Individual differences in fear memory expression engage distinct functional brain networks

  1. Barbara D Fontana
  2. Jacob Hudock
  3. Neha Rajput
  4. Dea Kanani
  5. Dinh Luong
  6. Justin W Kenney

  • Curated by eLife

    eLife logo

    eLife Assessment

    This important work sets out to identify the neural substrates of associative fear responses in adult zebrafish. Through a compelling and innovative paradigm and analysis, the authors identify brain regions associated with individual differences in fear memory expression. While most findings are well supported, there is limited evidence that the four behavioral clusters represent discrete forms of associative fear-memory expression and the related interpretations would benefit from additional analysis or more cautious framing. Nonetheless, this study showcases the strength of zebrafish for systems-level neuroscience and will be of broad interest to the neuroscience community.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Fearful stimuli elicit a mix of active (e.g., evasion) and passive (e.g., freezing) behaviors in a wide range of species, including zebrafish (Danio rerio). How these different responses are encoded in the brain, and the extent to which individuals vary in their fear responses, is not clear. To investigate this, we first developed a contextual fear conditioning paradigm in adult zebrafish where fish associate a specific tank with an aversive pheromone, conspecific alarm substance (CAS). We collected data from over 300 fish across four different strains (AB, TU, TL, and WIK) and both sexes finding that, as expected, fish exhibit a mix of active and passive responses during CAS exposure and during memory expression. We also found that behavior fell into four distinct groups: non-reactive, evaders, evading freezers, and freezers, with the evading freezer and freezer groups most clearly associated with memory formation. Background strain and sex also influenced how fish respond to CAS, with males more likely to increase evasive behaviors in response to CAS and the TU strain more likely to be non-reactive during recall. Finally, using whole-brain activity mapping, we identified the brain regions associated with active and passive fear responses during memory expression and how the functional brain networks of evading freezers and freezers differed. Freezing behavior was associated with widespread neural activity that was particularly strong in the cerebellum, reticular formation, and parts of the telencephalon associated with olfactory processing. Evasive behavior was associated with an increase in the activity of visual threat detection and a decrease in the activity of brain regions related to foraging and navigation. Network analysis revealed that animals with high freezing and low evasion (i.e., freezers) had strong interactions between the pallium and cerebellum as well as strong interconnectivity between visual (pretectal) and thalamic analog (preglomerular) nuclei. Animals that mix freezing with evasive behaviors (i.e., evading freezers) had higher subpallial connectivity to regions involved in autonomic function and stress responses such as the hypothalamus and preoptic areas.

Article activity feed

  1. eLife

    eLife Assessment

    This important work sets out to identify the neural substrates of associative fear responses in adult zebrafish. Through a compelling and innovative paradigm and analysis, the authors identify brain regions associated with individual differences in fear memory expression. While most findings are well supported, there is limited evidence that the four behavioral clusters represent discrete forms of associative fear-memory expression and the related interpretations would benefit from additional analysis or more cautious framing. Nonetheless, this study showcases the strength of zebrafish for systems-level neuroscience and will be of broad interest to the neuroscience community.

  2. eLife

    Reviewer #1 (Public review):

    Summary:

    This work provides a comprehensive analysis of how adult zebrafish show fear responses to conspecific alarm substances (CAS) and retain their associative memory. It shows that freezing is a more reliable measure of fear response and memory compared to evasive swimming, and that the reactivity and the type of responses depend on the zebrafish strain. It further suggests neuronal substrates of different fear responses based on c-Fos mapping.

    Strengths:

    The behavioral part is the most comprehensive and detailed yet in the zebrafish field, providing strong support for the authors' claim. The flow from Figure 1 to Figure 4 is very smooth. They provide extremely detailed, yet complementary and necessary, analyses of how different categories of behavior emerge over time during the CAS exposure and memory retrieval. I'm convinced that neuro researchers who study fear/stress responses will always refer to this paper to plan and interpret their future experiments.

    Comments on revised version:

    The authors successfully addressed my comments, including the addition of Figure S6-2, which gives us some intuition into the relationships between c-Fos levels in individual areas and the behavioral outputs.

  3. eLife

    Reviewer #2 (Public review):

    In this study, Fontana et al. develop a paradigm for associative conditioning by pairing exposure to alarm substance with a novel tank. Exposure to conspecific alarm substance (CAS) in the novel tank triggers freezing and what they characterize as evasive swimming behaviour, which are subsequently seen in a re-exposure to the novel tank without the CAS present. Importantly, these states are identified via automated processes including postural tracking and a random forest classification process, which could be very useful tools for subsequent studies.

    In their experiments they focus on the differences in behaviour among strains of zebrafish (both males and females), and among individual zebrafish. For males and females of different strains they find some differences, though the clearest message seems to be that the most robust measure of the behaviour in response to both the CAS and in the memory trials is the freezing behaviour, while evasive behaviour is more variable and not always seen. This may relate to their observation of significant "evasiveness" in vehicle control experiments (discussed further below).

    Moving on to individual variation from within this multi-strain male/female dataset, they first examine transition matrices between states, and find this is not dramatically altered by stimulus exposure. They then use clustering to identify 4 different "classes" of zebrafish that differ in their expression (or not) of two types of behaviour: freezing and/or evasive behaviour. They show that over the three exposure epochs of the experiment this classification is somewhat stable in an individual fish, though many fish change their behaviour -- e.g. evading + freezing -> only freezing.

    In the final set of experiments they move beyond behavioural analyses and perform whole-brain cFos mapping of these individual zebrafish, and perform analyses aimed at identifying correlations between individual behavioural expression and the number of cFos positive cells in different brain regions. Using partial least squares analysis they find areas associated with two types of behavioural contrasts, which differ in their weighting of different behavioural expression during the Memory trials. Covariation and network structure analysis within different classes of fish also find some differences in covariation among brain areas, providing hypotheses as to underlying network effects that may govern the expression of freezing and/or evasive behavior in the memory trial phases.

    Overall, I find this to be an interesting study that employs state of the art methods of behavioural analyses and whole-brain cFos analyses. The revision has clarified the take-home message considerably: the abstract is now more careful about which behavioural groups are memory-associated, and the causal language in the conclusions has been appropriately softened. Two of my three original main concerns have been addressed. The first is not and having looked at the data again I can now be more specific about what concerns me.

    Comments on revised version.

    (1) My first concern related to the claim that fear memory behaviour falls into four distinct groups, and specifically to the role of evasiveness in defining them. The authors give three reasons for retaining it, but I remain unconvinced.

