Brain-Wide Impacts of Sedation on Spontaneous Activity and Auditory Processing in Larval Zebrafish

Abstract

Despite their widespread use, we have limited knowledge of the mechanisms by which sedatives mediate their effects on brain-wide networks. This is, in part, due to the technical challenge of observing activity across large populations of neurons in normal and sedated brains. In this study, we examined the effects of the sedative dexmedetomidine, and its antagonist atipamezole, on spontaneous brain dynamics and auditory processing in zebrafish larvae, a stage when sex differentiation has not yet occurred. Our brain-wide, cellular-resolution calcium imaging reveals the brain regions involved in these network-scale dynamics and the individual neurons that are affected within those regions. Further analysis reveals a variety of dynamic changes in the brain at baseline, including marked reductions in spontaneous activity, correlation, and variance. The reductions in activity and variance represent a “quieter” brain state during sedation, an effect inducing highly correlated evoked activity in the auditory system to stand out more than it does in unsedated brains. We also observe a reduction in the persistence of auditory information across the brain during sedation, suggesting that the removal of spontaneous activity leaves the core auditory pathway free of impingement from other nonauditory information. Finally, we describe a less dynamic brain-wide network during sedation, with a higher energy barrier and a lower probability of brain state transitions during sedation. Overall, our brain-wide, cellular-resolution analysis shows that sedation leads to a quieter, more stable, and less dynamic brain and, that against this background, responses across the auditory processing pathway become sharper and more prominent.

Significance Statement

Animals’ brain states constantly fluctuate in response to their environment and context, leading to changes in perception and behavioral choices. Alterations in perception, sensorimotor gating, and behavioral selection are hallmarks of numerous neuropsychiatric disorders, but the circuit- and network-level underpinnings of these alterations are poorly understood. Pharmacological sedation alters perception and responsiveness and provides a controlled and repeatable manipulation for studying brain states and their underlying circuitry. Here, we show that sedation of larval zebrafish with dexmedetomidine reduces brain-wide spontaneous activity and locomotion but leaves portions of brain-wide auditory processing and behavior intact. We describe and computationally model changes at the levels of individual neurons, local circuits, and brain-wide networks leading to altered brain states and sensory processing during sedation.

Introduction

Sedatives are extensively used in medicine both to anesthetize patients and to treat psychological conditions such as acute agitation (Levy 1996; Boyer 2009; Bosch et al., 2022). While the biochemical modes of action for these drugs are often well known (Rudolph and Antkowiak 2004; Giovannitti et al., 2015), the details of their effects on brain-wide networks, and the mechanisms by which they cause altered perception and loss of consciousness, are incompletely understood. Our understanding of these mechanisms is limited by a lack of information about activity across the brain at the level of individual neurons and how the brain-wide networks comprising these neurons change during sedation.

As a means of addressing these open questions, we have explored the effects of the sedative dexmedetomidine (DEX), as well as its antagonist atipamezole (ATI), on spontaneous activity and auditory processing in zebrafish larvae, in which it is possible to perform brain-wide calcium imaging at cellular resolution (Ahrens et al., 2012). DEX induces sedation by decreasing the activity of noradrenergic neurons in the locus ceruleus (LC; Jorm and Stamford 1993; Guo et al., 1996). Its effects are reversed by the antagonist ATI, which blocks norepinephrine's feedback inhibition, thereby competing with DEX (Newman-Tancredi et al., 1998). In zebrafish, recent observations suggest that both mechanistic and behavioral effects of DEX and ATI are conserved between zebrafish and humans (Ruuskanen et al., 2005; Maximino and Herculano 2010).

Given zebrafish's unique advantages among vertebrates for the study of brain-wide spontaneous activity and sensory processing networks (V. Bedell et al., 2018; G. C. Vanwalleghem et al., 2018), we judged that zebrafish would be a suitable model for exploring the network-level mechanisms of DEX's and ATI's effects on wakefulness and consciousness. Several groups have used two-photon or light-sheet fluorescence imaging, combined with genetically encoded calcium indicators, to image whole larval zebrafish brains, in vivo, with cellular resolution (Ahrens et al., 2012, 2013; Wolf et al., 2015; Favre-Bulle et al., 2018; Migault et al., 2018; Taylor et al., 2018; Tunbak et al., 2020). Such studies have provided the details of individual neurons’ activity across the entire brain during spontaneous activity (Avitan et al., 2017) and during the processing of stimuli across the auditory (G. Vanwalleghem et al., 2017; Constantin et al., 2020; Poulsen et al., 2021), visual (Mancienne et al., 2021; Thompson and Scott 2016; X. Chen et al., 2018; Heap et al., 2018; Dragomir et al., 2020; Karpenko et al., 2020; Marquez Legorreta et al., 2022; Fotowat and Engert 2023), vestibular (Favre-Bulle et al., 2017, 2018, 2020; Migault et al., 2018), and water flow sensing (G. Vanwalleghem et al., 2020) systems.

In zebrafish, auditory processing presents a tractable sensory model in which stimuli can be precisely delivered, the responsive brain regions and neuron types have been described, and the resulting behavioral outcomes are simple. Starting with electrophysiology, past studies (Eaton et al., 1981; Sisneros and Bass 2005; Wright et al., 2010; Lu and DeSmidt 2013; Maruska and Sisneros 2016) have characterized sensory hair cell responses in the inner ear. Downstream neural responses have been described in response to diverse auditory stimuli such as pure tones of varying frequencies and white noise or dynamic changes such as volume ramps and frequency sweeps (G. Vanwalleghem et al., 2017; Poulsen et al., 2021). Key areas identified as auditory include the octavolateralis nucleus (ON), torus semicircularis (TS), thalamus (Th), cerebellum (Cb), and other parts of the hindbrain (excluding the ON and cerebellum; Privat et al., 2019; Poulsen et al., 2021). Strong acoustic stimuli generally elicit escape behavior in the form of powerful body bends followed by swim bouts (Kimmel et al., 1974; Burgess and Granato 2007a). The zebrafish auditory system therefore provides a simple and interpretable platform for studying sensory motor dynamics.

In this study, we examine how sedation influences brain dynamics in zebrafish larvae in response to acoustic stimuli and during periods of spontaneous activity. As expected from the literature, our results show a reduction of spontaneous behavioral activity (V. M. Bedell et al., 2020) but also in brain-wide neural activity during sedation. These changes in baseline activity have further effects on auditory processing. While the brain-wide auditory processing network appears to be essentially intact during sedation, with strong responses from neurons across regions that have previously been implicated in audition, we have observed reduced persistence of auditory processing during sedation. Behaviorally, escape responses occur with normal kinetics under sedation, but with an increase in the velocity and kinematic properties of the startle responses, suggesting a change in the sensorimotor gating or motor execution of this response when baseline activity is reduced during the sedated brain state.

