Behavioural responses of threespine stickleback with lateral line asymmetries to experimental mechanosensory stimuli

Behavioural laterality, the preferential use of one side of the body for a given task (Rogers et al., 2013), is widespread among vertebrates (Malashichev, 2006; Miletto Petrazzini et al., 2020; Rogers, 2002) and bilateral invertebrates (Frasnelli, 2013; Frasnelli et al., 2012). By specializing in the use of one side of the body, individuals can perform better in tasks such as object discrimination (Güntürkün et al., 2000; Matrai et al., 2019; Mehlis-Rick et al., 2018), predator evasion (Dadda et al., 2010; Heuts, 1999; Lippolis et al., 2002) and object manipulation (Magat and Brown, 2009; McGrew and Marchant, 1999), giving them a fitness advantage (Vallortigara and Rogers, 2020). Additionally, social individuals can influence the side preference of others (Bisazza and Dadda, 2005; Deng and Rogers, 2002; Karenina et al., 2018; Roux et al., 2016), e.g. mammalian mothers tend to orient with the mother on the left and the infant on the right (Karenina et al., 2017). Whether population laterality – a bias of the majority of individuals in a population to one side – arises through social or non-social mechanisms, it may incur a competitive advantage in interspecific interactions (Frasnelli and Vallortigara, 2018; Ghirlanda and Vallortigara, 2004; Lehman, 1981).

Behavioural laterality can be associated with morphological differences between the two hemispheres of the brain, in which one hemisphere becomes specialized for a given task (Rogers, 2000; Rogers and Andrew, 2002). This morphological asymmetry can improve the efficiency of neural tissue use (Levy, 1977), enable parallel processing (Rogers et al., 2004) and help resolve conflicting sensory information between the two sides (Vallortigara, 2000), all of which may have ramifications for behaviour. There is also a growing body of evidence suggesting that asymmetry in sensory structures influences behavioural laterality (Anfora et al., 2011; Fernandes et al., 2018; Hart et al., 2000; Krings et al., 2019; Lychakov et al., 2006, 2008; Werner and Seifan, 2006) and may play a role in determining which side becomes behaviourally dominant. The ubiquity of subtle morphological asymmetries across taxonomic groups (Clarke et al., 2000; Fey et al., 2020; Hart et al., 2000; Trokovic et al., 2012; Werner and Seifan, 2006) suggests that sensory receptor asymmetry may be a widespread mechanism for the development of behavioural laterality.

This study aimed to determine whether recently documented asymmetry in the lateral line of threespine stickleback (Gasterosteus aculeatus) influences behaviour. The lateral line is a mechanosensory organ composed of a series of neuromasts that detect displacement of the surrounding water over the body, either by acceleration-sensitive canal neuromasts or by velocity-sensitive superficial neuromasts (Coombs et al., 2014). Threespine stickleback only have superficial neuromasts (Wark and Peichel, 2010) and therefore are expected to be sensitive to velocity but not acceleration. The lateral line is important for rheotaxis in stickleback (Jiang et al., 2017) and is used by many species of fish during nocturnal foraging (Montgomery and Milton, 1993; Pohlmann et al., 2004; Schwarz et al., 2011); however, investigations of nocturnal foraging in stickleback have predominantly focused on olfaction (Baer et al., 2021; Mussen and Peeke, 2001). Threespine stickleback from ecologically distinct habitats differ in the number of neuromasts on their head and trunk (Ahnelt et al., 2021; Planidin and Reimchen, 2019; Wark and Peichel, 2010) as well as in the bilateral asymmetry of neuromast numbers along their trunk (Planidin and Reimchen, 2021). Stickleback can also exhibit directional asymmetry in bony structures – e.g. lateral plate numbers in a population exposed to asymmetric predator–prey interaction geometry (Bergstrom and Reimchen, 2002, 2003; Reimchen, 1997; Reimchen and Bergstrom, 2009; Reimchen and Nosil, 2001a,b) – and undergo extensive scaring and abrasion throughout their life history due to failed predation events and nest construction in males (Bergstrom and Reimchen, 2003; Reimchen, 1988). These aspects of stickleback ecology suggest that interactions between morphological asymmetry and behavioural laterality may have important functional consequences during their life history, regardless of the mechanism that produces asymmetry in the sensory structure.

Population-level laterality differs greatly among behaviours and contexts (Basile et al., 2009; Bisazza et al., 1997, 1998, 1999; Blois-Heulin et al., 2012; De Santi et al., 2001; Domenici et al., 2014; Lai et al., 2015; McLean and Morrell, 2021; Poyser et al., 2006; Yeater et al., 2014); therefore, testing a single population in different contexts will provide a more complete understanding of its laterality and how it relates among behaviours. To sample individual behaviour from a wide degree of contexts, we assessed behavioural laterality in simulations of three different behaviours: predator evasion, rheotaxis and prey localization by mechanosensation. These are known in other species to be associated with lateral line morphology or neuromast count (Yoshizawa et al., 2010; Olszewski et al., 2012; Jiang et al., 2017), making them candidate behaviours for testing the interaction between sensory structure asymmetry and behavioural laterality. Following the testing of each individual four times for each behaviour, we compared lateral line asymmetry with behavioural laterality. By assessing laterality in multiple kinematic metrics during these behaviours, we aimed to understand the degree to which individuals express an overall side preference and the relationship between sensory receptor asymmetry and laterality among behavioural contexts.

