clarified.

Reviewer #1 (Public Review):

This paper presents a highly compelling and novel hypothesis for how the brain could generate signals to guide navigation toward remembered goals. Under this hypothesis, which the authors call "Endotaxis", the brain co-opts its ancient ability to navigate up odor gradients (chemotaxis) by generating a "virtual odor" that grows stronger the closer the animal is to a goal location. This idea is compelling from an evolutionary perspective and a mechanistic perspective. The paper is well-written and delightful to read.

The authors develop a detailed model of how the brain may perform "Endotaxis", using a variety of interconnected cell types (point, map, and goal cells) to inform the chemotaxis system. They tested the ability of this model to navigate in several state spaces, representing both physical mazes and abstract cognitive tasks. The Endotaxis model performed reasonably well across different environments and different types of goals.

The authors further tested the model using parameter sweeps and discovered a critical level of network gain, beyond which task performance drops. This critical level approximately matched analytical derivations.

Overall, this paper provides a very compelling model for how neural circuits may have evolved the ability to navigate towards remembered goals, using ancient chemotaxis circuits.

This framework will likely be very important for understanding how the hippocampus (and other memory/navigation-related circuits) interfaces with other processes in the brain, giving rise to memory-guided behavior.

Reviewer #2 (Public Review):

The manuscript presents a computational model of how an organism might learn a map of the structure of its environment and the location of valuable resources through synaptic plasticity, and how this map could subsequently be used for goal-directed navigation.

The model is composed of 'map cells', which learn the structure of the environment in their recurrent connections, and 'goal-cell' which store the location of valued resources with respect to the map cell population. Each map cell corresponds to a particular location in the environment due to receiving external excitatory input at this location. The synaptic plasticity rule between map cells potentiates synapses when activity above a specified threshold at the pre-synaptic neuron is followed by above-threshold activity at the post-synaptic neuron. The threshold is set such that map neurons are only driven above this plasticity threshold by the external excitatory input, causing synapses to only be potentiated between a pair of map neurons when the organism moves directly between the locations they represent. This causes the weight matrix between the map neurons to learn the adjacency for the graph of locations in the environment, i.e. after learning the synaptic weight matrix matches the environment's adjacency matrix. Recurrent activity in the map neuron population then causes a bump of activity centred on the current location, which drops off exponentially with the diffusion distance on the graph. Each goal cell receives input from the map cells, and also from a 'resource cell' whose activity indicates the presence or absence of a given values resource at the current location. Synaptic plasticity potentiates map-cell to goal-cell synapses in proportion to the activity of the map cells at time-points when the resource cell is active. This causes goal cell activity to increase when the activity of the map cell population is similar to the activity where the resource was obtained. The upshot of all this is that after learning the activity of goal cells decreases exponentially with the diffusion distance from the corresponding goal location. The organism can therefore navigate to a given goal by doing gradient ascent on the activity of the corresponding goal cell. The process of evaluating these gradients and using them to select actions is not modelled explicitly, but the authors point to the similarity of this mechanism to chemotaxis (ascending a gradient of odour concentration to reach the odour source), and the widespread capacity for chemotaxis in the animal kingdom, to argue for its biological plausibility. The ideas are interesting and the presentation of the results in the manuscript is generally clear.

Closely related ideas have been explored in previous work, and there are some aspects of how the work relates to previous literature that it would be useful to clarify. Several lines of work have proposed learning long-range relationships between states in the environment, to enable navigation to rewarding goals by effectively descending distance gradients. The most well-known of these in the neuroscience literature is the Successor Representation (SR) (Dayan 1993), which is defined as the expected discounted future occupancy of each state given the current state. As noted in the discussion, this is closely related to the representation learnt by the map cells in the current model. The key difference is that the successor representation uses state-state transitions under a given policy (a mapping from states to actions), whereas the current model uses the adjacency matrix between states, which depends only on the environment and hence is independent of the policy followed while the representation is learnt (given sufficient exploration). This policy independence is useful, as the SR can fail to generate good routes to goals when these are very different from the policy under which it was learned (see Russek et al. https://doi.org/10.1371/journal.pcbi.1005768). However, there are several prior proposals for policy-independent SR-like mechanisms that it would be useful to discuss. Baram et al. (https://doi.org/10.1101/421461) propose navigating to goals by doing gradient descent on diffusion distances, computed as powers of the adjacency matrix as in the current work. One limitation of using the adjacency matrix is that it does not handle situations where transitions between states are probabilistic, which is not a big issue for navigation in physical space but is for applying the mechanism to cognitive tasks more broadly. There are prior ideas for learning policy-independent representations similar to the SR that do not have this limitation. Kaelbling (Learning to achieve goals, IJCAI, 1993) proposed using an off-policy learning rule similar to Q-learning, to learn shortest path distances between states. Piray and Daw https://doi.org/10.1038/s41467-021-25123-3) consider a default representation, which is a successor-like representation under a generic default policy, building on the Linear Markov Decision Process (LMDP) framework of Todorov (https://doi.org/10.1073/pnas.0710743106). Also relevant to the current study is the work of Fang et al. (https://doi.org/10.7554/eLife.80680) who, as in the current work, propose using recurrent network dynamics to compute a long-range representation (the SR) from synaptic weights that store local transition information.