Materials and Methods

Animals

High-speed behavioral recordings were performed with approval from the University of Melbourne Office of Research Ethics and Integrity (in accordance with ethics approval 2022-24987-35220-5). All other procedures were performed with approval from the University of Queensland Animal Welfare Unit (in accordance with ethics approval 2019/AE000341).

Zebrafish (Danio rerio) larvae, of both sexes, were maintained at 28.5°C on a 14 h ON/10 h OFF light cycle. Adult fish were maintained, fed, and mated as previously described (Westerfield 2000). Experiments were carried out in larvae of the TL strain (behavioral experiments) or mitf-mutant (nacre) elavl3:H2B-GCaMP6s larvae of the TLN strain (calcium imaging experiments; Lister et al., 1999; Freeman et al., 2014).

Free-swimming behavior and drug administration

Free-swimming behavioral experiments were performed using a custom-built behavioral rig. The rig consisted of an illumination system (infrared LED array), an elevated arena where fish were placed in custom-made acrylic well plates (16 wells in a 4 × 4 grid or 7 wells in a circle, with wells that were 20 mm in diameter, Extended Data Fig. 1.1). Acoustic stimuli were delivered using an amplifier (Dayton Audio DA30 2 × 15 W Class D Bridgeable Mini Amplifier) and a speaker (Dayton Audio DAEX19CT-4 4 Ω/5 W) affixed to the center of a circular array of wells (Extended Data Fig. 1.1). A high-speed camera (Ximea xiB-64, model CB019MG-LX-X8G3) equipped with an EF mount adapter and a Sigma 17−70 mm f/2.8–4 DC macro lens with an IR pass filter (750 nm) was used for imaging.

Zebrafish larvae at 6 d postfertilization (dpf) were placed in individual wells containing 1 ml of E3 zebrafish medium. Larvae were then transferred to the behavioral rig and allowed to acclimate for 15 min prior to recording.

Free-swimming activity was recorded at 50 frames per second (fps, Fig. 1A–C and Extended Data Fig. 1.2). During the recordings, 20 µl of a solution containing DEX, DMSO, and E3 was added with a micropipette to bring each well to the desired concentration of 0.02, 0.05, or 5 µM of DEX and <0.01% DMSO. For the control experiments, only DMSO plus E3 was used. Both spontaneous swimming and auditory responses were recorded for 25 min. During each auditory recording, stimuli (1 kHz tone for 10 ms at 83 dB SPL) were presented with an intertrial interval of 30 s. The parameters of the stimulus train were chosen based on behavioral analyses (Extended Data Fig. 2.1) as well as the analysis of auditory stimuli experienced by the animal in recent literature (Poulsen et al., 2021). Moreover, 1 kHz tones stimulate a large proportion of neurons in the auditory system (Constantin et al., 2020; Poulsen et al., 2021). and 83 dB SPL tones produce strong and reliable behavioral responses (Extended Data Fig. 2.1). Finally, a 30 s intertrial interval allows the fluorescence signal to return to baseline and does not produce appreciable habituation but is frequent enough to give numerous observations in an experiment of a feasible duration.

Figure 1.

Spontaneous activity under DEX. A, A representative trace of a free-swimming larva during a 5 min experiment. Free-swimming well configuration is shown in Extended Data Figure 1.1 and Materials and Methods. B, A representative trace of swimming speed over 10 s for a free-swimming larva. C, Movements of larvae (averaged across 22 animals for DMSO control and 25 animals for DEX) for a 15 min experiment, with the addition of DEX (red curve) or with the addition of DMSO control (black curve) at 5 min (dashed vertical line). DEX concentration tests in Extended Data Figure 1.2. D, A maximum-projection dorsal image of a larva expressing GCaMP6s in the nuclei of all neurons. Solid lines indicate the eyes, and dashed line indicates the outline of the brain. Colored dots show the locations of ROIs with fluorescence traces shown in E. E, Three examples of fluorescence traces, drawn from the ROIs indicated in D. F, Whole-brain Δf/f (averaged across 5 animals for DEX and DMSO control) for a 15 min experiment, with the addition of DEX (red curve) or DMSO control (black curve) at 5 min (dashed vertical line). Inset: Sum of the absolute value of the first derivative of the fluorescence signal, representing fluctuations in activity between 0–5 and 10–15 min in control (black, n.s. for a Friedman test) and DEX-treated animals (red, p = 0.02 for a Friedman test). R, rostral; C, caudal. Scale bars: A, 2 mm; D, 100 μm.

Figure 1-1

Well configuration for free-swimming experiments. A. 4 × 4 individual wells (20  mm diameter) with free swimming larvae. B. 4 individual wells (20  mm diameter) placed in a half circle with free swimming larvae. The speaker can be seen in the center between the wells. The wells’ centers (red dots) and borders (green outline) are automatically detected. Download Figure 1-1, TIF file ^{(917.4KB, tif)} .

Figure 1-2

Free swimming activity of larval zebrafish under variable DEX concentration. A. Average speed over time of 3 groups exposed to different DEX concentrations (values for each color in top right legend). DEX or DMSO control solutions are added at minute 5. No stimulus presented. B. Average speed over time of 4 groups exposed to different DEX concentrations (values for each color in top right legend) and auditory stimulation (1 kHz tone every 30  s, see Materials and Methods). DEX or DMSO control solutions are added at minute 5. Download Figure 1-2, TIF file ^{(830.8KB, tif)} .

For analyses of auditory responses at a higher temporal resolution, we recorded at 400 fps with a 0.5 ms exposure time (Fig. 2; Extended Data Figs. 2.1, 2.2). Videos were recorded from −2 to +2 s relative to stimulus onset stimulus (total of 4 s). Three sets of 5 min recordings were made. During each auditory recording, 10 auditory tones (1 kHz tone for 10 ms at 83 dB SPL) were presented, with an intertrial interval of 30 s. Immediately after the first recording, 20 µl of a solution containing DEX, DMSO, and E3 was added with a pipette to bring each well to a concentration of 0.05 µM of DEX and <0.0001% DMSO. After the introduction of 0.05 µM of DEX, we waited 5 min to allow for diffusion of the drug before the second recording began. Immediately after the second recording, each fish was carefully pipetted out of each well and placed in a well with fresh E3 (to eliminate contamination of DEX in the third recording). A solution of 20 µl ATI, DMSO, and E3 was added with a micropipette to bring each well to a concentration of 5 µM of ATI and <0.003% DMSO. The third recording started 10 min later.

Figure 2.

