From Schooling to Shoaling: Patterns of Collective Motion in Zebrafish (Danio rerio)

Animal groups on the move can take different configurations. For example, groups of fish can either be ‘shoals’ or ‘schools’: shoals are simply aggregations of individuals; schools are shoals exhibiting polarized, synchronized motion. Here we demonstrate that polarization distributions of groups of zebrafish (Danio rerio) are bimodal, showing two distinct modes of collective motion corresponding to the definitions of shoaling and schooling. Other features of the group's motion also vary consistently between the two modes: zebrafish schools are faster and less dense than zebrafish shoals. Habituation to an environment can also alter the proportion of time zebrafish groups spend schooling or shoaling. Models of collective motion suggest that the degree and stability of group polarization increases with the group's density. Examining zebrafish groups of different sizes from 5 to 50, we show that larger groups are less polarized than smaller groups. Decreased fearfulness in larger groups may function similarly to habituation, causing them to spend more time shoaling than schooling, contrary to most models' predictions.


Introduction
Living and traveling in groups confer multiple benefits in avoiding predation and improved foraging [1] and individuals of a majority of fish species spend some part of the their lives in groups [2].Groups of fish are commonly termed either shoals or schools and several authors have drawn a distinction between the two terms [3]: shoal refers to any group of fish that ''remain together for social reasons'' (i.e., not solely due to an external stimulus [4]) whilst schools are shoals that are ''polarized and coordinated'' [4].Polarization is a measure of the degree to which members of the group are moving in the same direction.Being in a polarized group may confer anti-predatory advantages beyond those available in a disordered shoal, for instance through increased predator confusion [5], and may make it easier to detect any sudden deviation in the heading of a conspecific, potentially a sign of impending danger [1].Polarized motion may also help fastmoving groups to stay together (or even be required to maintain the cohesion of fast groups) -particularly important for migrating species -and may improve the flow of information through the group [6,7].
Additionally, some theoretical models of collective motion predict a sharp density-dependent transition from a low polarization regime (corresponding to shoaling) to a high polarization regime (schooling [8]).However, these models do not consider indirect effects on collective motion resulting from changes in group size.For instance, members of larger groups may experience lower stress levels than members of smaller groups and these changes may themselves affect the polarization of the group.
Despite broad acceptance of the distinction between schooling and shoaling [4], no work that we are aware of has examined the empirical characteristics of each behavioral mode (as called for, e.g., by [9]) or sought to find what determines which of the two modes of motion groups adopt, other than group density.Here we show that polarization distributions of groups of zebrafish (Danio rerio) are bimodal, implying a behavioural distinction between highly polarized groups -schools -and weakly polarized groupsshoals -and that other features of collective motion, such as the mean speeds of group members and the mean spacing between them, correlate with polarization (Experiment 1).We also show that the dominant mode of motion exhibited by a group depends on several environmental conditions such as the group's habituation to the environment (Experiments 1 and 2) and the size of the group (Experiment 3).Interestingly, in our data polarization is negatively correlated with the size of the group, in contrast to the predictions of some models.We suggest that this may result from a failure of these models to consider indirect effects of changes in group size.

