Early-life social experience shapes social avoidance ... › content › 10.1101 ›...

Early-life social experience shapes social avoidancereactions in larval zebrafish

Antonia H. Groneberg, Joao C. Marques, A. Lucas Martins,Gonzalo G. de Polavieja∗, Michael B. Orger∗

Champalimaud Research, Champalimaud Centre for the Unknown, 1400-038 Lisbon, Portugal.

∗ Corresponding author, Email: [email protected]@neuro.fchampalimaud.org

1

.CC-BY-NC-ND 4.0 International license(which was not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprintthis version posted March 2, 2020. . https://doi.org/10.1101/2020.03.02.972612doi: bioRxiv preprint

https://doi.org/10.1101/2020.03.02.972612

http://creativecommons.org/licenses/by-nc-nd/4.0/

Highlights:

• Larval zebrafish raised in isolation show enhanced social avoidance reactions

• Enhanced avoidance is composed of increased avoidance distances and usage of highacceleration escape swims

• The lateral line sensory organ is necessary and sufficient for the increased usage of highacceleration escape swims

SummarySocial experiences greatly define successive social behavior. Lack of such experiences, espe-cially during critical phases of development, can severely impede the ability to behave ade-quately in social contexts. To date it is not well characterized how early-life social isolationleads to social deficits and impacts development. In many model species, it is challengingto fully control social experiences, because they depend on parental care. Moreover, com-plex social behaviors involve multiple sensory modalities, contexts, and actions. Hence, whenstudying social isolation effects, it is particularly important to parse apart social deficits fromgeneral developmental effects, such as abnormal motor learning. Here, we characterized howsocial experiences during early development of zebrafish larvae modulate their social behav-ior, at one week of age, when social avoidance reactions can be measured as discrete swimevents. We show that raising larvae in social isolation leads to enhanced social avoidance, interms of reaction distance and reaction strength. Specifically, larvae raised in isolation use ahigh-acceleration escape swim bout, the short latency C-start, more frequently during socialinteractions. These behavioral differences are absent in non-social contexts. By ablating thelateral line and presenting the fish with local water vibrations, we show that lateral line in-puts are both necessary and sufficient to drive enhanced social avoidance reactions. Takentogether, our results show that social experience during development is a critical factor inshaping mechanosensory avoidance reactions in larval zebrafish.

Keywords: Social experience, social avoidance, C-start, Zebrafish larvae, Lateral line

2


https://doi.org/10.1101/2020.03.02.972612


Introduction1

An important aspect of behavior is that it is adaptive and can change according to previous2

outcomes of the actions of the self or others. During early development, as the nervous sys-3

tem builds and refines its connections, an animal is particularly sensitive to external sensory4

input, or lack thereof (e.g. reviewed by Andersen [2003]). Social experiences during early5

development can have long lasting effects on social and other behaviors, as shown by the dev-6

astating phenotype of rhesus monkeys raised with social deprivation described by Harlow7

et al. [1965]. Since then, numerous studies have reported effects of social isolation raising in a8

variety of species, including rodents [Heidbreder et al., 2000, Lukkes et al., 2009], cockroaches9

[Lihoreau et al., 2009], lizards [Ballen et al., 2014], mites [Schausberger et al., 2017] and fish10

[Hesse et al., 2015, Shams et al., 2018]. Yet, the underlying mechanisms of how social experi-11

ences shape behavior and brain development have yet to be fully understood. One reason for12

this is a lack of full control over social experiences in species with parental care, where social13

isolation along with maternal separation can interfere with regular development. Moreover, in14

most animals, social behavior is complex, being triggered by multiple interacting sensory cues15

[Chen and Hong, 2018] and leading to behavioral outputs which constitute composite actions16

that can be difficult to isolate [Anderson and Perona, 2014]. Thus, the resulting phenotype of17

developmental social isolation is composed of various symptoms, making social deficits hard18

to differentiate from general impairments in motor development.19

Zebrafish are a social species, aggregating in groups in the wild [Engeszer et al., 2007,20

Spence and Smith, 2007] and in the laboratory [Arganda et al., 2012, Dreosti et al., 2015, Sted-21

nitz et al., 2018]. They develop oviparously, without parental care, making them an ideal22

vertebrate species to control precisely the social experiences during development. An early23

study from the 1940s reported that adult zebrafish show social attraction even without prior24

social experience [Breder and Halpern, 1946]. Similarly, juvenile zebrafish raised in social25

isolation show regular levels of visual attraction to a projected, naturalistically moving dot26

[Larsch and Baier, 2018]. This suggests that social attraction develops without the need for27

social experiences. However, it has also been shown that zebrafish prefer familiar over un-28

familiar visual social cues [Engeszer et al., 2004] and that olfactory kin preference requires29

visual and olfactory social experiences [Hinz et al., 2013]. Moreover, group cohesion in adult30

zebrafish was shown to be lower after social isolation-raising [Shams et al., 2018]. Therefore,31

unlike the drive for social attraction, fine-tuned aspects of social behavior seems to be acquired32

through social experiences.33

Apart from social attraction, moving in social groups also requires keeping the right dis-34

tance to neighbors. This has been formalized as a rule of avoidance, according to which an35

individual aims at maintaining a minimum distance towards others [Couzin et al., 2002, In-36

ada and Kawachi, 2002]. In developing zebrafish, social attraction develops gradually over37

the course of the second and third week of development [Hinz and de Polavieja, 2017, Dreosti38

et al., 2015, Larsch and Baier, 2018, Stednitz and Washbourne, 2020]. Social avoidance, on the39

other hand, can be observed robustly in one-week-old larvae [Hinz and de Polavieja, 2017,40

Marques et al., 2018, Mirat et al., 2013].41

Despite having a simplified social behavior, there are advantages to studying one-week-old42

zebrafish larvae. First, at this age fish are amenable to non-invasive manipulations, including43

3


https://doi.org/10.1101/2020.03.02.972612


bath application of psychoactive drugs [Rihel et al., 2010], genetic and optical ablation of44

defined sets of neurons [Asakawa and Kawakami, 2008, Sternberg et al., 2016], and optogenetic45

control of neural activity [Douglass et al., 2008, Friedrich et al., 2010]. Second, larvae are46

small and transparent enough to enable whole brain imaging with single neuron resolution47

in behaving animals [Feierstein et al., 2015], including freely moving fish [Kim et al., 2017,48

Marques et al., 2020, Cong et al., 2017], an ideal situation where animals can interact with49

each other. Finally, larval zebrafish organize their behaviors, including social interactions,50

in sequences of discrete bouts [Budick and O’Malley, 2000, Fero et al., 2011], which can be51

detected automatically and classified into types [Mirat et al., 2013, Marques et al., 2018, Mearns52

et al., 2020, Johnson et al., 2020], enabling precise behavioral phenotyping.53

Here, we characterized how the social experience during early development affects social54

avoidance behavior in one-week-old larval zebrafish. Using high-speed video tracking and55

unsupervised clustering of swim bout types in freely interacting larvae [Marques et al., 2018],56

we found that isolation-raised larvae avoid conspecifics at larger distances by executing short57

latency C-starts [Burgess and Granato, 2007a] with higher probability. These behavioral differ-58

ences are absent in non-social contexts, including upon the presentation of visual and acoustic59

stimuli that elicit escape reactions. Ablation of the lateral line reduces enhanced avoidance60

reactions, and the presentation of local water vibrations, which mimic the movements of an-61

other fish, elicits them. Therefore, lateral line inputs are both necessary and sufficient to drive62

enhanced social avoidance reactions. In summary, our results show that lack of social expe-63

rience during development affects the type of avoidance response that fish execute, and is64

therefore a critical factor in shaping mechanosensory avoidance reactions in larval zebrafish.65

Results66

Isolation raised larval zebrafish show enhanced social avoidance reactions67

To measure social interactions in one-week-old larvae, we tracked the positions and tail move-68

ments of groups of seven wild-type zebrafish larvae swimming freely in a circular arena under69

constant, homogeneous illumination (Fig. 1 A). Testing larvae were either raised in groups70

(GR) or in social isolation from 0 days post fertilization (dpf) until testing (ISO). We computed71

the density distribution of positions of all other larvae relative to a focal larva and compared72

it to a position density of a randomized distribution. In line with previous reports [Hinz and73

de Polavieja, 2017, Marques et al., 2018] there is a low density region centered around the focal74

larva where it is less likely to find other larvae swimming (Fig. 1 B). This region has been75

termed ’avoidance area’ and appears to be larger across the ISO groups than the GR groups.76

In order to quantify the avoidance area for each replicate per recording group, we detected77

the contour of the central area as defined by a common cutoff in density ratio and computed78

its area (Fig. 1 B right panel). Thereby we found that ISO groups have a larger avoidance area79

than GR fish (Fig. 1 C; p < 0.0001, N=9).80

In order to study in greater detail how larvae react to one another, we adjusted the behav-81

ioral assay to test pairs of larvae. We used a recording chamber with four separate swimming82

arenas (Fig. S1 A). In line with the results from the group recordings, the low density re-83

gion of neighboring positions in pairs of ISO larvae is larger than in GR larvae (Fig. S1 B).84

4


https://doi.org/10.1101/2020.03.02.972612


-100%

100%

D

ensi

ty D

iffer

ence

A B

C5 mm

***

Avo

idan

ce A

rea

(mm

2 )

150

100

50

0GroupRaised

IsolationRaised

Isol

atio

n R

aise

dG

roup

Rai

sed

5 mm 5 mm25 mm

Avoidance AreaContour

NGroups=9NLarvae=63

Figure 1: Isolation-raised larvae have a larger social avoidance area. (A) Groups of seven larvaeraised either in groups or in isolation were video recorded for 1 h while swimming in a circular arena50 mm in diameter. (B) Density distribution between real and temporally shuffled data for group-raised(top row) and isolation-raised larvae (bottom row). The center panel shows a zoom in section of thedensity distribution. The dot in the center indicates the position of the focal fish facing upwards. Theavoidance area was quantified as the area where the data point density distribution is below 60% ofthe shuffled data. Shown on the left are the contours of the avoidance area calculated for each replicateof groups of seven larvae overlaid per raising condition. (C) Box plots depict the 25-75-percentile ofavoidance area measures comparing the effect of the raising condition. Whiskers signal the range ofdata and + indicates the mean per condition. Sample size is shown in the inset and asterisks signal theresults of a two-tailed unsigned Wilcoxon test between raising conditions (alpha level 0.05).

Because the density distribution for pairs of larvae was more sparse, we computed a one di-85

mensional measure of avoidance by subtracting the distance between the two fish in a pair86

(henceforth referred to as the inter-fish distance) from a random distribution (see Methods for87

detail). Thereby we quantified that, also when tested in pairs, ISO fish show a larger avoidance88

distance than GR fish (Fig. S1 C, p = 0.0069, N = 43, see Table S1).89

Locomotion in larval zebrafish is composed of discontinuous bouts of sequential tail de-90

flections, which can be classified in 13 kinematically different swim bout types [Marques et al.,91

2018]. We tracked the tail movements (Fig. S1 D) of freely swimming pairs of larvae and ap-92

plied this bout classification based on 73 kinematic parameters per swim bout, as described93

previously [Marques et al., 2018]. For simplicity, we show here only two of the kinematic pa-94

rameters extracted from the tail movements that define a swim bout (Fig. 2 A). The C1 angle95

measures the change in heading caused by the first tail beat, while the displacement sums96

over the distance swum by the larva when integrating the path of the bout. Based on these97

two parameters, we can distinguish between three major classes of movements. Forward dis-98

placing bout types, such as slow1, slow2, the short capture swim (short CS) and the approach99

swim (AS) have a low C1 angle and variable, but relatively low levels of displacement. Reori-100

enting bout types, such as the J-turn, high angle turn (HAT) and routine turn (RT) also show101

a relatively low displacement, but combined with a larger C1 angle than forward displacing102

bouts. And, lastly, there are six bout types that displace the larvae one body length (∼4.2 mm)103

5


https://doi.org/10.1101/2020.03.02.972612


or more and have variable C1 angles. This last set of bouts comprises the long capture swim104

(long CS), burst swim (BS) , the shadow avoidance turn (SAT) , the O-bend and the long and105

short latency C-starts (LLC and SLC, respectively) .106

To understand if and how all 13 bout types are being used during social interactions, we107

calculated the median inter-fish distance per bout type for each pair of fish. While the majority108

of bout types show a median value similar to the radius of the area, five bout types are shifted109

towards smaller inter-fish distances, suggesting that they are used during close-range social110

interactions. Multiple comparison of raising condition for each bout type revealed that three111

of these close-range bout types (O-bend, LLC and SLC) are performed at increased inter-fish112

distances by ISO larvae (Fig. 2 B, see Table S2). Interestingly, the close-range bout types show113

the largest displacement (Fig. 2 C), suggesting that larvae use those large displacing bouts to114

avoid one another.115

We next compared the probability of using these close-range bouts as a function of inter-116

fish distance (Fig. 2 D). As expected there is an inverse relationship between bout type usage117

and inter-fish distance. Yet, there was a striking difference between the two raising conditions.118

While both show a preference for C-starts over the other close-range bout types, the relative119

choice of C-start bout type differs. We found that GR larvae use more LLCs (p = 0.0003), while120

ISO larvae are more likely to use SLCs (p = 0.0005). The usage of long CS, BS and O-bend121

were not different between raising conditions (see Table S3 for full statistical comparison).122

To test if ISO fish are generally more reactive to a social stimulus, we compared the overall123

number of C-starts (sum over LLC and SLC), normalized to the total number of bouts swum.124

There was no difference between raising conditions (Fig. 2 E, p = 0.28, see Table S1), but125

specifically the proportion of SLCs of all C-starts was increased in ISO larvae (Fig. 2 F, p <126

0.0001, see Table S1).127

Long and short latency C-starts have originally been described in response acoustic startles128

[Burgess and Granato, 2007a]. As evident from their respective tail angle traces, first two tail129

beats occur with a larger acceleration in SLCs than LLCs (Fig. S1 D). Thus, the SLC can be130

considered a high-acceleration version of the C-start. Moreover, as suggested by their names,131

the two C-starts occur at a different latency after a sudden acoustic stimulus. We therefore132

tested if we can also find a difference in latency in the socially induced C-starts. Because133

there is no clearly defined stimulus onset for socially triggered C-starts, we measured the134

time delay between the onset of the C-start of the focal larvae and the last preceding bout135

of the non-focal larva. The frequency distribution of these measures of reaction latency are136

not as clearly separated as expected for acoustic startles. Nonetheless, the median latency per137

larval pair was higher for LLCs than for SLCs in both raising conditions (see Table S4), hence138

matching their attributed names of long and short latency bouts (Fig. 2 G-H). Note that there139

was no significant difference in latency between GR and ISO, suggesting that ISO larvae react140

similarly, but at a larger inter-fish distance (see Fig. 2 B) and with a farther displacing bout141

type.142

In order to confirm that the two C-start types produced by GR and ISO larvae belong143

to equivalent bout types, we mapped the bouts in the original principal component space144

of kinematic parameters that was used for the classification (Fig. S1 E). LLCs and SLCs of145

both raising conditions were overlapping in the principal component space, showing that146

6


https://doi.org/10.1101/2020.03.02.972612


Displacement

C1 angle

-200 200

25

C1 Angle (º)

Dis

plac

em

ent (

mm

)

LLC SLC

AS BS

HAT RT SAT O-bend

Slow 1 Slow 2 Short CS Long CS

J-turn

Group Raised N = 43Isolation Raised N = 43

15

10 *

*

*

*

0

Inte

r-fis

h D

ista

nce

(mm

)

5

Inter-fish Distance (mm)0 2 4 6 8 10 12D

ispl

ace

men

t (m

m)

10

0

5

A

B C Avoidance bout types

GroupRaised

IsolationRaised

0

0.2

0.4

0.6

0.8

1.0P

rop

ortio

n of

SL

C

n .s .