Auditory behavioral responses under DEX and ATI. A, Distribution of 1,994 bouts (across baseline, DEX, and ATI periods), measured as each bout's maximum speed. B, Average number of spontaneous forward swims, measured during the 2 s period prior to stimulus onset. C, Individual trials (top) and average speed (bottom) for auditory escape responses during baseline (red), DEX (blue), and ATI (orange). Stimuli loudness test in Extended Data Figure 2.1. D, Distributions (individual points), means, and SEMs for maximum escape velocities in each condition. p-values are shown for the Wilcoxon ranked-sum test. E, Forward swims during baseline (red), DEX (blue), and ATI (orange). The top row shows forward swim timings for all trials, and the bottom row shows average forward swim frequency across all trials. F, Forward swim duration, mean, and standard error of the mean for all trials. n = 28. Additional statistics showing nonsignificant changes in Extended Data Figures 2.2 and 2.3.

Figure 2-1

Free swimming behavioural responses to variable intensity of auditory tones. A. Average number of bouts (forward and escape) per seconds across auditory tone intensities. Blue is control (pre-DEX). Red is during DEX. Error bars are SEM. B. Average swimming speed pre-DEX during auditory stimulation for variable intensities (lighter colours for louder intensities, intensities from A). C. Average swimming speed during DEX and during auditory stimulation for variable intensities (lighter colours for louder intensities, intensities from A). N = 40 for each condition. Download Figure 2-1, TIF file ^{(509.5KB, tif)} .

Figure 2-2

Reaction time distribution for baseline (top), DEX (centre) and ATI (bottom). Distributions and median values are indicated. Download Figure 2-2, TIF file ^{(260.8KB, tif)} .

Figure 2-3

Effects of DEX condition on behaviour. A. Mean forward swims during DEX against mean forward swims at baseline. Each dot represents one fish. Straight line representing line of equivalence. R square of 0.07. B. Number of evoked forward swims in DEX minus the number of evoked forward swims during baseline against the number of spontaneous forward swims in DEX minus the number of spontaneous forward swims during baseline. Each dot represents one fish. R square of 0.06. N = 28. Download Figure 2-3, TIF file ^{(326.4KB, tif)} .

Concentrations of DEX and ATI were chosen based on behaviorally effective doses described in the literature (V. M. Bedell et al., 2020) and on the testing of drug concentrations shown in Extended Data Figure 1.2. Together, the results support the choice of a DEX concentration of 0.05 µM for effective sedation within minutes of administration and 0.5 µM ATI for effective recovery.

Extraction and analysis of free-swimming behavior with DeepLabCut

To extract the pose of each fish automatically, we employed DeepLabCut (DLC; Mathis et al., 2018). DLC is a deep learning model based on transfer learning, which means that the training starts using pretrained weights from the ImageNet dataset and residual network (ResNet) model, thus avoiding the need to train a model from scratch and reducing the size of dataset required. Specifically, we adapted this toolset, applied a 50-layer ResNet network architecture, and trained two pose models: a simple model with 3 key points (center of the left and right eyes and one in the center of the swim bladder) and a more detailed one with 15 key points: left and right eyes (2 key points per eye), swim bladder (1 key point), and tail (10 key points). Errors reported are of the Euclidean distance in pixels between the annotated positions and the positions predicted by the model.

The first model (three key points) was trained on 82 frames from videos obtained in the same experimental conditions in a 95%/5% train/validation split. This achieved an error of 0.88 pixels in the training dataset and 0.77 pixels in the validation dataset (95% train, 5% test split, reached after a total of 1,030,000 iterations).

The detailed 15-key point model was trained using 1,437 frames from 23 different videos. The videos used were acquired from pilot data from other experiments in our laboratory. The videos varied in lighting conditions, resolution, and acquisition frame rate. After training using a 90%/10% train/validation split, the model achieved a train error of 1.09 pixels and an error of 1.43 pixels in the validation dataset after 1,350,000 iterations.

The fish positions obtained from DLC were further analyzed in MATLAB using custom scripts (available on the public GitHub repository: https://github.com/ItiaFB/ZfishBeh_DLC/tree/main).

Light-sheet calcium imaging and drug administration

Separate groups of larvae at 6 dpf were used for whole-brain calcium imaging. A total of n = 10 larvae were imaged for the analyses in Figure 1 (n = 5 for DEX and n = 5 for DMSO control). The analyses shown in Figure 3 through Figure 6 included a single group of n = 7 larvae.

Figure 3.

Behavioral and brain-wide responses to acoustic stimuli under DEX and ATI. A, Time trace of average Δf/f (top) and simultaneous average tail speed (bottom) across seven larvae. See individual traces and controls in Extended Data Figures 3.1 and 3.2. Dashed vertical lines show the timings for DEX (5 min) and ATI (15 min) addition. Responses to acoustic stimuli are clear at 30 s intervals. B, Averaged fluorescence traces of whole-brain responses to acoustic stimuli during baseline (red, sampled between 3 and 5 min), DEX (blue, between 13 and 15 min), and ATI (orange, between 28 and 30 min). C, Decay rates of whole-brain average responses to tones for the data shown in B (each dot indicates one animal; bars indicate mean and SD). p = 0.009 across all three conditions (Friedman test), and with post hoc analysis, p = 0.03 (baseline vs DEX), 0.03 (DEX vs ATI), and 0.3 (baseline vs ATI). D, Standard deviation of whole-brain average response to tones for the same data. p-value is shown for a Wilcoxon signed-rank test.

Figure 6.

Lagged calcium responses. A, Example of lagged response (top). Example of GCamp6s trace moved to different onset times for cross-correlation with lagged response (bottom), with the darker color indicating a stronger cross-correlation. B, Histogram of the lags at which the peak cross-correlation values occurred, calculated across all stimulus events in whole brains, for baseline (red), DEX (blue), and ATI (orange) across seven fish. C, Region-by-region differences in lags at which the cross-correlation peaked for DEX versus baseline and ATI. Negative values indicate higher relative proportions of ROIs at a given lag for the DEX treatment versus baseline (red) or ATI (orange). Tel, telencephalon; OT, optic tectum; Pt, pretectum; Teg, tegmentum. D, Spatiotemporal maps of lagged evoked calcium response for baseline, DEX, and ATI (r > 0.95 for visualization). E, Energy (logarithm of inverse state probability of mean-squared displacement distribution, see Materials and Methods) against mean-squared displacement (MSD) of calcium traces and consecutive frame (Δt), across all fish. This represents the dynamic stability of brain activity across the three conditions: baseline (i), DEX (ii), and ATI (iii). R, rostral; C, caudal.

Figure 3-1

Individual brain activity and tail responses under DEX and ATI exposure. Left column shows average Δf/f over time of all ROIs in each fish undergoing DEX and ATI exposure. Right column shows tail movement for each fish over time. Vertical dashed lines show times where DEX (min 5) and ATI (min 15) were added. Download Figure 3-1, TIF file ^{(923.5KB, tif)} .