    The first is that variable evasion in response to alarm substance is a long-standing observation (von Frisch; Suboski et al.), and that dissecting this individual variation is the purpose of the paper. I agree with the motivation, and it is a good reason to measure evasion. But it does not establish that evasion on memory day reflects fear memory, and memory day is the only day used for the clustering and neural activity mapping. The manuscript's own results point the other way: relative to pre-exposure, no strain or sex increased evasion on memory day, and relative to vehicle only female TUs did. The temporal profiles show evasion on memory day to be largely similar between vehicle and CAS-treated fish. Historical observations of variable evasion during CAS exposure do not carry over to the memory phase.

    The second is that the clustering itself reveals two kinds of freezing fish - one freezing between bouts of normal swimming, the other between bouts of evasion - demonstrating that a subset of fish increase evasion. In absolute terms, this does not match the data. In Figure 4B, evading freezers are below the population mean for absolute evasion, as are freezers. The text describes evading freezers as "high in freezing and evasive behaviors," and I do not think Figure 4B supports this.

    What actually separates the two freezing groups is the third measure, evasion as a percentage of active time. And this is where I have difficulty, because that measure is not an independent behavioural readout. The classifier assigns every window to normal, evasive or freezing, and active time is simply non-freezing time, so evasion-as-percent-of-active is fully determined once the other two are known.

    This matters for the clustering specifically. Distance-based methods weight each input dimension equally, so a variable that carries no information beyond the other two nonetheless contributes a full third of the distance between any two fish - and it contributes it in a way that counts freezing twice, once directly and once through the denominator of the derived measure. The space is nonetheless described as three-dimensional throughout, including in the Methods and the Figure 4 legend, when there are only two independent behaviours in it.

    The consequences fall hardest on exactly the animals at issue. Both freezing groups sit at 65-70% freezing, so there is very little active time to divide by, and small absolute differences in evasion - together with any noise in estimating them from a couple of minutes of non-frozen behaviour - are inflated into large differences on the rescaled measure. In terms of what the fish actually did, the two groups differ by a few percent of trial time. That is the boundary on which much of the rest of the paper rests.

    I recognise that evasion as a proportion of active time is in some respects the more biologically meaningful quantity, and the authors are right that a fish freezing 70% of the time has limited opportunity to do anything else. But that is an argument for reporting it as a descriptive measure, not for entering it into the clustering alongside the two variables from which it is computed.

    This impression is reinforced by Figure 4A itself. While the freezer group occupies a reasonably distinct region, the non-reactive, evader and evading freezer groups appear as a single continuous distribution with cluster boundaries drawn through it rather than around visible gaps. I appreciate that UMAP is a projection and that visual separation is not required for genuine structure, but this is the figure by which most readers will judge whether four discrete types exist, and it does not obviously support that reading - particularly given that the embedding is built from the same variables, including the rescaled measure, that most favour the separation.

    I would suggest that the authors re-run the clustering using only the two directly measured behaviours, percent freezing and percent evasion of total time, and report whether four groups still emerge and, in particular, whether the evading freezer / freezer split survives.

    The third is that the two groups have distinct functional networks despite equally high freezing, so the behavioural difference is real and is manifesting in the brain. This is the strongest of the three arguments, and I accept part of it: something about how a frozen fish spends its remaining active time does appear to be neurally meaningful, which is interesting in its own right. But it does not establish that these are two distinct types, nor that the difference has anything to do with the conditioning. Fish taken from either side of a cut through a continuous distribution will differ neurally if that continuum tracks brain state, so the network result is equally compatible with graded variation. More importantly, Figure 5A shows that a substantial proportion of fish are classified as evaders in the vehicle condition and at pre-exposure, before any CAS has been given. This suggests a pre-existing individual tendency toward evasive behaviour that is independent of the alarm substance, and one would expect such a tendency to persist into the memory trial. If so, the distinction the network analysis is drawing between freezers and evading freezers may simply reflect that baseline trait, and its neural correlates would be correlates of the trait rather than of fear memory. I am therefore not convinced that this distinction is related to CAS or to memory.

    (2) This concern is fully resolved. I had misread the CAS preparation: it was pooled from eight donors spanning all four strains and both sexes, so every fish received identical material and the strain and sex differences cannot be attributed to donor variability. The clarification now added to the Results will prevent other readers making the same error. The addition of FDR correction to the Figure 2 comparisons also addresses my related concern about multiple testing.

    (3) Somewhat resolved. The conclusion no longer states that behavioural variation is "driven by" activity in particular regions, and the added caveat that neural activity was not directly manipulated sets the right expectation for a mapping study. The scatterplots in Figure S6-2 are a useful addition and give a much better intuition for what the PLS contrasts represent. My remaining reservation is the one above: a great deal of the neural story rests on the evading freezer / freezer contrast, and I am not persuaded that this contrast marks a boundary relevant to fear memory.

  4. eLife

    Reviewer #3 (Public review):

    This revised manuscript by Fontana et al. aims to study how animals respond to fearful stimuli, with a specific focus on brain regions involved in predicting animals that passively freeze or those that actively evade the threat. I continue to be enthusiastic about the study. The study addresses an important question regarding individual variation in fear-related behavior and links these behavioral phenotypes to whole-brain activity patterns in adult zebrafish. The combination of a contextual fear conditioning paradigm, strain/sex comparisons, behavioral clustering, and AZBA-based c-Fos mapping makes this a valuable contribution to the field, not just in answering the question posed by the authors, but also in formulating a framework for using adult zebrafish for whole brain analysis of complex behaviors. Overall, I find the authors have responded to my concerns:

    (1) I still think that separating memory acquisition and consolidation is an interesting question, and further use of the framework will need to eventually solve that; however, I also appreciate that this may be beyond the scope of the current study, and I appreciate the authors acknowledging this in the manuscript.

    (2) Regarding Figure 3, I also agree that this is difficult to present differently, and I appreciate the authors adding text to the body to clarify things. My one request is that the sentence (lines 214-215) that reads: "This increase in evasion in the vehicle group likely represents a response to the water disturbance that occurs when solution is added to the tank." Be changed to: "This increase in evasion in the vehicle group may represent a response to the water disturbance that occurs when solution is added to the tank." While it is entirely possible, there are no concrete data to support that this is "likely."

    (3) I appreciate the clarification regarding the PLS-derived contrasts in Figure 6A and in the body.

    Overall, this is a really interesting paper that will have a wide-ranging impact. All of my concerns have been addressed.

  5. eLife

    Author response:

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

    We thank the reviewer’s for their thoughtful comments that have significantly strengthened the paper. Below, we have outlined our responses to both the public reviews and recommendations.