GroupRaised

IsolationRaised

0

0.1

0.4

0.3

0.2

C-s

tart

s pe

r bo

ut s

wum

D E F

0

0.1

0

0.1

5 10 15

Bou

t typ

e pr

oba

bilit

y

Inter-fish Distance (mm)

GroupRaised

IsolationRaised

0

0.1

0.2

0.1

0.2

Pro

bab

ility

0 500 1000Latency (ms)

0 50 100 150C1 Angle (º)

0 10 20 30Max. angular speed (º/ms)

G H I J

GroupRaised

IsolationRaised

LLCSLC

Med

ian

Late

ncy

(ms)

0

100

200

300

GR ISO

n .s .

7


https://doi.org/10.1101/2020.03.02.972612


they were classified correctly and that GR and ISO C-starts do not represent different bout147

types. To validate that the bout types are similar to the originally described acoustic startle148

C-starts, we next compared the C1 angle and the angular speed. Along with the latency, these149

two measures were originally used to differentiate LLCs and SLCs [Burgess and Granato,150

2007a]. For both measures we found a significant effect of C-start bout type, but not of raising151

condition (Fig. 2 I-J, see Table S4). These results indicate that the bout types themselves are152

not different between GR and ISO larvae, but rather only their preferred usage in a social153

context.154

Isolation-raising affects general locomotion155

Next, we tested for general effects of isolation raising on locomotion. For this we compared156

GR and ISO larvae swimming alone or in pairs in the same testing arena as described above.157

The overall distribution of bout type choices in pairs was similar between GR and ISO larvae158

(Fig. S2 A top panel), with the exception of the SAT, BS, O-bend, LLC and SLC (see Table S5).159

However, these differences disappear when testing the larvae alone in the recording arena160

(Fig. S2 A bottom panel, see Table S5). These results suggest that ISO larvae use more large161

displacement bouts specifically in social contexts. Independent of the social testing context,162

ISO larvae swim fewer and longer bouts (Fig. S2 B-E, see Table S6). We observed similar163

Figure 2 (preceding page): Isolation raised larvae perform large displacement bouts at larger distanceand use more SLCs. (A) Overview of the 13 swim bout types of larval zebrafish. C1 angle measuresthe change in heading caused by the first half beat of the tail relative to the axis prior to the bout.Displacement measures the sum over the distance swum by the larva by integrating the path of thebout. Shown are scatter plots of 1500 bouts per bout type category randomly chosen from 43 pairsof group-raised larvae. (B) The median distance between the two larvae per pair of each bout type iscompared between group and isolation-raised pairs. Bout types are color-coded, markers and errorbarsshow mean ± standard deviation, for group raised (circles) and isolation raised larvae (squares). As-terisks signal significant results of Bonferroni-Holm corrected p-values from unsigned Wlicoxon testsbetween raising conditions for each bout type (alpha level 0.05). (C) Displacement per bout type isplotted against the mean inter-fish distance, revealing that large displacing bout types tend to occurat small inter-fish distances. (D) Probability of usage of the close-range bout types (O-bend, BS, longCS, LLC and SLC) is expressed in proportion of each bout type over all 13 bout types. This proportionis plotted for inter-fish distance bins of 2mm. Shown are mean values across 43 pairs of larvae perraising condition. (E) The number of C-starts is calculated as the sum over LLC and SLC, normalizedto the total number of bouts performed by each larval pair. (F) The proportion of SLC of all C-startsas measured per testing pair for each raising condition. (E,F) Light gray open circles show every repli-cate per condition, dark gray filled circles and error-bars signal mean ± standard deviation. Asteriskspecify the results of unsigned Wlicoxon tests between raising conditions. (G) Latency was calculatedas the time difference between the C-start onset and the onset of the previous bout performed by thenon-focal larva. Shown are median latency measures per larval pair for group raised (GR) and isolationraised larvae (ISO), color-coded as red for SLC and yellow for LLC. Asterisks indicate the summarizedresults of unsigned Wilcoxon tests between raising conditions (horizontal) and signed Wilcoxon testsbetween C-start types (vertical), after Bonferroni-Holm correction for multiple comparison; alpha level0.05. (H-J) Histograms of latency, C1 angle and maximum angular speed per C-start type, as pooledover animals per raising conditions; group raised top row; isolation raised bottom row. Red bars showprobability values for SLCs, while yellow bars depict LLC values. AS: Approach Swim; Long CS: Longcapture swim; BS: Burst swim; HAT: High angle turn; RT: Routine turn; SAT: Shadow avoidance turn;LLC: Long latency C-start; SLC: Short latency C-start.

8


https://doi.org/10.1101/2020.03.02.972612


overall locomotion effects when treating GR larvae after exposure to the generic dopamine164

agonist, apomorphin (Fig. S2 F-G, see Table S6). This treatment, however, had no effect on165

the social avoidance measures (Fig. S2 H-I, see Table S7), suggesting that general locomotion166

and social avoidance effects are dissociated. Indeed, we found that isolating larvae only from167

3dpf until testing at 6dpf reduced general locomotion effects (Fig. S2 J-K, see Table S6), while168

maintaining the social avoidance phenotype (Fig. S2 L-M, see Table S7). For subsequent169

experiments we applied this adjusted protocol of social isolation from 3dpf onward.170

Vision and mechanosensation contribute differently to social avoidance reactions171

As an entry point to the mechanism of social avoidance reactions, we analyzed the angle172

of the non-focal larva position upon C-start onset (Fig. 3 A, see Table S8). The angular173

distribution was skewed towards the tail direction of the focal larva, in the blind spot of the174

larval vision. Fish can sense proximate water motion through their mechanosensory lateral175

line (LL), composed of neuromasts distributed laterally along the body axis and around the176

head [Bleckmann and Zelick, 2009]. We next aimed at testing the relative contribution of177

vision and LL sensing to social avoidance reactions and the isolation phenotyope. First, we178

tested larval pairs in darkness, where, as expected, the proportion of C-starts with the non-179

focal fish positioned behind the tail increased for both GR and ISO larvae (Fig. 3 D, see Table180

S8). The avoidance area was reduced when testing in darkness and compared to light, in181

line with previous reports [Marques et al., 2018]. Furthermore, we no longer observed an182

increased avoidance area in ISO larvae (Fig. 3 E, p = 0.235), while the difference in SLC183

proportion between raising conditions persisted (Fig. 3 F, p < 0.0001; see Table S9). These184

results suggest that LL mechanosensation contributes to the choice of C-start bout type. We185

therefore next ablated the LL using neomycin, an ototoxic agent that was shown to ablate the186

hair cells of the LL, but not the inner ear [Buck et al., 2012]. In line with reduced LL sensing,187

the angular distribution of non-focal fish position upon C-start onsets was no longer skewed188

towards the tail direction after neomycin treatment (Fig. 3 G, see Table S8). The avoidance189

area was of similar magnitude as in control fish, but the difference between raising conditions190

was diminished (Fig. 3 H, p = 0.281). Also the difference in SLC proportion between GR191

and ISO was reduced in magnitude, albeit still present after neomycin treatment (Fig. 3 I,192

p = 0.003, see Table S9). To validate these findings, we used a second ablation reagent,193

CuSO4, which yielded similar results (Fig. S3 A-C), while a control incubation with regular194

fish medium maintained the observed social avoidance phenotype of ISO larvae (Fig. S3 D-F).195

To summarize these results, we calculated the effect size of the difference between GR and196

ISO testing animals per treatment (Fig. 3 J). We hypothesise that the escape distance during197

social avoidance is informed by input from the visual and the LL system, while the choice of198

C-start escape bout type only depends on LL mechanosenation (Fig. 3 K).199

Controlled water vibrations reproduce social avoidance reactions200

To further decipher the contribution of vision and LL sensing to social avoidance reactions201

and the ISO avoidance phenotype, we next tested behavioural responses to controlled, single202

sensory-modality stimuli presented in closed-loop to larvae swimming alone in the testing203

9


https://doi.org/10.1101/2020.03.02.972612


0.20.2

A B

D

GroupRaised

IsolationRaised

Vis

ion

and

LL

Late

ral l

ine

onl

yV

isio

n on

ly

C

0.20.2

E F

G H I

0.20.2

0

0.5

1

1.5

2

2.5

0

0.2

0.4

0.6

0.8

1

Pro

port

ion

of S

LC

GroupRaised

IsolationRaised

GroupRaised

IsolationRaised

0

0.5

1

1.5

2

2.5

Avo

ida

nce

are

a (

a.u

.)n.s.

0

0.2

0.4

0.6

0.8

1

Pro

po

rtio

n o

f S

LC

Avo

ida

nce

are

a (

a.u

.)

GroupRaised

IsolationRaised

GroupDark

IsolationDark

GroupDark

IsolationDark

GroupRaised

IsolationRaised

0

0.5

1

1.5

2

2.5

Avo

ida

nce

are

a (

a.u

.)

n.s.

0

0.2

0.4

0.6

0.8

1

Pro

po

rtio

n o

f S

LC

tested in darkness

after LL ablation

GroupNeom.

IsolationNeom.

GroupNeom.

IsolationNeom.

Non-focal fish position upon C-start onset

Contr.Dark

Neom.

CuSO4

0

0.2

0.4

0

0.1

0.2

0.3

Contr.Dark

Neom.

CuSO4

Me

an D

iff.(

Gr

- Is

o)

Avoidance area SLC proportionJEscape bout type choice

Escape distance

Lateral line

Lateral line & Vision

K

tested in light

10


https://doi.org/10.1101/2020.03.02.972612


arena. First, we presented radially expanding dark shadows, which have previously been204

shown to produce escape responses [Temizer et al., 2015, Dunn et al., 2016, Bhattacharyya205

et al., 2017, Marques et al., 2018]. Such stimuli were effective at triggering O-bend and SAT206

bout type, but not C-starts in GR (Fig. 4 A), as well as ISO fish (Fig. S4 A). The response207

probability of these two bout types was similar for GR and ISO fish (Fig. 4 B, p = 0.43)208

and there was no difference in the relative bout choice between raising conditions (Fig. 4 C,209

p = 0.9; see Table S10). We then tested a dark spot of constant size approaching the larva at210

a steady speed of 0.5 cm/sec. Unlike the expanding stimuli, this approaching dot protocol211

proved to be highly effective in triggering SLCs in GR (Fig. 4 D) and ISO fish (Fig. S4 B). We212

found that ISO fish have a lower overall response rate of C-starts with the approaching dot213

(Fig. 4 E, p = 0.0003), but that when they respond they do so with a similar proportion of SLC214

bout types as GR fish (Fig. 4 F, p = 0.24). Furthermore, when comparing the C-start latency,215

equivalent to a measure of reaction distance from the dot, GR and ISO fish were similar (Fig.216

S4 D, p = 0.77 and p = 0.78 for SLC and LLC, see Table S11). These results are in line with217

our hypothesis that a visual stimulus alone is insufficient to produce the avoidance distance218

phenotype and does not account for the different C-start choice.219

To test the contribution of the LL to social avoidance reactions, we designed a stimulus to220

mimic the water disturbances caused by a swimming larva. While sudden acoustic stimuli221

are effective at triggering C-start responses, they lack a directional component and are quite222

unlike a social stimulus. De Marco et al. [2013] showed that local water vibrations can act223

as a stressor for larval zebrafish and enhance their level of cortisol in an intensity-dependent224

manner. To cause these vibrations, a rigid capillary was attached to a piezo bender and225

semi-submerged into the water of the swimming arena. We replicated a similar setup and226

programmed the triggering of the piezo displacements in closed-loop with the swimming227

pattern of the larva. In order to mimic the conditions during social avoidance reactions, piezo228

displacements were triggered when the larva was facing away from the stimulus source at a229

distance of 5mm. To avoid triggering the stimulus as the larva is performing a bout, we added230

a 400ms delay between the crossing of the distance threshold and the onset of the piezo. This231

stimulus protocol was effective as triggering both types of C-starts (Fig. 4 G; Fig. S4 C) in232

GR and ISO larvae. While both raising conditions were comparable in their overall response233

Figure 3 (preceding page): Contribution of vision and lateral line senses to enhanced social avoidancereactions in isolation-raised larvae. (A-C) Pairs of larvae raised in groups or in isolation were testedwhile swimming freely with homogeneous background illumination; NGR = 31; NISO = 39. (D-F)Larval pairs were tested in complete darkness; NGR = 27; NISO = 28. (G-I) Larvae were incubatedfor 1h in neomycin and given a 2 hour recovery period prior to testing in pairs with homogeneousbackground illumination; NGR = 19; NISO = 18. (A,D,E) Circular histograms of the position of thenon-focal larva upon the onset of the focal C-start. Data is pooled over fish per raising condition andtreatment. Red and yellow bars show data for SLC and LLC, respectively. (B,E,H) Social avoidancearea as calculated from the distribution of distances between the two fish throughout the recordingperiod. (C,F,I) The proportion of SLC bouts over the sum of SLC and LLC. Light gray open circlesshow each replicate per condition, dark gray filled circles and error-bars signal mean ± standarddeviation. Asterisks specify the results of unsigned Wilcoxon tests between raising conditions, alphalevel 0.05. (J) Effect sizes of the difference between group and isolation raised larvae, calculated as themean difference calculated with a bootstrapping method. (K) Summary of the proposed contribution ofvision and lateral line mechanosensation to the social avoidance phenotype of isolation-raised larvae.

11


https://doi.org/10.1101/2020.03.02.972612


probability (Fig. 4 H, p = 0.25), ISO larvae showed a higher probability of using SLCs (Fig.234

4 I, p = 0.0016, see Table S10), matching our findings during social interactions. Notably, the235

latency profile of LLC and SLC upon piezo stimuli remains consistent with the naming of the236

two C-starts, although they are not as clearly separated as with acoustic startle stimuli (Fig.237

S4 E-F). This resembles our findings of LLC and SLC latency during social interactions.238

To validate that the piezo-induced water motions are indeed a LL stimulus we measured239

the responses of GR larvae after neomycin treatment to ablate the LL hair cells. The overall240

response probability (p < 0.0001), as well as the proportion of SLC responses (p = 0.0002;241

Fig. S4 G-I, see Table S10) were significantly reduced although not entirely abolished with the242

ablation. These results show that LL sensing is the main, albeit not sole source of processing243

for water motion stimuli. Overall, we here showed that LL dependent stimuli, composed244

to local water vibrations, recapitulate the phenotype of social isolation raising during social245

avoidance reactions. Hence, indicating that LL sensing is sufficient to drive the differences246

between GR and ISO.247

Discussion248

Social experiences during early life greatly impact the development of social and non-social249

behaviors [Harlow et al., 1965]. To date, it is not fully understood what the underlying mech-250

anisms are by which social experiences shape appropriate behavior and development. Here251

we set up to test if social isolation effects can be studied in larval zebrafish whose social252

experiences can be precisely controlled and whose social behavior can be studied as discrete,253

quantifiable events. We found that isolation raising enhances LL mediated avoidance reactions254

to another fish, along with general effects on locomotion.255

The effects of social isolation on social interactions can already be observed in256

one-week-old larvae257

Previous studies have reported that social experience is not necessary for the development of258

social attraction in zebrafish [Breder and Halpern, 1946, Larsch and Baier, 2018]. However,259

social isolation led to reduced group cohesion and a larger average distance between groups260

of freely swimming adult zebrafish [Shams et al., 2018]. In line with this, we found that261

the avoidance distance is larger in larval zebrafish after isolation raising. To the best of our262

knowledge, we here showed for the first time that the effects of social isolation can already be263

observed during social interactions in one-week-old zebrafish larvae when social attraction is264

yet to be developed. Together these results suggest that social attraction develops without the265

need for social experience, while fine-tuning locomotion for regular collective motion requires266

the presence of conspecifics during development.267

Isolation affects general locomotion268

We also observed that ISO larvae have general, albeit small, deficits in locomotion. They pro-269

duce longer and fewer bouts than GR larvae, irrespective of the social context during testing.270

Hypoactivity in isolation raised zebrafish larvae has previously been reported upon exposure271

12


https://doi.org/10.1101/2020.03.02.972612


0 2 40

0.2

0.4

0.6B

ou

t ty

pe

Pb

0

0.2

0.4

0.6

0.8

1

0

0.2

0.4

0.6

0.8

1

n.s.