Figure 3-2

Individual brain activity and tail responses with DMSO control. Left column shows average Δf/f over time of all ROIs in each fish undergoing DMSO control. Right column shows the respective tail movement for each fish over time. Bottom line is the average across fish. Download Figure 3-2, TIF file ^{(790KB, tif)} .

Larvae were immobilized dorsal side up in 2% low melting point agarose (Sigma-Aldrich) on microscope slides. The agarose on the rostral and left side of the fish was cut vertically, parallel to the fish, with a scalpel, and removed. The embedded fish was transferred to a 3D-printed chamber (Marquez Legorreta et al., 2022). The agarose surrounding the tail was freed by removing segments of agarose perpendicular to the tail caudal to the swim bladder, and the chamber was filled with E3 media. Larvae were then transferred to a custom-built light-sheet microscope (M. A. Taylor et al., 2018) and allowed to acclimate for 15 min prior to imaging.

A mini speaker (Dayton Audio DAEX-9-4SM Skinny Mini Exciter Audio, Haptic Item Number 295-256), attached to the back wall of the chamber as previously described (Poulsen et al., 2021), was connected to an amplifier and to the computer to produce the auditory tones (83 dB, 1 kHz tone for 10 ms every 30 s) for the duration of the recording.

Volumetric calcium imaging was performed as previously described (Favre-Bulle et al., 2018). An exposure time of 30 ms was chosen for each plane during volumetric imaging, with a laser power output of 60 mW, which was attenuated to 1.5 mW for each of the two planes at the sample. A total Z scan of 125 μm was performed with 5 μm steps. This resulted in volumetric acquisition at 1.33 Hz. We commenced laser scanning 5 min prior to each new neural activity recording to eliminate responses to the onset of this off-target visual stimulus.

During the first set of recordings, 200 µl of DEX, DMSO, and E3 was carefully added manually with a pipette to bring the chamber to 0.05 µM of DEX and <0.0001% DMSO. During the second set of recordings, 200 µl of ATI, DMSO, and E3 was carefully added manually with a pipette to bring the chamber to 5 µM of ATI and <0.02% DMSO. Recordings were done in two sets to allow the removal of DEX in between recordings. A fluidic pump (Adelab Scientific, model NE-1000X) was connected to the chamber to permit media exchanges between recordings, and 15 ml (three chamber volumes) was pumped out while 15 ml of fresh E3 was delivered. A delay of approximately 10 min (5 min of slow water flow plus 5 min of laser scanning) separated the first and second recordings for each fish. Only fish that did not drift in the z-axis during the course of the whole experiment were kept for analysis. Therefore, the same neurons within the same fish have been imaged over 30 min through baseline and exposures to DEX and ATI.

Extraction and analysis of fluorescence traces

We used the Suite2p package to extract fluorescence traces from our raw images (Pachitariu et al., 2016). Most parameters used were default parameters. For “tau,” related to GCaMP6s dynamics, we used 1.4. Other parameters related to the recordings were determined from the recording conditions. The output regions of interest (ROIs) from Suite2p, generally representing individual neurons (G. Vanwalleghem et al., 2021), along with their corresponding fluorescent traces, were further analyzed in MATLAB.

Registration to a reference brain

For visualization in a common reference brain, we used Advanced Normalization Tools (ANTs, https://github.com/ANTsX/ANTs) to register our results to the H2B-RFP reference of Z Brain (Avants et al., 2011; Randlett et al., 2015) as previously described (Wong et al., 2020). A set of high-resolution image volumes with 1 µm between planes was used to build a common template before registering this template to the Z Brain Atlas. Motion-corrected, time-averaged stacks from Suite2p were then used to register individual fish to the common template and then to the Zbrain atlas as described (Marquez Legorreta et al., 2022). The resulting warps were sequentially applied to the centroids of extracted ROIs to map them all in the same frame of reference. The warped ROI coordinates were then placed in each of the 294 brain regions defined in the Z Brain Atlas (Randlett et al., 2015). ROIs that were located outside of the reference brain's boundaries after registration were discarded.

Linear regressions to acoustic stimuli

For linear regressions, regressors were built for each stimulus onset, with a typical GCaMP6s response occurring for each presentation of the stimulus as previously described (Favre-Bulle et al., 2018; Poulsen et al., 2021). The coefficient of determination (r²) of the linear regression models was used to select stimulus-responsive ROIs, and we chose a threshold based on the r² distribution of our models (0.6 for high selectivity).

To gauge auditory networks at baseline and during DEX and ATI exposure, we performed linear regressions to acoustic stimuli for three 5 min time windows: 0–5 min for baseline, 10–15 min for DEX, and 25–30 min for ATI.

Note that each stimulus was accompanied by motor activity; therefore, we have not differentiated between linear regression of motor and sensory activity.

Selection of ROIs with reduced spontaneous activity

For the detection of ROIs with reduced spontaneous activity during DEX administration, the sum of the derivative of Δf/f of each ROI trace was calculated for baseline (0–5 min), DEX (10–15 min) and ATI (25–30 min). Values smaller than 75% from baseline and ATI to DEX were included in our analysis. A comparison of the results yielded by a range of thresholds is shown in Extended Data Figure 4.1.

Energy landscapes

To quantify dynamic modes of variance and estimate the stability of the brain states, we analyzed stimulus-locked calcium traces using an energy landscape analysis (Munn et al., 2021; N. L. Taylor et al., 2022). Briefly, we formulated an energy landscape by first computing a one-dimensional measure of trajectories on the fluorescence time series data, namely, the mean-squared displacement (MSD), which is defined as follows:

$$\mathrm{MS}{\mathrm{D}}_{t,\tau }={⟨{|X_{t+\tau }-X_t|}^2⟩}_k,$$

averaged over all $k$ cells. The probability of observing a given MSD across the entire time series was then calculated using a Gaussian kernel density estimation:

$$P(\mathrm{MSD},t)={\displaystyle \frac{1}{4N}\sum _{i=1}^{n}K\left({\displaystyle \frac{\mathrm{MS}{\mathrm{D}}_{t,i}}{4}}\right),}$$

where $K(u)=\frac{1}{2\sqrt{\pi }}{e}^{-\frac{1}{2}{u}^2}$. As is typical in statistical mechanics, the energy of a given state, $E_{\sigma }$, and its probability are related by $P(\sigma )=\frac{1}{Z}{e}^{-E_{\sigma }/T}$ where $Z$ is the normalization function and $T$ is the scaling factor equivalent to temperature in thermodynamics (Tkačik et al., 2015). In our analysis, ${\sum }_{\sigma }{P}_{\sigma }=1\to Z=1$ by construction, and we can set $T=1$ for the observed data. Thus, the energy (E) of each MSD at a given time lag $t$ is then equal to the natural logarithm of the inverse probability, $P(\mathrm{MSD},t)$, of its occurrence:

$$E=\ln \left({\displaystyle \frac{1}{P(\mathrm{MSD},t)}}\right).$$

Using this approach, we analyzed dynamics in brain activity and the energies of those dynamics across space (ROIs) and time.

