A behavioral paradigm for investigating innate decision-making in mice.

(A) Schematic of the behavioral assay (3D and top-down views). (B) Top: arena occupancy patterns for five example mice in one session. Bottom: visit frequency and average duration for each visit. Colors mark the mouse’s identity. (C) Visit frequency and duration under exploration and threat conditions. Error bars represent the standard deviation. N = 46 mice, paired t-test. (D) The pipeline for behavioral classification. (E) Distribution of decisions across 3862 trials from 140 mice. (F) Left: distance to the safe zone over time for four decision types (10 example trials each). Red dashed lines mark the onset of each stimulus repetition; solid lines mark the end of the last repetition. Grey shade indicates the safe zone. Right: locomotion speed towards the safe zone over time for the same trials. Positive speed indicates movement towards the safe zone. (G) Distribution of the longest stationary time for “freezing” and “assessment+escape”. N = 224 and 1667 trials, Mann-Whitney-Wilcoxon test. Blue lines mark the median; red solid lines mark the end of the last repetition. (H) Distribution of latency to flee for “escape” and “assessment+escape”. N = 458 and 1667 trials, Mann-Whitney-Wilcoxon test. Blue lines mark the median; red dashed and solid lines mark the end of the first and last repetitions, respectively. (I) Distribution of peak speed for “escape” and “assessment+escape”. N = 458 and 1667, Mann-Whitney-Wilcoxon test. (J) Distribution of hiding latency in the safe zone for “escape” and “assessment+escape”. N = 455 and 1563, Mann-Whitney-Wilcoxon test. (K) Distribution of post-decision behavioral states. (L) Distribution of fear recovery time for “escape”, “assessment+escape”, and “freezing”. N = 74, 458, and 205, Kruskal-Wallis test followed by Dunn’s post-hoc test with Holm correction. *p < 0.05, **p < 0.01, ***p < 0.001.

Mice learn quickly from experience.

(A) Schematic of the behavioral assay for studying the economic modulation of innate decision-making. (B-C) Distance to the safe zone and locomotion speed over time for the 1st and 10th trials in response to low- and high-contrast looming stimuli across different reward conditions. Right inset in (B) is the distribution of distance to the safe zone at the end of each trial. Dashed lines mark the start of each stimulus, and solid lines mark the stimulus offset. N = 11 (no reward, low), 13 (water, low), 10 (sucrose, low), 10 (no reward, high), 8 (water, high), 10 (sucrose, high) mice. (D) Summary of the decisions across the first 10 trials under different risk and reward conditions. N = 11, 13, 10, 10, 8, 10 mice. (E-H) Escape distance under threat, duration in reward zone, peak escape speed, and latency to flee across trials in all conditions. Shade denotes the standard error of the mean. (I) Transition trials marking the shift between phases across conditions. Red lines indicate the median; boxes span the interquartile range (IQR); whiskers extend to 1.5 × IQR beyond the box.

Mice make economic decisions modulated by vigilance.

(A) Distribution of decisions in the early phase for six experimental conditions. N = 43 trials from 11 mice (none, low), 42 trials from 13 mice (water, low), 29 trials from 10 mice (sucrose, low), 31 trials from 10 mice (none, high), 21 trials from 8 mice (water, high) and 32 trials from 10 mice (sucrose, high), Chi-squared test. (B-G) Escape distance under threat, duration in the reward zone, latency to flee, foraging interval, foraging speed, and peak escape speed in the early phase across all conditions. Scheirer–Ray–Hare test with post hoc Dunn’s test (Holm correction). N = 43, 42, 29, 31, 21, 32 trials for B, C, F, G; N = 35, 37, 24, 31, 21, 32 trials for D; N = 116, 84, 54, 58, 36, 48 intervals for E. (H) Distribution of decisions in the late phase for six experimental conditions. N = 59 trials from 11 mice (none, low), 88 trials from 13 mice (water, low), 71 trials from 10 mice (sucrose, low), 67 trials from 10 mice (none, high), 59 trials from 8 mice (water, high) and 48 trials from 9 mice (sucrose, high), Chi-squared test. (I-N) Escape distance under threat, duration in the reward zone, latency to flee, foraging interval, foraging speed, and peak escape speed in the early phase across all conditions. Scheirer–Ray–Hare test with post hoc Dunn’s test (Holm correction). N = 59, 88, 71, 67, 59, 48 trials for I, J, M, N; N = 43, 49, 15, 66, 59, 47 trials for K; N = 116, 182, 114, 137, 71, 57 intervals for L. For all panels: *p < 0.05, **p < 0.01, ***p < 0.001.

Influence of social hierarchy on innate decision-making.