One other area where I felt the work could be better integrated with the existing literature was the discussion of mapping the model onto brain circuits. An interesting and attractive aspect of the work is the idea that the relatively high-level operation of goal-directed navigation could be built on top of evolutionarily older mechanisms for ascending odour gradients. Given this framing, I was expecting the discussion of brain circuits to consider interactions between spatial mapping systems and regions involved in olfactory processing. However the discussion of mammalian brains focussed exclusively on the hippocampus without any link to olfaction, which feels like a missed opportunity. I am not an expert on olfaction, but one region that seems particularly interesting in this context is the olfactory tubercle (see Wesson & Wilson https://doi.org/10.1016/j.neubiorev.2010.08.004 for a review). This region is contiguous with the ventral striatum and has similar local circuitry, receives strong input from olfactory regions, but also input from the hippocampal formation, and a strong dopaminergic innervation from VTA. This suggests a mapping of the model to brain circuits in which map cells in the hippocampal formation project to goal cells in the olfactory tubercle, with the dopaminergic input acting as resource cells (note that different dopamine neuron populations appear to respond to different reward types, see e.g. https://doi.org/10.1038/s41586-022-04954-0, https://doi.org/10.1101/2023.05.09.540067). I was also surprised not to see any discussion of internally generated sequential activity in the hippocampus as a possible mechanism for the look-ahead needed to evaluate the goal distance gradient, particularly given the authors suggest that vicarious trial and error (VTE) is a behavioural signature of this gradient sampling, and it is known that during VTE hippocampus plays out internally generated sequences of possible future locations (see Redish https://doi.org/10.1038/nrn.2015.30).

Reviewer #3 (Public Review):

This paper describes an algorithm that provides a general mechanism for goal-directed behaviour in a biologically plausible neural form.

The method depends on substantial simplifying assumptions. The simulated animal effectively moves through an environment consisting of discrete locations and can reliably detect when it is in each location. Whenever it moves from one location to an adjacent location, it perfectly learns the connectivity between these two locations (changes the value in an adjacency matrix to 1). This creates a graph of connections that reflects the explored environment. In this graph, the current location gets input activation and this spreads to all connected nodes multiplied by a constant decay (adjusted to the branching number of the graph) so that as the number of connection steps increases the activation decreases. Some locations will be marked as goals through experiencing a resource of a specific identity there and subsequently will be activated by an amount proportional to their distance in the graph from the current location, i.e., their activation will increase if the agent moves a step closer and decrease if it moves a step further away. Hence by making such exploratory movements, the animal can decide which way to move to obtain a specified goal.

Although the algorithm is presented within a conceptual framework of chemotaxis, I.e., making movements to sample a local gradient and move up it, the approach relates closely to previous models of exploration, learning, and navigation that similarly establish (through experience) a graph structure to represent how locations are connected and use some form of activity-propagation from the current node or goal node to identify a (shortest) route between them. Many of these similarly claim to be plausible neural circuits. The current authors argue that the current algorithm has several desirable features with respect to such previous work: for example, the 'readout' of the path does not require explicit 'look-up' and activation of the goal node (although it does require a choice of which goal node is currently connected to behavior); and does not require any separate control or rules for learning vs. navigation phases. By comparison to the successor representation method used in RL, which also appears related, they note that the gain (decay) factor is not equivalent to a temporal discount and that their method learns only state-state transitions, allowing the value of actions to be externalised, I.e., calculated by trying alternative actions to see which increases the activation at the goal node the most. On the other hand, it should be noted that some issues addressed in previous models, such as uncertainty over the current state or probabilistic state(-action) transitions are not addressed in this work.

The algorithm presents some elegant features with respect to previous work such as conceptually separating the 'goal' nodes from the state (location) graph (I.e. 'goals' are not just special target states within the graph) so that a small number of goals can become associated to (potentially) multiple regions of the state graph where they are satisfied, or near to being satisfied. This architecture is suggested, in the discussion, to resemble the insect mushroom body (MB), where it is known that a small number of output neurons (MBONs, putative goal neurons) are activated by plastic connections from Kenyon cells (KCs, putative state neurons). However, it goes substantially beyond any available evidence to claim that KC connectivity could support the acquisition of a graph (in the form of an adjacency matrix) representing the structure of the environment: KCs show sparse distributed activity (not one active node per state); it seems unlikely that any two arbitrary KCs can (rapidly) become connected; and as yet has not been demonstrated that KC connectivity is plastic at all.

The results presented are fairly straightforward given the simplification of the tasks, as described above. They show 1) in practical terms, the spreading signal travels further for a larger decay but becomes erratic as the decay parameter (map neuron gain) approaches its theoretical upper bound and decreases below noise levels beyond a certain distance. Both follow the theory but it is perhaps helpful to see that there is a viable range of values of the gain for which the mechanism works, that is, it is not highly dependent on precise tuning. 2) That different graph structures can be acquired and used to approach goal locations (not surprising). 3) That simultaneous learning and exploitation of the graph only minimally affects the performance over starting with perfect knowledge of the graph. 4) That the parameters interact in expected ways. 5) That the separation of goals from states can be used flexibly e.g. the homing behaviour (a goal state is learned before any of the map is learned) and the patrolling behaviour (a goal cell that monitors all states for how recently they were visited). It is also interesting to link the mechanism of exploration of neighbouring states to observed scanning behaviours in navigating animals. It would have been interesting to explore whether the parameters could be dynamically tuned, based on the overall graph activity.

Reviewer #3 (Public Review):

In this manuscript, Lewis et al. investigate the role of tetraspanins in the formation of discs- the key structure of vertebrate photoreceptors essential for light reception. Two tetraspanin proteins play a role in this process: PRPH2 and ROM1. The critical contribution of PRPH2 has been well established and loss of its function is not tolerated and result in gross anatomical pathology and degeneration in both mice and humans. However, the role of ROM1 is much less understood and has been considered somewhat redundant. This paper provides a definitive answer about the long-standing uncertainty regarding the contribution of ROM1 firmly establishing its role in outer segment morphogenesis. First, using ingenious quantitative proteomic technique the authors show PRPH2 compensatory increase in ROM1 knockout explaining the redundancy of its function. Second, they uncover that despite this compensation, ROM1 is still needed and its loss delays disc enclosure and result in the failure to form incisures. Third, the authors used a transgenic mouse model and show that deficits seen in ROM1 KO could be completely compensated by the overexpression of PRPH2. Finally, they analyzed yet another mouse model based on double manipulation with both ROM1 loss and expression of PRPH2 mutant unable to form dimerizing disulfide bonds further arguing that PRPH2-ROM1 interactions are not required for disc enclosure. To top it off the authors complement their in vivo studies by series of biochemical assays done upon reconstitution of tetraspanins in transfected cultured cell as well as fractionations of native retinas. This report is timely, addresses significant questions in cell biology of photoreceptors and pushes the field forward in a classical area of photoreceptor biology and mechanics of membrane structure as well. The manuscript is executed at the top level of technical standard, exceptionally well written and does not leave much more to desire. It also pushes standards of the field- one such domain is quantitative approach to analysis of the EM images which is notoriously open to alternative interpretations - yet this study does an exceptional job unbiasing this approach.