Experiments 1 and 2
Using detailed swimming trajectories of groups of 8 zebrafish each (see Methods), we constructed the mean distributions of the polarization of the group.Polarization distributions were bimodal (Figure 1), implying that groups of zebrafish often form either highly polarized schools or weakly polarized shoals but take intermediate forms less frequently.We explicitly tested each distribution for bimodality (see Methods): of 118 analyzed distributions, all but 21 were bimodal and the peaks of most distributions that were unimodal fell clearly into one or the other behavioral form (Figure S2).
Polarization distributions changed dramatically with repeated exposure to the testing tank (Figure 1; Table S1), becoming increasingly biased towards shoaling.A similar effect was observed across multiple hours of a single extended exposure to the testing tank (Figure 2) and these effects were highly significant (Tables S1,  S2).In other words, zebrafish groups spent more time shoaling and less time schooling as they habituated to the testing environment (Figure 3a).The degree of polarization of schools (the position of the peak corresponding to schooling in the distribution) did not change across repeated exposures to the tank (Repeated Measures ANOVA, F(4, 24) = 1.994, p = 0.128) but the degree of polarization of shoals decreased, i.e., shoals became more disorganized (F(4,24) = 2.816, p = 0.048) as fish habituated to the tank (Figure 3b).This result implies that schooling is a fixed behavioral pattern whereas the characteristics of shoals may depend on environmental factors (such as habituation).
Next, we segmented all the trajectories by which mode they fell into: each frame of data was designated either a schooling or shoaling frame by the polarization of the group at that time-point (unimodal distributions were excluded from this analysis).Most segments lasted only a few seconds, reflecting rapid changes in the groups' polarization, but groups spent over a third of their time (36.4%) in schooling or shoaling segments lasting longer than 30 sec.2.45% of schooling and 8.87% of shoaling segments lasted for more than 1 min.Shoaling segments were longer, on average, than schooling segments (Figure S3; Table S3).
Finally, we compared the two modes of behaviour on other measures of collective motion: for each frame in each behavioral mode we measured the mean Nearest Neighbor Distance (NND), mean Inter-Individual Distance (IID; the mean distance between an individual and all other members of the group), and mean individual speed (Figure 4).Fish were further apart when in schools than when in shoals (paired-sample t-test; NND, t = 8.51, p,0.0001;IID, t = 9.59, p,0.0001) and swam faster (t = 14.92, p,0.0001).The latter distinction, a positive correlation between speed and polarization, has been previously noted in other fish species [10].

Experiment 3
We next tested how changes in group size affect which behavioral mode the group inhabits.Some models of collective motion [8] suggest that the polarization of a group depends on its Summed polarization distributions for repeated exposures to the testing tank on consecutive days.Distributions are averaged across the entire session (1800 frames per session) and over all groups (24 groups of 8 fish each).As fish habituate to the tank across days they spend more time shoaling (low polarization) and less time schooling (high polarization; K-S test, days 1-2 vs. days 3-5, all p,0.0001; see Table S1).The inset shows a sample distribution from one complete session (1800 frames) (red) and its decomposition into schooling and shoaling modes (black), by fitting a Gaussian mixed model (see text).Note that each session's distribution was tested for bimodality independently.doi:10.1371/journal.pone.0048865.g001) and all groups (8 groups of 8 fish each).A similar effect of habituation is seen to that observed in Figure 1, with increased shoaling and decreased schooling as time passes (K-S test, hour 1 vs. hours 2-4, all p,0.001; see Table S2).doi:10.1371/journal.pone.0048865.g002size or density, an effect that has also been observed in some behavioral data [11,12].In these situations, both theoretical and empirical, small groups have widely and rapidly varying polarization; as group size increases polarization both increases and becomes more stable, so that large groups are highly polarized and rarely change direction.We tested groups of zebrafish over a range of group sizes (N = 5, 10, 20, 30, and 50) and applied the same analyses to the data as above (sample polarization time-series' are presented in Figure S4). Figure 5 shows the summed polarization distributions for all group sizes.Contrary to our expectations, and the predictions of the models, larger zebrafish groups were significantly less polarized, on average, than smaller groups (Table S4), spending more of their time shoaling and less time schooling.NND did not vary consistently across group sizes (Figure S5A).Larger shoals were somewhat slower on average than smaller shoals (Figure S5B), though this may simply reflect the greater amount of time they spent shoaling.Note that the mean NND of groups in this experiment was lower than that of groups in Experiment 1.

