For prey animals, avoiding approaching predators is crucial to survival, but unnecessary escape behavior is both energetically costly and disruptive to normal behavioral routines. As a consequence, there has been strong selective pressure to escape from genuinely threatening stimuli while ignoring innocuous stimuli and stimuli that occur repetitively without consequence. In larval zebrafish, input from the auditory (Burgess and Granato, 2007; Marques et al., 2018), somatosensory (Eaton et al., 1977; Douglass et al., 2008), and lateral line systems (McHenry et al., 2009) can all drive startle responses, but vision, which is uniquely informative about an incoming object’s spatial characteristics, has received particular attention in recent years.

In an experimental context, predators can be simulated on a screen with looming stimuli (looms): dark shapes that grow rapidly from a single point (Luca and Gerlai, 2012). If the loom is sufficiently salient, it elicits a stereotyped escape response, which in larval zebrafish takes the form of a rapid body bend followed by a powerful swim sequence (Dunn et al., 2016; Bhattacharyya et al., 2017; Marques et al., 2018). The core sensorimotor circuitry responsible for visual escape in larval zebrafish has been described (reviewed in Marquez-Legorreta et al., 2020), involving the perception of motion and luminance by specialized retinal ganglion cells (RGCs) (Temizer et al., 2015), widespread and powerful loom responses in neurons belonging to the tectum (called the superior colliculus in mammals) contralateral to the stimulus (Helmbrecht et al., 2018), and downstream activation of reticulospinal neurons in the hindbrain (Sato et al., 2007; Yao et al., 2016). However, many details about how particular components of the loom stimulus are detected and integrated, and how behaviors are gated, remain mysterious.

A dark looming stimulus comprises two components: a decrease in luminance and expanding moving edges at the periphery. These components can be separated experimentally with the presentation of a dimming circle (providing decreased luminance without moving edges) or an expanding isoluminant checkerboard on a gray background (moving edges without a change in overall luminance). Behavioral experiments have shown that expanding checkerboards, and therefore moving edges, are sufficient to elicit escape responses, albeit at a lower rate than dark loom stimuli (Dunn et al., 2016; Heap L. A. L. et al., 2018). The neural circuitry detecting these moving edges is incompletely understood, but it has been shown that specific laminae of the tectal neuropil receive edge-specific information, and that RGCs targeting these specific neuropil layers appear to project exclusively to the tectum and not to other retinal arborization fields (Robles et al., 2014; Temizer et al., 2015). On the other hand, the dimming component involves a retino-thalamo-tectal circuit, which increases the probability of an escape response (though dimming alone elicits startles infrequently), and also determines the direction of this startle (Temizer et al., 2015; Heap L. A. L. et al., 2018). These observations suggest that different components of the looming stimulus are processed in distinct pathways and mediate different facets of the behavioral response.

Habituation is a form of non-associative learning that is important for survival, as it minimizes energy expenditure and disruption from unnecessary responses to repetitive stimuli (Thompson and Spencer, 1966; Rankin et al., 2009). The visual escape response and O-bends habituate in larval zebrafish, with the startle probability decreasing with repeated presentations of loom (Marquez-Legorreta et al., 2019) or dark flash (Randlett et al., 2019) stimuli. It has been proposed that this habituation results from the tectum’s “uncoupling” the perceptual visual circuits in the retina and thalamus from premotor circuitry in the hindbrain, thus shifting sensorimotor gating toward non-responsiveness (Marquez-Legorreta et al., 2019). The tectum is well positioned to integrate movement and luminance information, and has been shown to contain populations of neurons whose responses habituate in parallel with behavioral habituation (Marquez-Legorreta et al., 2019). Nonetheless, the neural circuitry responsible for visual habituation, and the specific ways in which luminance and movement information interact in these circuits, remain incompletely understood.