    In addition to the alterations to the manuscript based on the reviews, during our review of the data analysis we uncovered some small errors that we have now corrected. In looking back over the image registration, we identified three animals whose olfactory bulbs did not register properly and one with poor cell counting in the telencephalon. To account for these issues, we imputed the missing data using an iterative soft-threshold singular value decomposition (described on lines 779-783 of the updated manuscript). This update had little impact on the results. We also identified a small error in how we determined ‘unique’ and ‘overlapping’ edges in the network analysis (Figure 8). In the previous analysis we had incorrectly noted that all ‘unique’ edges did not have an overlapping confidence interval with the two other networks (i.e., the networks for evading freezers, freezers, and non-reactive). Instead, the ‘unique’ edges in the prior version of the manuscript did not have an overlap with at least one other network. We have now updated the analysis so the reader can distinguish between edges that are truly ‘unique’ versus those with ‘1 overlapping confidence interval’ or ‘2 overlapping confidence intervals’ with other networks. As before, this update and change to the analysis does not materially affect the results or conclusions.

    Public Reviews:

    Reviewer #1 (Public review):

    Weaknesses:

    The neural analysis part is very comprehensive. Figure 5 and Figure 6 are independent but complement each other very well. They together support that the cerebellar system is the key brain component for a freezing response. Their extreme focus on high-level analyses, however, came at the expense of biological intuitions. I suggest adding some figure panels and result/discussion paragraphs to help with that aspect.

    Thank you for the suggestion. We have made extensive edits to the manuscript to include additional discussion and biological intuition. Specifically:

    We added a supplemental figure (Figure S6-2) that has scatterplots showing how cfos levels vary with the different behavioral contrasts. Although the PLS analysis is multivariate, this univariate analysis should help give readers a better intuition of how the behavior relates to brain function.

    We have also rewritten the results sections for both the PLS analysis (lines 303-361) and network analysis (lines 396-437) to incorporate more of a discussion about the biological context of different regions identified. Thank you for this suggestion, we feel that this significantly strengthens the biological interpretation of the data for the reader.

    Reviewer #2 (Public review):

    (1) My first concern relates to the claim in the abstract that "We found that fear memory behavior fell into four distinct groups: non-reactive, evaders, evading freezers, and freezers".

    In my opinion, the "freezing" aspect is well supported as being both triggered by the CAS and for memory effect upon re-exposure to the tank, but I am less convinced about the "evasive" behaviour. In Figure 2, it appears that "evasiveness" is generally not increased in both the Exposure or Memory phases for many groups, and in Figure 5, it appears that "evasiveness" is expressed by nearly 50% of the fish in the pre-exposure condition before CAS addition and in all phases in the vehicle condition. Therefore, it appears that most of the expression of this behaviour is independent of any memorybased effect.

    We thank the reviewer for this suggestion and we agree that this line in the abstract was unintentionally misleading. We have now altered this line in the abstract (lines 34-36) to read:

    “We also found that that behavior fell into four distinct groups: non-reactive, evaders, evading freezers, and freezers with the evading freezer and freezer groups most clearly associated with memory formation.”

    On the larger point of the inclusion of evasion as part of the fear response, we believe this is warranted for the following reasons: (1) evasive behavior has long been acknowledged as a highly variable aspect of how fish respond to alarm substance where some fish exhibit evasion and others do not. This observation goes back to the original work from Karl von Frisch in minnows (von Frisch, 1938), and others in zebrafish (e.g., Suboski et al, 1990). One goal of our paper (and the work from the lab in general) is to try dissecting out this individual variation that can get lost when only considering population averages. (2) The unsupervised clustering also suggests that there are two distinct types of freezing clusters (Figure 4B) where some fish freeze intermittently with normal swimming and others freeze intermittently with evasive behavior. This suggests that evasion is increased in response to CAS, but only in a subset of fish. (3) The brain networks from the evading freezer and freezer groups are distinct (Figure 8A) despite having equally high levels of freezing behavior (Figures 4B and C). This means the difference we’re able to distinguish behaviorally is also manifesting in the brain, suggesting that it is not anomalous. Thus, while we agree that freezing is definitely the strongest and clearest behavioral response to CAS, we believe the analysis of this large dataset supports the interpretation that, in a subset of fish, increased evasive behavior in response to CAS is also a part of the response.

    (2) My second concern relates to the claim in the abstract that "background strain and sex influenced how fish respond to CAS, with males more likely to increase evasive behaviors than females and the TU strain more likely to be non-reactive."

    My understanding, based on the introduction and on the methods, is that it is likely important that the CAS be prepared from conspecifics of the same strain and sex, and for this reason, they prepared different CAS specific for each strain and each sex. Therefore, the "CAS" that is applied is necessarily different for each condition, and I am concerned about if the differences observed could relate more to variation in the quality, purity, concentration, etc. of the specific CAS samples for different groups, rather than their reactivity to the substance or their ability to form memories based on such experiences.

    The CAS was prepared by mixing extracts from all four strains and both sexes (so 8 fish per batch). Thus, all the fish were exposed to the same CAS mix derived from the same donors. This is described in the methods (lines 626-629). However, to ensure that this is clear to readers, we’ve now included a line indicating this in the results section (lines 123-124).

    (3) My third concern relates to the interpretation of the cFos data.

    As I mentioned above, I feel as though the behavioural analysis is perhaps more complex than is warranted via the inclusion of evasiveness, and I wonder if the conclusions from the experiments would be simpler if analyzed only from the perspective of freezing.

    We agree that the freezing response is driving the majority of the neural cfos response that we are seeing (e.g., Figure 6A-C). However, we feel that the network analysis (Figure 8) justifies the distinction between freezers and evading freezers. This is because the brain networks for these two groups (freezers and evading freezers) are quite distinct, even though these groups both have the same levels of freezing behavior (Figure 4). This stark difference in patterns of neural activity suggests the brain of a freezer and an evading freezer are engaging with the world in two distinct ways that is worth noting. We’ve updated the abstract to make this point clearer (abstract: lines 39-48) and discuss the biological interpretations of patterns of brain activity unique to evasion or evading freezers in more depth (lines 303-361; lines 396-437).

    Reviewer #3 (Public review):

    (1) The three-day contextual fear paradigm, as implemented - one CAS pairing on day 2 followed by a single recall test on day 3 - inevitably conflates acquisition and long-term memory, making it impossible to know whether strains like TU truly recall the association poorly or simply learn it more slowly. For example, given that TU fish extinguish fear faster than AB or TL strains in extended protocols, they may simply require additional or repeated CAS pairings to achieve the same asymptotic performance. To disentangle learning kinetics from recall strength, the assay could be revised to include multiple acquisition trials (e.g., conditioning on two or more consecutive days) with an immediate post-conditioning probe to assess acquisition independent of consolidation, and continuous measurement of freezing and evasive behaviors across each trial to fit learning curves for each strain. Such refinements - even if on a subset of the strains - would reveal whether "non-reactive" phenotypes reflect genuine recall deficits or merely delayed acquisition.

