Fish perform rapid escape responses to avoid sudden predatory attacks. During escape responses, fish bend their bodies into a C-shape and quickly turn away from the predator and accelerate. The escape trajectory is determined by the initial turn (stage 1) and a contralateral bend (stage 2). Previous studies have used a single threat or model predator as a stimulus. In nature, however, multiple predators may attack from different directions simultaneously or in close succession. It is unknown whether fish are able to change the course of their escape response when startled by multiple stimuli at various time intervals. Pacific staghorn sculpin (Leptocottus armatus) were startled with a left and right visual stimulus in close succession. By varying the timing of the second stimulus, we were able to determine when and how a second stimulus could affect the escape response direction. Four treatments were used: a single visual stimulus (control); or two stimuli coming from opposite sides separated by a 0 ms (simultaneous treatment), 33 ms or 83 ms time interval. The 33 ms and 83 ms time intervals were chosen to occur either side of a predicted 60 ms visual escape latency (i.e. during stage 1). The 0 ms and 33 ms treatments influenced both the escape trajectory and the stage 1 turning angle, compared with a single stimulation, whereas the 83 ms treatment had no effect on the escape trajectory. We conclude that Pacific staghorn sculpin can modulate their escape trajectory only between stimulation and the onset of the response, but the escape trajectory cannot be modulated after the body motion has started.
Fish avoid predators by performing sudden accelerations, i.e. fast start escape responses (Domenici and Blake, 1997). The kinematics and neural control of escape responses have been widely investigated (Domenici and Blake, 1993; Domenici and Hale, 2019; Eaton et al., 2001; Stewart et al., 2014). Fish escape responses typically consist of C- or S-starts, based on the shape of the fish at the end of the first contraction (Domenici and Hale, 2019). In C-starts, fish bend their bodies into a C-shape during the initial muscle contraction (stage 1), usually away from the threat, while a subsequent return flip of the tail (when present; Domenici and Hale, 2019) can produce further acceleration (stage 2) (Fleuren et al., 2018).
Fish can be startled using a variety of stimuli, from mechano-acoustic to tactile and visual stimuli (Domenici and Hale, 2019). The shortest latencies are typically associated with the stimulation of the mechano-acoustic sensory system, leading to the activation of the Mauthner cells (M-cells) (Korn and Faber, 2005), whereas visual stimuli tend to show longer latencies because of the longer neural pathway (Mirjany and Faber, 2011) from the optic nerve to the M-cell via the optic tectum (Temizer et al., 2015). M-cell ablation was shown to delay the escape response and to decrease survival in predator–prey encounters (Hecker et al., 2020).
Most previous studies on fish escape responses have focused on a single threat such as a model or a real predator approaching, resulting in an escape response directed away from the threat (Domenici and Blake, 1993; Stewart et al., 2013, 2014; Walker et al., 2005). However, in nature, multiple predators may attack prey from two or more directions simultaneously or in close succession (Amo et al., 2004; Bshary et al., 2006; Stander, 1992; Steinegger et al., 2018). Multiple co-occurring threats are known to affect the prey's escape directions; for example, lizards escape ∼180 deg away from single predators but perpendicular to the predators when attacked simultaneously from two opposite directions (Cooper et al., 2007).
Previous work has investigated the possibility that a modification of the escape trajectory can occur after initial stimulation. Importantly, inhibition of the mechanosensory input occurs during stage 1 in both C- and S-starts (Russell, 1976), leading Eaton et al. (1981) and Eaton and Emberley (1991) to suggest that the neural command underlying the escape response is ballistic once the movement has begun (i.e. without further sensory information to compute its trajectory). Indeed, Eaton et al. (1988) found that the stage 2 command of the goldfish is preprogrammed and not dependent on sensory feedback; however, it remains unknown whether sensory feedback can occur before or after the initiation of stage 1.
Some taxa are able to modulate escape responses to multiple successive attacks. Certain crickets, for instance, were found to use two escape modes (i.e. running and jumping) with different degrees of flexibility: when crickets escaped from an initial predator attack by running, they were able to modulate their trajectory in response to a second attack; however, this was not that case for crickets that escaped from the initial attack by jumping (i.e. a ballistic response) (Sato et al., 2019). Recent work has shown that larval zebrafish may be able to integrate sensory information from multiple threats during delayed escape responses through a cluster of 38 prepontine neurons that are not part of the fast escape neural pathway (Marquart et al., 2019). This finding suggests that during the initial escape latency (i.e. before the onset of stage 1 contraction), fish may have the potential to integrate sensory information from multiple threats. Additionally, Domenici and Blake (1993) suggested that sensory feedback may occur after the onset of stage 1, resulting in a correction of escape trajectories during stage 2. Hence, the extent of stage 2 may, at least in part, be controlled by a feedback system (Domenici and Blake, 1993).