We collected 77 adult threespine stickleback (Gasterosteus aculeatus L.) from Eagle's Lake, BC, Canada (latitude: 48.5088, longitude: −123.4633) during multiple trips to the field in September of 2021, after the reproductive season for this population. They were captured using minnow traps and transported to the University of Victoria Aquatics Facility in aerated 19 l buckets. All stickleback were housed together in a single 189 l tank at 15°C, with artificial habitat enrichment, on a 12 h:12 h light:dark cycle, and fed on a diet of bloodworms once daily. Fish were haphazardly selected from the housing tank by dip net and transported to the experimental tanks in an aerated 19 l bucket.

We tested the evasion of a simulated predator attack, hereafter referred to as predator evasion, by releasing a stream of water adjacent to stickleback in a circular arena (Fig. S1A, Movie 1). Arenas consisted of 19 l buckets with the base cut off, mounted on Plexiglas sheets with silicon. We inserted a mesh 2 cm from the edge of the bucket to allow us to provide stimuli from both sides, even if the stickleback was against the side of the arena (Fig. S1A). We placed a GoPro Hero4 and SJcam4000 below the left and right tanks, respectively (from the observer's perspective), recording 1280×720 footage at 60 frames s⁻¹. We mounted a white sheet above the tanks to provide a consistent background, and after the first four groups, we mounted a mirror at 45 deg above each tank to see the fish without looming over the tank. No behavioural metrics were significantly different between trials conducted with and without the mirror (all χ²₁≤2.43, P≥0.119), so their results were pooled for analysis.

We placed fish in the apparatus, and allowed them to acclimate for 5 min. Following acclimation, we dropped 10 ml of water from a stick-mounted turkey baster from a height of 18 cm. We bored a small hole into the baster and attached a rubber hose, so water was consistently released in a continuous stream by removing one's thumb from the top of the hose (Fig. S1A). A ‘drop’ was repeated every 90 s for 11 drops, with the sixth drop being a control with no water released. We alternated positioning between sides, with the first side counterbalanced within pairs and distance haphazardly determined, usually placed about 2 cm from the fish's midsection. The leftmost tank always received the drop first.

We initially tested 10 stickleback to determine the best frequencies for assessing vibration attraction behaviour (VAB), the tendency for fish to approach an oscillating glass rod in the absence of light, first described by Yoshizawa et al. (2010). The VAB tank consisted of the midsection of a 19 l bucket affixed to a piece of Plexiglas with silicone. The tank was placed on top of cinderblocks, separated by vibration-absorbing foam, illuminated with an ITT IR Illuminator (850 nm) and viewed by an ITT mini Monocular NIGHT-VISION SCOPE mounted onto the lens of a GoPro Hero4, taking a time-lapse video at 0.667 images s⁻¹ (Fig. S2A, Movie 2). Oscillations were generated by a Speaker Craft MTR1C, connected via a Yamaha RS-V395 receiver and a Scarlet 2i2 audio interface to a MacBook Pro, generating tones with the oscillator plugin within Ableton Live 9. Sound pressure was converted to mechanical displacement by affixing a glass rod to the speaker diaphragm dust cap, which we centred above the tank. We filled the tank with 6 cm of water and extended the glass rod 1 cm into the water.

We placed a plastic lid on the tank and turned off the room's lights for 5 min prior to testing to allow the stickleback to acclimate to the setup. The infrared light shone through the plastic lid. Each stickleback was exposed to 10 Hz intervals from 20 to 100 Hz for 3 min in random order and 3 min of silence in between stimuli. We selected this frequency range because our speaker was unable to produce oscillations lower than 20 Hz without distortion and as superficial neuromasts are unlikely to be sensitive to stimuli above 100 Hz (Coombs et al., 2014). Following the fifth trial, we refreshed the water, and gave the stickleback another 5 min to re-acclimate.

We calibrated the amplitude and frequency of oscillations with fast Fourier transform analysis (R library GeneCycle; https://CRAN.R-project.org/package=GeneCycle) of 240 frames s⁻¹ video footage of the glass rod, verifying the rod was oscillating at the same frequency as the speaker and adjusting speaker volume so that displacement was 3 mm for all frequencies: 20 and 60 Hz elicited similar numbers of approaches and more than other frequencies, so they were both used in subsequent testing.

We tested the laterality of VAB using the same setup as for preliminary testing; however, we doubled up the apparatus after testing the first four groups (Fig. S2A). Stickleback were acclimated to the dark for 6 min, with behaviour recorded after the first 3 min as a control, followed by 3 min of exposure to either a 20 or 60 Hz oscillation. We then repeated the same procedure with the other frequency. We initiated video capture at the beginning of acclimation, and the room was kept dark throughout each VAB trial. No behavioural metrics were significantly different after doubling up the testing apparatus (all χ²₁≤3.40, P≥0.065), so their results were pooled for analysis.