Reviewer #2 (Public Review):

In this study, Lewis et al seek to further define the role of ROM1. ROM1 is a tetraspanin protein that oligomerizes with another tetraspanin, PRPH2, to shape the rims of the membrane discs that comprise the light sensitive outer segment of vertebrate photoreceptors. ROM1 knockout mice and several PRPH2 mutant mice are reexamined. The conclusion reached is that ROM1 is redundant to PRPH2 in regulating the size of newly forming discs, although excess PRPH2 is required to compensate for the loss of ROM1.

This replicates earlier findings, while adding rigor using a mass spectrometry-based approach to quantitate the ratio of ROM1 and PRPH2 to rhodopsin (the protein packed in the body of the disc membranes) and careful analysis of tannic acid labeled newly forming discs using transmission electron microscopy.

In ROM1 knockout mice PRPH2 expression was found to be increased so that the level of PRPH2 in those mice matches the combined amount of PRPH2 and ROM1 in wildtype mice. Despite this, there are defects in disc formation that are resolved when the ROM1 knockout is crossed to a PRPH2 overexpressing line. A weakness of the study is that the molar ratios between ROM1, PRPH2 and rhodopsin were not measured in the PRPH2 overexpressing mice. This would have allowed the authors to be more precise in their conclusion that a sufficient excess of PRPH2 can compensate for defects in ROM1.

Author Response

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

Public Reviews:

Reviewer #1 (Public Review):

Cell death plays a critical role on regulating organogenesis. During tooth morphogenesis, apoptosis of embryonic dental tissue plays critical roles on regulating tooth germ development. The current study focused on ferroptosis, another way of cell death which has rarely been investigated in tooth development, and showed it may also play an important role on regulating the tooth dimension. The topic is novel and interesting, but the experimental design has many flaws which significantly compromised the study.

1. The entire study was based on ex vivo tooth germ explant culture. Mandibular tooth germs of E15.5 (bell stage) were isolated for ex vivo culture. Most tooth germ explant culture experiments were actually using tooth germ of much earlier stages (E11.5-E13.5) for organ culture. After E16.5, both the large size and initially formed enamel/dentin could prevent nutrition from penetrating inside. Also, using tooth germ of earlier stage will help identify impact of ferroptosis upon early tooth development.

2. Due to limited penetration, the ex vivo culture in the study lasted for no more than 5 days. I would recommend the authors to perform kidney capsule transplantation as an alternative approach, which can support tooth germ development much longer even into root formation.

3. The major justification of using tooth germ ex vivo culture as the model in the study was to "conduct high-throughput analysis". However, the study could hardly be qualified as a high-throughput analysis. I would recommend the authors perform RNA sequencing for comparing tooth germs before/after erastin treatment. Such experiments won't take too much time or resource.

We are grateful for the insightful feedback on our ex vivo tooth germ culture model. We initially chose the E15.5 tooth germ over earlier stages due to peak Gpx4 expression and iron accumulation during molar development, which occurs between E15.5 and E17.5 (Figure 1A & 1B). This period may be the most sensitive to ferroptotic stress during tooth development. Our experiments also demonstrated that the tooth germ displays robust growth after seven days of ex vivo cultivation (Figure supplement 1B).

Kidney capsule transplantation is indeed an ideal method for ex vivo tooth germ culture. However, in our studies, we used erastin – a classic ferroptosis inducer – which exhibits instability in vivo, thereby constraining our investigation using kidney capsule transplantation.

Our results about Gpx4 expression in the tooth germ during development (Figure 1A) showed a spatiotemporal pattern. This pattern suggests that bulk RNA sequencing of the tooth germ might not yield accurate revelations about changes in ferroptosis-related genes. We are presently using transgenic mice to further study the impact of excessive in vivo ferroptotic stress on tooth development. In these experiments, we intend to conduct single-cell RNA sequencing to explore detailed alterations in the tooth germ.

1. Although the study mostly used molars as the model, the in vivo iron concentration was only demonstrated on incisors, but not molars (Figure 1).

We have updated Figure 1B to include images of molars, which illustrate the accumulation of iron during molar development. The iron concentration peaks at E17.5, then decreases at PN0. Interestingly, unlike Gpx4 expression, iron accumulation rebounds at PN3. To gain a more accurate understanding, further in vivo studies utilizing transgenic mice are required.

1. Phenotype analysis in Figure 2 is too superficial. Only dimensional information was provided. Cusps number, cusps distribution pattern and rooth/furcation formation were not evaluated. Differentiation of ameloblast/odontoblast was not evaluated. The proliferation rate in the dental epithelium/mesenchyme was not analyzed.

The cusps number/distribution pattern are not influenced by erastin treatment in recent model (Figure 2A & 2C). Recent ex vivo culture model of tooth germ is unable to investigate the possible function of ferroptotic stress in rooth/furcation formation since it mainly initiates from PN4 to PN7. The proliferation and differentiation of dental epithelium/mesenchyme will be analyzed using transgenic mice in vivo.

1. Low magnification images should be included in Figure 3 to display the entire tooth germs.

The emission spectrum of recent utilized iron probe will extend due to increasing concentration of iron. This property makes the counter staining of tissue samples unavailable. The structure of the ex vivo cultured tooth germ could only be recognized in high magnification. The calculation could represent the entire alternation.

1. In Figure 4, does ferroptotic inhibitor eliminate the iron accumulation in the tooth germ? How about the expression level of several target genes shown in Figure 3?