Here, we investigated the possibility that escape kinematics may vary depending on the time difference between the two visual threat stimuli coming from opposite sides. Visual looming stimuli are known to trigger an escape response once a given threshold (that depends on the size and speed of the approaching object) is reached (Cade et al., 2020; Hein et al., 2018). We hypothesized that if the escape response is fully ballistic from the time of the first stimulation, the escape kinematics will not be modified by a second stimulus delivered at any time interval >0 ms after the first one. If, in contrast, the escape kinematics is modified by a second stimulus, this indicates that sensory feedback is possible during that time interval.
MATERIALS AND METHODS
All animal care and experimental protocols followed the guidelines of the Institutional Animal Care and Use Committee at the University of Washington, Seattle, WA, USA (Protocol No. 4238-03).
Model species and housing conditions
Pacific staghorn sculpin [Leptocottus armatus Girard 1854, mean±s.d. total length (TL) 13.9±1.71 cm; n=71] were captured by beach seining at Jackson Beach, south of San Juan Island, WA, USA (48°31′11″N, 123°0′45″W) in July 2019. The fish were maintained in two acrylic tanks (87 cm length×57 cm width×14 cm depth) with flow-through seawater under a 14 h:10 h light:dark photoperiod, at 12.5±0.5°C. They were acclimatized for ≥24 h and fed shrimp pieces every second day. At the end of the experiment, they were released at Jackson Beach.
Experiments were conducted at Friday Harbor Laboratories, University of Washington, in an acrylic fish tank (125.5 cm length×57 cm width×35 cm depth; Fig. 1A) filled with seawater at 12.5±0.5°C. White plastic panels were placed on the tank walls and bottom. A white plastic panel with grid lines (48 cm×34 cm) was placed at the bottom center of the tank. Two 300 W halogen lamps were set above the tank to illuminate it. A high-speed camera (640×360 pixels, 240 frames s−1; Stylus TG-870, Olympus Corp., Tokyo, Japan) was positioned 110 cm above the tank and recorded the escape response.
Two looming stimuli were used. Each stimulus was played on a separate screen (1600×1200 pixels, 60 Hz; Dell 2000FP, Dell Inc., Round Rock, TX, USA) placed centrally on opposing sides of the experimental tank (Fig. 1A). Each stimulus simulated a black disk (24 cm diameter) approaching from 200 cm distance at a constant velocity of 1 m s−1. The movie of the looming stimulus (1600×1200 pixels; 60 frames s−1) was created with R v.3.6.1 (http://www.R-project.org/) using the package loomeR v.0.3.0 (doi:10.5281/zenodo.3279092). To control the delay of one movie from the other, two movies were stitched together horizontally using Shotcut Video Editor v.19.07 (Meltytech LLC, Walnut Creek, CA, USA) and each side of the movie was played on a separate screen by the extended dual display mode. For the 0 ms treatment, two identical movies were stitched together and played simultaneously. For the 33 ms treatment, one of the two movies (second stimulus) was played with a delay of two frames (∼33 ms) relative to the first stimulus. For the 83 ms treatment, the second stimulus was played with a delay of five frames (∼83 ms). For the control, only a single looming stimulus was played. The side of the first stimulus was randomized. The times for the delayed stimuli (33 ms and 83 ms) were selected based on the estimated visual escape latency of L. armatus (60 ms; Paglianti and Domenici, 2006). Hence, the stimulus delayed by 33 ms was assumed to be within the escape latency of L. armatus (defined as the time interval between the stimulus-reaching threshold and the fish response, corresponding to neurosensory processing prior to the visible response; Paglianti and Domenici, 2006) (Fig. 2), whereas the stimulus delayed by 83 ms was assumed to occur during stage 1 (Fig. 2).
Each fish was transferred to the experimental tank and placed in an opaque PVC shelter (15.5 cm diameter) where it was allowed to acclimatize for 15 min. A square panel under the tank bottom was used as a placement reference, ensuring that all fish were placed in the center of the tank at a distance >1.5 body lengths away from the walls to avoid any interference with their escape trajectory (Eaton and Emberley, 1991). The fish were placed perpendicular to the stimuli by carefully rotating the PVC shelter [86.76±10.93 deg; n=66].