We assessed rheotaxis behaviour in two circular tracks, with water flowing clockwise or counterclockwise (Fig. S3A, Movie 3). We placed two Sicce Mi-Mouse circulation pumps in a reservoir shared by both tanks, feeding into each track by a rubber hose, and we cut an outlet hole out of each track 90 deg upstream of the flow source and covered it with a fine mesh. We mounted a GoPro Hero3+ above each track and recorded a time-lapse video at 0.667 images s⁻¹. We filled the tracks to 6 cm depth and calibrated flow to average 0.08 m s⁻¹ in each track (∼0.2 m s⁻¹ at the outer edge and ∼0–0.04 m s⁻¹ at the inner edge), similar to the range used by Jiang et al. (2017), which was based on flow rate from a natural stream containing stickleback. We placed stickleback in each track and allowed them to acclimate to still water for 5 min. Then they were acclimated to flow for 5 min, and we then recorded 5 min of behavioural footage. We initiated video capture at the beginning of acclimation, and we started the pumps with a switch hidden from view so as not to disturb the fish's behaviour.

We tested 40 fish for each behaviour in four repeat trials in a counter-balanced design (example in Fig. 1; for details see Table S1). We conducted two trials each day for 2 days, with at least a 4 h rest period in between. We tested fish in groups of four (two pairs). We always tested VAB first; if the first pair received 20 Hz then 60 Hz, the other pair would receive 60 Hz then 20 Hz. We reversed the order of stimuli in the subsequent trial, and the day's sequence would be reversed on the second day. We tested one pair for rheotaxis first, with one receiving clockwise flow and the other counterclockwise flow. In the next trial, we reversed the order and tested rheotaxis after predator evasion. The next day we tested flow regimes in the opposite order and the alternative sequence relative to predation evasion testing. We did the same alternation for placement in the two predator evasion arenas and the first side to receive a drop. The sequence of trials was such that each fish completed each behavioural test in each combination of orders. In between experiments, fish were individually housed in 19 l tanks with a plastic aquarium plant and polyvinyl chloride (PVC) pipe shelter, a slow stream of fresh water feeding into the tank, and a plastic cover to reduce stress. We withheld feeding during the 2 day testing period, and the same investigator conducted all behavioural trials to minimize differences among trials.

Within 24 h of behavioural testing, the lateral line of live fish was viewed using fluorescence microscopy. We stained the lateral line using 2-[4-(dimethylamine)styryl]-N-ethylpyridinium iodide (DASPEI) using the procedure from Wark and Peichel (2010). We suspended DASPEI in dH₂O to a concentration of 0.038%, then diluted this solution to 0.025% concentration with aquatics facility water. Fish could freely swim in the staining solution for 15 min, then were rinsed with fresh water and anaesthetized in 0.016% tricaine methylsulfonate (MS-222) solution until motionless and breathing shallowly. Following anaesthesia, fish were placed in a deep Petri dish containing 0.005% MS-222 solution and viewed with an Olympus SZX-ILLD2-100 fluorescence microscope, illuminated by an Olympus U-LH100HGAPO broad-spectrum UV light source and filtered by an Olympus SZX-MGF filter (excitation 460–490 nm; emission 510 nm longpass). All individuals were kept under the microscope for less than 10 min, well within the decay time of DASPEI fluorescence.

Neuromasts were counted first on the left and then on the right side. Counts were divided into continuous rows of neuromasts, hereafter referred to as stitches, as described by Wark and Peichel (2010), but the mandibular (MD) stitch was divided into two sublines (Fig. 2). We scored MD1 on the dentary and MD2 on the preopercular bone separately as they were easily separable, and the MD1 neuromasts were often abraded (Fig. 2). We also extended the main anterior trunk (Ma) stitch to include the eighth lateral plate rather than stopping at the seventh and recorded neuromast counts on the fourth to eighth lateral plates individually. Following viewing, we euthanized the fish with an overdose of 0.025% MS-222, severed their isthmus, and preserved them in 70% ethanol. Following a few days of preservation, we scored fish for lateral plate count, standard length, sex and the presence of parasites. The same investigator conducted all scoring to minimize bias.

We converted VAB and rheotaxis images to video using ImageJ macros, cropping out the clips of interest from the full footage and dividing VAB videos into control and stimuli clips. Videos were also cropped to the test tank to minimize future computation. We manually annotated predator evasion videos for the first frame each drop contacted the water surface (impact frame). We then extracted the impact frames and 11 s of footage (1 s before and 10 s after) for each drop using a Python script. We converted impact frames into a video with ImageJ and manually annotated each drop's position using DLTdv8 (Hedrick, 2008).

We annotated the position of the tip of the snout and the midsection between the pectoral fins in all videos using DeepLabCut (Mathis et al., 2018). We manually annotated the initial training frames, selecting up to 20 frames using k-fold selection from 100 predator evasion clips, 50 VAB clips and 10 rheotaxis videos. We iteratively assessed each model's tracking quality and retrained with additional ‘outlier’ training frames. We corrected errors in the tracking of the final neural networks in DLTdv8.

We corrected for camera distortion and angle by using the edge of each tank as a reference. We annotated 30 points around the perimeter of each tank for each video and fitted an ellipse to these points; then, we compressed the fish's positions along the major axis of the ellipse to the length of the minor axis (see Fig. S1B).

For predator evasion videos, we divided drops into those on the left and right sides and calculated velocity for each frame (see Fig. S1B for equations). As velocity data are noisy, we used a rolling average with a window size of five frames to calculate maximum velocity. A fish was classified as having initiated an escape response if it reached at least 0.2 m s⁻¹. We chose this threshold as it matched the proportion of C-start escapes observed in previous predator evasion experiments. We also tabulated time to reach 0.2 m s⁻¹ (EscTime) using the rolling average velocity.