In Fig 5, Fer-1 reduced the iron accumulation in tooth germ. Different inhibitors suppressed ferroptosis via different ways, Lip-1 mainly inhibits lipid peroxidation, DFO is an iron chelator which reduces the labile iron pool, Fer-1 is reported to both inhibit lipid peroxidation and reduce the labile iron pool, their functions to the accumulation of iron might be varied. The core risk factors of ferroptosis are lipid peroxidation and iron accumulation, thus in Fig 5, we analyzed the expression of 4HNE and the accumulation of iron to illustrated the suppression o ferroptosis instead of detecting several regulatory genes.

1. The manuscript has many typos and grammar mistakes. All "submandibular" should be simply "mandibular". "eastin" should be "erastin" (line 92). "partly" should be "partially" (line 611).

We addressed all the gramma and typo errors.

Reviewer #2 (Recommendations for The Authors):

This is a very well done study. However, writing is absolutely substandard. The authors should check and review extensively for improvements to the use of English. This is not just about language but also about style of the paper and presentation. As written, the abstract is not concise at all, and the overall logic of the study is not well presented. Currently, the abstract reads like another introduction.

We improved our presentation.

Reviewer #3 (Recommendations for The Authors):

This is an interesting work reporting ferroptosis that is involved in the tooth morphogenesis. The authors showed that Gpx4, the core anti-lipid peroxidation enzyme in ferroptosis, is upregulated in tooth development using ex vivo culture system. They convincingly demonstrated that ferroptosis, but apoptosis, was present in tooth morphogenesis. The findings are interesting and novel. The work represents one of the earliest works studying Ferroptosis in tooth morphogenesis. There are several minor concerns.

1) The abstract is too long and should be shortened.

We modified the abstract to make it concise.

2) Can the Gpx4 quantitatively be measured by qRT-PCR?

3) How is Gpx4 regulated during development? If unknown, the authors should discuss it at least

4) Are there any tooth developmental defects associated with ferroptosis? If there is one, the authors should discuss it.

Our research on Gpx4 expression in the tooth germ during development (Figure 1A) highlights a specific spatiotemporal pattern. This pattern suggests that bulk RNA sequencing of the tooth germ may not provide accurate insight into changes in ferroptosis-related genes.

The developmental role of Gpx4 had been studied even before the ferroptosis was formally described (before 2012). In situ hybridization indicated expression of Gpx4 in all developing germ layers during gastrulation and in the somite stage in the developing central nervous system and in the heart, which made Gpx4 (-/-) mice die embryonically in utero by midgestation (E7.5) and are associated with a lack of normal structural compartmentalization. Specific deletion of Gpx4 during developmental process were found to participate in the maturation and survival of cerebral and photoreceptor cell. Recent years, more ferroptosis related function of Gpx4 were discovered in neutrophil and chondrocyte of adult mice, in which specific deletion will lead to ferroptosis-induced organ dysregulation and degeneration.

At present, no systematic study has been conducted on ferroptosis or ferroptotic stress in relation to tooth developmental defects. However, as early as the 1930s, pioneering dental biologists had already identified the presence of iron in the teeth of various animals. They also found that some enamel defects in mice were related to abnormal iron metabolism. Lipid metabolism and lipid peroxidation, which are other key risk factors of ferroptosis, were also described in the initial stages of dental biology research.

We are currently generating transgenic mice with dental epithelium/mesenchymal specific deletions of Gpx4. This will allow us to further investigate the developmental defects related to ferroptosis and ferroptotic stress.

eLife assessment

This important study time elegantly demonstrates that ferroptotic stress may play critical roles in regulating tooth germ development. The evidence presented is compelling, based on an explant model and providing novel mechanistic insights into tooth development.

Reviewer #1 (Public Review):

Cell death plays a critical role on regulating organogenesis. During tooth morphogenesis, apoptosis of embryonic dental tissue plays critical roles on regulating tooth germ development. The current study focused on ferroptosis, another way of cell death which has rarely been investigated in tooth development, and showed it may also play an important role on regulating the tooth dimension. The topic is novel and interesting, but the experimental design has some flaws which compromised the study.

The entire study was based on ex vivo tooth germ explant culture. I hope the authors can continue working on this direction with more convincing transgenic models.

Reviewer #2 (Public Review):

The present study by Ye et al. characterizes some of the major effects of ferroptotic stress on tooth morphogenesis.

The strengths of this study are its innovative nature and the beautiful histology. Mechanistic data are convincing Overall, the study is well done.

Reviewer #3 (Public Review):

This is an interesting work reporting ferroptosis that is involved in the tooth morphogenesis. The authors showed that Gpx4, the core anti-lipid peroxidation enzyme in ferroptosis, is upregulated in tooth development using ex vivo culture system.

Author Response

Reviewer #1 (Public Review):

The authors performed an RNAi screen to identify epigenetic regulators involved in oxygen-glucose deprivation (OGD)-induced neuronal injury using immortalized mouse hippocampal neuronal cell line HT-22. They identified PRMT5 as a novel negative regulator of neuronal cell survival after OGD. Both in vitro and in vivo experiments were then performed to evaluate the roles of PRMT5 in OGD and ischemic stroke-induced injury. The authors found that genetic and pharmacological inhibition of PRMT5 protected against neuronal cell death in both in vitro and in vivo models. Furthermore, they found that in response to OGD and ischemia, PRMT5 was translocated from the cytosol to the nucleus, where PRMT5 bound to the chromatin and promoter regions of targeted genes to repress the expression of downstream genes. Further, they showed that silencing PRMT5 significantly altered the OGD-induced changes for a large-scale of genes. In a mouse model of middle cerebral artery occlusion (MCAO), PRMT5 inhibitor EPZ015666 protected against neuronal death in vivo. This study reveals a potential therapeutic target for the treatment of ischemic stroke. Overall, the authors have done elegant work showing the role of PRMT5 in neuronal cell survival. However, the essential mechanisms underlying PRMT5 nuclear translocation have not been investigated, and the in vivo animal studies should be further strengthened.