After acclimatization to the experimental tank, the shelter was removed and fish were left undisturbed for an additional 2 min, after which they were startled. Each fish was exposed to each of the four treatments (in random order) only once. The side of the first looming stimulus was randomly selected. Between stimuli, the fish were returned to the PVC shelter to avoid stimulation prior to each treatment. Fish were allowed at least 2 min to recover from the previous stimulation before the next trial continued. If fish moved before the stimulation, they were returned to the PVC shelter and were acclimatized for an additional 2 min. If the ventilation rate was higher than at rest (i.e. a sign of an elevated stress level) by our visual observation, extra time was allocated until the ventilation rate decreased before the next stimulus was played.
Data and statistical analysis
The 240 frames s−1 video of the escape response was analyzed frame by frame with Logger Pro v.3.15 (Vernier Software & Technology, Beaverton, OR, USA). The only responses used were those in which the fish reacted to the stimuli and initiated an escape response to the first stimulus (total 66 responses: 0 ms treatment, 18 responses; 33 ms treatment, 13 responses; 83 ms treatment, 16 responses; control, 19 responses). The fish snout and center of mass (CoM; 35% of total body length; Paglianti and Domenici, 2006) were digitalized in each frame.
The time–distance variables maximum acceleration (m s−2), maximum speed (cm s−1) and cumulative distance (cm) were measured based on the CoM displacement. These variables were evaluated between the onset of stage 1 and the end of stage 2. Maximum acceleration and maximum speed were calculated by first- and second-order differentiation, respectively, of the cumulative distance for the time series. A Lanczos five-point quadratic moving regression method (Lanczos, 1956) was applied to calculate these last two values.
All estimations of the GMDs, comparisons with a control with Dunnett-corrected 95% confidence intervals, and the analysis of the I-T approach to find differences among treatments were performed in R v. 3.6.1 (http://www.R-project.org/) with the Mclust v.5.4.5 package (Scrucca et al., 2016). Because some complex models of each variable could not be calculated with the Mclust package as a consequence of a singularity in the covariance matrix (Scrucca et al., 2016), the analysis of the I-T approach was performed only on models that could be calculated. Although escape trajectories are circular variables which potentially span 360 deg (Domenici et al., 2011), most escape trajectories were distributed through a limited arc and the uniformity of the escape trajectories was not supported by Watson's goodness of fit test for a circular uniform distribution (U2 test; 0 ms: U2=0.86, P<0.01; 33 ms: U2=0.54, P<0.01; 83 ms: U2=0.60, P<0.01; control: U2=0.79, P<0.01) and escape trajectories were not distributed around 360 deg (Fig. 3A). Therefore, the distributions and the difference between treatments were analyzed using linear statistics (estimation of the GMD and the analysis of the I-T approach) as suggested by Batschelet (1981). Calculations of variables and statistical analyses were performed in R v. 3.6.1 with the circular v.0.4-93 package (https://r-forge.r-project.org/projects/circular/).
A summary of the statistical analyses on the kinematic variables is shown in Tables 2 and 3 and Tables S1 and S2. The escape trajectories, stage 1 turning angle, stage 1 turning duration and stage 1 turning rate of the 0 ms and 33 ms treatments were significantly smaller than those of the control, while those of the 83 ms treatment were not significantly different from those of the control (Table S1; Fig. 3A,B). The stage 2 turning rate of the 33 ms and 83 ms treatments was significantly lower than that of the control, while that of the 0 ms treatment was not significantly different from that of the control (Table S1). The apparent looming threshold of the 33 ms treatment was significantly higher than that of the control. In contrast, the threshold of the 0 ms and 83 ms treatments was not significantly different from that of the control (Table S1). The cumulative distance of the 0 ms and 83 ms treatments was significantly shorter than that of the control (Table S1). The best model of stage 2 turning rate in the I-T approach was 0 ms≠33 ms=83 ms, indicating that the stage 2 turning rate of the 0 ms treatment was lower than that of the 33 ms and 83 treatments. The best model of the apparent looming threshold was 33 ms≠0 ms=83 ms, indicating that value for the 33 ms treatment was lower than those for the 0 ms and 83 ms treatments. The best models of the other variables were 0 ms=33 ms =83 ms, suggesting no differences in the treatments.