Experimental design and statistical analysis

Animal numbers for calcium imaging experiments were constrained by the labor intensity of the data collection methods and were n = 5 or n = 7 for each group, depending on the experiment (the n for each experiment is reported in figure legends and main text). Free-swimming behavioral experiments had an n of 28. As detailed elsewhere in Materials and Methods, DMSO controls were used during experiments involving the application of drugs. The statistical tests used depended on the details of the experiments and datasets, and these tests included paired t tests, Friedman test, Wilcoxon sign-ranked tests, and Wilcoxon ranked-sum tests. The details of these statistical tests are reported in the figure legends and main text for each experiment. All statistical analyses were performed with individual animals as data points to avoid pseudoreplication.

Results

Reduction of spontaneous behavioral and brain-wide neural activity under DEX

Guided by past work (V. M. Bedell et al., 2020), we first explored the impact of DEX on spontaneous swimming behavior in zebrafish larvae. Larvae at 6 dpf were placed and tracked in individual wells (Fig. 1A and Extended Data Fig. 1.1), and their spontaneous free-swimming activity was recorded over time. An example of a spontaneous trace is shown in Figure 1B. Free-swimming activity was recorded before and after the administration of 5 µM DEX or DMSO control (Fig. 1C). Results showed a complete cessation of spontaneous activity under DEX, but no changes for DMSO controls. We observed this cessation across various concentrations of DEX (Extended Data Fig. 1.2).

We next performed brain-wide cellular-resolution calcium imaging using GCaMP6s and a custom-built light-sheet microscope, as described previously (Favre-Bulle et al., 2018). From these data, we identified regions of interest (ROIs) corresponding to individual neurons (see Materials and Methods) and used signals from these ROIs as the basis for our analyses of brain-wide activity. Examples of a fluorescence image and activity traces extracted from individual ROIs are shown in Figure 1, D and E, respectively. Brain-wide fluorescence traces were recorded before and after the administration of 5 µM DEX or DMSO control (Fig. 1F), and the average changes of fluorescence across all ROIs were quantified as readouts of brain-wide spontaneous activity. We found that the addition of 5 µM DEX led to, in the first instance, a burst of activity across the brain and, after 2 min, a decrease in average spontaneous activity below baseline and DMSO control (Fig. 1C). This initial burst of activity was likely a function of water flow stimuli introduced by our addition of DEX or control solutions (Thompson et al., 2016; G. Vanwalleghem et al., 2020). Therefore, we restricted our further analyses to the 5–10 min window after drug administration, after the water flow has ceased and DEX has taken effect. We calculated spontaneous neural activity during this interval by calculating the sum of the absolute value of the first derivative of Δf/f, which measures the fluctuations in fluorescence intensity as a proxy for activity across the brain. During this interval, spontaneous neural activity was significantly reduced in the DEX group (Fig. 1F, inset), but not in DMSO control, indicating a drop in spontaneous brain activity that coincides with the cessation of spontaneous swimming behavior (Fig. 1C).

Auditory responses of larvae exposed to DEX and ATI

To explore the effects of DEX and ATI on sensorimotor gating, we recorded free-swimming behavior under control, DEX, and ATI conditions while presenting auditory tap stimuli (83 dB, 1 kHz pure tone for 10 ms, every 30 s, see Materials and Methods, Extended Data Fig. 2.1). Responses were recorded for 5 min before drug treatment, followed by the application of 0.05 µM DEX for 10 min, and finally 5 µM of ATI for 15 min. Swim bouts were categorized as either forward swims or escape responses based on their location on a bimodal distribution of maximum velocities (with a threshold of 90 mm/s, Fig. 2A).

As expected from our results for spontaneous swimming (Fig. 1), we observed an overall decrease in the probability of forward swim bouts during periods without acoustic stimuli (the 2 s prior to stimulus onset) in DEX compared with baseline (Fig. 2B, p < 0.01, Wilcoxon signed-rank). This effect was reversed by the addition of ATI (p < 0.01, Wilcoxon signed-rank). A small number of fish exhibited an increase in swim activity after DEX application, raising the possibility of a distinct population of larvae with different dynamics. Exploring this possibility, we found no correlation between forward swims recorded before and during DEX in individual fish, indicating that responses observed under DEX are not predictable based on pre-DEX activity (Extended Data Fig. 2.3). As such, we attribute increased activity in these larvae to natural variation in the population rather than the existence of a distinct group.

In contrast to the drop in spontaneous swimming, behavioral responses to acoustic stimuli remained robust. Reaction times were not significantly altered by DEX or ATI (Extended Data Fig. 2.2), and in trials where escape responses occurred, the strengths of those responses, measured as peak velocity (Fig. 2C), were significantly higher in DEX than at baseline or in ATI (Fig. 2D, p < 0.001, Wilcoxon ranked-sum). When restricting our analysis to forward swims only (Fig. 2E), we observed a significantly prolonged poststimulus period during which forward swim responses were elevated in DEX compared with baseline (Fig. 2E, p < 0.05, Wilcoxon ranked-sum). These forward swim bouts were also significantly longer, in terms of their kinematics, than swim bouts following stimuli at baseline or in ATI (Fig. 2F).

Overall, our results suggest that DEX, while it reduced or eliminated spontaneous swimming, still allowed auditory escape responses. Additionally, responses elicited by acoustic stimuli were stronger in DEX than at baseline. In all of these regards, the application of ATI reversed the effects of DEX, returning the animals’ behavior to an approximate baseline state.

Brain-wide auditory response dynamics in larvae under DEX and ATI

To investigate the effects of DEX and ATI on brain-wide dynamics and sensory processing, we recorded brain-wide activity while presenting acoustic stimuli (as in Fig. 2). Brain-wide calcium activity was recorded in head-embedded larvae for 5 min before drug treatment as a baseline, for 10 min after the addition of 0.05 µM DEX, and, finally, for 15 min after the addition of 5 µM ATI. Since larvae were head-embedded, we simultaneously recorded tail movements to ensure that sensorimotor transduction was intact. Consistent with our free-swimming results (Fig. 2), we observed a reduction in spontaneous movements of the tail in DEX, while responses to acoustic stimuli persisted (Fig. 3A; individual traces and controls are shown in Extended Data Figs. 3.1 and 3.2, respectively). Consistent with our findings in Figure 1, we observed a burst of activity following the administration of DEX and ATI, likely attributable to the water flow involved. Here again, we limited our analysis to the 5–10 min window after administration.