Thank you very much for your comments and suggestions. While stroke stands as the second leading cause of death globally, and the burden of post-onset disability is substantial, particularly surging at a faster rate in low- and middle-income countries compared to high-income countries. The exploration of new drugs for stroke treatment holds profound societal implications. The concept of neuroprotective drug development is not novel; over the past half-century, considerable research and resources have been invested in this field. Yet, progress appears to be notably limited, and interest is currently waning.

Our research team is dedicated to devising rapid and cost-effective functional screening strategies grounded in the nervous system. Through this forward research approach, we aim to delve into potential neuroprotective targets across various neurological diseases. This endeavor not only bears significance for acute stroke but also holds potential application value for a spectrum of generalized nerve injuries.

Building on your insights, our upcoming studies will involve in vivo animal experiments, integrating the PRMT5 nuclear translocation mechanism. We anticipate that our continued research will benefit from further professional insights and guidance from your expertise.

Reviewer #2 (Public Review):

Haoyang Wu et al. have shown that the symmetric arginine methyltransferase PRMT5 binds to the promoter region of several essential genes and represses their expression, leading to neuronal cell death. Knocking down PRMT5 in HT-22 cells by shRNA leads to pertinent improvement in cell survival after oxygen-glucose deprivation (OGD) conditions. In another set of experiments, inhibition of the catalytic activity of PRMT5 by a specific inhibitor, EPZ015666, in a middle cerebral artery occlusion (MCAO) mice model also showed protective effects against neuronal cell death. In this manuscript, the authors have established the negative role of PRMT5 in cerebral ischemia both in vitro and in vivo.

However, my primary concern is the novelty of the manuscript. It has already been reported that inhibition of PRMT5 attenuates cerebral ischemia/reperfusion condition (Inhibition of PRMT5 attenuates cerebral ischemia/reperfusion-induced inflammation and pyroptosis through suppression of NF-κB/NLRP3 axis. Xiang Wu et al. Neuroscience Letters, Volume 776, 2022, 136576, ISSN 0304-3940, https://doi.org/10.1016/j.neulet.2022.136576.). Even these authors have also shown that treatment of PRMT5 specific catalytic inhibitor, LLY-283, could rescue ischemia-induced over-expression of inflammation-related factors.

However, it would be better to verify the specificity of the inhibitor, EPZ015666, using other methyltransferases to be sure that the rescue is indeed mediated by PRMT5 catalytic inhibition.

Thank you sincerely for dedicating time from your busy schedule to review our papers. Your comments and suggestions hold immense value for us, contributing significantly to the enhancement of our work. We acknowledge with honesty that this research journey has been a prolonged and challenging experience.

The major functional study, as indicated by the CHIP-seq data record, was concluded between 2017 and 2019. Since then, our efforts and resources have been devoted to conducting in-depth mechanism and regulation research for PRMT5. Notably, PRMT5 is involved in 4-5 types of histone arginine methylation, and it plays a role in complex modification effects for proteins in the cytoplasm. Despite employing a variety of investigative methods, understanding and controlling these intricate mechanisms in experimental design have proven quite challenging. This not only places us at a disadvantage compared to some competitors but also hinders the creative potential of our lab team.

We firmly believe that there is ample room for further research on the role of PRMT5 in the nervous system. We aspire to collaborate with other research teams to explore this area collectively.

eLife assessment

The authors performed a useful RNAi screen to identify epigenetic regulators involved in oxygen-glucose deprivation (OGD)-induced neuronal injury. PRMT5 was identified as a negative regulator of neuronal cell survival after OGD. Solid in vitro and in vivo data suggest that PRMT5 could be a novel therapeutic target for the treatment of ischemic stroke.

Reviewer #1 (Public Review):

The authors performed an RNAi screen to identify epigenetic regulators involved in oxygen-glucose deprivation (OGD)-induced neuronal injury using immortalized mouse hippocampal neuronal cell line HT-22. They identified PRMT5 as a novel negative regulator of neuronal cell survival after OGD. Both in vitro and in vivo experiments were then performed to evaluate the roles of PRMT5 in OGD and ischemic stroke-induced injury. The authors found that genetic and pharmacological inhibition of PRMT5 protected against neuronal cell death in both in vitro and in vivo models. Furthermore, they found that in response to OGD and ischemia, PRMT5 was translocated from the cytosol to the nucleus, where PRMT5 bound to the chromatin and promoter regions of targeted genes to repress the expression of downstream genes. Further, they showed that silencing PRMT5 significantly altered the OGD-induced changes for a large-scale of genes. In a mouse model of middle cerebral artery occlusion (MCAO), PRMT5 inhibitor EPZ015666 protected against neuronal death in vivo. This study reveals a potential therapeutic target for the treatment of ischemic stroke. Overall, the authors have done elegant work showing the role of PRMT5 in neuronal cell survival. However, the essential mechanisms underlying PRMT5 nuclear translocation have not been investigated, and the in vivo animal studies should be further strengthened.

Reviewer #2 (Public Review):

Haoyang Wu et al. have shown that the symmetric arginine methyltransferase PRMT5 binds to the promoter region of several essential genes and represses their expression, leading to neuronal cell death. Knocking down PRMT5 in HT-22 cells by shRNA leads to pertinent improvement in cell survival after oxygen-glucose deprivation (OGD) conditions. In another set of experiments, inhibition of the catalytic activity of PRMT5 by a specific inhibitor, EPZ015666, in a middle cerebral artery occlusion (MCAO) mice model also showed protective effects against neuronal cell death. In this manuscript, the authors have established the negative role of PRMT5 in cerebral ischemia both in vitro and in vivo.

However, my primary concern is the novelty of the manuscript. It has already been reported that inhibition of PRMT5 attenuates cerebral ischemia/reperfusion condition (Inhibition of PRMT5 attenuates cerebral ischemia/reperfusion-induced inflammation and pyroptosis through suppression of NF-κB/NLRP3 axis. Xiang Wu et al. Neuroscience Letters, Volume 776, 2022, 136576, ISSN 0304-3940, https://doi.org/10.1016/j.neulet.2022.136576.). Even these authors have also shown that treatment of PRMT5 specific catalytic inhibitor, LLY-283, could rescue ischemia-induced over-expression of inflammation-related factors.