    We thank the reviewer for this thoughtful comment. We agree that it is difficult to disentangle acquisition from consolidation. Indeed, the TU fish do appear to have lower levels of freezing in response to the CAS (Figure 2A), supporting the idea that reduced performance at memory day could be due to some sort of deficit at acquisition. However, pursuing a detailed examination of strain dependent differences in fear memory acquisition versus consolidation is beyond the scope of the current paper where we primarily focus on individual differences in behavior. Nonetheless, we have included this important point in the discussion (lines 470-471).

    (2) My second major question is with respect to Figure 3 panel B. This is a complex figure, and I can understand the gist of what the authors are attempting to show, but it is difficult to understand as it is. Can this be represented in a way that is clearer and explained a bit more easily?

    We agree that this figure is one of the more complex in the paper. However, we’ve struggled to come up with a better way to present it. We have improved the presentation based on other reviewer comments by making the vehicle and CAS groups more easily distinguishable by using open versus closed circles. We’ve also included additional interpretations of the data in the results, which we hope will help guide readers through this figure better (lines 208-223).

    (3) The brain mapping is by far one of the most interesting aspects of this study, and the methods that the group used are interesting. The brain mapping, however, relies on generating "contrasting" groups (Figure 6A), and I was not clear as to how these two groups were formed. Could the authors elaborate a bit?

    These contrasting groups (contrast 1, contrast 2) arise analytically from the partial least squares (PLS) analysis; they are not defined by the experimenter. In brief, PLS is a multivariate technique that identifies latent variables that capture axes of maximal covariation between two datasets: behavior and brain activity. As an analogy to a more widely known technique, principal components analysis (PCA) uncovers axes of maximal variance within a single dataset. PLS, in contrast, simultaneously analyzes the covariation in two datasets. The contrast groups in Figure 6A represent the behavioral weights of the latent variables that capture the most covariance, which illustrates how the four behaviors load onto these top two contrasts.

    Recommendations for the authors:

    Reviewer #1 (Recommendations for the authors):

    Major points:

    (1) The c-Fos analysis in Figure 5 is very comprehensive and convincing, but lacks intuitive presentations. In my understanding, the increase in c-Fos expression in red areas means increased freezing behavior for Contrast 1 for the PLS analysis? Do you have representative c-Fos expression images between different groups of fish?

    We decided not to include a representative cfos image because the data is derived from a large number of fish (N=87) and thus it can easily be cherry-picked to choose images that match the narrative. Instead, to more accurately capture the breadth of the data while providing a more intuitive presentation, we have included an additional supplemental figure that includes scatterplots of scaled cfos data against behavioral scores for each of the two contrasts (S6-2). We believe this more fully and accurately captures the relationship between behavior and brain activity. We included six different example brain regions and scatterplots for cfos activity against behavioral scores for contrasts 1 and 2, demonstrating a range of relationships. However, we should note that PLS is a multivariate technique, and so this univariate analysis does not fully capture the subtleties of the PLS analysis. Nonetheless, we think this will help give a more intuitive interpretation of the data to readers. We have also referenced this additional data in the manuscript (lines 307-309). We thank the reviewer for this excellent suggestion that improves the ability of readers to understand the paper.

    (2) Also related to Figure 5, the result section only describes the PLS statistics and does not try to describe the biological interpretation. Do the authors think the c-Fos expression directly represents lowlevel behavior, such as swimming, or a high-level behavioral state or learning? Maybe different areas mediate different aspects?

    For example, the medullary locomotor areas, which are usually highly correlated with swimming in terms of neural activity, seem to have higher c-Fos expression in freezing fish. I'm not saying this shouldn't be the case. c-Fos expression in this area was not elevated in larval fish during OMR in Shainer et al., 2023, indicating that it doesn't linearly reflect neural activity. But discussing a bit of intuition on the connection between c-Fos expression and biological process, rather than just saying "the cerebellum could regulate emotional states", would help us guide through this highly complex analysis.

    We have now added more interpretation of the data in both the PLS and network analysis sections (lines 303-361 and lines 396-437). Again, thank you for this excellent suggestion. This helps make the biological interpretation of the data clearer.

    Minor points:

    (1) Figure 2B titles: please write "memory" on the right side.

    We considered writing ‘memory’ on the right-hand side, but we thought this may add confusion because it would not apply to both graphs in the row. The left-hand graphs are the responses during ‘exposure’ and the right-hand graphs are the responses during the ‘memory’ phase. This is indicated by the titles above the left and right-hand sets of graphs.

    (2) Figure 2C: needs legend lines.

    We have now moved the legend lines from the top of the graphs to below the graph to make them more visible to readers.

    (3) Line 187: "aggregated" data.

    This has now been changed to ‘aggregated’ (now line 194).

    (4) Line 371: I'm not sure what "Beyond" means.

    We have now significantly changed this part of the paper and we no longer use the word ‘beyond’ here.

    Reviewer #2 (Recommendations for the authors):

    (1) Regarding point (1) in the Public Review:

    I would encourage the authors to consider whether this study might be better focused exclusively on the freezing behaviour, which does appear to be reliably expressed during CAS exposure and in the memory phases, and would significantly simplify the subsequent analyses of neural activity, and perhaps may lead to a more coherent conclusion.

    As noted in our response to the public review, we appreciate this suggestion, but we have decided to keep the inclusion of the evasive behavior. This is because (1) evasive behavior has long been acknowledged as a highly variable aspect of how fish respond to alarm substance where some fish exhibit evasion and others do not. This observation goes back to the original work from Karl von Frisch in minnows (von Frisch, 1938), and others in zebrafish (e.g., Suboski et al, 1990). One goal of our paper (and the work from the lab in general) is to try dissecting out this individual variation that can get lost when only considering population averages. (2) The unsupervised clustering also suggests that there are two distinct types of freezing clusters (Figure 4B) where some fish freeze intermittently with normal swimming and others freeze intermittently with evasive behavior. This suggests that evasion is increased in response to CAS, but only in a subset of fish. (3) The brain networks from the evading freezer and freezer groups are very distinct (Figure 8A) despite having equally high levels of freezing behavior (Figures 4B and C). This means the difference we’re able to distinguish behaviorally is also manifesting in the brain, suggesting that it is not anomalous. Thus, while we agree that freezing is definitely the strongest and clearest behavioral response to CAS, we believe the analysis of this large dataset supports the interpretation that, in a subset of fish, increased evasive behavior in response to CAS is also a part of the response.