The escape trajectories of the staghorn sculpin differed between single and dual threat stimuli when visual stimuli were applied from the left and right sides within 33 ms of one another. The mean of the escape trajectories for a single threat stimulus in the control was 110.60 deg (i.e. escaping away from a single threat stimulus). In contrast, the escape trajectories for dual threat stimuli in the 0 and 33 ms treatments were nearly 90 deg (i.e. perpendicular to the line of attack of dual threat stimuli). Hence, when attacked from two sides simultaneously or with a short delay (33 ms) between dual threat stimuli, fish tended to escape along a ‘compromise’ trajectory at a similar angle from both stimuli. However, the escape trajectories when the fish were attacked from two sides with a long delay (83 ms) between dual threat stimuli (i.e. when the delayed stimulus occurred after stage 1 initiation) did not differ from those of the single stimulus treatment. Consequently, our findings suggest that the escape trajectory of staghorn sculpin is not fully ballistic and sensory feedback may occur during the initial latency period of the escape responses. Once stage 1 of the escape response is initiated, the escape trajectory is set and is not affected by further feedback control.
Behavioral and neurophysiological studies have shown that the stage 1 turning angles are affected by stimulus direction (Domenici and Blake, 1993; Eaton and Emberley, 1991; Kimura and Kawabata, 2018). Stage 1 turning angles tend to be wide when the stimulus approaches the prey from the front and narrow when the stimulus approaches the prey from behind (Domenici and Blake, 1993; Eaton and Emberley, 1991; Kimura and Kawabata, 2018). The optimal escape trajectory is suggested to range between 90 deg and 180 deg depending on predator speed (Bhattacharyya et al., 2017; Domenici, 2002; Domenici et al., 2011; Weihs and Webb, 1984). Here, the fish were stimulated from the left and right sides. Thus, the resulting escape strategy consisted of remaining at an equal distance from the two threats by escaping at 90 deg when the two stimuli were simultaneous. The results of the present study suggest that the fish minimize their stage 1 turning angles when they are being attacked from the left and right sides simultaneously. Interestingly, control fish escaped at 110.60 deg (range 74.33–272.35 deg), which appears smaller than the mean escape trajectory (132 deg, range 98–175 deg) of the same species in the previous study (using a single visual stimulus with the same velocity as in this study, but using a nearly square tank) (Paglianti and Domenici, 2006). The rectangular shape of our experimental tank or other unknown factors may have caused this difference.
The mechanism allowing the modulation of stage 1 turning angles in the 0 ms and 33 ms treatments could be related to the activity of M-cells and associated neurons, as well as the prepontine neurons (Marquart et al., 2019). Prepontine neurons facilitate the integration of multiple sensory information (visual and auditory) and alter stage 1 of the escape response (Marquart et al., 2019). It is possible that a similar mechanism may occur in the presence of two visual stimuli, i.e. inputs from the two eyes might be integrated by prepontine neurons before the onset of the escape response. Additionally, the apparent looming threshold of the 33 ms treatment was higher than in the control, which corresponds to an approximately 57 ms delay in the escape latency of the 33 ms treatment. This result suggests that when the fish perceives the second stimulus during neural processing (i.e. between the first stimulation and the onset of stage1), it delays the process based on the single stimulus and integrates the second stimulus information, resulting in a 90 deg escape trajectory from both stimuli. Interestingly, this delay was not observed in the 0 ms treatment, suggesting that two simultaneous stimuli do not cause additional processing, although the mean escape trajectory was also nearly 90 deg (as in the 33 ms treatment), probably as a result of the symmetrical stimuli from both sides.
There were no differences among treatments in terms of their stage 2 turning angles. In stage 2, contraction of the body trunk muscles flips the caudal fin to the opposite side (Foreman and Eaton, 1993). As acceleration increases during stage 2, the body slightly rotates, and the final escape direction is determined. The accelerations and propulsive forces and jets are stronger in stage 2 than they are in stage 1 (Fleuren et al., 2018; Tytell and Lauder, 2008; Voesenek et al., 2019), and they differ in terms of the relative importance of the rotation or acceleration (propulsion) to their movement (Domenici and Blake, 1993; Domenici et al., 2004; Eaton et al., 1977; Eaton and Hackett, 1984; Tytell and Lauder, 2008; Weihs, 1973). Escape trajectories are related to stage 1 turning angles (Domenici and Hale, 2019) and the stage 2 turning angles and rates are smaller than those of stage 1 (Fleuren et al., 2018; Voesenek et al., 2019). Thus, stage 2 plays a relatively more important role in acceleration than it does in escape trajectory. The limited effect of stage 2 on the escape trajectory is in line with the lack of difference in stage 2 turning angles among treatments. Differences between treatments and control were found in cumulative distance, stage 1 turning duration and rate and the stage 2 turning rate. These may be related to the differences of the stage 1 and 2 turning angles.