For VAB videos, we calculated distance travelled as the difference in position of the snout between frames, and angles were calculated relative to the position of the glass rod (see Fig. S2B for equations). An ‘approach’ occurred when the stickleback got within 5 cm of the glass rod (roughly one body length), with a 3-frame buffer in between approaches to prevent double-counting from noise in position data. Approach side and wall side (the side facing the outer wall of the tank most often) were categorized as left or right, as both were highly bimodal.

As in Jiang et al. (2017), we looked at four metrics of rheotaxis behaviour; net displacement (+upstream, −downstream), cumulative upstream movement, time oriented upstream and flow regime selection (distance from outer edge) (see Fig. S3B for equations). However, we calculated displacement as the change in angle around the track and used distance from the outer edge, rather than calculating the proportion of the fish's movement tangential to the track and partitioning flow regime selection into bins.

We assessed four aspects of predator evasion behaviour for laterality: the side against the outer wall of the arena, whether escape response was initiated, EscTime and maximum velocity (see Table 1 for data transformations and error structures). We tested the effect of neuromast count [NC=Σ(R_s+L_s)] and directional neuromast count asymmetry [DA=Σ(R_s−L_s)] on laterality of wall side selection using backwards model selection by Wald's Chi-square test (α=0.05) with Eqn 1 and all other behaviours with Eqn 2 (Table 1; Murtaugh, 2009). R_s and L_s are the number of neuromasts on the right and left sides of stitch S, respectively. Previous analyses (Planidin, 2021) found no significant effect of sex, standard length and absolute neuromast count asymmetry [AA=Σ(|R_s−L_s|)], lateral plate count and lateral plate asymmetry, so we did not include these traits in the analysis (see Table S2 for a summary of population morphology). Group was not included as a random effect as its variance converged to zero and resulted in a singular fit for most models. We included drop number as a continuous predictor as differences among drop numbers were approximately linear for all behaviours. If NC or DA significantly influenced escape behaviour, we substituted the sum over all stitches with each individual stitch in the reduced model equation, e.g. if there was a significant effect of DA on EscTime, we would model EscTime using R−L neuromast counts for each stitch individually plus any other factors that had a significant effect on EscTime. P-values from individual stitch analysis were Bonferroni corrected (n=13) to account for multiple comparisons. Models were fitted with either the lme4 (Bates et al., 2015) or glmmTMB (Brooks et al., 2017) packages within R 3.6.3 (http://www.R-project.org/; see Table 1).

We tested two aspects of VAB behaviour for the effect of 20 and 60 Hz stimuli: distance travelled and the number of approaches; and two metrics of laterality: wall side and approach side (see Table 1 for data transformations and error structure). We compared non-lateralized and lateralized behaviours with morphological characteristics using Eqns 3 and 4, respectively (Table 1). Analysis of lateralized and non-lateralized behaviours had to be separated for VAB analysis as laterality could not be encoded as an interaction effect, e.g. drop side in the predator evasion experiment. We tested individual stitches the same way as with predator evasion.

We examined four metrics of rheotaxis behaviour for laterality (see Table 1 for data transformations and error structure). We tested the effect of morphology on lateralized and non-lateralized behaviour using Eqn 7 (Table 1). All models underwent the same backwards model selection procedure, and we tested individual stitches in the same way as predator evasions and VAB trials.

We tested the consistency of non-lateralized and lateralized behaviours across the four repeat trials using the rptR package (Stoffel et al., 2017). We included trial as a fixed effect and individual ID as the random effect for which repeatability was assessed in all rptR models. For the predator evasion experiment, we tested the repeatability of initiating predator evasion, wall side selection and the proportion of escapes initiated in response to drops from the right as binary characteristics. We modelled EscTime and maximum velocity as Gaussian and did not calculate the repeatability of laterality for these behaviours. For the VAB experiment, we tested the repeatability of distance travelled and the number of approaches with Gaussian and Poisson models, respectively. We tested VAB laterality by modelling the number of right-hugging 3 min intervals and right side approaches as proportions. We modelled all rheotaxis behaviours as Gaussian and did not test for repeatability of rheotaxis laterality. We tested correlation in laterality among behaviours exhibiting the greatest laterality with the psych package (Revelle, 2020) after summing or averaging the behaviour of each individual.

This research was undertaken and approved through the University of Victoria Animal Use for Research Protocol 2019-021(1), which follows the guidelines set by the Canadian Council for Animal Care. All individuals were kept in captivity for a maximum of 3 months. They were housed in the same tank and allowed to school in an enriched habitat. All individuals were euthanized as part of experimental procedure.

Statistical output of model selection for all behaviours is presented in Table 2.

Fish ‘hugged’ the arena wall with their right side in 56.7% of drops and hugged the left in 43.3% of drops. Fish with fewer neuromasts were more likely to hug the right wall (Fig. 3A), and the preference for hugging the right wall was greatest during trial a (Fig. 3B). Twelve of thirteen stitches negatively correlated with a bias towards right wall hugs, and the main posterior trunk (Mp) stitch had the strongest and only significant correlation (P_adj=0.013; β+s.e.=−0.37+0.11; Table S3).