    A more minor concern related to the analyses in Figure 2: in the figure legend, it is stated that "*-P < 0.05 compared to vehicle treated fish via t-tests". How are the authors dealing with the multiple comparisons problem? Would something like an ANOVA not be more appropriate?

    Thank you for bringing this point up. We did not initially correct for multiple comparisons because we considered each of these experiments across sex and strain separate since we did not compare across strains. However, the way we’ve grouped the data together in figure 2 makes it appear as if they are one large experiment. To alleviate any concern about multiple testing, we have now corrected for multiple comparisons using the false discover rate (FDR) correction. The statistics in the figure and captions have now been updated.

    (2) Regarding point (2) in the Public Review:

    If the authors agree with my concern regarding potential variability in the CAS samples, I would suggest either testing for differences among strains using the same batch of CAS, or including and explaining this caveat in the text.

    As noted in our response to the public review, the CAS was the same for all the fish. Each batch was derived from 8 donor fish, one fish from each strain and sex (described in lines 123-124 of the results and lines 626-629 of the methods).

    (3) Regarding point (3) in the Public Review:

    I feel like the standard in the field for such conclusions would be after

    (a) Direct analyses of the activity states in these areas. I was surprised not to see a direct analysis of the cFos stainings in the cerebellum relative to freezing behaviour, for example, ideally in a different animal cohort.

    The PLS analysis does relate activity in the cerebellum (and other brain regions) to specific behaviors via the the behavioral contrasts (Figure 6A). We believe this approach (instead of dividing fish into ‘high and low freezers’) is a more powerful way to leverage the data from all the animals tested (87 fish). However, we appreciate that the interpretation of the PLS analysis is not as intuitive as seeing scatterplots or bar charts comparing neural activity. For this reason (and in response to a comment from reviewer 1), we have included as a supplementary figure (Figure S6-2) scatterplots showing how standardized c-fos activity varies with the behavioral scores from the contrasts identified from the PLS analysis. Given that contrast 1 weights heavily in the positive direction on freezing, these figures can essentially be read as looking at cfos activity as a function of freezing levels. What can clearly be seen is that for regions of the cerebelleum (E.g., the LCa and CC) there is a clear positive relationship between cfos activity and the behavior scores for contrast 1.

    (b) Some kind of manipulation of the brain area resulting in the relevant behavioural modification.

    We completely agree with the reviewer. However, at the moment, we do not have the tools to do this in adult zebrafish. It is something we’re actively working on.

    Of course, I appreciate that such experiments might not be possible or feasible, and in which case I would suggest adjusting the claims accordingly and highlighting the caveats to their interpretations.

    We have incorporated the caveat that we have not directly altered neural activity into the discussion (lines 542-543) and adjusted how we discuss our findings in the abstract (lines 39-41) to more accurately represent the type of evidence we provide. Hopefully we’ll be able to do so in the near future!

    MINOR CONCERNS:

    (1) In Figure 3, how is the end of a behavioural epoch defined? I am surprised to see that you consider transitions between the same behavioural state. How does erratic swimming -> erratic swimming differ from a longer single epoch of erratic swimming? In general, I find this analysis confusing, and I am not sure if it adds significantly to the message of the paper.

    Thank you for this question as it prompted us to realize we were missing this in our methods section. We have now updated the methods to include how we calculated the behavioral transitions (lines 644-650). In short, we used a 750 ms behavioral epoch time that corresponds to the size of the sliding window we used for the random forest model.

    We have also updated the description of this analysis in the results to indicate the main finding from it (lines 207-223). In brief, the main finding is that exposure to CAS results in longer bouts of evasive behavior without increasing its frequency. Whereas CAS induced freezing arises from both longer bouts and likelihood of occuring. While we agree that this is a relatively minor finding in the paper, one of our goals is to provide as comprehensive analysis of fear behavior as possible to help guide future researchers interested in using fish for understanding different aspects of fear-related behaviors.

    (2) In the PLS analyses, two measures of evasion are used: evasion time, and evasion as a percent of active behavior. I don't understand the justification for both of these being used rather than one. Again, my overall recommendation is to reduce the focus on the analysis of evasion behaviour, but if you do not choose to do this, I think the rationale of how both measures are used and why needs explanation.

    We chose to incorporate two different measures of evasion throughout the study because the high levels of freezing in some animals results in little opportunity to express other behaviors (like evasion). Thus, to better capture what fish may be doing in the absence of freezing (i.e., when they are active) we also calculate the amount of active time spent performing evasive behaviors (instead of normal swimming). We have now included an explanation for this earlier in the results section when we first use this metric (lines 149-152).

    (3) In the methods, I don't understand this: "Animals that were assigned the wrong sex were removed from data analysis, as well as its paired fish (< 2%)".

    We determine the sex of fish when we set them up for dual housing. However, we occasionally make errors in sex determination. To ensure we properly sexed the fish, at the end of experiments, we euthanize the fish and check for the presence of eggs. If we incorrectly assigned the sex to a fish, they are removed from the experiment alongside the other fish they were dual housed with. This is because we want to ensure all fish are housed in the same way (i.e., a male fish with a female fish).

    (4) How was this determined differently from the first time, resulting in exclusion?

    After experiments, fish were euthanized and we checked for the presence of eggs (line 599-601). We’ve now added a line in this other part of the methods referring back to where we describe this (lines 676678).

    Reviewer #3 (Recommendations for the authors):

    Here are some minor concerns and errors found in the manuscript:

    (1) For Figure 2B and Figure 3B, can the group make the lines solid and dotted? The circle or triangle designation is difficult to see, and since the crux of the figure depends on comparing Veh and CAS, it would be easier to see if the lines were altered.

    Thank you for this suggestion. Instead of making the lines solid and dotted, we decided to make both the CAS and vehicle group circles and then have open and closed circles. We believe this solves the issue of being able to distinguish these groups and makes the data more readable.

    (2) Figure 2C: It appears that the line colors in the legend are missing.

    We have moved the line colors below the graphs to make them more obvious.

    (3) Figure 8A: Same thing here - could the text be enlarged? It's really difficult to make out each node, and when I zoom the text becomes pixelated. This is an important figure and one that will likely be referenced, and making it clear would be helpful.

    This one is difficult. We have made the network images as large as would fit on a page. We have now uploaded vectorized versions of the images so that they do not become pixelated when zooming in. As part of our supplemental materials we also include a cystoscope file that can be explored in greater depth as well.

    (4) The paper is really well written: I found a few typos, though:

    (a) Line 529: "Institutional Cara and Use Committee" should be "Institutional Animal Care and Use Committee" (Change cara to care and add animal).