Paralleling these behavioral results, calcium imaging in these animals revealed continued brain-wide responses to acoustic stimuli during DEX exposure (Fig. 3A, top; individual animals’ responses and DMSO controls are shown in Extended Data Figs. 3.1 and 3.2, respectively). To analyze these auditory responses in more detail, we focused on responses occurring at times of full drug effects (3–5 min into the experiment for baseline, 13–15 min for DEX, and 28–30 min for ATI). We found that neural auditory responses under the influence of DEX are sharper (with significantly faster decay rate; Fig. 3B,C; Friedman test p = 0.009) and more stereotyped (with significantly lower standard deviation, Fig. 3D, p < 0.001, Wilcoxon signed-rank) than during baseline and that both of these effects are reversed with the application of ATI (Fig. 2C,D).

These results suggest that DEX had multiple effects on neural activity and behavior. It reduced spontaneous neural activity and swimming, but left auditory activity and responses intact and may, in fact, lead to clearer, sharper, and less variable responses to acoustic stimuli.

Identifying neurons with decreased activity in DEX

We have discovered that DEX treatment significantly reduces spontaneous brain-wide activity and nearly eliminates spontaneous behaviors, while preserving auditory processing and related behaviors. To further investigate the selective changes in neuronal activity, we calculated the overall fluorescence activity changes independently of auditory responses between baseline and DEX conditions (Fig. 4A,B; see Materials and Methods).

Figure 4.

Identification of ROIs with reduced activity under sedation. A, Distribution of changes in fluorescence activity from baseline to DEX conditions (see Materials and Methods). B, Spatial distribution of ROIs in A (colors from A). R, rostral; C, caudal. C, Raster plot of fluorescence traces for ROIs with at least a 50% reduction in fluorescence activity in DEX. Separation of auditory and nonauditory ROIs in Extended Data Figure 4.1. D, Average Δf/f through time of the fluorescence traces shown in C. E, Fourier transform of average Δf/f time trace (in D) for different time windows (2.5 min in duration) across the timeline. Dashed vertical lines show the timings for DEX (5 min) and ATI (15 min) addition. Examples of individual ROI Fourier transform in Extended Data Figure 4.2. n = 7 larvae.

Figure 4-1

Auditory and non-auditory ROIs with responses suppressed by DEX. A. A Δf/f heatmap for all ROIs characterized as auditory (r2 > 0.1 after linear regression), among the ROIs shown in Figure 4C. B. Average Δf/f trace for the ROIs in A. C. Locations of the ROIs represented in A. D. A Δf/f heatmap for all ROIs characterized as non-auditory (r2 < 0.1 after linear regression), among the ROIs shown in Figure 4C. E. Average Δf/f for the ROIs in D. F. Location of non-auditory ROIs. R: rostral; C: caudal. Download Figure 4-1, TIF file ^{(4.8MB, tif)} .

Figure 4-2

Examples of individual ROIs with decreased activity during DEX. Δf/f of 5 randomly chosen ROIs with decreased activity when exposed to DEX (left), drawn from the population of ROIs represented in Figure 4. Right: the same ROIs’ Fourier transforms for different time windows across the timeline. Download Figure 4-2, TIF file ^{(1MB, tif)} .

We found that a majority of ROIs exhibited minor fluctuations in overall fluorescence activity across conditions (Fig. 4A, white bars). However, a substantial number of ROIs showed notable increases or decreases in fluorescence activity, with most decreasing during DEX exposure (Fig. 4A, blue bars). Mapping these changes across the brain (Fig. 4B), we observed that ROIs with decreased activity are distributed across the telencephalon, lateral cerebellum, and multiple rhombomeres.

Focusing on ROIs exhibiting decreased fluorescence activity, we identified a large population (>1,700 ROIs across seven fish) in which activity dropped by >50% during DEX exposure (Fig. 4C). Examining the average fluorescence activity throughout the experiment, we found that responses to auditory stimuli only became evident within this population with the reduced background activity after DEX administration (Fig. 4D and Extended Data Fig. 4.1).

To assess whether auditory responses were present (if masked by spontaneous activity) in these neurons during baseline or ATI exposure, we performed a Fourier transform (Fig. 4E). Fourier transforms highlight repetitive patterns over time, and given our acoustic stimulus frequency (every 30 s, 0.033 Hz), we expected to detect the stimulus frequency (Fig. 4E, green arrow) and its harmonics (Fig. 4E, blue arrows). While clearer signals were evident during DEX treatment, we also observed the primary frequency and weaker harmonics during baseline and ATI, albeit with more noise (examples of individual ROIs’ Fourier transforms shown in Extended Data Fig. 4.2).

These results reinforced the idea that DEX's primary impact on the auditory network was a reduction in activity at times when acoustic stimuli were not present. The effect, especially in neurons whose spontaneous activity was most dramatically reduced, was to elevate the prominence of the auditory responses that these neurons have. This, in turn, had implications for how the brain-wide processing of acoustic stimuli may occur in DEX and may explain some of the effects of DEX that we have seen on the animals’ auditory behavior.

Auditory networks at baseline and during sedation

To identify the neurons composing the auditory processing network, we performed a linear regression between the acoustic stimuli and each ROI's activity (Bzdok and Ioannidis 2019). Given that each stimulus was accompanied by motor activity, we have not differentiated between linear regression of motor and sensory activity. In this analysis, ROIs whose responses are strongly correlated with stimuli have high r² values and are deemed to be involved in the perception or processing of the stimulus. We found that the distribution of r² values was shifted toward higher values during DEX treatment (Fig. 5A), indicating stronger correlations between the stimuli and the responses of ROIs across the brain as compared with baseline and ATI conditions.

Figure 5.

Detection of brain-wide auditory networks. A, Distributions of r² values for all ROIs’ responses to acoustic stimuli (red, baseline; blue, DEX; orange, ATI). Dashed vertical lines show a threshold of 0.6 used for panels C–E. B, Blue curve, Proportion (in %) of ROIs identified as auditory during DEX among ROIs previously shown to be auditory during baseline. Orange and red curves, Proportion (in %) of ROIs found to be auditory during baseline and ATI, respectively, within ROIs previously found to be auditory during DEX. C, Spatial distribution of auditory ROIs from A at (i) baseline, (ii) in DEX, and (iii) in ATI at high selectivity r² > 0.6 for clarity. Auditory responses were prominent in the ON, TS, Th, Cb, and the remainder of the hindbrain (Hb). R, rostral; C, caudal. D, Venn diagram showing the overlap of auditory ROIs from A at baseline, in DEX, and in ATI for r² > 0.6. E, Average Δf/f of ROIs found auditory during DEX with r² > 0.6 across different time windows (red for baseline, blue for DEX, and orange for ATI). See C, D, and E for other r² values in Extended Data Figure 5.1. n = 7 larvae.

