Fear memory paradigm and automated behavioral assessment.

A) Adult zebrafish were trained in contextual fear conditioning over three days using CAS as the unconditioned stimulus. B) Automated behavioral assessment began with tracking using DeepLabCut, followed by feature extraction and then training of a random forest model for classification. C) Confusion matrix for the random forest model using a model trained on 80% of the data and tested on 20% of the data not used for training.

Behavioral responses during fear conditioning.

A) Percent freezing (above) and % evasive behavior as a proportion of active behavior (below). Data are means ± SEMs. *-P < 0.05 compared to vehicle treated fish via t-tests. B) Change (in comparison to pre-exposure) in percent freezing (above) and percent evasive behavior (below) during exposure to CAS (left) and during memory day (right). *-P < 0.05 compared to zero via one-sample t-tests. Boxplot center is the median, hinges are interquartile range, and whiskers are the hinge ± 1.5 times the interquartile range. C) Freezing (above) and evasive behavior as proportion of active (below) behavior over time during pre-exposure, exposure, and memory days. Lines are means and ribbons are 95% confidence intervals. Vehicle treated groups, n’s = 12; CAS treated groups, n’s = 40-42. Part C includes all the CAS treated animals (N = 331).

Transitions between behavioral states.

A) Transition matrices capturing how often zebrafish change from the source to the destination state at different stages of the contextual fear memory behavioral paradigm. B) Transitions between source and destination states over time for normal, evasive and freezing behavior. Error bars are 95% confidence intervals.

Clustering of zebrafish behavior during fear memory recall.

A) Two-dimensional representation of the three-dimensional behavioral space using a uniform manifold approximation (UMAP). The outer circles delineate the clusters as defined using Louvain clustering applied to a k-nearest neighbor network (k = 74). Each point represents an individual fish (N=331). B) Behaviors associated with the identified clusters presented as z-scores. Box plots indicate median (center line), interquartile range (box ends) and ± 1.5 times the interquartile range (whiskers). C) Behaviors across time for the different clusters; ribbons indicate 95% confidence intervals. D) Percentage of animals falling into each behavioral cluster across strain and sex. Striped bars (P < 0.05) represent under/over representation using permutation tests with FDR corrections; n’s = 40-42 per strain and sex.

Transitions between behavioral states over time.

A) Alluvial plot indicating the number of fish in each cluster during pre-exposure, exposure, and memory day for animals exposed to vehicle (left) or CAS (right). B) Confusion matrix of animals falling into different clusters during exposure and memory day. Numbers indicate the quantity of animals falling into each group on exposure and memory days. Vehicle treated fish; N = 96; CAS treated: N = 331.

Partial least squares (PLS) analysis to identify brain regions that covary with behavior.

A) Top two contrasts and related behavioral saliences identified from PLS analysis. B) Bootstrap ratios of brain saliences for the first contrast on a sagittal image of the adult zebrafish brain. Only bootstrap ratios of above 2.5 are depicted. C) Coronal slices with same coloration as part B. D) Bootstrap ratios of brain saliences for the second contrast on a sagittal image. E) Coronal slices of bootstrap ratios correspond to the second contrast. Brain region abbreviations can be found in Table S2. N = 87.

Covariation between brain regions across behavioral groups.

A) Correlation matrices of cfos activity between brain regions. Each row/column is an individual brain region, and each entry is the Pearson correlation of cfos activity across animals in a group. Colored bars represent the ontological level of each brain region. B) Pearson’s correlations of matrices from each group with one another. Error bars are bootstrapped 95% confidence intervals. C) Correlation matrices with only suprathreshold correlations (FDR < 0.001). Habituation: n = 19; non-reactive, n = 17; evading freezer: n = 16; freezer: n = 21.

Functional brain networks underlying fear memory expression.

A) Functional networks where nodes represent brain regions, and edges the presence of suprathreshold correlations (FDR < 0.001). Node colors correspond to ontological levels from figure 7. Lighter dashed edges correspond to suprathreshold correlations where confidence intervals overlap with the correlation from at least one other network (evading freezer, freezer, or non-reactive). Solid lines correspond to suprathreshold correlations with no overlaps with correlations in other matrices. B) Degree centrality of the top 15 nodes. Bars are colored according to the number of edges that arise from unique (light gray) or overlapping (black) confidence intervals as in part A. C) The relative degree score for nodes enriched in the evading freezer (positive) or freezer (negative) functional networks. Only those nodes with a relative degree score of more than ± 0.5 and degree greater than 12 are shown.