    (b) Line 274: "hybdridization" should read hybridization.

    Thank you for catching these typos. They have now been fixed.

  6. Version published to 10.7554/elife.107908.2 on eLife
  7. Version published to 10.7554/elife.107908 on eLife
  8. Version published to 10.1101/2025.05.12.653531v3 on bioRxiv
  9. Version published to 10.1101/2025.05.12.653531v4 on bioRxiv
  10. Version published to 10.7554/elife.107908.1 on eLife
  11. eLife

    eLife Assessment

    This important work sets out to identify the neural substrates of associative fear responses in adult zebrafish. Through a compelling and innovative paradigm and analysis, the authors suggest brain regions associated with individual differences in fear memory. While several findings are well supported, aspects of the interpretation and presentation are partially incomplete, and the manuscript would benefit from adjusting key claims or including additional experiments. Nonetheless, this study showcases the strength of zebrafish for systems-level neuroscience and will be of broad interest to the neuroscience community.

  12. eLife

    Reviewer #1 (Public review):

    Summary:

    This work provides a comprehensive analysis of how adult zebrafish show fear responses to conspecific alarm substances (CAS) and retain their associative memory. It shows that freezing is a more reliable measure of fear response and memory compared to evasive swimming, and that the reactivity and the type of responses depend on the zebrafish strain. It further suggests neuronal substrates of different fear responses based on c-Fos mapping.

    Strengths:

    The behavioral part is the most comprehensive and detailed yet in the zebrafish field, providing strong support for the authors' claim. The flow from Figure 1 to Figure 4 is very smooth. They provide extremely detailed, yet complementary and necessary, analyses of how different categories of behavior emerge over time during the CAS exposure and memory retrieval. I'm convinced that neuro researchers who study fear/stress responses will always refer to this paper to plan and interpret their future experiments.

    Weaknesses:

    The neural analysis part is very comprehensive. Figure 5 and Figure 6 are independent but complement each other very well. They together support that the cerebellar system is the key brain component for a freezing response. Their extreme focus on high-level analyses, however, came at the expense of biological intuitions. I suggest adding some figure panels and result/discussion paragraphs to help with that aspect.

  13. eLife

    Reviewer #2 (Public review):

    In this study, Fontana et al. develop a paradigm for associative conditioning by pairing exposure to an alarm substance with a novel tank. Exposure to conspecific alarm substance (CAS) in the novel tank triggers freezing and what they characterize as evasive swimming behaviour, which is subsequently seen in a re-exposure to the novel tank without the CAS present. Importantly, these states are identified via automated processes, including postural tracking and a random forest classification process, which could be very useful tools for subsequent studies.

    In their experiments, they focus on the differences in behaviour among strains of zebrafish (both males and females), and among individual zebrafish. For males and females of different strains, they find some differences, though the clearest message seems to be that the most robust measure of the behaviour in response to both the CAS and in the memory trials is the freezing behaviour, while evasive behaviour is more variable. and not always seen. This may relate to their observation of significant "evasiveness" in vehicle control experiments (discussed further below).

    Moving on to individual variation from within this multi-strain male/female dataset, they first examine transition matrices between states and find tthat his is not dramatically altered by stimulus exposure. They then use clustering to identify 4 different "classes" of zebrafish that differ in their expression (or not) of two types of behaviour: freezing and/or evasive behaviour. They show that over the three exposure epochs of the experiment, this classification is somewhat stable in an individual fish, though many fish change their behaviour - e.g., evading + freezing -> only freezing.

    In the final set of experiments, the authors move beyond behavioural analyses and perform whole-brain cFos mapping of these individual zebrafish. They perform analyses aimed at identifying correlations between individual behavioural expression and the number of cFos-positive cells in different brain regions. Using partial least squares analysis, they find areas associated with two types of behavioural contrasts, which differ in their weighting of different behavioural expression during the Memory trials. Covariation and network structure analysis within different classes of larvae also find some differences in covariation among brain areas, providing hypotheses as to underlying network effects that may govern the expression of freezing and/or evasive behavior in the memory trial phases.

    Overall, I find this to be an interesting study that employs state of the are methods of behavioural analyses and whole-brain cFos analyses, but I am left a little bit confused as to what the take home message is and what can be concluded from this complex study that mixes in analyses of strain, sex, and individuality within a quite complex assay with multiple behavioural parameters.

    My suggestions are as follows:

    (1) My first concern relates to the claim in the abstract that "We found that fear memory behavior fell into four distinct groups: non-reactive, evaders, evading freezers, and freezers".

    In my opinion, the "freezing" aspect is well supported as being both triggered by the CAS and for memory effect upon re-exposure to the tank, but I am less convinced about the "evasive" behaviour. In Figure 2, it appears that "evasiveness" is generally not increased in both the Exposure or Memory phases for many groups, and in Figure 5, it appears that "evasiveness" is expressed by nearly 50% of the fish in the pre-exposure condition before CAS addition and in all phases in the vehicle condition. Therefore, it appears that most of the expression of this behaviour is independent of any memory-based effect.

    (2) My second concern relates to the claim in the abstract that "background strain and sex influenced how fish respond to CAS, with males more likely to increase evasive behaviors than females and the TU strain more likely to be non-reactive."

    My understanding, based on the introduction and on the methods, is that it is likely important that the CAS be prepared from conspecifics of the same strain and sex, and for this reason, they prepared different CAS specific for each strain and each sex. Therefore, the "CAS" that is applied is necessarily different for each condition, and I am concerned about if the differences observed could relate more to variation in the quality, purity, concentration, etc. of the specific CAS samples for different groups, rather than their reactivity to the substance or their ability to form memories based on such experiences.

    (3) My third concern relates to the interpretation of the cFos data.

    As I mentioned above, I feel as though the behavioural analysis is perhaps more complex than is warranted via the inclusion of evasiveness, and I wonder if the conclusions from the experiments would be simpler if analyzed only from the perspective of freezing.

    But considering the presented analyses: while I dont think there is anything wrong with the partial least squares approach and the network analyses, I am concerned that the simple messaging in the text does not reflect the complexity of this analysis combining different weightings of different behavioural characteristics in a behavioural contrast, or covariations among many regions and what such analyses mean at the level of brain function. For these reasons, I feel like statements along the lines of "Behavioral variation is driven by differences in the activity of brain regions outside the telencephalon, such as the cerebellum, preglomerular nuclei, preoptic area and hypothalamus" are not well supported.