(A) Schematic of the behavioral assay for studying the social modulation of innate decision-making. Top, experimental timeline. Each session lasted 2 hours per mouse pair during the pre-looming, looming, and post-looming phases. Bottom, schematic of the tube test. (B) Arena occupancy for an example pair of mice during the three sessions. (C) Visit frequency and total visit duration for dominant and subordinate mice during the pre-threat, threat, and post-threat sessions. N = 5 pairs (pre), 5 pairs (threat), 4 pairs (post), respectively; paired t-test. (D) Proportion of nest-return trials in escape trials for dominant and subordinate mice. N = 9 pairs, paired t-test. (E) Distance to the safe zone over seven days for an example pair. Looming stimuli were presented on days 4 and 5. (F) Percentage of time spent in the reward zone across days. Error bars represent SEM. N = 9 pairs, paired t-test. (G) Distance to the safe zone (left) and locomotion speed (right) for an example pair. (H) Behavioral decisions across the first 10 trials for 9 mouse pairs. (I) Pie charts showing decision distributions for dominant and subordinate mice. N = 90, 90 trials, Stuart-Maxwell test. (J) Latency to flee for dominant and subordinate mice. N = 78 trials; paired t test. (K) Foraging interval for dominant and subordinate mice. N = 75 trials. (L) Foraging speed for dominant and subordinate mice. N = 53 trials. (M–O) Escape distance under threat, peak escape speed, and duration in the reward zone for dominant and subordinate mice. N = 90 trials. *p < 0.05, **p < 0.01, ***p < 0.001.

Drift-diffusion leaky integrator model for escape decisions.

(A-F) Simulated accumulation of escape evidence, along with predicted latencies to flee and decision distributions across six threat and reward conditions. Green dashed horizontal lines mark the x = 0; blue dashed horizontal lines mark the x = xthr; red dashed vertical lines mark the onset of each looming stimulus; blue vertical lines mark the median. (G) Heatmap of decision scores as a function of threat gain and reward value. White and red stars indicate fitted parameters for the early and late phases of the reward–threat paradigm, respectively; upward and downward pink triangles indicate fitted parameters for dominant and subordinate mice in the social-threat paradigm, respectively. (H) Heatmap of latencies to flee as a function of threat gain and reward value. (I) Vigilance as a function of input strength, illustrating how the indirect effect of reward on defensive decisions via vigilance depends on the baseline vigilance level. (J) Schematic showing how threat intensity, reward value, and vigilance jointly determine defensive decisions. Color saturation indicates the likelihood of defensive decisions.

Model validation and feature analysis for automated behavioral classification.

(A) Tracking accuracy of the mouse nose and tail base using DeepLabCut. (B) Weights of the 19 behavioral features in the random forest classifier. (C) Model performance evaluated by a confusion matrix on the test dataset.

Effects of threat position, prey–threat distance, and prey–safety distance on defensive responses to looming stimuli.

(A) Schematic of the experiment in which looming stimuli were presented either in front of or behind mice during foraging. (B) Distribution of escape direction for front and behind stimulus positions. N = 97, 16 trials. (C) Schematic of the experiment in which looming stimuli were presented at the end of the linear arena with varying distances between the mouse and the threat. (D) Distance to the safe zone over time for four prey-threat distances. Red dashed lines mark the onset of each stimulus repetition; solid lines mark the end of the last repetition. Grey shade indicates the safe zone. Blue lines mark the mouse’s position when the looming stimulus was triggered; orange lines mark the location of the looming stimulus. N = 7 (15 cm), 7 (35 cm), 9 (55 cm), and 8 (75 cm) trials from 5 mice. (E) Distribution of decisions across prey-threat distances. (F–G) Latency to flee and peak escape speed across prey-threat distances. Kruskal-Wallis test with post hoc Dunn’s test (Holm correction). (H) Schematic of the experiment in which looming stimuli were presented in front of the mouse at varying distances between the mouse and the safe zone. (I) Distribution of decisions across prey-safety distances. N = 4, 4 trials. (J) Schematic of the experiment in which low-contrast looming stimuli were presented at varying distances between the mouse and the safe zone in the linear arena with barriers. (K) Distribution of decisions across prey-safety distances in the barrier condition. N = 8, 7 trials. *p < 0.05, **p < 0.01.

(A) Water or sucrose consumption during exploration and looming experiments at low and high contrasts. N = 10, 10, 5, 5, 5, 5 sessions; Mann–Whitney–Wilcoxon test. Boxes represent the interquartile range (IQR), and whiskers show the full data range. (B) Pie chart showing the proportion of looming stimuli detected by mice in no-response trials under low- and high-threat conditions. Low contrast: N = 117 trials from 29 mice; high contrast: N = 2 trials from 2 mice (high contrast). (C) Proportion of total behavioral variance explained by principal component 1 (PC1) for individual mice across all conditions. N = 11, 13, 10, 10, 8, 10. (D) Cumulative PC1 score for an example mouse in each condition. Red dashed lines mark the transition from the early to the late phase. *p < 0.05, **p < 0.01.

Distribution of foraging speed.