In this study, we have used light-sheet microscopy and genetically encoded calcium indicators to track activity in individual neurons across the brains of larval zebrafish while presenting repetitive loom, checkerboard, and dimming stimuli. The goal has been to describe each neuron’s anatomical position; selective responsiveness to movement, luminance, and full loom stimuli; and degree of habituation to repeated stimuli. By providing links between these anatomical and functional readouts, we have aimed to determine the stimulus properties and brain regions responsible for behavioral habituation to repetitive threatening visual stimuli. Behaviorally, we have found that the visual escape response of zebrafish larvae does not habituate to checkerboard stimuli, and that the habituation to looms requires both the movement and luminance components. We have identified the neuronal substrates for the visual habituation to luminance, movement, and looms, and show that loom habituation at the circuit level also requires both components. We have not identified neurons responding only to full looms, suggesting that the integration of the two components drives the sensorimotor gating.

It has previously been shown that drops in luminance (dim), moving edges (checkerboard), and loom (combining dim and edges) stimuli lead to different types and probabilities of visual startle behavior in zebrafish larvae (Dunn et al., 2016; Heap L. A. L. et al., 2018), and that startle behavior habituates in response to repeated loom stimuli (Marquez-Legorreta et al., 2019). This leaves two open questions about the relationships between visual stimulus components and habituation. First, it is unknown whether behavioral responses to repeated dimming or checkerboard stimuli habituate, as they do in response to repeated loom stimuli. Second, it is unclear whether prior exposure to repeated dims or checkerboard stimuli is sufficient to habituate animals to loom stimuli.

To address both questions, we designed a behavioral apparatus (Figure 1A) and stimulus train (Figures 1B,C) in which larvae were presented with 30 visual stimuli in close succession (20–35 s interstimulus interval, ISI). The first 20 stimuli were either checkerboards or dims, followed by 10 looms. A third group of larvae was shown 30 looms. As previously described, loom stimuli initially caused a high rate of visual startle behavior (Temizer et al., 2015; Dunn et al., 2016; Heap L. A. L. et al., 2018) and these responses gradually habituated to repeated loom stimuli (Figure 1D; Marquez-Legorreta et al., 2019). By comparison, larvae startled to looming checkerboards at a lower rate, and these responses did not undergo significant habituation over the course of 20 trials. Dimming stimuli seldom drove escape behavior, and therefore could not be assessed for habituation. When 20 trials of either dims or checkerboards preceded loom trials, no significant reduction was seen in the responses to the first loom (the 21st trial), as compared to naïve animals seeing a loom stimulus as their first trials (Figure 1E). These results provide no evidence for behavioral habituation to checkerboard stimuli, and suggest that the circuits mediating loom habituation may depend on both components of the loom stimuli, rather than either moving edges or dimming in isolation.

To probe the functional structure of the circuits underlying these behavioral effects, we showed the equivalent stimulus trains during whole-brain calcium imaging of larvae expressing nuclear-targeted GCaMP6s in all neurons, using a custom-built light-sheet microscope (Figure 2F, see section “Materials and Methods”). We collected volumetric data using 50 planes spanning the brain at 5 μm intervals across the dorso-ventral axis, and tracked activity in individual regions of interest (ROIs) generally corresponding to individual neurons.

In contrast to the behavioral results, we see marked habituation in the responses of dim- and edge-sensitive ROIs during repeated stimulus presentations (Figure 2E). Brain-wide loom-responsive ROIs also undergo pronounced habituation (Figure 2E), in line with startle behavior (Figure 1C). In terms of their abundance and distribution, we identified relatively few dim-sensitive ROIs (1,154 ROIs across 12 animals), and these were primarily in the tectum (Figure 2A). Checkerboard-responsive ROIs were more numerous (4,873 ROIs, 11 animals), and were spread broadly across several brain regions (Figure 2B). Loom stimuli elicited the most numerous (11,050 ROIs, 11 animals) and widespread responses, with an overall distribution resembling that of checkerboard-responsive ROIs (Figure 2C).