  14. eLife

    Reviewer #3 (Public review):

    Summary:

    This manuscript by Fontana et al. sets out to fill a critical gap in our understanding of how individuality in fear responses corresponds to changes in brain activity. Previous work has shown in myriad species that fear behaviors are highly variable, and these variabilities correlate with sex and strain, with epigenetic modifications, and neural activity in specific regions of the brain, such as the amygdala. However, a whole-brain functional assessment of whether activity in different regions of the brain is associated with fear behavior has been difficult to assess, in part due to the large size and opacity of the brain. The Kenney group overcomes these limitations using the zebrafish, together with powerful behavioral and brain imaging approaches pioneered by their lab. To overcome the technical obstacles of delivering a reproducible unconditioned stimulus in water and quantifying nuanced behavioral responses, the authors developed a three-day conditioning paradigm in which fish were repeatedly exposed to CAS in one tank context and to control water in another. Leveraging automated cluster analysis across over 300 individuals from four inbred strains, they identified four distinct memory-recall phenotypes - non-reactive, evaders, evading freezers, and freezers - demonstrating both the robustness of their assay and the influence of genetic background and sex on fear learning. Finally, whole-brain imaging using the AZBA atlas (Kenney et al. eLife) and cfos mapping coupled with multivariate analysis revealed that although all fish reengaged telencephalic regions during recall, high-freezing phenotypes uniquely recruited cerebellar, preglomerular, and pretectal nuclei, whereas mixed evasion-freezing fish showed preferential activation of preoptic and hypothalamic areas - a finding that lays the groundwork for dissecting the distributed neural substrates of associative fear in zebrafish.

    Strengths:

    The strengths of the study lie in the use of zeberarish and the innovative behavioral, modeling, and brain imaging tools applied to address this question. The question of how brain-wide activity correlates with variations in fear behavior is fundamental, and arguably, this system is the only system that could be used to address this. The statistics are appropriate, and the study is well reasoned. Overall, I like this manuscript very much and think it adds invaluable information to the field of fear/anxiety.

    Weaknesses:

    I have a few questions and suggestions.

    (1) The three-day contextual fear paradigm, as implemented - one CAS pairing on day 2 followed by a single recall test on day 3 - inevitably conflates acquisition and long-term memory, making it impossible to know whether strains like TU truly recall the association poorly or simply learn it more slowly. For example, given that TU fish extinguish fear faster than AB or TL strains in extended protocols, they may simply require additional or repeated CAS pairings to achieve the same asymptotic performance. To disentangle learning kinetics from recall strength, the assay could be revised to include multiple acquisition trials (e.g., conditioning on two or more consecutive days) with an immediate post-conditioning probe to assess acquisition independent of consolidation, and continuous measurement of freezing and evasive behaviors across each trial to fit learning curves for each strain. Such refinements - even if on a subset of the strains - would reveal whether "non-reactive" phenotypes reflect genuine recall deficits or merely delayed acquisition.

    (2) My second major question is with respect to Figure 3 panel B. This is a complex figure, and I can understand the gist of what the authors are attempting to show, but it is difficult to understand as it is. Can this be represented in a way that is clearer and explained a bit more easily?

    (3) The brain mapping is by far one of the most interesting aspects of this study, and the methods that the group used are interesting. The brain mapping, however, relies on generating "contrasting" groups (Figure 6A), and I was not clear as to how these two groups were formed. Could the authors elaborate a bit?

  15. eLife

    eLife Assessment

    This important study uses a combination of eye-tracking and computational models based on Active Inference to explain behavior in a gaze-contingent cued-reversal paradigm with 6 - 10-month-old infants. The study demonstrates solid evidence that the same rigorous computational modeling standards commonly applied in studies in adults can also be applied in studies of infants' learning, and a cluster analysis reveals that the parameters of the winning model provide better pattern separation between identified subgroups than behavior or questionnaire data alone. However, the evidence for some specific claims is incomplete, due to poor behavioral performance, unclear significance of the pupil data, and complexity of the model fitting; the claims regarding implications for psychiatry were also considered to be too strong and unsupported by evidence. This work will be of interest to developmental psychologists and cognitive neuroscientists.

  16. eLife

    Reviewer #1 (Public review):

    Summary:

    The authors developed a new gaze-based reversal task to study 6 - 10-month-old infants, in what would typically be a very challenging age group to study behavior related to learning, exploration, and perseveration. Here, the research question is excellently motivated by pointing out the limitation of past work that has typically studied adult clinical populations using similar approaches, which presents only the endpoint of the developmental process. Thus, there is important clinical and scientific value in studying much earlier stages in the developmental process. Here, the authors accomplish this with a new gaze-based paradigm that allows them to fit a variety of complex computational models to data from 41 infants. The main advantage of their winning model is that the parameters provide better pattern separation between two identified clusters of participants compared to behavioral variables alone.

    Strengths:

    Overall, the paper is well-written, and the models and analyses are applied in a principled and thorough fashion. The authors do an excellent job of both motivating their research question and addressing it through their task and set of computational models. The scope is also quite ambitious, modeling both choices and pupillary responses, while also using the models to generate behavior that is comparable to the experimental data and performing a cluster analysis to compare the suitability of the model parameters vs. other behavioral/questionnaire data in performing pattern separation between participants.

    Weaknesses:

    However, despite these strengths, I had a number of concerns that may limit the reliability of the findings.

    First, given the fact that the rewards for the initial pre-reversal setting are defined by the first choice of the infants, it was unclear to me whether the behavioral patterns in Figure 2 really support the fact that there was in fact, (prediction-error-based) learning in the task at all. The behavioral analyses proceed very briskly without really addressing this question, before rapidly jumping off the complexity cliff to present the models. However, even with the models, the winning model only had free parameters for preference (c) and a left-right dominance (epsilon), which don't really capture mechanisms related to learning. The epistemic and extrinsic components included in the model at the 2nd stage could potentially help shed light on this question, but (unless I've misunderstood) they seem to be all-or-nothing parts of the model, and thus don't reappear in later analyses (e.g., cluster analysis) because they are not individual-specific parameters. Thus, the main learning-relevant aspects of the model seem divorced from the ability to perform clustering or other clinically relevant diagnoses downstream. Thus, it was unclear to me whether the results really capture mechanisms related to cognitive flexibility that motivate the manuscript in the introduction.