However, it would be better to verify the specificity of the inhibitor, EPZ015666, using other methyltransferases to be sure that the rescue is indeed mediated by PRMT5 catalytic inhibition.

Reviewer #2 (Public Review):

Summary:<br /> Here the authors address the idea that postural and movement control are differentially impacted with stroke. Specifically, they examined whether resting postural forces influenced several metrics of sensorimotor control (e.g., initial reach angle, maximum lateral hand deviation following a perturbation, etc.) during movement or posture. The authors found that resting postural forces influenced control only following the posture perturbation for the paretic arm of stroke patients, but not during movement. They also found that resting postural forces were greater when the arm was unsupported, which correlated with abnormal synergies (as assessed by the Fugl-Meyer). The authors suggest that these findings can be explained by the idea that the neural circuitry associated with posture is relatively more impacted by stroke than the neural circuitry associated with movement. They also propose a conceptual model that differentially weights the reticulospinal tract (RST) and corticospinal tract (CST) to explain greater relative impairments with posture control relative to movement control, due to abnormal synergies, in those with stroke.

Strengths:<br /> The strength of the paper is that they clearly demonstrate with the posture task (i.e., active holding against a load) that the resting postural forces influence subsequent control (i.e., the path to stabilize, time to stabilize, max. deviation) following a sudden perturbation (i.e., suddenly removal of the load). Further, they can explain their findings with a conceptual model, which is depicted in Figure 9.

Weaknesses:<br /> Current weaknesses and potential concerns relate to i) not displaying or reporting the results of healthy controls and non-paretic arm in Experiment 2 and ii) large differences in force perturbation waveforms between movement (sudden onset) and posture (sudden release), which could potentially influence the results and or interpretation.

Larger concerns<br /> 1. Additional analyses to further support the interpretation. In Experiment 1 the authors present the results for the paretic arm, non-paretic arm, and controls. However, in Experiment 2 for several key analyses, they only report summary statistics for the paretic arm (Figure 5D-I; Figure 6D-E; Figure 7F). It is understood that the controls have much smaller resting postural force biases, but they are still present (Figure 3B). It would strengthen the position of the paper to show that controls and the non-paretic arm are not influenced by resting postural force biases during movement and particularly during posture, while acknowledging the caveat that the resting positional forces are smaller in these groups. It is recommended that the authors report and display the results shown in Figure 5D-I; Figure 6D-E; Figure 7F for the controls and non-paretic arm. If these results are all null, the authors could alternatively place these results in an additional supplementary.

Further, the results could be further boosted by reporting/displaying additional analyses. In Figure 6D the authors performed a correlation analysis. Can they also display the same analysis for initial deviation and endpoint deviation for the data shown in Figure 5D-F & 5G-I, as well for 7F for the path to stabilization, time to stabilization, and max deviation? This will also create consistency in the analyses performed for each dependent variable across the paper.

2. Inconsistency in perturbations that would differentially impact muscle and limb states during movement and posture. It is well known that differences in muscle state (activation / preloaded, muscle fiber length and velocity) and limb state (position and velocity) impact sensorimotor control (Pruszynski, J. A., & Scott, S. H. (2012). Experimental brain research, 218, 341-359.). Of course, it is appreciated that it is not possible to completely control all states when comparing movement and posture (i.e., muscle and limb velocity). However, using different perturbations differentially impacts muscle and limb states. Within this paper, the authors used very different force waveforms for movement perturbations (i.e., 12 N peak, bell-shaped, 0.7ms duration -> sudden force onset to push the limb; Figure 6A) and posture perturbations (i.e., 6N, 2s ramp up -> 3s hold -> sudden force release that resulted in limb movement; Figure 4) that would differentially impact muscle (and limb) states. Preloaded muscle (as in the posture perturbation) has a very different response compared to muscle that has little preload (as in the movement perturbations, where muscles that would resist a sudden lateral perturbation would likely be less activated since they are not contributing to the forward movement). Would the results hold if the same perturbation had been used for both posture and movement (e.g., 12 N pulse for both experiments)? It is recommended that the authors comment and discuss in the paper why they chose different perturbations and how that might impact the results.

Relatedly, an alternative interpretation of the results is that preloading muscle for stroke patients, whether by supporting the weight of one's arm (experiment 1) or statically resisting a load prior to force release (experiment 2), leads to a greater postural force bias that can subsequently influence control. It is recommended that the authors comment on this.

eLife assessment

This valuable study extends the previous interesting work of this group to address the potentially differential control of movement and posture. Through experiments in which stroke participants used a robotic manipulandum, the authors provide evidence supporting a lack of a relation between the resting force postural bias they measure (closely related to the flexor synergy in stroke) and kinematic deficits during movement. Based on these results, the authors propose a conceptual framework that differentially weights the two main descending pathways (corticospinal tract and reticulospinal tract) for neurologically intact and stroke patients. The reviewers point out that some of the evidence supporting the authors' conclusions is incomplete, and that the study would benefit from considering alternative explanations involving other mechanisms, which could be addressed with additional experiments and analyses.

Reviewer #1 (Public Review):

This study extends the previous interesting work of this group to address the potentially differential control of movement and posture. Their earlier work explored a broad range of data to make the case for a downstream neural integrator hypothesized to convert descending velocity movement commands into postural holding commands. Included in that data were observations from people with hemiparesis due to stroke. The current study uses similar data but pushes into a different, but closely related direction, suggesting that these data may address the independence of these two fundamental components of motor control. I find the logic laid out in the second sentence of the abstract ("The paretic arm after stroke is notable for abnormalities both at rest and during movement, thus it provides an opportunity to address the relationships between control of reaching, stopping, and stabilizing") less than compelling, but the study does make some interesting observations. Foremost among them, is the relation between the resting force postural bias and the effect of force perturbations during the target hold periods, but not during movement. While this interesting observation is consistent with the central mechanism the authors suggest, it seems hard to me to rule out other mechanisms, including peripheral ones.