Fish never initiated an escape response during control drops, but 6% of escape responses were initiated before drop impact during drops where water was released. The rate of escape behaviour initiation was highly variable among individuals, ranging from individuals that never initiated an escape response to others which initiated an escape response more than half of the time (Fig. 4A). On average, an escape response was initiated 9.5% [95% confidence interval (CI): 6.6%, 13.3%] of the time. Escape was initiated less frequently in subsequent drops within trials (log-odds+s.e.=−0.25+0.03) and in later trials (Fig. 4C). Fish initiated escape behaviour 10.7% [95% CI: 7.6%, 14.7%] of the time when we placed the drop on the right and 8.2% [95% CI: 5.7%, 11.6%] of the time when the drop was on the left. Fish with more neuromasts initiated an escape response less frequently, with a stronger relationship for drops on the left side (Fig. 4A); however, there was a wide range of response rates for fish with any given neuromast count. Escape responses occurred more frequently with the arena wall on the left and the drop on the right, relative to other wall and drop positions (Fig. 4B). There was a right bias in escape response in all trials except trial d (Fig. 4C). The fish's probability of initiating an escape response did not differ with their DA. Neuromasts counts in 12 of 13 stitches had a negative association with escape behaviour, and 11 of 13 had a stronger negative association when drops came from the left side, but there were no significant individual stitches (Table S3).

Average EscTime was 1.2 s [95% CI: 0.98 s, 1.49 s] and ranged from 0.02 to 9.92 s for escape responses initiated after impact. EscTime increased in later trials [log(EscTime)+s.e.: trial a=0.77+0.04; trial b=1.09+0.05; trial c=1.14+0.06; trial d=1.21+0.07; Table 2] and by log(EscTime)+s.e.=0.015+0.07 per drop. EscTime did not differ with wall side, neuromast count or DA, and no predictors had a significant interaction with drop side.

Maximum velocity was 0.39 m s⁻¹ [95% CI: 0.34 m s⁻¹, 0.44 m s⁻¹] on average, ranged from 0.2 to 2.2 m s⁻¹ and was highly variable both within and among individuals (Fig. 5A). Maximum velocity decreased in later trials (Fig. 5C) but did not change with drop number. Fish with more neuromasts reached a lower maximum velocity on average (Fig. 5A), and fish with more neuromasts on the right reached a higher maximum velocity on average (Fig. 5B), with a relatively small effect size compared with the range of maximum velocities observed. Maximum velocity did not differ with drop side or wall side, and no interactions with drop side were significant. Increased neuromast count in 12 of 13 stitches had a negative association with maximum velocity, with significant associations from the supratemporal (ST) (P_adj=0.046; β+s.e.=−0.10+0.04) and Ma (P_adj=0.007; β+s.e.=−0.13+0.04) stitches (Table S3). The effect of DA across stitches was inconsistent, with a right bias in eight of 13 lines increasing maximum velocity, including the significantly associated infraorbital (IO) stitch (P_adj=0.021; β+s.e.=−0.11+0.04; Table S3).

In each VAB video, 2.2 approaches [95% CI: 1.9, 2.6] were made on average. There were fewer approaches after trial a [log(approaches)+s.e.: a=1.23+0.10; b=0.63+0.11; c=0.77+0.11; d=0.63+0.11]. Fish travelled 2.24 m [95% CI: 1.77 m, 2.80 m] per 3 min VAB video on average. However, the distance travelled was much higher during trial a relative to other trials [log(distance travelled)+s.e.: a=1.70+0.05; b=1.24+0.05; c=1.20+0.05; d=1.10+0.05]. Neither the 60 Hz stimulus nor the 20 Hz stimulus induced greater travel distance or number of approaches when compared with controls (all t_437–622≤1.67, P≥0.218). Neuromast count did not affect distance travelled or number of approaches. Oscillatory stimuli did not interact with neuromast count for distance travelled or number of approaches.

Fish tended to orient themselves with the wall on the right more often than the left (proportion wall on right+s.e.=0.56+0.02). Fish with few neuromasts were more likely to orient with the wall on their left, whereas fish with many neuromasts were more likely to orient with the tank wall on their right, with a wide degree of intra-individual variation (Fig. 6). DA did not affect positioning. Nine of the 13 stitches were associated with an increase in the number of right wall hugs, including the significantly associated ethmoid (ET) (P_adj=0.029; β+s.e.=0.45+0.15) and Ma (P_adj=0.023; β+s.e.=−0.41+0.13) stitches (Table S3).

The proportion of left and right approaches to the centre was the same (proportion right approaches+s.e.=0.522+0.19). Fish with more neuromasts tended to approach the centre from the left more frequently (log-odds+s.e.=−0.277+0.151), and DA had no effect.

The position of fish after the rheotaxis trial was on average 0.7 m [95% CI: 0.23 m, 1.65 m] downstream from their initial position. There was less net downstream displacement in trial b compared with all other trials (Fig. 7A). The average distance from the outer edge was 4.70 cm [95% CI: 4.32 cm, 5.08 cm]. Fish tended to position closer to the outer edge in trial a than in other trials (Fig. 7B). Fish faced upstream 78.5% [95% CI: 75.0%, 81.6%] of the time, which did not vary among trials. Cumulative upstream movement was 3.54 m [95% CI: 2.72 m, 4.59 m] on average. Upstream movement was greatest in trial a (Fig. 7C).