Discussion
Above, we show that zebrafish more frequently form either highly or weakly polarized groups and rarely take intermediate forms.We call these modes schooling and shoaling, respectively, following the terminology common in the literature [4].However, it might be argued that the distinction between schools and shoals should refer only to changes that persist for longer periods of time and that the fluctuations in polarization in our data, often lasting only a few seconds (Figure S3), represent no more than noise around one mode or the other.Nonetheless, our data demonstrate that groups of zebrafish exist predominantly in two statistically distinct modes of collective motion rather than intermediate forms.We also show that both social environment and internal states can affect the distribution of group polarization which, in some cases, might lead to long periods spent strictly in one mode or the other.Such situations would more closely correspond to the traditional distinction between schooling and shoaling but would, according to this account, represent merely extreme manifestations of the same mechanisms revealed in our data.
Additionally, the walls of the testing tank might have had different effects on large vs. small or on habituated vs. stressed groups.Zebrafish (and other species) tend to remain close to the   Larger groups are significantly less polarized than smaller groups (K-S test, all p,0.0001; see Table S3).doi:10.1371/journal.pone.0048865.g005walls of novel environments and this thigmotactic response might have acted to linearize, and thus polarize, the fish in our groups.Larger (more numerous) groups may have filled a larger portion of the tank and thus been unable to remain sufficiently close to the walls of the tank for any polarizing effect to be felt throughout the group.However, we note that in our data, large and habituated groups were still able to achieve high levels of polarization, equal to those demonstrated by smaller or more stressed groups, they simply did so less often (or for shorter periods of time).Thus, as shown in Figure 1, habituating and habituated groups still displayed schooling behavior which was as polarized as their schooling on earlier days (Figure 3B) but they performed this behavior less often.
Partridge [9] suggested that the collective motion of different species of fish might best be characterized by the proportion of time that they spend schooling.Our data (e.g., Figure 2B) demonstrate that factors other than species also contribute to this ratio.Both group size and habituation to the environment affect the proportion of time zebrafish groups spend in each mode.Rather than smoothing out the polarization distribution and eliminating its bimodality, changes in these variables serve primarily to alter the relative amplitudes of the two modes, suggesting that they mark two distinct methods of forming a group.This suggestion is supported by differences in other characteristics, such as the speed and density of the group, between the two modes.
The changes we observed in polarization and in the proportion of time spent in each of the two modes of motion may be explained by changes in the level of fear (or stress) of the fish.As fish habituate to the testing environment, either over repeated exposures to it (Experiment 1) or simply over time during a single exposure (Experiment 2), or when fish are in a larger group (Experiment 3), they might be less fearful and therefore more likely to adopt a shoaling configuration rather than schooling.This would further suggest that schooling behaviour is maintained, under stressful conditions, at some energetic or cognitive cost over shoaling behaviour.This cost might be associated with greater attention required to school [5] or, indirectly, with the metabolic costs of swimming faster (as we and others have shown that individuals in schools swim more quickly than those in shoals).
Schooling may provide greater protection from predation than shoaling but may hinder individual foraging, particularly for individuals near the back of the school.We [13] have previously shown that the density of zebrafish groups oscillates with a characteristic frequency and suggested that this may represent a solution to the trade-off between protection, which is enhanced by closeness to conspecifics, and foraging, which is easier at some distance from the rest of the group [14].The present data demonstrate another way that zebrafish balance these conflicting requirements, by adjusting the proportion of time they spend shoaling or schooling to match their current social conditions.
Most models of collective motion assign a set of simple rules to each individual, dependent on its local environment [8,[15][16][17][18][19][20][21][22] and the combination of many such local interactions gives rise to globally coordinated motion, closely simulating the movement of real groups of fish.Many of these models exhibit both shoaling and schooling under different parameter regimes (e.g., [15]).In some cases increasing the size of the simulated group, and thus the number of local interactions, leads to an increase in polarization and stabilization of the polarization [8,17], a result that is supported by some empirical data [11,12].In our data, larger groups of zebrafish were more likely to shoal, displaying low polarization, than smaller groups, contrary to the predictions of such models.This may reflect an additional behavioral 'rule' that is not accounted for by current models of collective motion and plays a smaller role in the species studied to date under varying densities [11,12], possibly one that depends on the stress or fear levels of the animals, as suggested above.This effect, in our data, is sufficiently powerful to completely overshadow any increase in polarization resulting from larger group sizes.Recently, a possible mechanism has been suggested [5] by which an increase in stress levels may lead to individuals increasing the rate at which they examine their relative position and orientation to the group and this increased 'update rate' may result in increased polarization.
Our data provide quantitative support for the long-standing definition in the literature between schooling and shoaling and demonstrate how some environmental variables can affect the proportion of time zebrafish groups spend in each mode.To the best of our knowledge, this distinction has been studied only in groups of fish.However, it is likely that similar processes of habituation and similar group-size effects to those observed here operate in other group-living taxa.Do herds of mammals or flocks of birds also have two (or more) distinct modes of collective motion?Examining the polarization distributions of other species may reveal interesting similarities and differences in their methods of collective motion.