GroupRaised

IsolationRaised

n.s.

GroupRaised

IsolationRaised

Pro

por

tion

of O

-ben

d

Rea

ctio

n pr

oba

bilit

y

Chasing dot stimulus

A B C

D E F

0

0.2

0.4

0.6

0.8

1

0

0.2

0.4

0.6

0.8

1

GroupRaised

IsolationRaised

GroupRaised

IsolationRaised

n.s.

Pro

por

tion

of S

LC

Rea

ctio

n pr

oba

bilit

y

LLCSLC

SATO-bend

G H I

0

0.2

0.4

0.6

0.8

1

n.s.

0

0.2

0.4

0.6

0.8

1

GroupRaised

IsolationRaised

GroupRaised

IsolationRaised

Pro

por

tion

of S

LC

Rea

ctio

n pr

oba

bilit

y

Piezo stimulus

GroupRaised

IsolationRaised

Semi-submergedcapillary attached to piezo

5 mm

0 2 4

Time (sec)

0

0.2

0.4

0.6

Bo

ut ty

pe

Pb

Looming stimulus

13


https://doi.org/10.1101/2020.03.02.972612


to a sudden dark period during the larva’s day time [Zellner et al., 2011, Steenbergen et al.,272

2011]. While our findings are in line with these results, there is a noteworthy difference. We273

here tested spontaneous locomotion with continuous background illumination, while in the274

referred publications larvae were dark-adapted and then tested in alternating light patterns.275

Exposing a dark-adapted larva to sudden changes in illumination leads to increased levels276

of the stress hormone, cortisol [De Marco et al., 2013]. Therefore protocols using light dark277

alternation may test specifically for locomotion under acute stress. A previous study has re-278

ported increased locomotor activity in juvenile, but not larval zebrafish after isolation raising279

[Shams et al., 2018]. We have observed a similar effect in a pilot experiment, when raising280

ISO fish to three weeks of age (data not shown). However, we also noted that ISO fish were281

growing faster than GR fish once they were fed, thus the change in direction of the isolation282

effects on locomotion with age, may result from an interaction with feeding. Nonetheless,283

general effects of social isolation on locomotor activity have been observed in several species284

and can vary in sign and magnitude (e.g. Wilkinson et al. [1994] and Archer [1969]). This285

further strengthens the point that when studying the impact of social isolation on social be-286

havior, it is essential to thoroughly distinguish general locomotion and social-specific effects.287

Figure 4 (preceding page): Avoidance bout type responses upon controlled closed-loop stimuli. (A)A radially expanding dark spot (1 cm/s) was projected 4 cm away from the larva. The outer edgeof the stimulus reaches the larva’s center of mass after 4 sec (shown as inset). As the dark spotexpands, the likelihood of performing O-bend and SAT bout types increases, but not SLC and LLC.(B) The probability of performing an O-bend or SAT during the visual stimulation period per stimuluspresentation was calculated for each larva tested for group and isolation raised larvae (NGR = 16,NISO = 17, p = 0.3209). (C) The proportion of O-bend responses calculated as the number of O-bendsperformed during looming stimulus presentation divided by the total number of stimulus-inducedO-bends and SATs (NGR = 16, NISO = 17, p = 0.3678). (D) A non-expansive dark spot (0.5 to 2.5 mmin diameter) was projected at 2 cm away from the test larva and approached at a speed of 0.5 cm/s.Hence, the outer edge of the stimulus reaches the center of mass of the larva after 4 sec (shown asinset). Such stimulus increases the probability of performing SLCs. (E) Reaction probability is definedas the number of SLC or LLC performed per total number of chasing dot stimulus presentations perlarvae or either raising condition (NGR = 14, NISO = 15, p = 0.0022). (F) The proportion of SLCswas calculated as the number of stimulus induced SLCs divided by the sum over stimulus-inducedSLCs and LLCs (NGR = 14, NISO = 15, p = 0.2384). (A,D) Shown are the overlaid image sequencesof an O-bend (A) and an SLC (D) performed by a group raised larvae, acquired at 90 frames persecond. The bout type probability was calculated as the number of bouts of a given bout categoryperformed within a 500ms time window divided by the total number of bouts performed. (G) Localwater vibrations per elicited by submerging the tip of a rod into the swimming area. The rod wasattached to a piezo bender which would trigger a pulse train of five 2ms pulses with an input voltageof 40V at a minimum inter-stimulus interval of 2 min. Stimuli were triggered when the larva was facingaway from the source at a distance of 5mm from the tip of the rod. Shown are the trajectories of thefirst bout after piezo stimulus presentation for five randomly chosen group raised (left) and isolationraised larvae (right). (H) Reaction probability is defined as the number of SLC or LLC performed pertotal number of chasing dot stimulus presentations per larvae or either raising condition (NGR = 15,NISO = 15, p = 0.7714). (I) The proportion of SLCs was calculated as the number of stimulus inducedSLCs divided by the sum over stimulus-induced SLCs and LLCs (NGR = 15, NISO = 15, p = 0.0017).(B,C,E,F,H,I) Light gray open circles show every replicate per condition, dark gray filled circles anderror-bars signal mean ± standard deviation. Asterisks specify the results of unsigned Wilcoxon testsbetween raising conditions, alpha level 0.05. SAT: Shadow avoidance turn; LLC: Long latency C-start;SLC: Short latency C-start.

14


https://doi.org/10.1101/2020.03.02.972612


We here designed our experiments to parse general effects on spontaneous locomotion apart288

from social effects of isolation raising, by comparing locomotion parameters across differ-289

ent social testing contexts and isolation time windows, as well as manipulating locomotion290

pharmacologically. Therefore, two lines of evidence in our data support the hypothesis that291

general and social locomotor effects are distinct. First, the increased bout duration and social292

avoidance phenotype have dissimilar critical developmental windows. Isolating larvae at the293

third day of development abolishes the general increase of bout duration, but does not de-294

tract the behavioral phenotype during social avoidance. Secondly, treatment with the generic295

dopamine agonist, apomorphin, recapitulates the general locomotion of ISO fish in GR fish296

without affecting the social avoidance measures.297

Isolated fish show enhanced use of SLCs in social context298

An advantage of using one-week-old zebrafish is that the repertoire of their swims has been299

described and can be clustered into movement types through automatic, unsupervised meth-300

ods [Marques et al., 2018, Johnson et al., 2020, Mearns et al., 2020], enabling exact phenotyping301

of behavioral events. Using this technology, we found that larvae avoid each other by execut-302

ing four types of large displacement bouts: burst swims [Severi et al., 2014], long capture303

swims [Marques et al., 2018], O-bends [Burgess and Granato, 2007b], and C-starts (LLCs and304

SLCs) [Burgess and Granato, 2007a]. Strikingly, although the use of C-starts as a whole is305

unaltered, ISO fish use SLCs twice as frequently as LLCs when compared to GR fish. SLCs306

are a high acceleration version of the C-start and lead to a larger displacement. Therefore,307

the imbalance in usage of the two C-start types could be a major contributor to the enhanced308

avoidance area seen in ISO larvae. Given the larger displacement, SLCs increase the inter-309

fish distance faster than LLCs. In line with this argument, after ablation of the lateral line, a310

manipulation that balances the use of C-starts, the avoidance area difference between raising311

conditions disappears.312

Larval social avoidance is multimodal and isolation impacts lateral line mediated313

C-starts314

The social stimuli during collective swimming are multimodal and fish rely mostly on vision315

and the LL. Only upon removal of both of these, fish fail to swim in a coordinated fashion316

[Pitcher et al., 1976, Partridge, 1982]. It has been proposed that the contribution of these317

two sensory systems to schooling is distinct; vision is thought to be primarily important318

for maintaining position and angle between neighboring fish, while the LL contributes to319

monitoring swim speed and direction of swimming neighbors [Partridge and Pitcher, 1980].320

Presumably due to do the technical difficulty of experimentally mimicking water wakes of a321

swimming fish, to date only visual stimuli have been shown to be sufficient as a sole motion322

cue to induce following behavior in adult zebrafish [Lemasson et al., 2018], as well as juvenile323

zebrafish [Larsch and Baier, 2018]. The social stimuli that trigger social avoidance in larval324

zebrafish are also multimodal and previous work has shown that vision plays a critical role.325

First, the avoidance area is decreased when fish are interacting in the dark [Marques et al.,326

2018], suggesting that vision helps larvae avoid each other at larger distances. Secondly,327

15


https://doi.org/10.1101/2020.03.02.972612


through visual inputs larvae and juvenile fish synchronize their movements with other fish328

[Dreosti et al., 2015, Marques et al., 2018, Stednitz and Washbourne, 2020].329

To parse apart the contribution of vision and LL sensing, we tested larvae with manipu-330

lations that selectively blocked each modality. While testing social avoidance in darkness has331

a strong effect on the avoidance distance, it has no impact on the C-start choice, suggesting332

that vision defines the avoidance distance, but not the type of swim bout used. The avoidance333

distance in darkness is similarly low for GR and ISO fish, pointing towards mixed effects of334

isolation raising and the availability of visual cues. The LL, on the contrary, seems to con-335

tribute specifically to the effect of isolation on the balance of the two C-start types. In line336

with this, we did not observe enhanced reactions to visual stimuli. To probe sufficiency of337

mechanosensory stimuli, we designed a social-like LL stimulus inspired by previous work338

that used a piezo bender to produce local water vibrations [De Marco et al., 2013, Groneberg339

et al., 2015]. The stimulus parameters here were chosen to mimic the conditions we observed340

during social escape reactions in freely interacting fish. To ensure that all larvae receive com-341

parable sensory input, we triggered the stimuli in closed-loop with the behavior of the testing342

larva. Thereby we found that ISO fish also showed an enhanced proportion of SLC usage. This343

assay was performed in darkness, but we additionally confirmed through neomycin treatment344

that it requires the LL. Hence, LL stimuli are sufficient to trigger enhanced social avoidance345

reactions in ISO larvae.346

Notably, it has previously been reported that during the acoustic startle assay, larvae raised347

at a lower density are more likely to perform SLCs reactions [Burgess and Granato, 2008].348

Generally, this report is in accordance with the results shown here. However, we believe that349

the social isolation effects during social avoidance represent a different scenario. Through350

the ablation experiments we learned that the LL sensing is necessary for the increased SLC351

proportion in ISO larvae. Acoustic startles mainly depend on the inner ear and not the LL352

[Lacoste et al., 2015] and are strong, non-localized stimuli, unlike the water disturbances353

caused by another larva swimming in proximity. Accordingly, we also observed differences354

in the latency profile of the C-starts between acoustic and piezo stimuli. This suggests that355

although the bout types are kinematically the same, the responses to minute water vibrations,356

artificial or socially-borne, are differently triggered from the highly stereotyped acoustic startle357

reactions.358

Neural control of startle reactions and the social isolation phenotype359

The neural circuits underlying escape reactions in larval zebrafish have been studied for over360

two decades [Eaton et al., 1991, Liu and Fetcho, 1999, O’Malley et al., 1996, Kohashi and Oda,361

2008, Burgess and Granato, 2007a]. Most of the studies have used acoustic or tactile stimuli362

upon which the neural populations controlling SLCs and LLCs appear to be distinct. The363

prominent Mauthner escape neurons (M-cells), a pair of large cells in the dorsal hindbrain364

segment (rhombomere) 4 [Kimmel et al., 1981], have been studied in great detail and were365

shown to be necessary for SLCs but not LLCs [Burgess and Granato, 2007a, Takahashi et al.,366

2017]. More recently, a study identified a cluster of prepontine neurons, in hindbrain rhom-367

bomere 1, as necessary and sufficient to drive LLCs [Marquart et al., 2019]. In line with their368

delayed response outputs, these neurons neither project directly to the spinal cord, nor receive369

16


https://doi.org/10.1101/2020.03.02.972612


direct auditory input. Moreover, prepontine neurons are only active when the larva performs370

an ipsilateral LLC, but not when it performs an SLC, suggesting that the circuits driving the371

two C-start types are active in a mutually exclusive manner [Marquart et al., 2019].372

M-cell excitability is regulated through a network of excitatory and inhibitory feedback373

and feed-forward interneurons [Takahashi et al., 2002, Satou et al., 2009, Koyama et al., 2011,374

Lacoste et al., 2015, Shimazaki et al., 2019]. Behavioral switching from SLC to LLC during con-375

ditioning or habituation is associated with reduced excitability of the M-cell [Takahashi et al.,376

2017, Marsden and Granato, 2015]. Furthermore, forward genetic screens have shown that377

mutants of altered SLC-LLC reaction balance are associated with altered M-cell excitability378

[Marsden et al., 2018, Jain et al., 2018].379

Therefore, one explanation of why ISO larvae perform more SLCs during social interac-380

tions, is an increased M-cell excitability. However, in this case, one would expect to also see381

enhanced responses to other stimuli that trigger M-cell dependent escapes, such as the ap-382

proaching dark spot [Dunn et al., 2016] and acoustic startles [Burgess and Granato, 2007a].383

An alternative mechanism would be that the LL in ISO larvae is more sensitive to the water384

motions caused by a conspecific swimming in proximity, as these represent a novel stimu-385

lus. Although it has been shown that the connections between afferent neurons and hair-cell386

bundles of the LL neuromast develop independently of experience in larval zebrafish [Nagiel387

et al., 2009], this sensory organ shows several filtering properties whose development might388

depend on experience. For instance, efferent copies of forward motion suppress synaptic389

transmission in lateral line hair cells [Pichler and Lagnado, 2020], and canal neuromasts are390

capable of sensing water vibrations even against a background of unidirectional water flow391

[Engelmann et al., 2002]. Such fine-tuning of LL sensing under specialized conditions could392

be shaped through the sensory experiences of the presence of conspecifics.393

In support of this latter hypothesis, it has previously shown that a hypoactivity response394

to subtle, minute water vibrations was more sensitive in larvae that were raised in isolation395

[Groneberg et al., 2015]. Notably, in this behavioral assay the larvae did not perform an396

escape reaction to the stimulus presentation and therefore an enhanced response indicates397

higher sensory sensitivity rather than reduced startle thresholds.398

To fully probe these two possible mechanisms of enhanced social avoidance reactions in399