(A) Distribution of foraging speed relative to distance from the safe zone in the early phase across different risk and reward conditions. (B) Distribution of foraging speed in the late phase. Data are from 11 mice (no reward, low contrast), 13 mice (water, low contrast), 10 mice (sucrose, low contrast), 10 mice (no reward, high contrast), 8 mice (water, high contrast), and 10 mice (sucrose, high contrast).

Behavioral responses to looming stimuli across risk and reward conditions in the first trial.

(A) Distribution of decisions for each mouse under six conditions. N = 11 (no reward, low contrast), 13 (water, low contrast), 10 (sucrose, low contrast), 10 (no reward, high contrast), 8 (water, high contrast), 10 (sucrose, high contrast); Chi-squared test. (B–F) Escape distance under threat, duration in the reward zone, latency to flee, foraging speed, and peak escape speed across conditions. Scheirer–Ray–Hare test with post hoc Dunn’s test (Holm correction). N = 11, 13, 10, 10, 8, 10 trials for B, C, E, F; N = 11, 11, 10, 10, 8, 10 trials for D. *p < 0.05, ***p < 0.001.

Reward modulation of defensive behavior to looming stimuli in the same mouse.

(A) Experimental timeline illustrating how water reward influences innate decision-making within the same animal. (B) Distance to the safe zone and locomotion speed towards the safe zone across trials in response to low-contrast looming stimuli with and without water reward in two example mice. Left: no-reward condition followed by water-reward condition; Right: water-reward condition followed by no-reward condition. (C) Decision patterns of nine mice across two sessions with 5 trials for each. Gray squares indicate trials where the mouse did not enter the arena within 30 minutes. (D) Defensive probability in no-reward and water-reward conditions. N = 9 mice, paired t-test. (E) Escape distance under threat, duration in the reward zone, latency to flee, and peak escape speed in no-reward and water-reward conditions. N = 5 mice for latency to flee and 9 mice for other measures, paired t-test, two-sided. #p < 0.1, *p < 0.05, **p < 0.01.

Behavioral responses to looming stimuli for dominant and subordinate mice in different phases.

(A) Transition trial marking the start of the late phase for dominant and subordinate mice. N = 9 pairs. (B) Behavioral decisions for dominant and subordinate mice during their first exposure to threat. Stuart-Maxwell test. N = 9 pairs. (C) Latency to flee, foraging speed, escape distance under threat, peak escape speed, and duration in the reward zone during the first threat exposure. N = 4 pairs for foraging speed; N = 9 pairs for other features. Paired t test (D) Behavioral decisions for dominant and subordinate mice in the early phase. Chi-squared test. N = 34, 43 trials, respectively. (E) Violin plots showing latency to flee, average foraging interval, foraging speed, escape distance under threat, peak escape speed, and duration in the reward zone for dominant and subordinate mice in the early phase. N = 34, 43 trials, respectively. Mann-Whitney-Wilcoxon test. (F) Behavioral decisions for dominant and subordinate mice in the late phase. Chi-squared test. N = 56, 47 trials, respectively. (G) Violin plots showing latency to flee, average foraging interval, foraging speed, escape distance under threat, peak escape speed, and duration in the reward zone for dominant and subordinate mice in the late phase. N = 56, 35 trials for latency to flee; N = 56, 47 trials for other measures. Mann-Whitney-Wilcoxon test. *p < 0.05, **p < 0.01, ***p < 0.001.

Comparison of behavioral responses to looming stimuli before and after the tube test.

(A) Schematic timeline of the looming experiments before and after the tube test. (B) Behavioral decisions for dominant and subordinate mice before and after the tube test. Chi-squared test. Dominant: N = 16 (pre) and 20 (post) trials from 4 mice; Subordinate: N = 20 (pre) and 16 (post) trials from 4 mice. (C) Violin plots showing latency to flee, escape distance under threat, peak escape speed, and duration in the reward zone for dominant (top) and subordinate (bottom) mice before and after the tube test. Mann-Whitney-Wilcoxon test.

Fitting loss in the drifting–diffusion leaky integrator model.

(A) Temporal profile of the normalized looming stimulus diameter. (B) Loss landscapes across the parameter space for estimating the leakage rate, high-contrast threat gain, diffusion rate, and decision threshold during the first-stage fitting. Red stars indicate the optimal parameter estimates based on experimental data in the late phase of the reward-related paradigm under the no-reward condition. (C) Loss landscapes across the parameter space for estimating the reward value and high-contrast threat gain under the water-reward condition. Red stars mark the optimal fits to experimental data in the late phase of the reward-related paradigm. (D) Same as (C), for the sucrose-reward condition. (E) Model-predicted distribution of decisions across threat gain and reward values. (F) Predicted latency to flee as a function of threat gain for varying reward values. (G) Predicted latency to flee as a function of reward value for varying threat gain. (H–I) Decisions and latency to flee across threat gain and reward values predicted by a simplified deterministic model. (J–K) Same as (F–G), but for the simplified deterministic model.