A pretectal command system controls hunting behaviour

For many species, hunting is an innate behaviour that is crucial for survival, yet the circuits that control predatory action sequences are poorly understood. We used larval zebrafish to identify a command system that controls hunting. By combining calcium imaging with a virtual hunting assay, we identified a discrete pretectal region that is selectively active when animals initiate hunting. Targeted genetic labelling allowed us to examine the function and morphology of individual cells and identify two classes of pretectal neuron that project to ipsilateral optic tectum or the contralateral tegmentum. Optogenetic stimulation of single neurons of either class was able to induce sustained hunting sequences, in the absence of prey. Furthermore, laser ablation of these neurons impaired prey-catching and prevented induction of hunting by optogenetic stimulation of the anterior-ventral tectum. In sum, we define a specific population of pretectal neurons that functions as a command system to drive predatory behaviour. Key findings Pretectal neurons are recruited during hunting initiation Optogenetic stimulation of single pretectal neurons can induce predatory behaviour Ablation of pretectal neurons impairs hunting Pretectal cells comprise a command system controlling hunting behaviour


Introduction 17
In response to sensory information and internal states, animals select specific actions from a

37
In this study, we sought to identify a command system for control of predatory 38 behaviour, using larval zebrafish as a vertebrate model system. Command systems comprise 39 interneurons that are activated in association with a specific behaviour and whose activation 3 is able to induce that behaviour (Kupfermann and Weiss, 1978; Yoshihara and Yoshihara, 41 2018). In contrast to modulatory circuits, the presence of the releasing stimulus should not be 42 required for experimental activation of command neurons to induce the behavioural 43 response. In larval zebrafish, hunting is an innate, visually guided behaviour, which involves 44 a sequence of specialised oculomotor and locomotor actions. A defining characteristic is that 45 larvae initiate hunting by rapidly converging their eyes, which substantially increases their 46 binocular visual field (Bianco et al., 2011). A high vergence angle is maintained during prey pursuit, which entails a sequence of discrete orienting turns and approach swims, which 48 culminate in binocular fixation of prey followed by a kinematically distinct capture swim

96
To define groups of neurons with consistent functional properties related to the first 97 stages of hunting behaviour, we first computed, for every cell, a visuomotor vector (VMV) 98 that quantified its sensory and motor-related activity ( Figure 1G and Materials and Methods).

113
This analysis identified neurons that were recruited during hunting initiation.

114
Specifically, four clusters showed activity highly correlated with eye convergence (clusters 115 25-28; Figure 1I) yet exhibited little activity in response to visual cues (including prey-like 116 moving spots; Figure 1J and Figure 1-

126
We computed a `hunting index` (HIx) for individual neurons as a direct means to 127 distinguish neural activity associated with hunting initiation from `sensory` activity evoked 128 by prey-like visual stimuli. Briefly, for each hunting response, GCaMP fluorescence in a time 129 window (±1 s) surrounding the convergent saccade was compared to activity at the same time 130 in non-response trials during which the same visual stimulus was presented ( Figure 1K, left).

131
The mean of these difference measures across all response events represents the HIx score for 132 the cell and quantifies neural activity attributable to hunting initiation while accounting for 133 any visually evoked response. To account for directional tuning, we separately computed HIx 134 for hunting responses paired with leftwards, rightwards or symmetrical/no tail movements.

135
This analysis revealed that neurons in clusters 25-28 showed considerably higher HIx scores 136 than other cells, including those in prey-responsive clusters (1 and 4, Figure 1K). Moreover, 137 tail directional preferences were consistent with those determined by regression modelling.

138
Overall, our functional analyses identified four clusters of neurons with activity specifically 139 associated with the specialised motor outputs that characterise initiation of hunting behaviour 140 and showed little activity in response to prey-like visual cues. Thus, we will refer to these as 141 `hunting-initiation` clusters.

159
To confirm the response properties indicated by the VMV representations and HIx 160 scores and further examine visuomotor tuning, we computed visual stimulus-aligned and 161 convergence-aligned activity profiles for left and right hemisphere neurons in prey-162 responsive and hunting-initiation clusters ( Figure 1N,O). This confirmed that prey-responsive 163 neurons in clusters 1 and 4 showed direction-selective activity in response to small dark 164 6 moving spots, but minimal activity associated with eye convergence ( Figure 1N,O, left 165 columns). On the other hand, hunting-initiation neurons (clusters 25-28) showed weak visual 166 responses -as shown by moving spot-triggered activity during non-response trials -but 167 substantial activity triggered on hunting initiation. For clusters 26 and 28, neurons showed 168 stronger activation when convergent saccades were paired with left and right-sided turns, 169 respectively ( Figure 1N,O, right columns).

170
In summary, we identified populations of neurons in AF7-pretectum that are recruited 171 in association with two distinct components of hunting -visual responses to prey and 172 initiation of predatory behaviour. We subsequently examined whether cells with hunting-173 initiation activity are directly involved in inducing hunting behaviour.

195
We identified three morphological classes of pretectal neuron labelled by KalTA4u508.

196
One class projects to the ipsilateral optic tectum ( Figure 2F

229
KalTA4u508 cells assigned to hunting-initiation clusters had higher HIx scores than those 230 assigned to other clusters, supporting the hunting-response specificity of their activity ( Figure   231 2L,N).