ISO larvae, future experiments will need to measure neural activity in the LL ganglion cells,400

as well as the M-cells upon controlled stimuli of graded intensity, along with social stimuli in401

larvae where the LL is free and intact.402

Conclusion403

Understanding how the brain implements adequate, fine-tuned social behaviors according to404

social experiences, requires precise experimental control of such experiences along with de-405

tailed behavioral analysis. Another critical choice is a model that will allow studying etholog-406

ically relevant social interactions, along with the capacity to measure and manipulate activity407

in neural circuits. Here we showed that a relatively simple social behavior, the avoidance408

reactions to a minute, local social-like stimuli, is shaped by experience and therefore more409

complex than previously thought. Moreover, we show that the effects of early life social iso-410

lation can be distinguished between general locomotor effects and social behavior, at a young411

17


https://doi.org/10.1101/2020.03.02.972612


age in a model which allows whole-brain neural activity measure and manipulation.412

Author contributions413

AHG, MBO and GGdP conceived the project. AHG and JCM designed experiments. JCM414

and MBO implemented the recording apparatus and AHG performed all experiments. ALM415

programmed the piezo stimulus. AHG and JCM analyzed data and all authors interpreted416

the results. AHG and JCM wrote the manuscript with contributions from all authors. MBO417

and GGdP supervised the project.418

Acknowledgements419

We thank Alexandre Laborde for assistance in setting up the experiments, and Mattia Bergomi420

and Sabine Renninger for discussions on the project. We also thank the Champalimaud Fish421

Facility team for excellent fish care, and Paulo Carrico and the Champalimaud Hardware422

platform for logistic support.423

Funding424

This work was realised through funding from an ERC Consolidator Grant (Neurofish) to MBO425

and from the Fundacao para a Ciencia e Tecnologia (Portugal) to GGdP (project PTDC/ NEU-426

SCC/0948/2014). AHG was founded by a PhD scholarship (PD/BD/52444/2013) granted by427

the Fundacao para a Ciencia e Tecnologia (Portugal). This work was developed with support428

from the research infrastructure CONGENTO, co-financed by Lisboa Regional Operational429

Programme (Lisboa2020), under the PORTUGAL 2020 Partnership Agreement, through the430

European Regional Development Fund (ERDF) and Fundacao para a Ciencia e Tecnologia431

(Portugal) under the project LISBOA-01-0145-FEDER-022170.432

18


https://doi.org/10.1101/2020.03.02.972612


References433

S. L. Andersen. Trajectories of brain development: Point of vulnerability or window of op-434

portunity? In Neurosci. Biobehav. Rev., volume 27, pages 3–18. Elsevier Ltd, 2003. doi:435

10.1016/S0149-7634(03)00005-8.436

D. J. Anderson and P. Perona. Toward a science of computational ethology, 2014. ISSN437

10974199.438

J. Archer. Contrasting effects of group housing and isolation on subsequent open field explo-439

ration in laboratory rats. Psychon. Sci., 1969. ISSN 00333131. doi: 10.3758/BF03332812.440

S. Arganda, A. Perez-Escudero, and G. G. De Polavieja. A common rule for decision making441

in animal collectives across species. Proc. Natl. Acad. Sci. U. S. A., 2012. ISSN 00278424. doi:442

10.1073/pnas.1210664109.443

K. Asakawa and K. Kawakami. Targeted gene expression by the Gal4-UAS system in zebrafish,444

2008. ISSN 00121592.445

C. Ballen, R. Shine, and M. Olsson. Effects of early social isolation on the behaviour and446

performance of juvenile lizards, chamaeleo calyptratus. Anim. Behav., 88:1–6, feb 2014. ISSN447

00033472. doi: 10.1016/j.anbehav.2013.11.010.448

P. Berens. CircStat : A MATLAB Toolbox for Circular Statistics . J. Stat. Softw., 2009. ISSN449

1548-7660. doi: 10.18637/jss.v031.i10.450

K. Bhattacharyya, D. L. McLean, and M. A. MacIver. Visual Threat Assessment and Retic-451

ulospinal Encoding of Calibrated Responses in Larval Zebrafish. Curr. Biol., 2017. ISSN452

09609822. doi: 10.1016/j.cub.2017.08.012.453

H. Bleckmann and R. Zelick. Lateral line system of fish. Integr. Zool., 4(1):13–25, mar 2009.454

ISSN 1749-4877. doi: 10.1111/j.1749-4877.2008.00131.x.455

C. M. Breder and F. Halpern. Innate and Acquired Behavior Affecting the Aggregation of456

Fishes. Physiol. Zool., 19(2):154–190, apr 1946. ISSN 0031-935X. doi: 10.1086/physzool.19.2.457

30151891.458

L. M. Buck, M. J. Winter, W. S. Redfern, and T. T. Whitfield. Ototoxin-induced cellular damage459

in neuromasts disrupts lateral line function in larval zebrafish. Hear. Res., 284(1-2):67–81,460

feb 2012. ISSN 0378-5955. doi: 10.1016/J.HEARES.2011.12.001.461

S. A. Budick and D. M. O’Malley. Locomotor repertoire of the larval zebrafish: Swimming,462

turning and prey capture. J. Exp. Biol., 2000. ISSN 00220949.463

H. A. Burgess and M. Granato. Sensorimotor gating in larval zebrafish. J. Neurosci., 27(18):464

4984–4994, 2007a. ISSN 1529-2401. doi: 10.1523/JNEUROSCI.0615-07.2007.465

H. A. Burgess and M. Granato. Modulation of locomotor activity in larval zebrafish during466

light adaptation. J. Exp. Biol., 210(Pt 14):2526–39, jul 2007b. ISSN 0022-0949. doi: 10.1242/467

jeb.003939.468

19


https://doi.org/10.1101/2020.03.02.972612


H. A. Burgess and M. Granato. The neurogenetic frontier–lessons from misbehaving zebrafish.469

Briefings Funct. Genomics Proteomics, 7(6):474–482, oct 2008. ISSN 1473-9550. doi: 10.1093/470

bfgp/eln039.471

P. Chen and W. Hong. Neural Circuit Mechanisms of Social Behavior, 2018. ISSN 10974199.472

L. Cong, Z. Wang, Y. Chai, W. Hang, C. Shang, W. Yang, L. Bai, J. Du, K. Wang, and Q. Wen.473

Rapid whole brain imaging of neural activity in freely behaving larval zebrafish (Danio474

rerio). Elife, 2017. doi: 10.7554/elife.28158.475

I. D. Couzin, J. Krause, R. James, G. D. Ruxton, and N. R. Franks. Collective Memory and476

Spatial Sorting in Animal Groups. J. Theor. Biol., 218(1):1–11, sep 2002. doi: 10.1006/JTBI.477

2002.3065.478

R. J. De Marco, A. H. Groneberg, C.-M. Yeh, L. A. Castillo Ramırez, and S. Ryu. Optogenetic479

elevation of endogenous glucocorticoid level in larval zebrafish. Front. Neural Circuits, 7480

(APR 2013), 2013. ISSN 1662-5110. doi: 10.3389/fncir.2013.00082.481

A. D. Douglass, S. Kraves, K. Deisseroth, A. F. Schier, and F. Engert. Escape Behavior Elicited482

by Single, Channelrhodopsin-2-Evoked Spikes in Zebrafish Somatosensory Neurons. Curr.483

Biol., 2008. ISSN 09609822. doi: 10.1016/j.cub.2008.06.077.484

E. Dreosti, G. Lopes, A. R. Kampff, and S. W. Wilson. Development of social behavior in485

young zebrafish. Front. Neural Circuits, 9:39, aug 2015. ISSN 1662-5110. doi: 10.3389/fncir.486

2015.00039.487

T. W. Dunn, C. Gebhardt, E. A. Naumann, C. Riegler, M. B. Ahrens, F. Engert, and F. Del488

Bene. Neural Circuits Underlying Visually Evoked Escapes in Larval Zebrafish. Neuron, 89489

(3):613–628, 2016. ISSN 10974199. doi: 10.1016/j.neuron.2015.12.021.490

R. C. Eaton, R. DiDomenico, and J. Nissanov. Role of the Mauthner cell in sensorimotor491

integration by the brain stem escape network., 1991. ISSN 00068977.492

J. Engelmann, W. Hanke, and H. Bleckmann. Lateral line reception in still- and running493

water. J. Comp. Physiol. A Neuroethol. Sensory, Neural, Behav. Physiol., 2002. ISSN 03407594.494

doi: 10.1007/s00359-002-0326-6.495

R. E. Engeszer, M. J. Ryan, and D. M. Parichy. Learned social preference in zebrafish. Curr.496

Biol., 14(10):881–884, 2004. doi: 10.1016/j.cub.2004.04.042.497

R. E. Engeszer, L. B. Patterson, A. A. Rao, and D. M. Parichy. Zebrafish in The Wild: A Review498

of Natural History And New Notes from The Field. Zebrafish, 4(1):21–40, mar 2007. ISSN499

1545-8547. doi: 10.1089/zeb.2006.9997.500

C. E. Feierstein, R. Portugues, and M. B. Orger. Seeing the whole picture: A comprehensive501

imaging approach to functional mapping of circuits in behaving zebrafish, 2015. ISSN502

18737544.503

K. Fero, T. Yokogawa, and H. A. Burgess. Zebrafish Models in Neurobehavioral Research, vol-504

ume 52. 2011. ISBN 978-1-60761-921-5. doi: 10.1007/978-1-60761-922-2.505

20


https://doi.org/10.1101/2020.03.02.972612


R. W. Friedrich, G. A. Jacobson, and P. Zhu. Circuit Neuroscience in Zebrafish. Curr. Biol., 20506

(8):R371–R381, apr 2010. ISSN 0960-9822. doi: 10.1016/J.CUB.2010.02.039.507

A. H. Groneberg, U. Herget, S. Ryu, and R. J. De Marco. Positive taxis and sustained respon-508

siveness to water motions in larval zebrafish. Front. Neural Circuits, 9(MAR), 2015. ISSN509

1662-5110. doi: 10.3389/fncir.2015.00009.510

H. F. Harlow, R. O. Dodsworth, and M. K. Harlow. Total social isolation in monkeys. Proc.511

Natl. Acad. Sci. U. S. A., 54(1):90–7, jul 1965. ISSN 0027-8424. doi: 10.1073/PNAS.54.1.90.512

C. Heidbreder, I. Weiss, A. Domeney, C. Pryce, J. Homberg, G. Hedou, J. Feldon, M. Moran,513

and P. Nelson. Behavioral, neurochemical and endocrinological characterization of the early514

social isolation syndrome. Neuroscience, 100(4):749–768, 2000. doi: 10.1016/S0306-4522(00)515

00336-5.516

H. Hentschke and M. C. Stuttgen. Computation of measures of effect size for neuroscience517

data sets. Eur. J. Neurosci., 34:1887–1894, 2011. doi: 10.1111/j.1460-9568.2011.07902.x.518

S. Hesse, J. M. Anaya-Rojas, J. G. Frommen, and T. Thunken. Social deprivation affects coop-519

erative predator inspection in a cichlid fish. R. Soc. Open Sci., 2(3), mar 2015. ISSN 20545703.520

doi: 10.1098/rsos.140451.521

C. Hinz, S. Kobbenbring, S. Kress, L. Sigman, A. Muller, and G. Gerlach. Kin recognition in522

zebrafish, Danio rerio, is based on imprinting on olfactory and visual stimuli. Anim. Behav.,523

85(5):925–930, 2013. ISSN 00033472. doi: 10.1016/j.anbehav.2013.02.010.524

R. C. Hinz and G. G. de Polavieja. Ontogeny of collective behavior reveals a simple attraction525

rule. Proc. Natl. Acad. Sci., 114(9):2295–2300, feb 2017. ISSN 0027-8424. doi: 10.1073/pnas.526

1616926114.527

Y. Inada and K. Kawachi. Order and Flexibility in the Motion of Fish Schools. J. Theor. Biol.,528

214(3):371–387, feb 2002. doi: 10.1006/JTBI.2001.2449.529

R. A. Jain, M. A. Wolman, K. C. Marsden, J. C. Nelson, H. Shoenhard, F. A. Echeverry, C. Szi,530

H. Bell, J. Skinner, E. N. Cobbs, K. Sawada, A. D. Zamora, A. E. Pereda, and M. Granato.531

A Forward Genetic Screen in Zebrafish Identifies the G-Protein-Coupled Receptor CaSR532

as a Modulator of Sensorimotor Decision Making. Curr. Biol., 2018. ISSN 09609822. doi:533

10.1016/j.cub.2018.03.025.534

R. E. Johnson, S. Linderman, T. Panier, C. L. Wee, E. Song, K. J. Herrera, A. Miller, and535

F. Engert. Probabilistic Models of Larval Zebrafish Behavior Reveal Structure on Many536

Scales. Curr. Biol., 2020. ISSN 09609822. doi: 10.1016/j.cub.2019.11.026.537

D. H. Kim, J. Kim, J. C. Marques, A. Grama, D. G. C. Hildebrand, W. Gu, J. M. Li, and538

D. N. Robson. Pan-neuronal calcium imaging with cellular resolution in freely swimming539

zebrafish. Nat. Methods, 2017. ISSN 1548-7105. doi: 10.1038/nmeth.4429.540

21


https://doi.org/10.1101/2020.03.02.972612


C. B. Kimmel, S. K. Sessions, and R. J. Kimmel. Morphogenesis and synaptogenesis of the541

zebrafish mauthner neuron. J. Comp. Neurol., 1981. ISSN 10969861. doi: 10.1002/cne.542

901980110.543

T. Kohashi and Y. Oda. Initiation of Mauthner- or Non-Mauthner-Mediated Fast Escape544

Evoked by Different Modes of Sensory Input. J. Neurosci., 28(42):10641–10653, oct 2008.545

ISSN 0270-6474. doi: 10.1523/JNEUROSCI.1435-08.2008.546

M. Koyama, A. Kinkhabwala, C. Satou, S.-i. Higashijima, and J. Fetcho. Mapping a sensory-547

motor network onto a structural and functional ground plan in the hindbrain. Proc. Natl.548

Acad. Sci., 2011. ISSN 0027-8424. doi: 10.1073/pnas.1012189108.549

A. M. B. Lacoste, D. Schoppik, D. N. Robson, M. Haesemeyer, R. Portugues, J. M. Li, O. Ran-550

dlett, C. L. Wee, F. Engert, and A. F. Schier. A convergent and essential interneuron pathway551

for Mauthner-cell-mediated escapes. Curr. Biol., 25(11):1526–34, jun 2015. ISSN 1879-0445.552

doi: 10.1016/j.cub.2015.04.025.553

J. Larsch and H. Baier. Biological Motion as an Innate Perceptual Mechanism Driving Social554

Affiliation. Curr. Biol., 2018. ISSN 09609822. doi: 10.1016/j.cub.2018.09.014.555

B. Lemasson, C. Tanner, C. Woodley, T. Threadgill, S. Qarqish, and D. Smith. Motion cues556

tune social influence in shoaling fish. Sci. Rep., 8(1):9785, dec 2018. ISSN 2045-2322. doi:557

10.1038/s41598-018-27807-1.558

M. Lihoreau, L. Brepson, and C. Rivault. The weight of the clan: Even in insects, social559

isolation can induce a behavioural syndrome. Behav. Processes, 82(1):81–84, 2009. ISSN560