Average distance from the outer edge of the tank was greater when water flow was clockwise (clockwise: 5.11 cm [95% CI: 4.73 cm, 5.48 cm]; counterclockwise: 4.48 cm [95% CI: 4.11 cm, 4.86 cm]). Neuromast count and DA did not affect distance from the outer edge or interact with flow direction. Net downstream displacement, cumulative upstream movement and time facing upstream were not influenced by neuromast count or DA, including interactions with flow direction.

During predator evasion trials, the probability of escape (R=0.08 [95% CI: 0.02, 0.11], P<0.001) and maximum velocity were repeatable (R=0.25 [95% CI: 0.09, 0.45], P< 0.001) but EscTime was not (R=0.03 [95% CI: 0, 0.16], P=1). Laterality of escape probability (R=0.0 [95% CI: 0, 0], P=1) and wall side selection (R=0.0 [95% CI: 0, 0] P=1) were not repeatable. During VAB trials, distance travelled (R=0.33 [95% CI: 0.20, 0.44], P<0.001) and number of approaches (R=0.19 [95% CI: 0.04, 0.23], P<0.001) were repeatable; however, wall side selection (R=0.0 [95% CI: 0, 0.01], P=0.5) and approach laterality (R=0.0 [95% CI: 0, 0.01], P=0.5) was not repeatable. No rheotaxis behaviours were repeatable, i.e. net displacement (R=0.10 [95% CI: 0, 0.29], P=0.34), cumulative upstream movement (R=0.09 [95% CI: 0, 0.29], P=0.41), time oriented upstream (R=0 [95% CI: 0, 0.17], P=1) and distance from the outer edge (R=0.12 [95% CI: 0, 0.33], P=0.26).

Laterality of individuals during predator avoidance, VAB and rheotaxis trials were not correlated. However, the laterality of wall side selection was correlated with escape laterality, and wall selection during VAB trials was strongly correlated with approach side (Table 3).

We found associations between neuromast count and laterality in multiple behaviours. When exposed to an abrupt perturbation of the water simulating a predatory strike, stickleback were more likely to rest with their left side facing the arena wall between drops, and they were more likely to initiate escape behaviour when the stimulus came from the right. Furthermore, stickleback with few neuromasts were more likely to initiate an escape response when the stimuli came from the left side. Stickleback with more neuromasts escaped more slowly, and stickleback with more neuromasts on the right had higher maximum escape velocity. Stickleback did not display VAB towards 20 or 60 Hz stimuli. Despite the lack of VAB, stickleback tended to face the arena's outer wall with their right side, with a stronger right bias in stickleback with more neuromasts. Variance in neuromast count was not associated with rheotaxis behaviour; however, stickleback occupied a slower flow regime when in clockwise flow. No individual stitches stood out as a consistent predictor of predator evasion, VAB or rheotaxis behaviour; rather, most stitches tended to have a similar effect on behaviour. No lateralized behaviours were repeatable, nor was time to reach 0.2 m s⁻¹ (EscTime) or any rheotaxis metrics. Many behaviours changed over the course of experimentation, as fish tended to be less active and less lateralized in later trials.

We observed laterality in responses to a simulated predator strike. Stickleback were most likely to respond to drops on the right; however, they were also more likely to place their right side against the arena wall. The preference to expose the left side and respond to stimuli from the right may suggest that the stickleback were more effective at determining that the simulated predatory stimulus was not a true threat when the stimuli came from their left. The notion that initiating escape behaviour was due to a misidentification of the mechanosensory stimuli is further supported by fish with fewer neuromasts responding more often and more vigorously to stimuli. Greater responsiveness to stimulation of neuromasts on the left side would explain why neuromasts influenced escape probability more when drops came from the left and the reduced intensity of response when individuals have more neuromasts on the left. Interestingly, a left bias in neuromasts is also associated with feeding behaviour in blind Mexican cavefish (Fernandes et al., 2018). However, the trend we observed differs from the most directly comparable study of predator evasion laterality, in which lungfish were more likely to escape to the left in response to a vertically plunging model predator (Lippolis et al., 2009). The direction and extent of laterality in response to predators is generally variable, differing among species (Heuts, 1999) and life history stages (Cantalupo et al., 1995). There is also disagreement among studies whether lateralized fish initiate escape behaviour faster (Agrillo et al., 2009; Dadda et al., 2010). Overall, there does not seem to be a consistent population-level side bias in escape behaviour among fishes, a trend which occurs in other potentially lateralized behaviours (Bisazza et al., 1997; Bisazza et al., 2000; Sovrano et al., 2001) and across taxa (Jozet-Alves et al., 2012; Lippolis et al., 2002; Lippolis et al., 2005; Romano et al., 2017). It has been suggested that the wide degree of variability in morphological and behavioural laterality among species is adaptive and highly plastic (Hori et al., 2017; Vallortigara et al., 1999), but it is unclear whether this is true of predator evasion laterality. Methodological differences in studies of predator evasion laterality may be inflating observed differences among species and obfuscating true and potentially adaptive differences among species.