Figure 5-1

Detection of auditory neurons under DEX and ATI using linear regression. Venn diagram (left) showing the overlap of ROIs deemed to be auditory during baseline (red), DEX (blue), and ATI (yellow). Corresponding spatial distributions of ROIs (center) from these populations. Corresponding average Δf/f of ROIs found auditory during DEX across different time windows (red for baseline, blue for DEX and orange for ATI). Results are shown for different r² thresholds: A. r²> 0.3. B. r²> 0.5. C. r²> 0.7. Download Figure 5-1, TIF file ^{(3MB, tif)} .

We observed that across a wide range of r² values (0.1 to 0.8), the majority (70 ± 5%) of ROIs identified as auditory responsive during baseline were also found to be auditory responsive during DEX (Fig. 5B, blue curve). On the other hand, we found a lower identity overlap between ROIs found to be auditory responsive during DEX compared with baseline or ATI (Fig. 5B, orange and red curves).

Using a stringent r² threshold (0.6) to define auditory responsive ROIs, we conducted a comparative analysis of their spatial distributions (Fig. 5C), identity overlap (Fig. 5D), and fluorescence traces (Fig. 5E) across experimental conditions (see Extended Data Fig. 5.1 for results using lower and higher thresholds). The spatial distributions of these ROIs (Fig. 5C) covered well-established auditory brain regions including the ON, TS, Th, Cb, and the remainder of the hindbrain. Notably, this analysis covered the Cb as well as the anterior and lateral sections of the hindbrain—previously identified motor-associated regions (Marquez Legorreta et al., 2022). The spatial distributions of these ROIs (Fig. 5C) revealed a notable increase in the number of ROIs identified as auditory during DEX. This appeared to be an essentially uniform increase across auditory brain regions. Examining the identity overlap (Fig. 5D), we found that ROIs uniquely associated with baseline, DEX, and ATI conditions account for 49, 71, and 12%, respectively, of their auditory ROIs. This highlighted DEX as particularly effective in capturing a significantly expanded population of auditory responsive ROIs. In Figure 5E, calcium responses of ROIs identified as auditory during the DEX condition were compared with those during baseline and ATI, illustrating changes in neural activity under DEX and consistently lower fluorescence levels in between stimuli.

Both the overlap in auditory ROIs (Fig. 5B,D; Extended Data Fig. 5.1) and the locations of these ROIs (Fig. 5C, Extended Data Fig. 5.1) suggest that, in DEX, the core auditory system could be expanded to include additional neurons, but these additional ROIs did not point to any particular functionally or anatomically distinct population of neurons that joined the auditory network as a result of sedation.

In total, the results from our behavioral and brain imaging experiments suggest that DEX reduced baseline activity in the brain without interfering with the processing of or response to acoustic stimuli. The reduction in spontaneous neural activity caused by DEX led to (1) reduced spontaneous swimming and (2) increased salience of auditory responses in the brain, since they are occurring over a quieter baseline. This increased salience appeared to be the reason for our detecting more auditory responses in DEX than at baseline, and this effect appeared to be roughly evenly distributed across the auditory network. All of these effects were reversed when ATI was applied, suggesting an adrenergic mechanism for the behavioral and network effects that we observed.

Temporal properties of auditory responses in DEX and ATI

In terms of the network-scale mechanisms, it was not immediately evident how DEX strongly suppressed spontaneous activity but permitted sensory processing. It was also not clear why the strengths of behavioral responses were increased in DEX. In order to better understand whole-brain dynamics, we next explored the temporal properties of auditory processing at baseline, in DEX, and in ATI, and described the relative stability of brain-wide networks under each of these conditions.

We calculated delayed correlations to explore auditory response dynamics in greater temporal detail. Specifically, for this study, we used a cross-correlation function to calculate the correlation value between the signals through time for each stimulus event to an ideal GCaMP6s response function and found the temporal lag that maximizes this activity for each ROI (Fig. 6A). Each ROI's response delay (or lag, given in number of successive measurements or frames) was thereby calculated for each stimulus event. This method, while highly sensitive to time delays, would also capture spontaneous activity that occurs randomly during the experiments. Neurons with such activity are prominent in the telencephalon at all time points, and this region has previously been shown to display high levels of spontaneous activity (Vargas et al., 2012). Such neurons’ inclusion in Figure 6D is presumed to result from this spontaneous activity rather than an involvement in auditory processing.

We generated distributions of the response lags for ROIs across the brain at baseline and in DEX and ATI (Fig. 6B). We found more stereotyped auditory responses at smaller lags (Δt = 0–1 frame, equal to 0–0.75 s) in the DEX condition than at baseline or in ATI, which had more heterogeneous and less synchronous responses shifted toward longer lags (peak at Δt = 2 frames, equal to 1.5 s). By restricting analyses to the ROIs belonging to specific auditory brain regions, we gauged how these lags are manifested in different parts of the auditory network (Fig. 6C). While responses in the telencephalon seemed consistent across different lag values, in all other brain regions studied, responses in DEX occurred earlier than the equivalent responses at baseline or in ATI. This is represented by negative values for (baseline-DEX) and (ATI-DEX) at short lags and positive values at longer lags (Fig. 6C). While the temporal details of these quicker responses were slightly different across regions, there was a consistent preponderance of shorter lag responses in DEX and longer lag responses both at baseline and in ATI. These results suggest that this brain-wide auditory network processed information more quickly when DEX was present.

Viewing the spatial distribution of all ROIs for all three conditions at each lag (Fig. 6D), we could see the spatiotemporal distribution of auditory responses, illustrating both the earlier occurrence of responses in DEX (consistent with Fig. 6B: 1–2 frames, representing 0.75–1.5 s) and the brain regions in which these temporal effects were most dramatic (consistent with Fig. 6C: all auditory brain regions except for the telencephalon). The brain-wide correlations returned to a near-baseline state after three frames (2.25 s) for DEX, while this took roughly four frames (3 s) for baseline and ATI.

