Consistent Individual Differences Drive Collective Behavior and Group Functioning of Schooling Fish

Abstract

The ubiquity of consistent inter-individual differences in behavior ("animal personalities") [1, 2] suggests that they might play a fundamental role in driving the movements and functioning of animal groups [3, 4], including their collective decision-making, foraging performance, and predator avoidance. Despite increasing evidence that highlights their importance [5-16], we still lack a unified mechanistic framework to explain and to predict how consistent inter-individual differences may drive collective behavior. Here we investigate how the structure, leadership, movement dynamics, and foraging performance of groups can emerge from inter-individual differences by high-resolution tracking of known behavioral types in free-swimming stickleback (Gasterosteus aculeatus) shoals. We show that individual's propensity to stay near others, measured by a classic "sociability" assay, was negatively linked to swim speed across a range of contexts, and predicted spatial positioning and leadership within groups as well as differences in structure and movement dynamics between groups. In turn, this trait, together with individual's exploratory tendency, measured by a classic "boldness" assay, explained individual and group foraging performance. These effects of consistent individual differences on group-level states emerged naturally from a generic model of self-organizing groups composed of individuals differing in speed and goal-orientedness. Our study provides experimental and theoretical evidence for a simple mechanism to explain the emergence of collective behavior from consistent individual differences, including variation in the structure, leadership, movement dynamics, and functional capabilities of groups, across social and ecological scales. In addition, we demonstrate individual performance is conditional on group composition, indicating how social selection may drive behavioral differentiation between individuals.

Keywords: animal grouping; animal personality; collective behavior; consistent individual differences; group performance; group phenotypic composition; leadership; schooling; sociality; stickleback.

Figure 1

Group Shoaling Experiments (A–C) Schematics of (A) the free-schooling context, (B) the open foraging context with patches of food, and (C) the semi-covered foraging context with patches of food and plant cover. Schematics show tracking segments of one randomly selected group, with colors corresponding to the individual fish. Triangles point in the direction of motion. (D) Graphic illustrating key spatial and movement characteristics with arrows depicting movement vectors. For the individual assays, see Figure S1.

Figure 2

Effect of Social Proximity Tendency on Spatial Positioning and Leadership (A) Fish nearest neighbor distance in groups as a function of their social proximity tendency, shown in five equally sized categories (mean ± 2 SEM; n = 120 fish). (B) Proportion of time fish occupied the most central to the most peripheral position in the group, calculated for each frame and averaged per individual across all frames (mean ± 2 SEM). (C) Density plot of the proportion of time individuals spent in front of the group center for the full 30 min trial. (D) Visualization of a leadership network in terms of propagation of speeding changes of one randomly selected group. Numbers indicate the average temporal delay in seconds and arrows point in the direction of propagation; see Figure S3. For plots (B) and (C), individuals were evenly distributed into three categories, with the intermediate category not shown for clarity. All data were analyzed as a continuous variable. See also Figures S2 and S4 for model simulations.

Figure 3

Group Structure and Movement Dynamics in Relation to Group Mean Social Proximity Tendency (A–C) Heatmaps showing the distribution and link between the three key components of collective motion for groups with a low mean social proximity tendency (n = 13) relative to groups with a high mean social proximity tendency (n = 12). Groups with a relatively high social proximity tendency were more likely to be found in the bluer regions of the plots, whereas groups with relatively low social proximity tendency were more likely to be found in the redder regions of the plots. Group speed depicts the mean median swimming speed of the individuals in a group and is qualitatively similar to the speed of the group centroid. Plots are based on frame-by-frame data at time steps of 1/24^th s, with groups evenly allocated to two categories based on their mean social proximity tendency. Units are in mean body length (BL; 40.6 mm), and contours represent iso-levels in percentage of the highest bin for all groups combined; see Figure S2. (D) Proportion of time groups were schooling, characterized based on the raw distributions of group speed, cohesion, and polarization (see STAR Methods). Solid gray line and dashed gray lines indicate a linear fit to the data with 95% confidence intervals.

Figure 4

Effects of Individual Social Proximity and Exploratory Tendencies on Group Foraging Dynamics (A) Total number of foraging areas discovered during the open foraging context trials (out of 295 discoveries). (B) Inverted survival plot with confidence intervals of fish's likelihood to feed in the open and semi-covered foraging context. (C) Boxplots depicting total time spent out of plant cover alone in the semi-covered foraging context when food was still available. (D) Density plot of the mean number of food items eaten per trial across both foraging contexts. For plots (A)–(D), individual tendencies were evenly distributed into low, medium, and high categories (n = 42, n = 42, and n = 41 fish, respectively), with the intermediate category not shown for clarity. (E) Group foraging speed in the open (top) and semi-covered foraging cover context (bottom) in terms of the latency to consume each food item (15 provided per trial). The plot shows latencies averaged across trials for each group, and groups split into four categories based on their mean exploration and social proximity tendencies (low-low, low-high, high-low, and high-high: n = 5, 8, 8, and 4, respectively). (F) Surface plot of the mean number of food items eaten (log transformed) in the open foraging context (points indicate individual fish), based on a generalized linear mixed model (GLMM) fit to the data, cropped to 90% to show the effect excluding fish with the most extreme tendencies (n = 12 fish). Relative social proximity tendency is shown inverted such that faster fish are on the right and slower fish on the left, directly comparable with the model simulations of speed (see Figure S4).