The association between escape behaviour and neuromast count that we observed may be influenced by visual sensory information. Despite using water to minimize predator-like visual stimuli during testing, visual and mechanosensory information were present during escape response trials, and at least 6% of escape responses were entirely visually mediated, i.e. initiated prior to impact at the water's surface. Irregularities in the posture of the observer and reflections from the bottom of the tank or water surface also may have introduced variability into the perceived position of the predatory stimuli, reducing the power of our experimental observations. During preliminary testing, we were unable to elicit escape responses from stickleback in the dark. Similarly, goldfish rarely initiate escape responses to only auditory stimuli; however, the integration of auditory and visual stimuli increases their escape response beyond what would be expected if the two sensory modalities were operating independently (McIntyre and Preuss, 2019). The integration of both mechanosensory and visual information in the initiation of an escape response affects our interpretation of side dominance. Innervation of Mauthner cells by mechanosensory afferents is predominantly ipsilateral (Mirjany and Faber, 2011), whereas innervation of the optic tectum (Schwassmann and Kruger, 1965) and nucleus isthmi (Northmore, 1991) by the optic nerve is predominantly contralateral. These opposing innervation patterns mean that an individual with a dominant right brain hemisphere would have a dominant left eye and or a dominant right side of the lateral line, putting laterality in the two sensory modalities at odds. Furthermore, both modalities show evidence of integration of both ipsilateral and contralateral stimuli (Gebhardt et al., 2019; Mirjany and Faber, 2011), suggesting that laterality in these sensory modalities cannot be explained by simple dominance of function in one side of the brain. Laterality in different sensory modalities can be independent (McGreevy and Rogers, 2005; Oltedal and Hugdahl, 2017) or related (Harshaw et al., 2021). To understand the role that sensory structures play in the development of a lateralized escape response, we require a detailed understanding of how sensory information is integrated between the two hemispheres of the brain because of the multi-sensory nature of this behaviour.

Lateralized behaviour may directly produce asymmetry in neuromast counts. During fluorescence microscopy, it was clear that the MD1 and Mp stitches often had damaged or missing neuromasts, suggesting that fish that tended to escape towards the arena's outer wall would have lost neuromasts on that side. If abrasion of the MD1 and Mp lines drove all associations between laterality and neuromast count, we would expect them to be outliers in the individual stitch analyses; however, as the MD1 and Mp stitches did not consistently drive associations between neuromast count and behaviour, abrasion is not likely to be the sole cause of the observed correlation between morphology and behaviour. Behaviourally induced receptor asymmetry may further bias the behaviour of that individual to escape with the same trajectory – positive feedback – or bias the individual to respond in the opposite direction – negative feedback – in the future. These sensory–behaviour feedback loops may be particularly important for long-lived populations of stickleback, which often incur scars and abrasions during interaction with lateralized predators (Bergstrom and Reimchen, 2003; Reimchen and Bergstrom, 2009). Asymmetry in other morphological characteristics, such as trunk muscle volume (Heuts, 1999), may also influence asymmetry in neuromasts, whether by behavioural or developmental interactions. By incorporating a feedback mechanism between sensory structures and behavioural laterality, we may better understand how lateralized behaviour changes over the life history of individuals.

Despite marine stickleback nocturnally foraging on brine shrimp (Mussen and Peeke, 2001), which produce regular oscillations by beating their legs (Lent, 1971), we found that stickleback do not exhibit VAB behaviour at the frequencies we tested. However, the primary frequency of oscillations produced by prey may be outside of the range we examined during preliminary trials (20–100 Hz), as is the case with brine shrimp, which produce oscillations between 5 and 10 Hz (Lent, 1971). There is also the possibility that reflections off the outer wall of the tank obfuscated the signal being produced by the centre rod; however, Yoshizawa et al. (2010) utilized a similar assay to demonstrate VAB in a species with a greater proliferation of neuromasts, which is likely to be more sensitive to these weaker reflections. The lack of attraction to vibrations by stickleback is not wholly unexpected, given that only cave varieties of Astyanax mexicanus exhibit this behaviour (Yoshizawa et al., 2010). While we might expect that stickleback in low light environments, e.g. stained lakes, would have a similar adaptation for finding prey in the absence of visual information, these populations typically have undergone a reduction in the number of neuromasts (Planidin and Reimchen, 2019) rather than the proliferation seen in cavefish and deep sea species (Marranzino and Webb, 2018; Marshall, 1996; Yoshizawa et al., 2010). These same low-light populations of stickleback also exhibit increased levels of asymmetry in the lateral line (Planidin and Reimchen, 2021), suggesting that morphological asymmetry in sensory structures may exert a greater influence on the laterality of these fish.

When navigating in the dark, stickleback tended to hug the wall with their right sides. This side bias is the same as in the predator avoidance test and matches the laterality of cavefish exploring novel objects in the dark (Burt de Perera and Braithwaite, 2005). The relationship between neuromast count and right-biased ‘wall hugs’ is opposite in predator evasion and VAB trials, in which the environment differed in light regime and the presence of a threat. As early light exposure (Dadda and Bisazza, 2012; George et al., 2021; Rogers and Krebs, 1996; Sovrano et al., 2016) and the presence of predatory stimuli (De Santi et al., 2000; Jozet-Alves and Hebert, 2013) can influence laterality, differences in both of these environmental factors are plausible causes for changes in the relationship between neuromast count and laterality between predator evasion and VAB experiments. Cavefish utilize mechanosensation for navigation in the dark (Holzman et al., 2014) and may have a greater potential to develop lateralized mechanosensory navigation behaviour than stickleback, as a result of having a higher density of facial neuromasts. Eagle's Lake stickleback, a population with a relatively high number of neuromasts, do not exhibit specialized mechanosensory-mediated foraging, suggesting that stickleback lack this behaviour.