When the second stimulus reached its threshold after the initiation of the escape response (i.e. in the 83 ms treatment), there was no change in the escape trajectory compared with the control. This is in line with a study on fathead minnows attempting to escape tentacled snakes (Catania, 2009). When fathead minnows were at a strike distance, the tentacled snakes generated a water flow with their bodies and induced a C-start in the fish, directed away from their body but into the snake's mouth. Prey fish cannot modify their escape response after their reaction to the body-generated water flow because one of two M-cells, which fires first, stimulates the body trunk muscle to initiate an escape response but inhibits activation of the opposite body trunk muscle (Faber et al., 1991; Korn and Faber, 2005). In the 83 ms treatment here, it is likely that the M-cell on the stimulus side was activated, with feedback inhibition preventing the activation of the opposing M-cell (Korn and Faber, 2005) as the latter would result in poor escape response performance. As a result, the escape trajectory of the 83 ms treatment did not differ from that of the control. However, our 33 ms treatment suggests that, if stimulated during the neural processing of the first stimulus (i.e. during the escape latency), fish can modify their escape trajectory. Hence, our results demonstrate that fish are capable of receiving additional sensory information during the neurosensory process that involves the circuitry from the optic tectum to the M-cells (Zottoli et al., 1987).
In conclusion, we suggest that the escape response consists of a flexible phase (from stimulation until the onset of stage 1) where sensory feedback is possible, and a ballistic phase (from the onset of stage 1 onwards). The ballistic phase is likely to occur in fish fast escape responses as a result of inhibition of the M-cell to trigger a further contraction during stage 1 (Faber et al., 1991). Specifically, a feed-forward inhibitory network guarantees that when one M-cell is excited, (1) it only generates a single action potential (preventing repetitive firing of one M-cell) and (2) the contralateral M-cell will not be activated. As suggested by Faber et al. (1991), this prevents the occurrence of ineffective escape behaviors, characterized by multiple fast body bends, and bilateral muscle contraction, which would lead to minimal displacement of the fish. While such ineffective motion patterns are prevented, so is the integration of multiple threats within the time interval that corresponds to stage 1. Hence, effective escape is ensured at the cost of eliminating the flexibility that would be associated with sensory feedback during the early phase of the escape response.
We found that Pacific staghorn sculpin can modulate their escape response only between stimulation and the onset of the response, but that escape responses are ballistic after the body motion has started. Although the flexible phase of Pacific staghorn sculpin lasts for at least 33 ms after stimulation, other fish species may show a different relative timing of the flexible and ballistic phases. Identifying which patterns are employed and during which phase may depend on the species and a phylogenetic analysis of escape flexibility would help us understand the evolution of fish escape response patterns. Furthermore, future research integrating behavioral experiments with neurophysiological measures (e.g. calcium imaging) could enable us to understand how the behavioral patterns of the escape response are related to the neural activity when fish are startled by multiple threats.
We would like to thank the 2019 Fish Swimming course for support and assistance with fish collections, and Friday Harbor Laboratories, University of Washington, for facilities and technical support. H.K. would like to thank his sponsor, Japan/U.S.–E.S. Morse Scholar Exchange Program and M.L. would like to thank the ‘Adopt a Student’ program, the Dudley Fund, and the GRIL (Interuniversity Research Group in Limnology and Aquatic Environments) for their financial support to attend the 2019 Fish Swimming course.
Conceptualization: P.D.; Methodology: J.F.S., J.L.J., P.D.; Software: H.K.; Formal analysis: H.K., T.P., M.L., Y.K., P.D.; Investigation: H.K., T.P., M.L.; Resources: H.K., T.P., M.L., J.F.S., J.L.J., P.D.; Data curation: H.K., T.P., M.L.; Writing - original draft: H.K.; Writing - review & editing: H.K., T.P., M.L., Y.K., J.L.J., P.D.; Visualization: H.K.; Supervision: Y.K., P.D.; Project administration: P.D.; Funding acquisition: H.K., M.L., Y.K.
This study was funded by Grants-in-Aid for Scientific Research, Japan Society for the Promotion of Science, to Y.K. (19H04936). H.K. was supported by the Japan/U.S.–E.S. Morse Scholar Exchange Program. M.L. was supported by the University of Washington and by a grant from the GRIL (Groupe de recherche interuniversitaire en limnologie) [a Fonds de Recherche du Québec - Nature et Technologies (FRQNT) funded strategic research network].
Data are available from figshare: https://figshare.com/articles/dataset/dataframe_csv/19519033
The authors declare no competing or financial interests.