03766357. doi: 10.1016/j.beproc.2009.03.008.561

K. S. Liu and J. R. Fetcho. Laser Ablations Reveal Functional Relationships of Segmental562

Hindbrain Neurons in Zebrafish. Neuron, 23(2):325–335, jun 1999. ISSN 0896-6273. doi:563

10.1016/S0896-6273(00)80783-7.564

J. L. Lukkes, M. V. Mokin, J. L. Scholl, and G. L. Forster. Adult rats exposed to early-life social565

isolation exhibit increased anxiety and conditioned fear behavior, and altered hormonal566

stress responses. Horm. Behav., 55(1):248–256, jan 2009. ISSN 0018-506X. doi: 10.1016/J.567

YHBEH.2008.10.014.568

G. D. Marquart, K. M. Tabor, S. A. Bergeron, K. L. Briggman, and H. A. Burgess. Prepon-569

tine non-giant neurons drive flexible escape behavior in zebrafish. PLoS Biol., 2019. ISSN570

15457885. doi: 10.1371/journal.pbio.3000480.571

J. C. Marques and M. B. Orger. Clusterdv: A simple density-based clustering method that572

is robust, general and automatic. Bioinformatics, 2019. ISSN 14602059. doi: 10.1093/573

bioinformatics/bty932.574

J. C. Marques, S. Lackner, R. Felix, and M. B. Orger. Structure of the Zebrafish Locomotor575

Repertoire Revealed with Unsupervised Behavioral Clustering. Curr. Biol., 28(2):181–195.e5,576

jan 2018. ISSN 1879-0445. doi: 10.1016/j.cub.2017.12.002.577

22


https://doi.org/10.1101/2020.03.02.972612


J. C. Marques, M. Li, D. Schaak, D. N. Robson, and J. M. Li. Internal state dynamics shape578

brainwide activity and foraging behaviour. Nature, 2020. ISSN 14764687. doi: 10.1038/579

s41586-019-1858-z.580

K. C. Marsden and M. Granato. In Vivo Ca2+ Imaging Reveals that Decreased Dendritic581

Excitability Drives Startle Habituation. Cell Rep., 13(9):1733–1740, dec 2015. ISSN 22111247.582

doi: 10.1016/j.celrep.2015.10.060.583

K. C. Marsden, R. A. Jain, M. A. Wolman, F. A. Echeverry, J. C. Nelson, K. E. Hayer, B. Mil-584

tenberg, A. E. Pereda, and M. Granato. A Cyfip2-Dependent Excitatory Interneuron Path-585

way Establishes the Innate Startle Threshold. Cell Rep., 23(3):878–887, apr 2018. ISSN586

22111247. doi: 10.1016/j.celrep.2018.03.095.587

S. Martins, J. F. Monteiro, M. Vito, D. Weintraub, J. Almeida, and A. C. Certal. Toward an588

Integrated Zebrafish Health Management Program Supporting Cancer and Neuroscience589

Research. Zebrafish, 13(S1):S–47–S–55, jul 2016. ISSN 1545-8547. doi: 10.1089/zeb.2015.1198.590

D. S. Mearns, J. C. Donovan, A. M. Fernandes, J. L. Semmelhack, and H. Baier. Deconstructing591

Hunting Behavior Reveals a Tightly Coupled Stimulus-Response Loop. Curr. Biol., 2020.592

ISSN 09609822. doi: 10.1016/j.cub.2019.11.022.593

O. Mirat, J. R. Sternberg, K. E. Severi, and C. Wyart. ZebraZoom: An automated program for594

high-throughput behavioral analysis and categorization. Front. Neural Circuits, 2013. ISSN595

16625110. doi: 10.3389/fncir.2013.00107.596

A. Nagiel, S. H. Patel, D. Andor-Ardo, and A. J. Hudspeth. Activity-independent specification597

of synaptic targets in the posterior lateral line of the larval zebrafish. Proc. Natl. Acad. Sci.598

U. S. A., 2009. ISSN 00278424. doi: 10.1073/pnas.0912082106.599

D. M. O’Malley, Y.-H. Kao, and J. R. Fetcho. Imaging the Functional Organization of Zebrafish600

Hindbrain Segments during Escape Behaviors. Neuron, 17(6):1145–1155, dec 1996. ISSN601

0896-6273. doi: 10.1016/S0896-6273(00)80246-9.602

B. L. Partridge. The structure and function of fish schools. Sci. Am., 246(6):114–23, jun 1982.603

ISSN 0036-8733. doi: 10.1038/scientificamerican0682-114.604

B. L. Partridge and T. J. Pitcher. The Sensory Basis of Fish Schools: Relative Roles of Lateral605

Line and Vision. J. Comp. Physiol, 135:315–325, 1980.606

P. Pichler and L. Lagnado. Motor Behavior Selectively Inhibits Hair Cells Activated by For-607

ward Motion in the Lateral Line of Zebrafish. Curr. Biol., 2020. ISSN 09609822. doi:608

10.1016/j.cub.2019.11.020.609

T. J. Pitcher, B. L. Partridge, and C. S. Wardle. A blind fish can school. Science (80-. )., 194610

(4268):963–965, 1976. ISSN 00368075. doi: 10.1126/science.982056.611

J. Rihel, D. A. Prober, A. Arvanites, K. Lam, S. Zimmerman, S. Jang, S. J. Haggarty, D. Kokel,612

L. L. Rubin, R. T. Peterson, and A. F. Schier. Zebrafish behavioral profiling links drugs613

23


https://doi.org/10.1101/2020.03.02.972612


to biological targets and rest/wake regulation. Science (80-. )., 2010. ISSN 00368075. doi:614

10.1126/science.1183090.615

C. Satou, Y. Kimura, T. Kohashi, K. Horikawa, H. Takeda, Y. Oda, and S.-i. Higashijima.616

Functional Role of a Specialized Class of Spinal Commissural Inhibitory Neurons during617

Fast Escapes in Zebrafish. J. Neurosci., 29(21):6780–6793, 2009. ISSN 0270-6474. doi: 10.618

1523/JNEUROSCI.0801-09.2009.619

P. Schausberger, M. Gratzer, and M. A. Strodl. Early social isolation impairs development,620

mate choice and grouping behaviour of predatory mites. Anim. Behav., 127:15–21, may 2017.621

ISSN 00033472. doi: 10.1016/j.anbehav.2017.02.024.622

K. E. Severi, R. Portugues, J. C. Marques, D. M. O’Malley, M. B. Orger, and F. Engert. Neural623

Control and Modulation of Swimming Speed in the Larval Zebrafish. Neuron, 2014. ISSN624

10974199. doi: 10.1016/j.neuron.2014.06.032.625

S. Shams, S. Amlani, C. Buske, D. Chatterjee, and R. Gerlai. Developmental social isolation626

affects adult behavior, social interaction, and dopamine metabolite levels in zebrafish. Dev.627

Psychobiol., 2018. ISSN 10982302. doi: 10.1002/dev.21581.628

T. Shimazaki, M. Tanimoto, Y. Oda, and S. I. Higashijima. Behavioral role of the reciprocal629

inhibition between a pair of mauthner cells during fast escapes in Zebrafish. J. Neurosci.,630

2019. ISSN 15292401. doi: 10.1523/JNEUROSCI.1964-18.2018.631

R. Spence and C. Smith. The role of early learning in determining shoaling preferences based632

on visual cues in the zebrafish, Danio rerio. Ethology, 113(1):62–67, jan 2007. ISSN 01791613.633

doi: 10.1111/j.1439-0310.2006.01295.x.634

S. J. Stednitz and P. Washbourne. Rapid Progressive Social Development of Zebrafish. Ze-635

brafish, 2020. ISSN 15578542. doi: 10.1089/zeb.2019.1815.636

S. J. Stednitz, E. M. McDermott, D. Ncube, A. Tallafuss, J. S. Eisen, and P. Washbourne. Fore-637

brain Control of Behaviorally Driven Social Orienting in Zebrafish. Curr. Biol., 28(15):2445–638

2451.e3, aug 2018. ISSN 0960-9822. doi: 10.1016/J.CUB.2018.06.016.639

P. J. Steenbergen, M. K. Richardson, and D. L. Champagne. The use of the zebrafish model in640

stress research. Prog. Neuro-Psychopharmacology Biol. Psychiatry, 35(6):1432–1451, aug 2011.641

ISSN 0278-5846. doi: 10.1016/J.PNPBP.2010.10.010.642

J. R. Sternberg, K. E. Severi, K. Fidelin, J. Gomez, H. Ihara, Y. Alcheikh, J. M. Hubbard,643

K. Kawakami, M. Suster, and C. Wyart. Optimization of a Neurotoxin to Investigate the644

Contribution of Excitatory Interneurons to Speed Modulation In Vivo. Curr. Biol., 26(17):645

2319–2328, sep 2016. ISSN 0960-9822. doi: 10.1016/J.CUB.2016.06.037.646

M. Takahashi, M. Narushima, and Y. Oda. In vivo imaging of functional inhibitory networks647

on the mauthner cell of larval zebrafish. J. Neurosci., 22(10):3929–38, may 2002. ISSN 1529-648

2401. doi: 20026396.649

24


https://doi.org/10.1101/2020.03.02.972612


M. Takahashi, M. Inoue, M. Tanimoto, T. Kohashi, and Y. Oda. Short-term desensitization650

of fast escape behavior associated with suppression of Mauthner cell activity in larval ze-651

brafish. Neurosci. Res., 121:29–36, aug 2017. ISSN 0168-0102. doi: 10.1016/J.NEURES.2017.652

03.008.653

I. Temizer, J. C. Donovan, H. Baier, and J. L. Semmelhack. A Visual Pathway for Looming-654

Evoked Escape in Larval Zebrafish. Curr. Biol., 2015. ISSN 09609822. doi: 10.1016/j.cub.655

2015.06.002.656

M. Westerfield. The Zebrafish Book. A Guide for the Laboratory Use of Zebrafish (Danio657

rerio), 5th Edition. Univ. Oregon Press. Eugene, 385, mar 2007.658

L. S. Wilkinson, S. S. Killcross, T. Humby, F. S. Hall, M. A. Geyer, and T. W. Robbins. Social659

isolation in the rat produces developmentally specific deficits in prepulse inhibition of the660

acoustic startle response without disrupting latent inhibition. Neuropsychopharmacology, 10661

(1):61–72, 1994. ISSN 1740634X. doi: 10.1038/npp.1994.8.662

D. Zellner, B. Padnos, D. L. Hunter, R. C. MacPhail, and S. Padilla. Rearing conditions663

differentially affect the locomotor behavior of larval zebrafish, but not their response to664

valproate-induced developmental neurotoxicity. Neurotoxicol. Teratol., 33(6):674–679, 2011.665

ISSN 08920362. doi: 10.1016/j.ntt.2011.06.007.666

25


https://doi.org/10.1101/2020.03.02.972612


METHODS667

Experimental model and subject details668

Zebrafish breeding and maintenance was performed under standard conditions [Westerfield,669

2007, Martins et al., 2016]. The sex of the larvae is not defined at this early stage of develop-670

ment. Fertilized wild type embryos of the Tubingen (TU) strain were collected in the morning671

and placed in petri dishes containing E3 medium [Westerfield, 2007]. Embryos were kept in672

10 cm petri dishes in groups of 20 for the first three days of development. Then at 3 dpf,673

larvae were randomly assigned to a raising condition (isolation raised or group raised) and674

placed either alone or in groups of 8 into 35 mm petri dishes containing fresh E3 medium.675

All dishes were kept at 28◦C and a 14/10 h light-dark cycle, starting at 8 am. Larvae were676

tested at 6 dpf. All behavioral experiments were performed in a randomized order during677

the larva’s day period, between 9 am and 8 pm. Animal handling and experimental proce-678

dures were approved by the Champalimaud Center for the Unknown Bioethics Committee679

and the Portuguese Veterinary General Board (Direcao Geral Veterinaria), in accordance with680

the European Union Directive 86/609/EU.681

Behavioral recording apparatus682

The behavioral setup has previously been described by [Marques et al., 2018]. Inside of a light-683

proof enclosure, a high-speed infrared sensitive camera (MC1362, Mikroton) with a Schneider684

apo-Xenoplan 2.0/35 lens and a 790 nm long pass filter was mounted above a stage holder685

for the fish recording chamber. Larvae were illuminated from below by a 10 x 10 cm infrared686

LED-based diffusive backlight (850 nm). Tracking of the larva’s position and equally spaced687

tail segments of 0.3 mm was performed online during video acquisition at 700 frames per688

second using a custom-made algorithm in C#, as previously described by [Severi et al., 2014,689

Burgess and Granato, 2007a]. The tail data was interpolated to a standardized set of points in690

the tail spaced at 310 µm increments, starting at a point 1345 µm behind the midpoint between691

the two eyes.692

Swim bout detection693

We applied the bout detection algorithm described in [Marques et al., 2018]. To detect bouts, a694

measure of change in tail curvature was used to capture tail movements reliably. To compute695

this value, we applied spatiotemporal smoothing to the raw tail angles using a boxcar filter696

(3 segments, 10 frames) and calculated the frame-to-frame change of this smoothed measure697

at each point along the tail. For each segment, we then calculated the cumulative sum (from698

rostral segment to current segment), took the absolute value, and then summed the results699

over all segments to obtain a scalar measure of tail curvature. This helps to ignore local tail700

fluctuations and highlights extended regions of curvature. To exclude effects of slow drifts in701

tail position, we subtracted the minimum of the previously obtained curvature measure over702

a 571 ms window. To smooth out the half-beats that compose individual bouts, we applied a703

maximum filter with a 30 ms window. For each fish the maximum of this measure was cal-704

culated and divided by 20, and this value was used as a threshold to detect individual bouts.705

26


https://doi.org/10.1101/2020.03.02.972612


Short fluctuations of the tail curvature measure that did not correspond to tail movements706

were excluded by ensuring that bouts are at least 43 ms long. To ensure that individual bouts707

were not broken into parts, the interbouts must also have a minimum duration of 43 ms. The708

exact start and end of bouts was calculated by the start of the first half-beat and the end of the709

last half-beat (see half-beat detection).710

Half-beat detection711

Half-beats were detected as previously described [Marques et al., 2018]. A dynamic threshold712

was computed for each bout by using a 43 ms boxcar filter to smooth the tail curvature values.713

The beginning and end of each half beat, at each segment, were found by locating the points714

where the tail crosses this threshold. The half-beat peak at each segment was defined as715

the maximum absolute value of tail angle. Some large amplitude bouts had a first half-beat716

that was a small passive deflection in the opposite direction of the second larger half-beat.717

These situations are detected by finding small first half-beats (10% higher than the baseline718

value) that were followed by larger movements in the opposite direction. To maintain the719

consistency of the half-beat numbering, for these bouts, the first small half-beat was excluded720

and the second larger half-beat was considered the first.721

Bout type categorization722

For each swim bout, we computed 73 kinematic parameters based on the bout and half-beat723

detection, as described previously in depth [Marques et al., 2018]. After, we embedded all724

the data into a previously computed principal component analysis (PCA) space and included725

for further analysis the first 20 Principal Components (PCs). This space was obtained by726

performing PCA over the 73 kinematic parameters on a dataset with 3 million bouts, that727

fish executed while performing 9 behaviors over 255 stimuli, that included social avoidance in728

the light and dark, acoustic startles, and responses to looming stimuli [Marques et al., 2018].729