There are several interpretations to be drawn from these results. First, we see habituation, as measured by brain-wide calcium responses, to repeated presentations of both dimming and checkerboard stimuli. This is difficult to register to behavior in the case of dim-responsive ROIs, since dimming stimuli seldom trigger escape behavior even in naïve larvae, but it suggests that for motion stimuli, animals sustain behavioral sensitivity to moving edges even as motion-sensitive ROIs (at least as a population) decrease their response strength. It is worth noting the caveat, however, that the stimuli are presented in different orientations (from the bottom for behavioral experiments and from the side during calcium imaging) due to the spatial constraints of the two setups. As has been shown previously (Marquez-Legorreta et al., 2019), brain-wide responses to loom stimuli habituate, paralleling behavioral habituation to looms. Also paralleling our behavioral results (Figure 1), these results provide no evidence of pre-habituation of full loom responses by either checkerboards or dim stimuli, suggesting that only full loom stimuli can cause habituation in the visual escape circuit.

In terms of saliency, at least as judged by the number of brain-wide responses, we find many more and more widespread checkerboard-responsive ROIs than dim-responsive ROIs. Loom responses are more numerous than either, and indeed outnumber the combined dim- and checkerboard-responsive ROIs from the other stimulus trains. This suggests that motion may be the more salient component of the loom, but that there may also be neurons that respond only when a full loom stimulus, with both dimming and motion components, is present.

Because one habituation experiment would likely impact subsequent ones, the stimulus trains used in Figure 2 were played separately, and each animal only observed one of the trains. This, in turn, prevents us from judging how much overlap there is among dim-, checkerboard-, and loom-responsive ROIs. The specificity or overlap of these responses, however, have important implications for the circuit-level processing underlying network-wide and behavioral habituation. In order to determine the degree to which neurons respond to more than one of these stimuli, while minimizing the impacts of habituation, we presented all three stimuli to the same animals with only two repetitions of each stimulus, and with long intervals between stimuli (Figure 3).

To address the degree to which individual ROIs respond to dim, checkerboard, and/or full loom stimuli, we then used a k-means approach (see section “Materials and Methods”) to identify functional categories of ROIs with distinct response properties to each of the stimuli. This yielded three clusters that were responsive to (1) dims and full looms (but not checkerboards), (2) to checkerboards and full looms (but not dims), and (3) to all three stimuli.

A dim-sensitive cluster (blue, Figure 3A), responsive both to dims and full looms, contained the fewest ROIs, and these were principally spread across the habenulae, thalamus, tectum, and pallium, consistent with previous observations (Cheng et al., 2017; Zhang et al., 2017; Chen et al., 2018; Heap L. A. L. et al., 2018; Robles et al., 2021). On average, these ROIs also had very weak responses to checkerboard stimuli. Another cluster, sensitive to the moving edge component of the loom stimulus (yellow, Figure 3B), comprised checkerboard- and loom-responsive ROIs. Among our three clusters, this was the most abundant in terms of ROI number, and contributed a majority of the ROIs in each of the brain regions that we studied except for the tectum (balanced among the three clusters), and the habenulae (with mostly dim-sensitive ROIs). The final cluster (green, Figure 3C) comprising ROIs sensitive to both dims and checkerboards (and unsurprisingly to looms), was principally contained within the tectum.

These results have a few notable implications. First, neurons responding to moving edges (from checkerboard and full loom stimuli) are much more numerous and broadly distributed in the brain than the luminance signals (from the dim and full loom stimuli), suggesting a rather targeted perception and use of dimming information, at least as manifested by this particular set of stimuli. Second, neurons responding to all three stimuli are numerous in, and largely restricted to, the tectum contralateral to the stimulus (the right tectum in Figure 3). This reinforces the idea that the tectum (superior colliculus) is a key structure in loom perception and likely in visual escape behavior.