Perhaps our most surprising finding is the lack of association between neuromast count and rheotaxis behaviour. Jiang et al. (2017) demonstrated that ablation of the entire lateral line in threespine stickleback affects rheotaxis behaviour and that natural variation in neuromast count within populations also correlates with rheotaxis behaviour. In contrast, we found that natural variation in neuromast count does not correlate with rheotaxis behaviour under similar flow regimes to those used by Jiang et al. (2017). There are several mechanisms through which these differences may have arisen. Firstly, habitat differences – e.g. flow regime of inlet streams – in the populations assessed may have led to divergent results. However, the range of natural variation in neuromast counts examined is comparable between studies (Jiang et al., 2017; Fig. 2). Secondly, experimental protocol may have influenced the results; for example, if there were major differences in the turbulence of the flow regime, which may influence the ability of superficial neuromasts to detect abiotic flows (Coombs et al., 2014). Jiang et al. (2017) also tested for a different kind of association, using canonical correlation analysis (CCA) to examine the multivariate interaction between natural variation in neuromast count and rheotaxis behaviour. In contrast, we looked for a linear relationship with total neuromast count. While the lateral line is important for rheotaxis in stickleback, natural variation in neuromast count within populations likely does not have a consistent linear relationship with rheotaxis.

We found that population-level laterality changed in repeat tests at different rates among behaviours. Wall hug laterality was lost after the first predator evasion trial (illuminated) but remained consistent during VAB trials (dark), and laterality in escape response initiation diminished slowly over time. Laterality differs in response to novel versus familiar stimuli across taxonomic groups (Burt de Perera and Braithwaite, 2005; Fourie et al., 2021; Hausberger et al., 2021; Regolin and Vallortigara, 1996; Robins and Rogers, 2006; Wilson and McLaughlin, 2007). Therefore, habituation to a stimulus consistently alters the laterality of related behaviours. As laterality is more prevalent during first exposures, laterality may be the default state of behaviour, and symmetry may be learned over time. The exception to this would be if there is a positive feedback mechanism supporting further lateralization. While the changes in laterality we observed cannot be viewed independently from the overall reduction in activity level in subsequent trials, our findings suggest that the valence of stimuli modulates the strength of this learning response, e.g. laterality was reduced over time when exposed to predatory stimuli but not when in the dark. Furthermore, changes in laterality between repeat tests of the same individual, whether within the same day or over the ontogeny of an individual (Bisazza et al., 1997, 1998; Burt de Perera and Braithwaite, 2005; Cantalupo et al., 1995; Sovrano et al., 2001), may emerge from habituation rather than developmental effects or genetic selection.

Our findings indicate that stickleback do not exhibit repeatable individual-level laterality across multiple behaviours. Further, individuals did not display consistent behavioural phenotypes for non-lateralized behaviours, i.e. time to initiate escape response and all of the rheotaxis assays. Furthermore, the repeatability metric we and others (Roche et al., 2020) used does not consider learning. A population that is 100% left-lateralized in the first of four trials, then exhibits randomized laterality in the following three trials generates zero repeatability. Roche et al. (2020) call into question the validity of the T-maze assay for lateralized behaviour, but the T-maze assay has typically been used as an assessment of an individual's innate laterality. Even with a highly novel stimulus, laterality may arise from learning from similar events in an individual's life history and not necessarily be truly innate. Our findings suggest that innate or previously learned laterality can only be observed in the first few encounters with a stimulus before habituation, with differences among stimuli of variable salience or novelty. This rapid habitation then raises the question of whether laterality prior to or after a novel encounter is more critical to the survival of an individual, to which the answer seems largely dependent on the specific behavioural context. In the evasion of a predator, success in the first encounter is pivotal, whereas foraging may be less important to succeed in the first attempt.

Laterality of individual sticklebacks did not correlate among any of the behavioural assays, but bias in how individuals oriented themselves during trials influenced the laterality of their subsequent behaviour. The effect of orientation on behavioural laterality suggests that environmental geometry that may bias individuals towards a certain orientation can largely determine the laterality of behaviours in the wild. This also means that temporal autocorrelation may be an important factor when determining the absolute degree of laterality. If individuals are slow moving and change their orientation infrequently, they will be more likely to exhibit consistently lateralized behaviours during testing. If individuals are fast moving and change their orientation frequently, they will be more likely to exhibit inconsistent laterality. Given that our fish's activity levels were reduced after the first trial in many of the behaviours we observed, changes in laterality in subsequent trials may be due to this mechanism.

We found evidence of laterality in the positioning of stickleback within a circular arena but no association between asymmetry in the lateral line and laterality in mechanosensory-related behaviours. Stickleback preferred to hover with the wall on their right, and neuromast count and light regime modulated this tendency. Thus, sensory structure morphology plays a role in the development of lateralized behaviour, potentially including interactions across multiple sensory modalities. We observed rapid habituation and subsequent loss of laterality across multiple behaviours, suggesting that laterality, while deeply rooted in evolutionary history (Wiper, 2017), can exhibit plasticity with respect to directionality. The structure of the lateral line in stickleback appears to have a nuanced association with their behavioural laterality, a trend which could possibly occur across sensory modalities and taxonomic groups.

We thank Mackenzie Woods and Micah Quindazzi for assistance with field work and discussion, John Taylor for the use of his fluorescence microscope, and Heather Down for assistance with microscopy.