Therefore, we expect that the new data gathered in this report, that is mainly constituted of730

avoidance bouts, to be well represented in this PCA space. To categorize the bouts into types,731

we used a dataset composed of equal number of bout types (Bout Map) [Marques et al., 2018],732

that were categorized using density valley clustering [Marques and Orger, 2019], and used k733

nearest neighbors (k = 50) to assigned every new bout to one of the 13 possible bout categories.734

Behavioral testing of spontaneous locomotion735

Larvae were carefully placed in a circular swimming arena with a depth of 3mm inside the736

recording setup. After 5 min of adjustment time the recording was started and lasted for737

1 hour. Pairs of larvae of the same raising condition were tested in circular arenas with a738

20 mm diameter. One recording chamber consisted of four arenas, hence eight larvae could739

be recorded at once. All arenas had a transparent bottom plate and black acrylic walls to740

avoid that larvae could see the pairs in the neighboring arenas. For recordings in light, a741

homogeneous background illumination of 1000 lux was projected onto an opal glass diffuser742

5 mm below the larvae by a DLP projector (BenQ). For recordings in darkness, the projector743

remained turned off, hence the illumination inside the light-proof enclosure was 0 lux.744

27


https://doi.org/10.1101/2020.03.02.972612


Avoidance area measure745

The avoidance area measure was adjusted from [Marques et al., 2018]. First, we computed746

the density distribution of the positions of the other larvae relative to a focal larva for each747

frame of the experiment. Then we created a reference density distribution by randomizing748

the order of the recording frames. Both, the data and the reference density distributions were749

then smoothed by a Gaussian filter with a kernel sigma of 0.36mm. To avoid arbitrarily high750

ratio values due to division by a value near zero, points at the edges where the randomized751

control data had a value less than 1 were excluded. The raw density data was divided by752

the control to generate a ratio image. The low density region in the center of the image was753

quantified by counting the pixels in a bounded region where the density of the real data was754

below 60% of the randomized control density. The contours of these detected areas are shown755

in the right panel of Fig. 1B.756

When testing pairs of larvae instead of groups of seven, the data is more sparse and the757

thresholding of the density distribution can lead to false high values in area. We therefore758

adjusted the avoidance measure to a one-dimensional version by considering the distance759

between the two fish rather than a two-dimensional map of positions. Similarly to the 2D760

avoidance area measure, we created a random distribution by shuffling the frame order so761

that the trajectories of both fish were no longer matched in time. We then made a histogram of762

the values of the real and the randomized data with a bin size of 0.5mm. The ratio between the763

randomized and the real data histogram was negative for close distances, indicating that the764

fish avoid each other. To quantify this avoidance we used a trapezoidal numerical integration765

of the distances smaller than the zero crossing.766

Comparison of SLC and LLC767

The proportion of SLCs is calculated by dividing the total number of SLCs performed by the768

sum over all C-starts (SLC and LLC) performed when the two fish were swimming within769

10mm of one another. Position of the non-focal fish upon C-start onset of the focal fish is770

considered for when the distance between the two fish was smaller than 10mm and the non-771

focal fish performed a bout within 1000ms prior to C-start onset.772

Cluster comparison between C-starts of group and isolation raised fish773

The LLC and SLC bouts were embedded in the same PCA space described in the “Bout774

categorization” section and the first four PCs were taken into account. If raising conditions775

modulate kinematically the bout types, we expect them to occupy a distinct place in this space.776

For visual comparison we plotted 200 randomly chosen LLCs and SLCs per raising conditions777

into the first four components of the PCA space.778

Pharmacology779

The general dopamine agonist, apomorphine (R-(-)-Apomorphine hydrochloride hemihydrate,780

Sigma-Aldrich, A4393), was prepared to a final concentration of 20 µM in buffered E3 medium.781

Larvae were incubated for 10min in the drug, and tested after three washes and a recovery782

28


https://doi.org/10.1101/2020.03.02.972612


period of 10min. This protocol is in line with what has been published in adult zebrafish783

[Stednitz et al., 2018].784

For lateral line ablation, CuSO4 (Copper(II) Sulfate pentahydrat, Acros Organics) was dis-785

solved as concentration of 1 mM in E3 medium. Larvae were incubated for 1h without a786

change in social context; isolation raised larvae were incubated in isolation and group raised787

larvae in groups. After incubation larvae were washed three times in fresh E3 medium and788

given a 1h recovery period prior to behavioral testing. Neomycin trisulfate (Sigma-Adrich)789

was dissolved in tris-buffered E3 medium (E3-tris, 1 mM tris) to a final concentration of 200790

µM. Larvae were incubated as described above for 1h. Thereafter larvae were washed three791

times in E3-tris and given a 2h recovery period before behavioral testing.792

Both of these incubation protocols have been shown to ablate hair cells with a regeneration793

time larger than our testing time window [Buck et al., 2012].794

Handled controlled for CuSO4 and neomycin treatments were exposed to the same treat-795

ment with incubation in E3 medium or E3-tris, respectively.796

Closed-loop presentation of visual stimuli797

Single larvae of either raising condition were swimming freely in a circular arena (5 cm in798

diameter) with a homogeneous background illumination projected onto an opal glass diffuser799

5 mm below the larvae by a DLP projector (BenQ). Both types of visual stimuli were presented800

with an inter-stimulus interval of 2 min. The walls of the arena were transparent, so that visual801

stimuli could be displayed even when the larva was swimming in the arena borders. For802

testing looming stimuli an expanding dark spot (1 cm/s) was projected 4 cm away from the803

larva, either placed to the front, the back, left or right of the larva [Marques et al., 2018]. In the804

approaching dark spot assay a spot of constant size was projected 2 cm away from the larva805

and approached it with a constant speed of 0.5 cm/s. The spot radius varied between trials806

from 0.5 to 2.5 mm and was approaching from the front, the back, left or right. All stimuli807

were tested twice per animal and presented in a randomized order. Each randomized order808

was presented to one group raised and one isolation raised larva to avoid any unbalanced809

order effects.810

Closed-loop presentation of mechanosensory stimuli811

To produce minute local water displacements we used a piezoelectric bender actuator (Thor-812

labs; voltage range: 150V; max. displacement: 450 µm). A 2 cm long capillary rod was813

attached to the body of the piezo bender using silicone. The rod was made of the tip of a814

loading-pipette tip (outer diameter: 0.3 mm). To give more rigidity to the rod a headless and815

stainless steel insect pin was pushed inside. The larger end of the capillary was melted into816

a flat circle via a heat-shrink gun to allow for better attachment to the piezo bender. The817

capillary rod attached to the piezo bender was mounted vertically at a 45◦ angle, so that 2818

mm of its tip would be submerged into the water of the swimming arena. We left a single819

larva of either raising condition swim inside this arena for one hour in complete darkness.820

Movements of the capillary tip were triggered in closed-loop depending on the position and821

movement of the larva with a minimum interval between capillary stimuli of 2min. We used822

29


https://doi.org/10.1101/2020.03.02.972612


the following criteria to trigger piezo deflections: 1) the larva had to be less than 4mm away823

from the tip of the capillary rod; 2) thereafter the larva had to be more than 5mm away from824

the tip; 3) after a 400ms delay the piezo was triggered. Unless otherwise stated the structure of825

the piezo pulse was composed of five 2ms pulses with a 2ms inter-pulse interval. The voltage826

applied stayed fixed at 40V for each pulse train. Each fish was only exposed to one set of827

fixed stimulus parameters. Different sets of stimulus parameters were tested across different828

fish. Testing times and fish clutches were interleaved to avoid any uncontrolled bias.829

Acoustic startle assay830

A single larva was swimming for 1 hour and 40 min in a custom-build chamber. with a circular831

swimming arena of 5 cm diameter. Outside of the swimming arena was a black acrylic sheet832

onto which four speaker membranes (Visaton) were glued with equal spacing. A single 100833

ms pure tone of 600 Hz was played by gating the sound output of the recording PC through a834

digital switch (Arduino One). Acoustic stimuli were presented with a minimum ISI of 6min.835

The experiment was interleaved with other visual stimuli, which were not analyzed for this836

manuscript. In total 16 acoustic startle stimuli were presented per larva.837

Quantification and statistical analysis838

All data are either presented as pooled events per raising condition (when plotted as his-839

tograms), or as single measure per replica and shown as mean with standard deviation, un-840

less otherwise stated in the figure legend. Only datasets that were collected in the same841

days of experimentation are considered for statistical testing. Datasets composed of pairs842

or groups of larvae per recording were first averaged between individuals and then this843

average was considered as the unit of comparison across replicates, so as to avoid pseudo-844

replication. We applied two-tailed unsigned Wilcoxon tests (Matlab: ranksum) for com-845

paring between raising conditions or treatments and two-tailed signed Wilcoxon tests (Mat-846

lab: signrank) when comparing repeated measures coming from the same larva or pair of847

larvae, i.e. comparison between bout types. When performing series of pairwise com-848

parisons, p-values were corrected for the number of comparisons using Holm’s sequential849

Bonferroni procedure (www.mathworks.com/matlabcentral/fileexchange/28303-bonferroni-850

holm-correction-for-multiple-comparisons). For comparing the distribution of angles of the851

position of the non-focal larvae upon C-start onset, we used the CircStat toolbox [Berens,852

2009]. As an addition to regular hypothesis tests, for all comparisons between raising con-853

ditions or treatments, we quantified the effect sizes of the observed differences using the854

Measures of Effect Size (MES) Toolbox [Hentschke and Stuttgen, 2011]. In the statistical sum-855

mary tables in the supplementary material, as well as in Fig. 3J we used the mean differences856

and its confidence intervals as calculated with 1000 bootstraps. All analysis and statistical857

tests were performed in Matlab 2018a.858

30


https://doi.org/10.1101/2020.03.02.972612


Supplementary Material859

31


https://doi.org/10.1101/2020.03.02.972612


-100%

100%

D

ensi

ty D

iffer

ence

A B

Isolation Raised

Group Raised

5 mm25 mm5 mm

C

Avo

idan

ce A

rea

(a.u

.)

0GroupRaised

IsolationRaised

0.5

1.0

1.5

2.0

D Principal components of bout kinematicsETail tracking

LLC

SLC

Group RaisedIsolation Raised

LLC

SLC

10

0°

300ms tail angle trace

Figure S1: Social avoidance area measures in pairs of larvae. (A) Pairs of larvae raised either ingroups or in isolation were video recorded for 1 h while swimming in a circular arena with a 20mmdiameter. (B) Density distributions, relative to a focal individual, of group raised (top) and isolationraised larvae (bottom). The focal fish is located in the center and facing upwards. Raw densities aredivided by a reference distribution created by choosing fish locations from random frames. Rightpanels are zoomed views of the black squares in the left panels. (C) For pairs of larvae the avoidancedistance was calculated as the negative area under the curve of histograms comparing the distancebetween the two fish to a control distribution of randomly shuffled recording frames. Light gray opencircles show every replicate per condition, dark gray filled circles and error-bars signal mean ± SD.Asterisks specify the results of a two-tailed unsigned Wilcoxon test; p ¡ 0.001; N=43). (D) Trackingof equally spaced tail segments at 700 Hz enables measuring the tail oscillations of each swim bout.Shown here are example traces of tail angles across a 300 ms sequence containing a long latency C-start(LLC, top) and a short latency C-start (SLC, bottom). The measures of all tail segments are overlaid andcolor-coded according to the position along the body of the fish show on the top. (E) Mapping of 200randomly selected LLCs (top) and SLCs (bottom) for group (black) and isolation-raised larvae (red), inthe space of the first four principal components that were used for the bout classification based on 73kinematic parameters. Gray scatter points display all bouts that are not C-starts. Group and isolationdata points were randomized so that the colors are not overlaid.

32


https://doi.org/10.1101/2020.03.02.972612


K

Bou

t dur

atio

n (m

s)

0

50

100

150

0

LLCSLC

AS

BS

HATRT

SATO-bend

Slow 1Slow 2 Short CS

Long CSJ-turn

Group Raised N=20Isolation Raised N=20

Swimming Alone

Pro

bab

ility

of B

out T

ype

Cho

ice

Swimming in Pairs

0.2

0.4

0

0.2

0.4

A B C

F H IG

D E

Group Isolation

Bou

ts s

wum

per

min

0

20

40

60

80

Group Isolation

Bou

t dur

atio

n (m

s)

0

50

100

150

PairsPairs

Group Isolation

Bou

ts s

wum

per

min

0

20

40

60

80

Alone

Group Isolation

Bou

t dur

atio

n (m

s)

0

50

100

150

Alone

0

0.5

1.0

1.5

Avo

ida

nce

are

a (

a.u

.)

Pro

po

rtio

n o

f S

LC

Control Apo.

Bou

ts s

wum

per

min

0

20

40

60

80

Bou

t dur

atio

n (m

s)

0

50

100

150

Control Apo. Control Apo. Control Apo.

2.0

0

0.2

0.4

0.6

0.8n.s.

n.s.

J

Bou

ts s

wum

per

min

0

20

40

60

80 n.s.

Group Iso 3dpf

L

0

0.5

1.0

1.5

Avo

ida

nce

are

a (

a.u

.)

2.0

M

Pro

po

rtio

n o

f S

LC

0

0.2

0.4

0.6

0.8

Group Iso 3dpf Group Iso 3dpf Group Iso 3dpf

Dopamine Agonist (Apomorphin)

Social Isolation from 3 dpf onward

* * ***

*

33


https://doi.org/10.1101/2020.03.02.972612


Figure S2 (preceding page): General locomotion effects of isolation raising. (A) Bout type probabilityduring 30min of spontaneous locomotion of larvae tested in pairs of the same raising condition (top) orswimming alone in the arena (bottom). Bout types are color coded, markers and errorbars mean ± SD,for group raised (circles) and isolation raised larvae (squares). Asterisks signal significant results ofbonferroni-holm corrected p-values from two-tailed unsigned Wilcoxon tests for each bout type (alphalevel 0.05). (B,D) Number of bouts swum per min over the 30 min recording period (N=20). (C,E)Average duration of each bout performed irrespective of the bout category (N=20). (F-I) Group raisedlarvae were treated for 10min with the generic dopamine agonist apomorphin (Apo.) (N=20). (J-M)Instead of a full-term isolation protocol, larvae were raised in groups until 3dpf and thereafter isolated(Iso 3dpf) or handled similarly and placed in a new dish remaining in the social context (Group)(NGR = 40, NISO = 39). Larvae were tested in pairs of the same treatment group. Shown are themeasures of number of bouts swum per min over the 30 min recording period (F,J), the average boutduration (G,K), the avoidance area as calculated by comparing the distribution of inter-fish distanceswith a random distribution (see methods; H,L) and the proportion of SLC out of all C-starts (LLC andSLC) performed as the larvae swam within 10mm of one another (I,M). Light gray open circles showevery replicate per condition, dark gray filled circles and error-bars signal mean ± SD. Asterisk specifythe results of statistical comparisons between raising conditions. AS: Approach Swim; Long CS: Longcapture swim; BS: Burst swim; HAT: High angle turn; RT: Routine turn; SAT: Shadow avoidance turn;LLC: Long latency C-start; SLC: Short latency C-start.

34


https://doi.org/10.1101/2020.03.02.972612


A

Vis

ion

only

0.20.2

B C

D E F

0.20.2

GroupRaised

IsolationRaised

GroupRaised

IsolationRaised

after CuSO4 treatment

after control treatment

Non-focal fish position upon C-start onset

Vis

ion

and

LL

0

0.2

0.4

0.6

0.8

1

Pro

port

ion

of S

LC n.s.