However, it is interesting to note that the checkerboard specific responses, and the dim responses to a lesser extent, show a bilateral distribution. This is especially noticeable in the thalamus and tectum, suggesting a position-independent representation of the stimuli. Finally, we did not identify a loom-specific cluster (responsive to looms but not to dims or checkerboards, see Supplementary Methods and Supplementary Figure 3). This suggests that the dim- and checkerboard-responsive ROIs’ responses (Figure 3C) represent the summed inputs of the luminance and movement components of the loom stimulus, rather an emergent signal specific to looms. This notion of the linear summation of the luminance and movement signals is supported by the form of the loom responses in these ROIs, which combines the short, broad double peak seen to dimming stimuli with the tall, sharp single peak elicited by a looming checkerboard (Figure 3C).

Comparing the spatial distributions of dim- and motion-sensitive ROIs (Figures 3A,B) to the distributions of moderately- and fast-habituating ROIs during repeated looms (as seen in Marquez-Legorreta et al., 2019) suggests possible links between these populations (Supplementary Figure 4). Specifically, fast-habituating ROIs are distributed similarly to those responding to motion, while moderately habituating ROIs’ distribution resembles that of dim-sensitive ROIs. To test whether these are, indeed, the same populations, we devised a stimulus train in which we could identify each ROI’s inherent responsiveness to dims and motion, and its habituation rate to repeated looms (Figure 4). This stimulus train begins with two dims and two checkerboard looms at the beginning of the train, followed by ten loom stimuli, and finishes with two more dims and checkerboard stimuli. This experiment permits us to categorize ROIs by their responsiveness to dims and checkerboards (using the first four trials, analogous to the analysis done in Figure 3), and separately to gauge their habituation rate using the ten looms, as previously performed (Marquez-Legorreta et al., 2019).

By selectively categorizing responses using customized set of criteria (see Supplementary Table 1), we identified functional clusters of ROIs that were introduced in Figure 3: dim- and loom- responsive (Figure 4A), checkerboard- and loom-responsive (Figure 4B), and ROIs responsive to all three stimuli (Figure 4C). Categorizing the responsive ROIs by their inherent stimulus sensitivity then allowed us to look for differences among these groups in their basic responses to looms and in their degree and rate of habituation to repeated loom stimuli. We approached this by fitting an exponential habituation curve to each ROI and using this curve to assess its rate of habituation. We then looked at the statistics of habituation rate across clusters to determine whether these rates differ across clusters.

We found that the rate of habituation during the 10x loom portion of the stimulus train (as judged by the decay constant (c in our exponential equation) was similar across all three groups (dim-sensitive median 0.6566, SD 0.2945; checkerboard-sensitive median 0.5794, SD 0.2012; sensitive to both median 0.6695, SD 0.4730; p = 0.6271, using the Friedman Test). As such, this experiment, which was designed to explore the possibility that neurons’ inherent sensitivity is linked to their habituation properties, has failed to find evidence of such a link. Indeed, the results suggest that the visual neurons’ habituation properties are very similar regardless of the stimulus properties that they detect.