Ethics statement
All animal procedures were approved by the University of Toronto at Mississauga Animal Care Committee and followed the guidelines set by the Canadian Council on Animal Care (CCAC).

Subjects
Subjects were all laboratory-reared sexually mature zebrafish 3-12 months of age and were 3-4 cm long.All fish were bred inhouse.In Experiment 1, 24 groups of 8 fish per group were tested for five consecutive days each; In Experiment 2, 8 groups of 8 fish each were tested once each; and in Experiment 3, 4 groups of each size (N = 5, 10, 20, 30, and 50 fish) were tested once each.All groups consisted of approximately equal numbers of males and females.Groups of fish to be tested together were housed together for at least one week prior to the start of each experiment.Fish were housed in 40-litre tanks containing 'system' water that was previously reverse osmosis purified and mixed with sea salt ('Instant Ocean' sea salt, Aquarium Systems Inc., OH) so that the conductivity of the water was between 900-1200 micro Siemens (576-768 TDS ppm).The water in the tanks was filtered (Aqueon PowerFilter 30, Franklin, WI), aerated, and maintained at a temperature of 2662uC.Lights in the room in which the fish were kept turned on at 7:00 h and off at 21:00 h.Fish were fed flake food (Tetramin Tropical Flakes, Tetra, USA) ad lib, 30 min before each testing session.

Apparatus
Fish were tested in a circular white plastic tank with a diameter of 91 cm, filled with system water to a depth of 10 cm (Figure S6), located in one corner of the room where they were housed.The temperature of the water in the tank was kept at 2662uC.A moveable black room divider visually separated the tank from the rest of the room.The tank was lit by two fluorescent light fixtures placed at opposite sides of the tank, ensuring an even level of illumination in all parts of the tank.Experiments were filmed using a Sony Handycam (model HDR-XR-520) attached to the ceiling with the lens of the camera 200 cm above the surface of the water, located so that the entire tank was visible in the frame of the video.

Figure 1 .
Figure1.Density distributions of polarization by testing day for Experiment 1. Summed polarization distributions for repeated exposures to the testing tank on consecutive days.Distributions are averaged across the entire session (1800 frames per session) and over all groups (24 groups of 8 fish each).As fish habituate to the tank across days they spend more time shoaling (low polarization) and less time schooling (high polarization; K-S test, days 1-2 vs. days 3-5, all p,0.0001; see TableS1).The inset shows a sample distribution from one complete session (1800 frames) (red) and its decomposition into schooling and shoaling modes (black), by fitting a Gaussian mixed model (see text).Note that each session's distribution was tested for bimodality independently.doi:10.1371/journal.pone.0048865.g001

Figure 2 .
Figure 2. Density distributions of polarization by hour for Experiment 2. Summed polarization distributions for each hour of a single, 4 hour, exposure to the testing tank.Distributions are averaged across the entire session (1800 frames per session) and all groups (8 groups of 8 fish each).A similar effect of habituation is seen to that observed in Figure1, with increased shoaling and decreased schooling as time passes (K-S test, hour 1 vs. hours 2-4, all p,0.001; see TableS2).doi:10.1371/journal.pone.0048865.g002

Figure 3 .
Figure 3. Ratio of the mean time spent schooling to shoaling (a) and mean modes of polarization distribution components (b) by testing day in Experiment 1. Unimodal data distributions were excluded.Error bars represent 6 SEM.The spike in time spent schooling on day 2 (a) is attributable to a few outlier sessions.doi:10.1371/journal.pone.0048865.g003