0

0.5

1

1.5

2

2.5

Avo

idan

ce a

rea

(a.u

.)

n.s.

GroupCuSO4

IsolationCuSO4

GroupCuSO4

IsolationCuSO4

0

0.5

1

1.5

2

2.5

Avo

idan

ce a

rea

(a.u

.)

0

0.2

0.4

0.6

0.8

1

Pro

port

ion

of S

LC

GroupContr.

IsolationContr.

GroupContr.

IsolationContr.

Figure S3: Effects of CuSO4 and control treatment of social avoidance reactions. (A-C) Larvae wereincubated for 1h in CuSO4 and given a 1 hour recovery period prior to testing with homogeneousbackground illumination; NGR = 27; NISO = 28. (D-F) Control treatment consisted of incubatinglarvae for 1h in regular E3 medium, performing the washing steps and then allowing for a 1 hourrecovery period prior to testing with homogeneous background illumination; NGR = 20; NISO = 20.(A,D) Circular histograms of the position of the non-focal larva upon the onset of the C-start. Data ispooled over fish per raising condition and treatment. Red and yellow bars show data for SLC and LLC,respectively. (B,E) Social avoidance area as calculated from the distribution of distances between thetwo fish throughout the recording period. (C,F) The proportion of SLC bouts over the sum of SLC andLLC. Light gray open circles show every replicate per condition, dark gray filled circles and error-barssignal mean ± SD. Asterisk specify the results of two-tailed unsigned Wilcoxon tests between raisingconditions (alpha level 0.05).

35


https://doi.org/10.1101/2020.03.02.972612


D E

G H I

20

40

Nu

mb

er

of e

ve

nts

0 2000 4000

Latency (ms)

0

20

40

20

40

0 500 1000

Latency (ms)

0

20

40

GroupRaised

IsolationRaised

Chasing dot stimulus Piezo stimulus

-2 0 20

0.2

0.4

0.6

Bo

ut

typ

e P

b

-2 0 2

Time (sec)

GroupControl

GroupNeomycin

0

0.2

0.4

0.6

0.8

1

0

0.2

0.4

0.6

0.8

1

GroupControl

Pro

por

tion

of S

LC

Rea

ctio

n pr

oba

bilit

y

GroupNeom.

GroupControl

GroupNeom.

0 2 4

Time (sec)

0

0.2

0.4

0.6

Bo

ut

typ

e P

b

0 2 4

IsolationRaised

Looming stimulus Chasing dot stimulusA

-2 0 2

Time (sec)

0

0.2

0.4

0.6

Bo

ut

typ

e P

b

-2 0 2

GroupRaised

IsolationRaised

Piezo Onset Piezo OnsetC

LLC

SLC

SAT

O-bend

Time (sec)

0

0.2

0.4

0.6

Bo

ut

typ

e P

b

B

Nu

mb

er

of e

ve

nts

0

Latency (ms)

100 2000

10

20

30

10

20

30

Nu

mb

er

of e

ve

nts

F Acoustic stimulus

GroupRaised

IsolationRaised

Piezo Onset Piezo Onset

36


https://doi.org/10.1101/2020.03.02.972612


Figure S4 (preceding page): Response comparison between controlled visual and lateral line stimuli.(A,B) Probability of performing an avoidance bout type out of all 13 bout types was calculated in 500ms time windows during presentation of a visual stimulus, which reaches the centroid of the larvaeafter 4 sec. Shown is the data of isolation raised larvae; N = 17 for looming stimuli (A) and N = 15 forchasing dot stimuli (B). Bout types are color-coded as shown in the legend. SAT: Shadow avoidanceturn; LLC: Long latency C-start; SLC: Short latency C-start. (C) Bout type probability before and afterpresentation of local water vibrations through a semi-submerged rod, attached to a piezo bender (40V).Shown are the mean values of group (left, N = 15) and isolation raised larvae (right, N = 15). (D-F)Histograms of C-start latency after stimulus presentation shown for group (top) and isolation raisedlarvae (bottom). SLCs and LLCs are color-coded in red and yellow, respectively. The acoustic stimulus(F) consisted of a 100ms long 600 Hz pure tone played through speakers directly attached to the wallsof the swimming arena. (G) Avoidance bout type probability across time before and after local watervibrations of 80V piezo input voltage is shown for control group raised larvae (left) and group raisedlarvae treated with neomycin to ablate the lateral line (right). (H) Reaction probability is defined as thenumber of SLC or LLC performed per total number of chasing dot stimulus presentations per larvae oreither raising condition. (I) The proportion of SLCs was calculated as the number of stimulus inducedSLCs divided by the sum over stimulus-induced SLCs and LLCs. Light gray open circles show everyreplicate per condition, dark gray filled circles and error-bars signal mean ± SD. Asterisk specify theresults of two-tailed unsigned Wilcoxon tests between raising conditions (alpha level 0.05).

37


https://doi.org/10.1101/2020.03.02.972612


Tables with statistical results860

FigureRef Measure zValue p MD CIlow CIhigh Ngr Niso

Fig. 1C Avoidance area NaN* 0.00078 -40.93 -55.8 -25.3 9 9Fig. S1C Avoidance area -2.7 0.0069 -0.24 -0.39 -0.08

43 43Fig. 2E C-starts per bout -1.09 0.2765 0.21 0.19 0.24Fig. 2F SLC proportion -5.63 <0.0001 -0.17 -0.22 -0.13

Table S1: Statistical comparison of group and isolation-raised larvae in their social avoid-ance measures. Figure reference indicates where the data is displayed. Avoidance area, C-Starts per bout and SLC proportion were calculated as described in the methods. Two-tailedunsigned Wilcoxon tests were applied for determining z and p values. MD and CI refer to themean difference and its confidence intervals of the difference between group and isolation-raised larvae, as determined by 1000 bootstraps using the measure of effect size toolbox inMatlab. *Note that with the comparison of avoidance area measure in groups of seven larvaeit was not possible to calculate z-values due to insufficient sample sizes.

38


https://doi.org/10.1101/2020.03.02.972612


Bout Type MeanGr SDGr MeanIso SDIso zValue p corr p

AS 10.983 1.959 11.175 1.24 -0.15 0.8833 1.1256Slow 1 11.673 0.827 11.274 1.15 1.58 0.114 0.5699Slow 2 11.821 0.778 11.231 1.328 2.73 0.0063 0.0571Short CS 12.088 0.864 11.172 1.416 3.53 0.0004 0.0049J turn 11.78 0.742 11.336 1.176 1.97 0.0489 0.3425HAT 11.689 0.71 11.44 1.037 1.2 0.2299 0.9197Routine turn 11.723 0.732 11.278 1.205 2.13 0.0329 0.2632SAT 11.521 1.226 11.914 1.31 -0.92 0.3599 1.0798Burst swim 7.62 2.109 7.922 2.184 -0.58 0.5628 1.1256Long CS 5.991 3.249 7.384 3.899 -1.72 0.0857 0.5139O bend 6.618 2.361 8.246 2.438 -2.87 0.0041 0.0455LLC 6.998 1.983 8.177 2.117 -2.82 0.0049 0.0487SLC 3.788 1.142 4.978 1.798 -4.1 0 0.0005

Table S2: Statistical comparison for data shown in Fig 2B. Statistical comparison per bouttype of the difference between group and isolation-raised larvae in their median inter-fishdistance per larval test pair. N = 43 for both group and isolation-raised data. Two-tailedunsigned Wilcoxon tests were applied for each bout type for determining z and p values.To correct for multiple comparison, p-values were corrected (corr p) for the number of com-parisons (13) using Holm’s sequential Bonferroni procedure. Mean and standard deviation(SD) are reported for each measure used for the comparison. AS: Approach Swim; Long CS:Long capture swim; BS: Burst swim; HAT: High angle turn; RT: Routine turn; SAT: Shadowavoidance turn; LLC: Long latency C-start; SLC: Short latency C-start.


Long CS 0.015 0.006 0.016 0.024 2.36 0.0184 0.0551Burst swim 0.023 0.016 0.024 0.026 0.5 0.6164 0.6164O bend 0.014 0.009 0.019 0.011 -2.36 0.0184 0.0551LLC 0.071 0.03 0.051 0.025 4.02 0.0001 0.0003SLC 0.036 0.014 0.064 0.037 -3.84 0.0001 0.0005

Table S3: Statistical comparison for data shown in Fig 2D. Statistical comparison per bouttype of the difference between group and isolation-raised larvae in their probability of usagewhen the fish are at least within a distance of 10mm of one another. Bout type probabilitywas determined as the number of the bouts per type divided by the total number of boutsswum of all 13 types. Two-tailed unsigned Wilcoxon tests were applied for each bout type fordetermining z and p values. To correct for multiple comparison, p-values were corrected (corrp) for the number of comparisons (5) using Holm’s sequential Bonferroni procedure. Meanand standard deviation (SD) are reported for each measure used for the comparison. N =43 for both group raised and isolation raised data. Long CS: Long capture swim; LLC: Longlatency C-start; SLC: Short latency C-start.

39


https://doi.org/10.1101/2020.03.02.972612


Figu

reR

efM

easu

reC

ompa

riso

nTyp

em

ean

1SD

1m

ean

2SD

2zV

alue

pco

rrp

Fig.

2GLa

tenc

y

GrS

LCvs

IsoS

LC95

.95

28.8

910

3.65

38.0

8-0

.92

0.35

540.

3554

GrL

LCvs

IsoL

LC19

5.2

70.1

223

5.85

108.

69-1

.90.

0568

0.11

37G

rSLC

vsG

rLLC

95.9

528

.89

195.

270

.12

-5.7

<0.

0001

<0.

0001

0Is

oSLC

vsIs

oLLC

103.

6538

.08

235.

8510

8.69

-5.5

<0.

0001

<0.

0001

Fig.

2IC

1an

gle

GrS

LCvs

IsoS

LC80

.93

7.96

83.4

9.67

-1.0

20.

3081

0.30

81G

rLLC

vsIs

oLLC

56.8

86.

2259

.98.

17-2

.24

0.02

530.

0506

GrS

LCvs

GrL

LC80

.93

7.96

56.8

86.

225.

7<

0.00

01<

0.00

01Is

oSLC

vsIs

oLLC

83.4

9.67

59.9

8.17

5.71

<0.

0001

<0.

0001

Fig.

2JA

ngul

arsp

eed

GrS

LCvs

IsoS

LC13

.36

0.75

13.3

91.

05-0

.32

0.74

930.

7493

GrL

LCvs

IsoL

LC7.

560.

897.

911.

11-2

.03

0.04

240.

0848

GrS

LCvs

GrL

LC13

.36

0.75

7.56

0.89

5.71

<0.

0001

<0.

0001

IsoS

LCvs

IsoL

LC13

.39

1.05

7.91

1.11

5.71

<0.

0001

<0.

0001

Tabl

eS4

:St

atis

tica

lco

mpa

riso

nof

C-s

tart

type

sbe

twee

ngr

oup

and

isol

atio

n-ra

ised

larv

ae.

Figu

rere

fere

nce

indi

cate

sw

here

the

data

isdi

spla

yed.

Two-

taile

dun

sign

edan

dsi

gned

Wilc

oxon

test

sw

ere

appl

ied

whe

nco

mpa

ring

betw

een

rais

ing

cond

itio

nsan

dbe

twee

nbo

utty

pes

wit

hin

rais

ing

cond

itio

n(r

epea

ted

mea

sure

),re

spec

tive

ly,f

orde

term

inin

gz

and

pva

lues

.To

corr

ect

for

mul

tipl

eco

mpa

riso

n,p-

valu

esw

ere

corr

ecte

d(c

orr

p)fo

rth

enu

mbe

rof

com

pari

sons

(4)

usin

gH

olm

’sse

quen

tial

Bonf

erro

nipr

oced

ure.

The

com

pari

son

type

isab

brev

iate

das

the

follo

win

g;G

rLLC

and

GrS

LC:L

ong

and

shor

tlat

ency

C-s

tart

spe

rfor

med

bygr

oup-

rais

edla

rvae

;Is

oLLC

and

IsoS

LC:L

ong

and

shor

tlat

ency

C-s

tart

spe

rfor

med

byis

olat

ion-

rais

edla

rvae

.Mea

n1

refe

rsto

the

first

men

tion

edm

easu

re(i

.e.G

rLLC

whe

nw

ritt

en:G

rLLC

vsG

rSLC

)and

likew

ise,

mea

n2

refe

rsto

the

seco

ndm

enti

oned

mea

sure

.Sim

ilarl

y,SD

1an

d2

refe

rto

the

stan

dard

devi

atio

nof

the

mea

sure

s.

40


https://doi.org/10.1101/2020.03.02.972612



Tested in pairs of larvaeAS 0.022 0.026 0.052 0.056 -1.69 0.0909 0.5151Slow 1 0.273 0.083 0.261 0.088 0.09 0.9246 1.1957Slow 2 0.265 0.075 0.206 0.089 2.34 0.0193 0.1543Short CS 0.023 0.01 0.019 0.009 1.58 0.1136 0.4545J turn 0.053 0.012 0.065 0.026 -1.72 0.0859 0.5151HAT 0.122 0.022 0.11 0.039 1.47 0.1404 0.4542Routine turn 0.19 0.068 0.173 0.059 0.53 0.5979 1.1957SAT 0.023 0.018 0.045 0.025 -3.29 0.001 0.0101Burst swim 0.003 0.003 0.013 0.013 -3.99 0.0001 0.0007Long CS 0.002 0.002 0.005 0.004 -2.02 0.0439 0.3072O bend 0.003 0.002 0.009 0.004 -4.34 <0.0001 0.0002LLC 0.014 0.015 0.025 0.016 -3.18 0.0015 0.0133SLC 0.006 0.003 0.017 0.011 -4.23 <0.0001 0.0003

Tested as single larvaAS 0.024 0.031 0.057 0.086 -0.58 0.5609 2.03Slow 1 0.249 0.122 0.236 0.136 0.58 0.5609 1.6826Slow 2 0.303 0.128 0.244 0.151 1.04 0.2977 2.3814Short CS 0.021 0.011 0.014 0.011 2.23 0.0256 0.3077J turn 0.057 0.03 0.066 0.085 0.8 0.4249 2.9742HAT 0.132 0.06 0.119 0.073 0.66 0.5075 2.4083Routine turn 0.178 0.076 0.184 0.098 -0.04 0.9676 1.1217SAT 0.024 0.04 0.046 0.046 -2.93 0.0033 0.0434Burst swim 0.001 0.001 0.001 0.002 2.04 0.0416 0.4575Long CS 0.003 0.005 0.003 0.006 0.7 0.4817 2.6071O bend 0.002 0.002 0.004 0.006 -1.37 0.1717 1.5454LLC 0.006 0.008 0.025 0.055 -1.56 0.1198 1.1984SLC 0.001 0.001 0.001 0.002 0.78 0.4345 2.9742

Table S5: Statistical comparison for data shown in Fig S2 A comparing raising conditionsfor social and non-social testing conditions. Statistical comparison of difference betweengroup and isolation raised larvae in the overall bout type Pb of all 13 bout types. N = 20for both group raised and isolation raised data, tested in pairs or as single larvae. Two-tailed unsigned Wilcoxon tests were applied between raising conditions for each bout typefor determining z and p values. To correct for multiple comparison, p-values were corrected(corr p) for the number of comparisons (13) using Holm’s sequential Bonferroni procedure.Mean and standard deviation (SD) are reported for each measure used for the comparison.AS: Approach Swim; Long CS: Long capture swim; BS: Burst swim; HAT: High angle turn;RT: Routine turn; SAT: Shadow avoidance turn; LLC: Long latency C-start; SLC: Short latencyC-start.