In this study, we have used brain-wide calcium imaging to identify the response profiles and locations of visual neurons across the larval zebrafish brain. Focusing on loom stimuli and the luminance changes that compose them, we have echoed previous results finding similar distribution of responses across the brain (Dunn et al., 2016; Yao et al., 2016; Cheng et al., 2017; Zhang et al., 2017; Chen et al., 2018; Heap L. A. L. et al., 2018; Helmbrecht et al., 2018; Marquez-Legorreta et al., 2019; Fernandes et al., 2021; Robles et al., 2021). A new interesting finding is that responses to motion are numerous and widespread, while responses to changes in luminance are more localized. The widespread response to checkerboard looms, with similar distribution to looms, and the fact that checkerboard looms produce escape responses while dim stimuli do not, suggest that the perception of expanding edges is a key feature for the detection of possible threats. As such, this stimulus recruits a broad network of neurons to elicit a motor response, but also extensive upstream processing regions in addition to the core startle circuit. Although it has been shown that tectal cells are sensitive to directional movements (Gabriel et al., 2012; Hunter et al., 2013; Wang et al., 2019), how this information is combined for the perception of the expanding edges of a loom remains unclear. Another possible explanation for the spatial spread and prevalence of motion-sensing neurons is the wide range of ethologically relevant moving stimuli that larval zebrafish encounter, along with several critical behaviors that rely on motion perception. These behaviors include predation, the optomotor and optokinetic responses, social behaviors, and visual escape (Semmelhack et al., 2014; Dreosti et al., 2015; Naumann et al., 2016; Larsch and Baier, 2018; Henriques et al., 2019; Marquez-Legorreta et al., 2020; Wang et al., 2020), and while the visual stimuli eliciting these various responses are diverse, a large salient stimulus like a checkerboard loom might elicit responses from a broad swathe of visual perceptual circuits. We also note checkerboard responses in regions such as the thalamus, where we have previously failed to observe them (Heap L. A. L. et al., 2018). We account for this difference based on the current study’s volumetric imaging, functional segmentation (Giovannucci et al., 2019), and altered thresholds for inclusion or ROIs as responsive.

Although some evidence exists for loom-specific RGCs in larval zebrafish (Temizer et al., 2015), recent studies suggest that such RGCs tend to be also sensitive to other moving objects or dimming stimuli (Förster et al., 2020; Kölsch et al., 2021). In general, these previous studies find a different distribution of movement sensitive and dimming sensitive RGC axons in the tectal neuropil, suggesting that different RGCs process this information (Robles et al., 2014; Kölsch et al., 2021). Tectal cells, however, seem to be capable of computing new features based on the RGCs input (Hunter et al., 2013; Förster et al., 2020). We have shown that neurons that respond to both the motion and dimming components of loom stimuli are concentrated in the tectum (Figure 3). This supports the notion that the tectum (and likely the superior colliculus in mammals) is the first stage in the visual pathway at which loom stimuli are specifically encoded (Heap L. A. L. et al., 2018). Regardless of whether loom encoding is an emergent property of the tectum, these loom responses put the tectum in a likely position to perform or direct sensorimotor gating in the visual escape pathway (Yao et al., 2016; Helmbrecht et al., 2018).

Convergent visual information from the eye and thalamus, combined with processing integral to tectal circuits, could encode the presence, location, and speed of looming objects. The resulting sensorimotor gating could then be modulated in the tectum by inputs from a number of brain regions with information on the larva’s context and recent history (Filosa et al., 2016; Lovett-Barron et al., 2017; Heap L. A. et al., 2018; Yáñez et al., 2018; Henriques et al., 2019; Marques et al., 2019; Fernandes et al., 2021). Outputs from the tectum to downstream premotor neurons, therefore, could direct appropriate responses, depending on the details of the loom stimulus and the animal’s circumstances. In all of these regards, the results presented here fit in with a model of the tectum as a key pivot point in the visual escape pathway (Marquez-Legorreta et al., 2020; Isa et al., 2021).

Another goal of this study was to explore the relationships that exist between neurons’ sensitivity to the individual components of loom stimuli and their possible roles in habituation. Behaviorally, we found that checkerboard stimuli elicited escapes responses, but that these responses did not habituate after repeated checkerboard stimuli, as occurs during trains of repeated loom stimuli. Furthermore, we found that neither repeated checkerboard stimuli nor repeated dim stimuli pre-habituated larvae to subsequent looms, which elicited escape responses like those seen in naïve animals exposed to looms. As such, at least at the behavioral level, there is no evidence for habituation in movement-sensitive escape circuits, nor is there evidence for contributions to loom habituation from either motion-specific or dim-specific pathways. This suggests that both movement and luminance properties of loom stimuli are required for habituation of the loom response.