    My other main concern was the complexity of the models and the way model comparison was performed using the three stages. First of all, the set of models is quite complex and risks alienating many developmental psychologists who would otherwise be very interested in these findings. Thus, I'm curious why the authors didn't consider including much simpler context-based RL models (e.g., Rescorla-Wagner/Q-learning models) that explicitly use prediction-error updates and whose simplicity might better match the simplicity of the behavior that 6-10 month infants are capable of displaying. Certainly, preference (as an inverse temperature parameter for a softmax policy) and left-right dominance (as a bias) could be implemented with these much simpler models. Second, while the three-stage model comparison seems somewhat principled, it left me questioning whether the 1st stage or 2nd stage results might be impacted by later stages. For instance, if the Simple-discard model were to still win in the first stage, once omega and eta have been eliminated as free parameters. Of course, I understand that there may be feasibility issues with testing all combinatorial variants of the model. But it was unclear why this specific order was chosen and what consequences this sequential dependency in the model fitting may have for the conclusions. And while model identifiability is stated in the abstract as one of the strengths of this approach, there don't seem to be any clear analyses supporting this fact. I would have loved to see a model recovery analysis (see Wilson & Collins et al., eLife 2019) to support this statement.

  17. eLife

    Reviewer #2 (Public review):

    Summary:

    This paper examines infants' learning in a novel gaze-contingent cued reversal learning task. The study provides strong evidence that infants learn in the task, and they characterize individual differences in learning using computational modeling. The best-fitting model of the set compared reflects a learning of mappings between context cues and outcomes that do not carry over across blocks. Infants are then clustered into two groups based on model parameter estimates capturing primacy bias and reward sensitivity. These groupings exhibited differences in infant temperament and other developmental measures. The modeling is rigorous, with model predictions accounting for substantial variance in infants' choices, and parameter estimates showing high recoverability. This study is important in that it demonstrates that such rigorous standards in computational modeling of behavior can be successfully deployed in infant studies.

    Strengths:

    The study provides evidence that infants exhibit cognitive flexibility within a reversal learning task and do not simply perseverate.

    The methods used within the novel gaze-contingent will be useful for other groups interested in studying learning and decision-making in infants.

    The study applies rigorous computational modeling approaches to infants' choices (inferred from gaze) and their physiological responses (i.e., pupil dilation) in the task, demonstrating that infants' reward learning is well-captured by an error-driven learning process.

    The authors conduct model comparison, posterior predictive checks, and parameter recoverability analyses and demonstrate that model parameters can be well estimated and that the model can recapitulate infant choice behavior.

    Physiological pupil dilation measures that correlate with prediction error signals from the model further validate the model as capturing the learning process.

    Weaknesses:

    It is not entirely clear that the individual differences in reversal learning identified between the two clusters of infants (ostensibly reflecting differences in cognitive flexibility) have construct validity or specificity for the associated developmental abilities that differ between groups (daily living, communication, motor function, and socialization).

    Similarly, it's not clear why the paper is framed as an advance for infant computational *psychiatry* rather than simply an advance in computational modeling of infant behavior. It seems to me that a more general framing is warranted. Basic cognitive development research can also benefit from cognitive hypothesis testing via computational model comparison and precise measurement of infants' behavior in reward learning tasks. Is there reason to believe that infants' behavior in this task might have construct validity for mental health problems related to cognitive flexibility later in development? Do the Vineland or IBQ-R-VSF prospectively predict clinical symptoms?

    A large proportion of the recruited infants (14 of 55) were excluded, but few details are provided on why and when they were excluded. Did the excluded infants differ on any of the non-task measures? This information would be helpful to understand limitations in the utility of the task or the generalizability of the findings.

    It is stated that: "The infants who completed at least three trials following the reversal were included in the analysis, as it is more likely that their expectations were violated in this interval." Are three trials post-reversal sufficient to obtain reliable estimates of model parameters? More details should be provided on the number of trials completed for all of the included/excluded infants.

  18. eLife

    Reviewer #3 (Public review):

    This paper used computational modeling of infants' performance in a reversal learning paradigm to identify two subgroups of infants, one that initially learned a bit faster but then perseverated more and failed to switch after the reversal (yellow cluster), and those who sampled more before the switch but then perseverated less/switched better (magenta cluster - though see below for comments about infants' overall weak performance). The authors describe magenta babies as showing a profile of greater cognitive flexibility, which they note in adults is linked to better outcomes and a lower incidence of psychiatric disorder. Indeed, the yellow cluster scored less well on several scales of the Vineland and showed lower surgency on the IBQ than the magenta cluster. The authors argue that this paper paves the way for the field of "infant computational neuropsychiatry."

    In general, I think this is a fun and intriguing paper. That said, I have a number of concerns with how it is currently written.

    First, the role of pupil dilation in the models was really unclear -- I've read it through a few times and came away with different impressions each time. I am now pretty sure the models were only based on infants' behavioural responses (e.g., choice for the correct versus incorrect location) rather than differences in pupil size, but pupil size kept popping up throughout, and so I initially thought the clusters were based on that. The authors should clarify this so other readers are not confused. (One thing that might help is avoiding the word "behaviour" on its own, unless it is further specified as looking behaviour or not, as I assume that some would characterize pupil dilation as a behaviour as well.)

    If clusters were NOT based on pupil size (e.g., reaction to prediction error), why not? Was this attempted, and did no clusters emerge? Did the yellow and magenta group also differ in reaction to prediction error, or not? It seems like the argument that this work will be the basis of infant computational psychiatry would require that there not simply be a link between behaviour in an infant study and other measurements of their functioning - because many other papers to date have demonstrated such relationships, many longitudinally - but instead with the link to something where the neurobiology of the behaviour being studied is better understood. I assume this is why pupil dilation kept coming up, but again, it didn't actually seem to be part of the modelling unless I missed something. That is, although I think that this is a nice finding, currently I think the novelty of the finding, as well as the suggestion that it will start a whole new field, may be overblown. I certainly think the pupillometry data has promise, as does the LUMO data, which the authors alluded to being in the works. But perhaps the implications should be toned down a bit in this paper, until those data are further along.

    My final substantial comment (a few more minimal ones below) is that overall, babies did quite poorly at this task. Even after 9 post-switch trials, the magenta group was still responding at chance, and the yellow group seemed not to switch at all. Infants then all seemed to perform very well again during block 2, which makes it seem like they still had the original contingency in mind. That said, from what I could see, no data was provided about how many babies looked to the original correct first during Block 2. But based on the data, I assume they basically all went back to predicting on the first side, as otherwise their return to high levels of successful trials would not make sense, unless they somehow forgot the entire thing. It would be good to know for sure, and to have that data (specifically, how many babies looked to the original side again at the start of block 2) in the main paper. Given this overall lack of sensitive performance in the paradigm, even despite the cues signaling where the rewarding video would be changing completely (that is, the contingency between cue and outcome did not itself switch, the cues themselves did), it seems odd to discuss things like statistical or even skillful learning alongside these data.

  19. Version published to 10.1101/2025.05.12.653531v2 on bioRxiv
  20. Version published to 10.1101/2025.05.12.653531v1 on bioRxiv