41


https://doi.org/10.1101/2020.03.02.972612


FigureRef Treatment Measure zValue p MD CIlow CIhigh N

Fig. S2BPairs

Bouts swum 4.1 <0.0001 24.78 15.43 32.91 Gr: 20Fig. S2C Bout dur. -3.02 0.0026 -13.35 -21.17 -6.15 Iso: 20Fig. S2D

AloneBouts swum 3.18 0.0015 25.41 11.89 38.75 Gr: 20

Fig. S2E Bout dur. -2.75 0.006 -17.52 -34.98 -0.13 Iso: 20Fig. S2F

Apo.Bouts swum 4.1 <0.0001 24.78 15.66 33.6 Ctr: 20

Fig. S2G Bout dur. -3.02 0.0026 -13.35 -20.62 -5.76 Apo: 20Fig. S2J

Iso3Bouts swum 6.55 <0.0001 26.22 20.69 32.14 Gr: 40

Fig. S2K Bout dur. -0.31 0.7574 -1.89 -6.11 2.23 Iso: 39

Table S6: Statistical comparison for data shown in Fig S2 B-G and J-K. Figure referenceindicates where the data is displayed. Total number of bouts swum and average bout durationwere compared between raising conditions or between control and apomorphin treatment(Apo). Two-tailed unsigned Wilcoxon tests were applied for determining z and p values. MDand CI refer to the mean difference and its confidence intervals of the difference betweengroup and isolation-raised larvae, as determined by 1000 bootstraps using the measure ofeffect size toolbox in Matlab.


Fig. S2HApo.

Avoid. area -0.61 0.5428 -0.08 -0.36 0.21 Ctr: 20Fig. S2I SLC prop. 0.74 0.457 0.02 -0.03 0.06 Apo: 20Fig. S2L

Iso3Avoid. area -3.45 0.0006 -0.29 -0.45 -0.13 Gr: 40

Fig. S2M SLC prop. -4.9 <0.0001 -0.14 -0.19 -0.09 Iso: 39

Table S7: Statistical comparison for data shown in Fig S2 H-I and L-M. Social avoidancemeasures after apomorphin treatment (Apo.) or larvae that were isolated from 3dpf untiltesting (Iso3) were compared to regular group raised control larvae. Figure reference indicateswhere the data is displayed. Two-tailed unsigned Wilcoxon tests were applied for determiningz and p values. MD and CI refer to the mean difference and its confidence intervals of thedifference between group and isolation-raised larvae, as determined by 1000 bootstraps usingthe measure of effect size toolbox in Matlab.

42


https://doi.org/10.1101/2020.03.02.972612


Figu

reR

efC

ompa

riso

nTy

pem

edia

n1

SD1

med

ian

2SD

2K

uipe

rk

pco

rrp

Fig.

3A

GrS

LCvs

IsoS

LC0.

830.

861.

220.

8820

0627

0.00

10.

004

GrL

LCvs

IsoL

LC1.

260.

911.

260.

8730

4129

0.00

50.

01G

rSLC

vsG

rLLC

0.83

0.86

1.26

0.91

2927

500.

001

0.00

4Is

oSLC

vsIs

oLLC

1.22

0.88

1.26

0.87

8249

9>

0.1

>0.

1

Fig.

3D

GrS

LCvs

IsoS

LC0.

90.

891.

010.

8618

555

>0.

1>

0.1

GrL

LCvs

IsoL

LC0.

950.

860.

90.

8254

285

>0.

1>

0.1

GrS

LCvs

GrL

LC0.

90.

890.

950.

8641

320

>0.

1>

0.1

IsoS

LCv

IsoL

LC1.

010.

860.

90.

8223

052

>0.

1>

0.1

Fig.

3G

GrS

LCvs

IsoS

LC1.

220.

841.

160.

7937

19>

0.1

>0.

1G

rLLC

vsIs

oLLC

1.32

0.81

1.33

0.71

2676

80.

05>

0.1

GrS

LCvs

GrL

LC1.

220.

841.

320.

8116

315

>0.

1>

0.1

IsoS

LCvs

IsoL

LC1.

160.

791.

330.

7160

12>

0.1

>0.

1

Fig.

S3A

GrS

LCvs

IsoS

LC1.

470.

841.

270.

7977

60>

0.1

>0.

1G

rLLC

vsIs

oLLC

1.3

0.79

1.18

0.77

1960

4>

0.1

>0.

1G

rSLC

vsG

rLLC

1.47

0.84

1.3

0.79

1564

4>

0.1

>0.

1Is

oSLC

vsIs

oLLC

1.27

0.79

1.18

0.77

1199

0>

0.1

>0.

1

Fig.

S3D

GrS

LCvs

IsoS

LC1.

070.

891.

130.

8936

543

>0.

1>

0.1

GrL

LCvs

IsoL

LC1.

170.

891.

140.

8663

758

>0.

1>

0.1

GrS

LCvs

GrL

LC1.

070.

891.

170.

8972

195

>0.

1>

0.1

IsoS

LCvs

IsoL

LC1.

130.

891.

140.

8633

006

>0.

1>

0.1

Tabl

eS8

:St

atis

tica

lco

mpa

riso

nof

angu

lar

dist

ribu

tion

sof

the

C-s

tart

prec

edin

gbo

utof

the

non-

foca

lla

rvae

.Fi

gure

refe

renc

ein

dica

tes

whe

reth

eda

tais

disp

laye

d.Tw

o-ta

iled

unsi

gned

and

sign

edW

ilcox

onte

sts

wer

eap

plie

dw

hen

com

pari

ngbe

twee

nra

isin

gco

ndit

ions

and

betw

een

bout

type

sw

ithi

nra

isin

gco

ndit

ion

(rep

eate

dm

easu

re),

resp

ecti

vely

,fo

rde

term

inin

gz

and

pva

lues

.To

corr

ectf

orm

ulti

ple

com

pari

son,

p-va

lues

wer

eco

rrec

ted

(cor

rp)

for

the

num

ber

ofco

mpa

riso

ns(4

)usi

ngH

olm

’sse

quen

tial

Bonf

erro

nipr

oced

ure.

The

com

pari

son

type

isab

brev

iate

das

the

follo

win

g;G

rLLC

and

GrS

LC:L

ong

and

shor

tla

tenc

yC

-sta

rts

perf

orm

edby

grou

p-ra

ised

larv

ae;

IsoL

LCan

dIs

oSLC

:Lon

gan

dsh

ort

late

ncy

C-s

tart

spe

rfor

med

byis

olat

ion-

rais

edla

rvae

.M

ean

1re

fers

toth

efir

stm

enti

oned

mea

sure

(i.e

.G

rLLC

whe

nw

ritt

en:

GrL

LCvs

GrS

LC)

and

likew

ise,

mea

n2

refe

rsto

the

seco

ndm

enti

oned

mea

sure

.Si

mila

rly,

SD1

and

2re

fer

toth

est

anda

rdde

viat

ion

ofth

em

easu

res.

Mat

lab’

sC

ircS

tat

tool

box

was

used

toca

lcul

ate

Kui

per

k.

43


https://doi.org/10.1101/2020.03.02.972612



Fig. 3BContr.

Avoid. area -2.61 0.009 -0.28 -0.47 -0.08 Gr: 31Fig. 3C SLC prop. -5.66 <0.0001 -0.19 -0.24 -0.14 Iso: 39Fig. 3E

DarkAvoid. area -1.19 0.2353 -0.06 -0.2 0.09 Gr: 27

Fig. 3F SLC prop. -4.82 <0.0001 -0.18 -0.23 -0.12 Iso: 28Fig. 3H

Neom.Avoid. area -1.08 0.2807 -0.12 -0.33 0.09 Gr: 19

Fig. 3I SLC prop. -3.02 0.0025 -0.11 -0.18 -0.04 Iso: 18Fig. S3B CuSO4

Avoid. area -0.6 0.5501 -0.07 -0.28 0.14 Gr: 27Fig. S3C SLC prop. 0.39 0.6986 0 -0.07 0.07 Iso: 28Fig. S3E

HandledAvoid. area -1.99 0.0468 -0.27 -0.5 -0.03 Gr: 20

Fig. S3F SLC prop. -3.39 0.0007 -0.14 -0.21 -0.07 Iso: 20

Table S9: Statistical comparison for data shown in Fig 3 and S3. Social avoidance mea-sures were compared between raising conditions after manipulation of the sensory input, i.e.Neomycin (Neom.) and CuSO4 treatment for ablation of the LL. Figure reference indicateswhere the data is displayed. Two-tailed unsigned Wilcoxon tests were applied for determin-ing z and p values. MD and CI refer to the mean difference and its confidence intervals of thedifference between group and isolation-raised larvae, as determined by 1000 bootstraps usingthe measure of effect size toolbox in Matlab.

FigureRef Stim type Measure zValue p MD CIlow CIhigh N

Fig. 4BLoom

ReactionPb 0.79 0.4278 0.030 -0.020 0.08 Gr: 16Fig. 4C O-bend Prop. 0.9 0.3678 0.06 -0.04 0.14 Iso: 17Fig. 4E

CDReactionPb 2.96 0.0031 0.20 0.10 0.32 Gr: 14

Fig. 4F SLC Prop. 1.18 0.2384 0.11 -0.06 0.28 Iso: 15Fig. 4H

PiezoReactionPb -1.14 0.254 -0.090 -0.23 0.07 Gr: 15

Fig. 4I SLC Prop. -3.15 0.0016 -0.34 -0.48 -0.18 Iso: 15Fig. S4H Piezo ReactionPb 4.36 <0.0001 0.36 0.24 0.46 Ctr: 15Fig. S4I Neom. SLC Prop. 3.73 0.0002 0.32 0.18 0.44 Neom: 18

Table S10: Statistical comparison for data shown in Fig. 4 and Fig. S4E and F. Differencebetween group and isolation raised larvae in their escape bout reactions to looming stimuli, anapproaching dark spot (CD) and local water vibrations caused by a semi-submerged capillarytip attached to a piezo bender (piezo). Piezo neom. refers to the comparison between controlfish and neomycin treated group raised larvae for ablation of the LL. Figure reference indicateswhere the data is displayed. Two-tailed unsigned Wilcoxon tests were applied for determiningz and p values. MD and CI refer to the mean difference and its confidence intervals of thedifference between group and isolation-raised larvae, as determined by 1000 bootstraps usingthe measure of effect size toolbox in Matlab.

44


https://doi.org/10.1101/2020.03.02.972612


Figu

reR

efSt

imty

peC

ompa

riso

nTyp

em

ean

1SD

1m

ean

2SD

2zV

alue

pco

rrp

N

Fig.

S4D

Cha

sing

dot

GrS

LCvs

IsoS

LC39

0031

7.25

3982

.65

347.

75-0

.30.

7652

0.76

52G

rSLC

:14

GrL

LCvs

IsoL

LC29

15.4

213

51.9

835

28.1

510

56.7

1-1

.13

0.26

020.

7807

IsoS

LC:1

4G

rSLC

vsG

rLLC

3900

317.

2529

15.4

213

51.9

8N

aN0.

0269

0.10

74G

rLLC

:12

IsoS

LCvs

IsoL

LC39

82.6

534

7.75

3528

.15

1056

.71

NaN

0.32

030.

7807

IsoL

LC:1

2

Fig.

S4E

Piez

o

GrS

LCvs

IsoS

LC60

.41

50.2

122.

1981

.72

-2.2

80.

0229

0.06

86G

rSLC

:14

GrL

LCvs

IsoL

LC15

1.1

137.

9912

7.64

56.0

3-0

.25

0.80

290.

8867

IsoS

LC:1

4G

rSLC

vsG

rLLC

60.4

150

.215

1.1

137.

99N

aN0.

0009

0.00

34G

rLLC

:15

IsoS

LCvs

IsoL

LC12

2.19

81.7

212

7.64

56.0

3N

aN0.

4434

0.88

67Is

oLLC

:10

Fig.

S4F

Aco

usti

c

GrS

LCvs

IsoS

LC11

.69.

7516

.88

25.5

9-1

.02

0.30

570.

6115

GrS

LC:1

7G

rLLC

vsIs

oLLC

64.8

127

.27

60.5

69.

58-0

.38

0.70

350.

7035

IsoS

LC:1

9G

rSLC

vsG

rLLC

11.6

9.75

64.8

127

.27

NaN

*0.

0313

0.12

5G

rLLC

:11

IsoS

LCvs

IsoL

LC16

.88

25.5

960

.56

9.58

NaN

*0.

0313

0.12

5Is

oLLC

:9

T abl

eS1

1:C

ompa

riso

nof

C-s

tart

late

ncy

afte

rst

imul

uspr

esen

tati

onsh

own

inFi

g.S4

DF.

Figu

rere

fere

nce

indi

cate

sw

here

the

data

isdi

spla

yed.

Two-

taile

dun

sign

edan

dsi

gned

Wilc

oxon

test

sw

ere

appl

ied

whe

nco

mpa

ring

betw

een

rais

ing

cond

itio

nsan

dbe

twee

nbo

utty

pes

wit

hin

rais

ing

cond

itio

n(r

epea

ted

mea

sure

),re

spec

tive

ly,f

orde

term

inin

gz

and

pva

lues

.To

corr

ect

for

mul

tipl

eco

mpa

riso

n,p-

valu

esw

ere

corr

ecte

d(c

orr

p)fo

rth

enu

mbe

rof

com

pari

sons

(4)

usin

gH

olm

’sse

quen

tial

Bonf

erro

nipr

oced

ure.

The

com

pari

son

type

isab

brev

iate

das

the

follo

win

g;G

rLLC

and

GrS

LC:L

ong

and

shor

tlat

ency

C-s

tart

spe

rfor

med

bygr

oup-

rais

edla

rvae

;Is

oLLC

and

IsoS

LC:L

ong

and

shor

tlat

ency

C-s

tart

spe

rfor

med

byis

olat

ion-

rais

edla

rvae

.Mea

n1

refe

rsto

the

first

men

tion

edm

easu

re(i

.e.G

rLLC

whe

nw

ritt

en:G

rLLC

vsG

rSLC

)and

likew

ise,

mea

n2

refe

rsto

the

seco

ndm

enti

oned

mea

sure

.Sim

ilarl

y,SD

1an

d2

refe

rto

the

stan

dard

devi

atio

nof

the

mea

sure

s.*N

ote

that

for

the

repe

ated

mea

sure

com

pari

son

betw

een

C-s

tart

bout

type

sw

ithi

nra

isin

gco

ndit

ion

itw

asno

tpo

ssib

leto

calc

ulat

ez-

valu

esdu

eto

insu

ffici

ent

sam

ple

size

s.Th

esa

mpl

esi

zes

may

diff

erfr

omth

eto

taln

umbe

rof

anim

als

test

ed,s

ince

only

anim

als

who

perf

orm

eda

C-s

tart

duri

ngth

est

imul

uspr

esen

tati

onco

uld

beta

ken

into

cons

ider

atio

n,se

eFi

g.4

E-F

and

H-I

.

45


https://doi.org/10.1101/2020.03.02.972612


Early-life social experience shapes social avoidance ... › content › 10.1101 ›...

Documents

Transcript of Early-life social experience shapes social avoidance ... › content › 10.1101 ›...