Using calcium imaging, we next explored whether neurons across the brain that respond to specific stimuli have characteristic habituation profiles. We found that all three functional categories that we identified (movement-responsive, luminance-responsive, and responsive to both) underwent habituation to repeated presentations of the relevant stimuli (checkerboards, dims, and full looms, respectively). While this was expected for full looms, which habituate behaviorally, it was more surprising for checkerboards, since there is no behavioral habituation to this stimulus (Figure 1). This uncoupling of our calcium imaging and behavioral data could result from the abundance and breadth of motion-sensitive neurons across the brain (discussed above), many of which may be involved in behaviors other than escape. This difference could be due to the stimulus’ always coming from the same visual location in the calcium imaging experiments, while in free-swimming larvae the visual stimuli varied depending on the position and orientation of the fish. The role of visual position, and the consistency of visual position in repeated loom stimuli, would be an interesting future avenue to pursue. Another feature that could explain habituation in the neuronal responses while not in the free-swimming behavior is the lack of sensory feedback after an attempted escape response. The lack of this feedback could affect the fish’s processing of the stimulus and facilitate habituation, as such absence of feedback has been shown to induce passivity (Mu et al., 2019). Finally, we explored the possibility that neurons with specific response profiles (to loom components) would have characteristic and differing habituation rates, based on similarities seen between these neurons’ distributions in this study and the distributions of neurons with different habituation rates from a prior study (Marquez-Legorreta et al., 2019). Refuting this possibility, we found that motion-sensitive neurons, dim-sensitive neurons, and neurons responsive to both components all had similar habituation profiles in animals presented with repeated loom stimuli.

The cellular and sub-cellular mechanisms underlying behavioral habituation are complex (reviewed in McDiarmid et al., 2019). As we have shown, the relationship between the amplitude of the response from stimulus-sensitive neurons and the probability of a behavioral response is not linear. We have shown habituation in responses at the cellular level to all three stimuli used. On the timescales that we are observing, this is more likely to result from reduction in neurotransmitter release at the presynaptic site than postsynaptic mechanisms (Castellucci and Kandel, 1974; Glanzman, 2009). However, we observe habituation across all the responsive structures in the brain, so the spatial origin of the decreases and the ways in which these attenuated responses are propagated through the brain are not clear. Importantly, we have shown that habituation to one component of the loom (either moving edges or dimming), does not generalize to habituate responses across the network of neurons responsive to the full loom.

Overall, the outcomes of this study, while incremental, shed light on specific details of the visual escape pathway and its habituation. Behaviorally, we show that escape responses only habituate during repeated full loom stimuli, and that neither the movement nor the luminance components of loom stimuli contribute to behavioral loom habituation by themselves. Our brain-wide calcium imaging shows broader and more numerous responses to movement than to dimming stimuli, and provides no evidence for loom-specific neurons (that is, neurons that respond to loom but no to movement or dimming alone). Furthermore, these results show that neurons responsive to both movement and dimming stimuli are concentrated in the tectum, further suggesting an important role for this structure in detecting and responding to threatening visual stimuli. Finally, we have failed to find a link between neurons’ initial response selectivity and habituation rate during repeated looms. Such relationships would have suggested that the processing of particular stimulus properties plays specific roles in brain-wide visual habituation. The absence of such relationships suggests that habituation may rest with local microcircuits in the tectum or in interactions across multiple functional brain-wide populations of neurons. The detailed mechanisms of these local or brain-wide networks provide a rich topic for future exploration.

The original contributions presented in the study are available at: https://doi.org/10.48610/3684089.

The animal study was reviewed and approved by the University of Queensland Animal Welfare Unit under approval SBS/341/19.

TM collected and analyzed data and wrote the manuscript. EM-L designed experiments, collected data, performed analyses, and wrote the manuscript. MW guided experimental analysis and wrote the manuscript. MP collected and analyzed data. IF-B designed, built, maintained the light-sheet microscope, and edited the manuscript. GV performed analyses and edited the manuscript. ES designed experiments, guided analyses, and wrote the manuscript. All authors contributed to the article and approved the submitted version.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.