On the other hand, the relation between force bias and the well-recognized flexor synergy seems rather self-evident, and I don't see that these results add much to that story. I am also struck by what seems to be a contradiction between the conclusions of the current and former studies: "These findings in stroke suggest that moving and holding still are functionally separable modes of control" and "the commands that hold the arm and finger at a target location depend on the mathematical integration of the commands that moved the limb to that location." The former study is mentioned here only in passing, in a single phrase in the discussion, with no consideration of the relation between the two studies. This is odd and should be addressed.

A minor wording concern I had is that the term "holding still" is frequently hard to parse. A couple of examples: "These findings in stroke suggest that moving and holding still are functionally separable modes of control." This example is easily read, "moving and holding [continue to be] functionally separable". Another: "...active reaching and holding still in the same workspace, " could be "...active reaching and holding [are] still in the same workspace." Simply "holding", "posture" or "posture maintenance" would all be better options.

Reviewer #3 (Public Review):

The authors attempt to dissociate differences in resting vs active vs perturbed movement biases in people with motor deficits resulting from stroke. The analysis of movement utilizes techniques that are similar to previous motor control in both humans and non-human primates, to assess impairments related to sensorimotor injuries. In this regard, the authors provide additional support to the extensive literature describing movement abnormalities in patients with hemiparesis both at rest and during active movement. The authors describe their intention to separate out the contribution of holding still at a position vs active movement as a demonstration that these two aspects of motor control are controlled by two separate control regimes.

Strengths:<br /> 1. The authors utilize a device that is the same or similar to devices previously used to investigate motor control of movement in normal and impaired conditions in humans and non-human primates. This allows comparisons to existing motor control studies.<br /> 2. Experiment 1 demonstrates resting flexion biases both in supported and unsupported forelimb conditions. These biases show a correlated relationship with FM-UE scores, suggesting that the degree of motor impairment and the degree of resting bias are related.<br /> 3. The stroke patient participant population had a wide range of both levels of impairment and time since stroke, including both sub-acute and chronic cases allowing the results to be compared across impairment levels.

The authors describe several results from their study: 1. Postural biases were systematically toward the body (flexion) and increased with distance from the body (when the arm was more extended) and were stronger when the arm was unsupported. 2. These postural biases were correlated with FM-UE score. 3. They found no evidence of postural biases impacting movement, even when that movement was perturbed. 4. When holding a position at the end of a movement, if the position was perturbed opposite of the direction of bias, movement back to the target was improved compared to the perturbation in the direction of bias. Taken together, the authors suggest that there are at least two separate motor controls for tasks at rest versus with motion. Further, the authors propose that these results indicate that there is an imbalance between cortical control of movement (through the corticospinal tracts) and postural control (through the reticulospinal tract). There are several weaknesses related to the interpretation of the results:

In Experiment 1, the participants are instructed to keep their limbs in a passive position after being moved. The authors show that, in the impaired limb, these resting biases are significantly higher when the limb is unsupported and increase when the arm is moved to a more extended position.

When supported by the air sled, the arm is in a purely passive position, not requiring the same anti-gravity response so will have less RST but also less CST involvement. While the unsupported task invokes more involvement of the reticulospinal tract (RST), it likely also has significantly higher CST involvement due to the increased difficulty and novelty of the task.

If there were an imbalance in CST regulating RST as proposed by the authors, the bias should be higher in the supported condition as there should be relatively less CST activation/involvement/modulation leading to less moderating input onto the RST and introducing postural biases. In the unsupported condition, there is likely more CST involvement, potentially leading to an increased modulatory effect on RST. If the proportion of CST involvement significantly outweighs the RST activation in the unsupported task, then it isn't obvious that there is a clear differentiation of motor control. As the degree of resting force bias and FM-UE score are correlated, an argument could be made that they are both measuring the impairment of the CST unrelated to any RST output. If it is purely the balance of CST integrity compared to RST, then the degree of bias should have been the same in both conditions. In this idea of controller vs modulator, it is unclear when this switch occurs or how to weigh individual contributions of CST vs. extrapyramidal tracts. Further, it isn't clear why less modulation on the RST would lead only to abnormal flexion.

This resting bias could be explained by an imbalance in the activation of flexors vs extensors which follows the results that this bias is larger as the arm is extended further, and/or in a disconnect in sensory integration that is overcome during active movement. Neither would necessitate separate motor control for holding vs active movement.

In Experiment 2, the participants are actively moving to and holding at targets for all trials while being supported by the air sled. Even with the support, the paretic participants all showed start- and end-point force biases around the movement despite not showing systematic deviations in force direction during active movement start or stop. There could be several factors that limit systematic deviations in force direction. The most obvious is that the measured biases are significantly higher when the limb is unsupported and by testing with a supported limb the authors are artificially limiting any effect of the bias. It is also possible that significant adaptation or plasticity with the CST or rubrospinal tracts could give rise to motor output that already accounts for any intrinsic resting bias. In any case, the results from the reaching phase of Experiment 2 do not definitively show that directional biases are not present during active reaching, just that the authors were unable to detect them with their design. The authors do acknowledge the limitations in this design (a 2D constrained task) in explaining motor impairment in 3D unconstrained tasks.

It would have been useful, in Experiment 2, to use FM-UE scores (and time from injury) as a factor to determine the relationship between movement and rest biases. Using a GLMM would have allowed a similar comparison to Experiment 1 of how impairment level is related to static perturbation responses. While not a surrogate for imaging tractography data showing a degree of CST involvement in stroke, FM-UE may serve as an appropriate proxy so that this perturbation at hold responses may be put into context relative to impairment.

It is not clear that even in the static perturbation trials that the hold (and subsequent move from perturbation) is being driven by reticulospinal projections. Given a task where ~20% of the trials are going to be perturbed, there is likely a significant amount of anticipatory or preparatory signaling from the CST. How does this balance with any proposed contribution that the RST may have with increased grip?

In general, the weakness of the interpretation of the results with respect to the CST/RST framework is that it is necessary to ascribe relative contributions of different tracts to different phases of movement and hold using limited or indirect measures. Barring any quantification of this data during these tasks, different investigators are likely to assess these contributions in different ways and proportions limiting the framework's utility.