232
In summary, KalTA4u508 provides genetic access to a subset of AF7-pretectal neurons 233 that are selectively active during initiation of hunting behaviour.

Optogenetic activation of single KalTA4u508 pretectal neurons induces hunting 235
To test whether KalTA4u508 pretectal cells are capable of inducing predatory behaviour, we 236 optogenetically stimulated single neurons while using high-speed tracking to monitor free-237 swimming behaviour ( Figure 3A). To do this, we used the same larvae described above in

288
To directly establish whether the KalTA4u508 pretectal neurons than can drive 289 predatory behaviour are the same cells that are recruited during visually evoked hunting, we 290 9 combined optogenetic stimulation and functional calcium imaging of single neurons. To 291 achieve this, we first established that optogenetic stimulation of a given KalTA4u508 pretectal 292 neuron could induce hunting and then tethered the larva in agarose and performed calcium 293 imaging of H2B-GCaMP6s, expressed in the nucleus of the same neuron, while the animal 294 engaged in the virtual hunting assay. Visuomotor fingerprinting and cluster assignment of 295 these neurons showed that they all belonged to hunting-initiation clusters (clusters 25-27) and 296 had high HIx scores (n = 6 cells from 6 fish; Figure 3T).

297
In summary, KalTA4u508 labels a specific group of pretectal neurons that are recruited 298 during hunting initiation and which are capable of inducing naturalistic predatory behaviour 299 in the absence of prey.

Ablation of KalTA4u508 pretectal neurons impairs hunting 301
To what extent are KalTA4u508 pretectal neurons required for hunting? To address this 302 question, we assessed hunting performance in freely swimming larvae provided with 303 Paramecia, both before and after laser-ablation of KalTA4u508 pretectal neurons ( Figure 4A-304 C). To enable evaluation of the specificity of behavioural phenotypes, we also presented 305 looming stimuli and drifting gratings to test visually evoked escape and optomotor response 306 (OMR), respectively. Ablations were performed at 6 dpf in KalTA4u508;UAS:mCherry larvae 307 and their efficacy was confirmed by reimaging the pretectum the following day. We estimated 308 that ~90% of the fluorescently labelled KalTA4u508 pretectal population was typically ablated 309 in both brain hemispheres ( Figure 4D,E). Behaviour was tested both before (6 dpf) and after 310 ablation (7 dpf) and control larvae underwent the same manipulations, other than laser-311 ablation, and were tested at the same time-points.

312
Analysis of prey consumption revealed that ablation of KalTA4u508 pretectal neurons 313 resulted in decreased hunting performance ( Figure 4F). Further analysis revealed that this 314 reduction in prey capture was associated with a reduced rate of hunting initiation ( Figure

346
In summary, these data indicate that KalTA4u508 pretectal neurons contribute to the 347 initiation and maintenance of natural hunting behaviour and are required for release of 348 predatory behaviour by circuits in the anterior optic tectum.

Discussion 350
In this study we combined multi-photon calcium imaging, single-cell optogenetic stimulation

384
Ablations targeting KalTA4u508 pretectal neurons impaired, but did not eliminate, 385 hunting. One contributing factor is likely to be that ablations were incomplete. We observed

394
Overall, our data identify a discrete population of pretectal neurons that comprise a 395 command system controlling predatory hunting.

443
We hypothesise that pretectal activity releases predatory behaviour and operates in parallel 444 with prey-directed steering signals, most likely from OT.                                                             The procedure was very similar to that described in Bianco and Engert (2015). Larval zebrafish 662 were mounted in 3% low-melting point agarose (Sigma-Aldrich) at 5 dpf or 6 dpf and allowed 663 to recover overnight before functional imaging at 6 dpf or 7 dpf. Imaging was performed using

697
We used multilinear regression to model the fluorescent timeseries of each imaged 698 neuron (ROI) in terms of simultaneously recorded kinematic predictors (`regressors`).

699
Regressors were generated for oculomotor and locomotor variables (7 eye and 3 tail

730
VMVs were generated as described above and the same assignment strategy and correlation 731 distance threshold (0.7) were used. Note that normalisation of components was performed 732 using the standard deviations computed for the initial matrix of VMVs.

733
Hunting Index (HIx) scores were calculated for each cell as follows. . Images were converted to the NRRD file format required by ANTs using ImageJ.

747
As an example, to register the 3D image volume in 'fish1_01.nrrd' to the reference brain

773
• The brain regions displayed in Figure 2F,H, and Figure

777
• For the registration displayed in Figure 2A,

789
• Imaging volumes related to photo-activation of PA-GFP were registered to a whole-

796
All registration steps were manually assessed for global and local alignment accuracy.

797
All brain regions referred to in this paper correspond to the volumetric binary image masks     selected by imaging the brain at 920 nm. Photo-activation was performed by continuously 925 scanning at 790 nm (5 mW at sample) for 4 min. Larvae were then unmounted and allowed to 926 recover. At 7 dpf, an image stack (1200×800 px, 0.38 µm/px, ∼200 µm z-extent) was acquired 927 at 920 nm covering a large portion of the midbrain, tegmentum and hindbrain. Axonal