Finally, we looked at the moment-to-moment stability of whole-brain dynamics by measuring brain state energy transitions (Fig. 6E, see Materials and Methods). To do so, we calculated the likelihood that neuronal activity (i.e., the brain state) would change by a prespecified amount in a specific window of time (Munn et al., 2021; see Materials and Methods). If arousal is thought to increase the general excitability of the brain, it is expected that the brain will change less through time during sedation as compared with baseline and ATI conditions. Low-probability brain state transitions are associated with a high energy barrier, whereas higher-probability state transitions are associated with a lower energy barrier. Using all time series from all seven fish, we found that the energy landscape under DEX was initially deeper relative to baseline, suggesting a stable attractor state (Fig. 6Ei), whereas baseline and ATI were shallow, reflecting a more unstable and temporally variable set of state transitions (Fig. 6Eii,iii). DEX exposure led to a deeper energy landscape, a more stable attractor state, and a lower probability of state transitions.

Discussion

We have found that the application of the sedative DEX, which indirectly inhibits noradrenergic neurons in the LC, led to a brain-wide suppression of spontaneous activity, a deepened network-wide energy landscape that made state transitions less likely, and a cessation of spontaneous swimming behavior. All of these effects were reversed by the application of ATI, which led to the restoration of noradrenergic signaling. We observed that the loss of spontaneous brain activity, which we presumed to be the root cause of the other network and behavioral effects, did not prevent elicited activity in the brain. Indeed, responses to acoustic stimuli became sharper and more widespread in the brain-wide auditory network and, stronger behaviorally, when DEX was applied.

These relationships among spontaneous brain activity, sensory processing, and behavior have important implications for how animals make context-dependent decisions. An alert animal navigating its environment must take its surroundings and recent experience into account when making behavioral decisions. Numerous external and internal factors impinge on sensory processing and sensorimotor gating. These range from simple mechanisms for sensorimotor gating, such as prepulse inhibition (PPI) and habituation, to more general context, including the presence of predators, social factors, and general arousal (Pereira and Moita 2016). We propose that many of these contextual features are encoded, at least at short time intervals, in the network's spontaneous activity. Our results using DEX suggest that, with the loss of baseline activity in the brain, the auditory processing pathway functions unimpeded by contextual considerations, resulting in stronger, broader, and sharper activity in response to acoustic stimuli and in more dramatic behavioral responses. It would be interesting in future work to explore which types of sensorimotor gating are impacted by DEX. For instance, it is possible that PPI is fundamental to the auditory processing pathway and would therefore still take place under DEX, while the effects of general arousal would depend on baseline activity and therefore be lost during sedation.

The current results also showed how the brain-wide auditory network changes with the removal of its noradrenergic components, especially from the LC. Specifically, we find that neurons whose spontaneous activity was most reduced by DEX are concentrated in the lateral cerebellum and elsewhere in the hindbrain, including in regions downstream of the LC (Wang et al., 2022). Unsurprisingly, these regions containing neurons with reduced spontaneous activity (Fig. 4) also showed an increase in neurons strongly correlated to acoustic stimuli (Fig. 5) in DEX. Caudal regions of the hindbrain, where premotor neurons are abundant, showed a particular enrichment of auditory responsiveness in DEX, consistent with stronger sensory transmission and reduced sensorimotor gating when noradrenergic signaling from the LC is inhibited.

Notably, the reversal of this blockade with ATI restored normal behavior and brain activity according to all of the analyses that we have applied. These reversals included a return to normal spontaneous swimming (Fig. 2) and spontaneous activity in the brain (Figs. 2, 4) and also a return to baseline values for acoustic escape responses and the decay rates of brain-wide auditory processing (Fig. 3). Closer analyses of the auditory network revealed that similar numbers of neurons with similar correlations to acoustic stimuli were observed at baseline and after the reversal of blockade with ATI and, importantly, that ATI restored a brain-wide distribution of auditory neurons that was spatially indistinguishable from the baseline distribution (Fig. 5). All of these results suggest that ATI, by releasing the blockade on noradrenergic signaling from the LC, precisely reversed the effects of DEX, rather than introducing a separate compensatory effect to restore behavior.

Looking more closely at the temporal properties of auditory responses, we observed that auditory processing was initiated at shorter lags and that it was abbreviated, under DEX exposure (Fig. 6). These effects occurred on the timescale of seconds, which was too long [even accounting for GCaMP kinetics (T. W. Chen et al., 2013)] to be directly involved in the generation of behavioral responses that typically occur in tens of milliseconds (Zeddies and Fay 2005; Burgess and Granato 2007b; McClenahan et al., 2012; Jain et al., 2018). We speculate that extended activity in the auditory network may play a role in establishing context for future decision-making. Future tests of this hypothesis could involve rapid repetitive acoustic stimulation in the presence of DEX to see whether habituation occurs during sedation. It would also be interesting to see whether recall of a habituated state (established during sedation) is possible after sedation is lifted. These tests would establish whether the neural mechanisms for habituation rest in the auditory processing pathway itself or whether they also depend on the formation of novel patterns of activity elsewhere in the brain. Overall, the data presented here provide a tantalizing glimpse of the connections that exist among baseline brain activity, sensory processing, ethological context, and behavior, leaving many mechanistic questions open to future exploration.

Our identification of auditory ROIs depended on their correlations to the acoustic stimuli (see Materials and Methods), and on this basis, we found a greater number of auditory ROIs during sedation (Fig. 5). The simplest explanation for this effect is that the removal of background activity led to higher correlations for all ROIs with roles in auditory processing. Based on their distributions and response profiles, there was no widespread evidence that new ROIs were recruited to the auditory network during sedation but rather that their responses became more salient in the absence of spontaneous activity, allowing some neurons to exceed the r² threshold for inclusion as auditory while affected by DEX. This interpretation was bolstered by our observations of activity in the Fourier domain (Fig. 4), in which temporal patterns of activity that were clear during sedation were nonetheless present (in a noisier form) at baseline and in ATI. As such, while sedation is an unnatural state, it is one in which sensory processing can be detected more sensitively, and where subtle network activity that would normally be missed might become discoverable. As an example from the current study, we observed auditory ROIs in the raphe nucleus under DEX exposure, but not at baseline or in ATI (Fig. 5C). The raphe has not been identified as part of the larval zebrafish auditory system in previous whole-brain analyses (G. Vanwalleghem et al., 2017; Privat et al., 2019; Poulsen et al., 2021), but past literature focusing on the raphe has shown that it modulates sensory information (Yokogawa et al., 2012) and is activated during acoustic stimuli (Pantoja et al., 2016), potentially giving it an important role in auditory processing or decision-making. While the contributions of such circuit elements would eventually have to be described in a natural, unsedated state, their initial identification could be aided by observing sensory processing under sedation.

The use of a sedated state to map sensory networks—particularly when employing regression-based techniques and optical physiology—holds promise for broader applications addressing various sensory modalities and model systems. As shown in this study, sedation significantly reduced background noise throughout the brain of larval zebrafish, which allowed us to see the auditory network with greater sensitivity. The same effect should be applicable to other sensory modalities in zebrafish and to sensory studies in other model systems.