Reviewer #3 (Public Review):

Summary:<br /> Ishii et al used molecular genetics and behavioral analyses in mice to study the functional role of a subset of MPOA neurons in the regulation of female sexual drive. They first employed a self-paced mating assay during which a female could control the amount of interaction time with a male to assess female sexual drive after completion of mating. The authors observed that after mating completion females spent significantly less time interacting with the mated males, indicating that their sexual drive was reduced. Next, the authors performed a brain-wide analysis of neurons activated during the completion of mating and identified the MPOA as a strong candidate region. However, their activity labeling was not exclusive to neurons activated during mating completion but included all neurons activated before, during, and after the mating encounter. This makes it difficult to interpret these data. Importantly, the authors do provide in vivo calcium imaging data showing that a subset of MPOA neurons responds significantly and specifically to mating completion and not other behaviors during the social encounter. The authors performed these studies in both excitatory and inhibitory populations of the MPOA. Their analysis identified a subpopulation of inhibitory neurons that exhibit sustained increased activity for 90 sec following mating completion. Finally, the authors used chemogenetics to activate MPOA neurons during home cage mating, condition place preference, pup retrieval, and the self-paced mating assay. They found that activation of these neurons significantly reduced mating behaviors and time spent interacting with a male during the self-paced mating assay. The authors suggest that their chemogenetic activation is restricted to neurons activated during mating completion, but their activity-dependent labeling strategy resulted in chemogenetic activation of all MPOA neurons activated either before, during, or after mating.

The authors' experimental execution is rigorous and well-performed. Their data identify inhibitory neurons in the female MPOA as a neural locus that is activated following the completion of mating and potentially a key neural population in the regulation of female sexual motivation. However, the conclusions and interpretation of the data extend beyond what is reasonable given the limitations of the activity-dependent labeling strategy employed.

Strengths:<br /> 1) The use of the self-paced mating assay in combination with neural imaging and manipulation to assess female sexual drive is innovative. The authors correctly assert that relatively little is known about how mating completion affects sexual motivation in females as compared to males. Therefore, the data collected from these studies is important and valuable.

2) The authors provide convincing histological data and analyses to verify and validate their brain-wide activity labeling, neural imaging, and chemogenetic studies.

3) The single-cell in vivo calcium imaging data are well performed and analyzed. They provide key insights into the activity profiles of both excitatory and inhibitory neurons in the female MPOA during mating encounters. The authors' identification of an inhibitory subpopulation of female MPOA neurons that are selectively activated following the completion of mating is fundamental for future experiments which could potentially find a molecular marker for this population and specifically manipulate these neurons to understand their role in female sexual motivation in greater detail.

Weaknesses:<br /> 1) Their activity-dependent labeling strategy is not exclusive to mating completion but instead includes all neurons active before, during, and after the social encounter. In the manuscript, the authors did not discuss the time course of Fos activation or the timeframe of the FosTRAP labeling strategy. Fos continues to be expressed and is detectable for hours following neural activation. Therefore, the FosTRAP strategy also labels neurons that were activated 3 hours before the injection of 4-OHT. The original FosTRAP2 paper which is cited in this manuscript (DeNardo et al, 2019) performed a detailed analysis of the labeling window in Supplementary Figure 2 of that paper. Here is quoted text from that paper: "Resultant patterns of tdTomato expression revealed that the majority of TRAPing occurred within a 6-hour window centered around the 4-OHT injection." Thus, the FosTRAP "mating completion" groups throughout this manuscript also include neurons activated 3 hours before mating completion, which includes neurons activated during appetitive and consummatory mating behaviors.

This makes all of the FosTRAP data very difficult to interpret. Compounding this is the issue that the two groups the authors compare in their experiments are females administered 4-OHT following appetitive investigation behaviors (with the male removed before mating behaviors occurred) and females administered 4-OHT following mating completion. The "appetitive" group labeled neurons activated only during appetitive investigation, but the "completion" group labeled neurons activated during appetitive investigations, consummatory mating bouts, and mating completion. Therefore, in the brain-wide analysis of Figure 2, it is impossible to identify brain regions that were activated exclusively by mating completion and not by consummatory mating behaviors. This could have been achieved if the "completion" group was compared to a group of females that had commenced consummatory mating behaviors but were separated from the male before mating was completed. Then, any neurons labeled by the "completion" FosTRAP but not the "consummatory" FosTRAP would be neurons specifically activated by mating completion. In the current brain-wide analysis experiments, neurons activated by consummatory behaviors and mating completion can not be disassociated.

This same issue is present in the interpretation of the chemogenetic activation data in Figure 6. In the experiments of Figure 6, the authors are activating neurons naturally activated during consummatory mating behaviors as well as those activated during mating completion.

2) This study does not definitively show that the female mice used in this study display decreased sexual motivation after the completion of mating. The females exhibit reduced interaction with males that had also just completed mating, but it is unclear if the females would continue to show reduced interaction time if given the choice to interact with a male that was not in the post-ejaculatory refractory period. Perhaps, these females have a natural preference to interact more with sexually motivated males compared to recently mated (not sexually motivated) males. To definitively show that these females exhibit decreased sexual motivation the authors should perform two control experiments: 1) provide the females with access to a fully sexually motivated male after the females have completed mating with a different male to see if interaction time changes, and 2) compare interaction time toward mated and non-mated males using the self-paced mating assay. These controls would show that the reduction in the interaction time is because the females have reduced sexual motivation and not because these females just naturally interact with sexually motivated males more than males in the post-ejaculatory refractory period.

3) It is unclear how the transient 90-second response of these MPOA neurons following the completion of mating causes the prolonged reduction in female sexual motivation that is at the minutes to hours timeframe. No molecular or cellular mechanism is discussed.

4) The authors discuss potential cell types and neural population markers within the MPOA and go into some detail in Figure S3. However, their experiments are performed with only the larger excitatory and inhibitory MPOA neural populations.