The lonely fish is not a loner fish: whole-brain mapping reveals abnormal activity in socially isolated zebrafish

The zebrafish is used to assess the impact of social isolation on behaviour and brain function. As in humans and other social species, early social deprivation reduces social preference in juvenile zebrafish. Whole-brain functional maps of anti-social isolated fish were distinct from anti-social fish found in the normal population. These isolation-induced activity changes revealed profound disruption of neural activity in brain areas linked to social behaviour, such as the preoptic area and hypothalamus. Several of these affected regions are modulated by serotonin, and we found that social preference in isolated fish could be rescued by acutely reducing serotonin levels.

Fish were placed in full isolation from fertilization to 3 weeks, as described in the Methods section, and then tested in the behavioural assay. Each experiment comprised two sessions, 15 minutes of acclimation to the chamber followed by 15 minutes of exposure to two size matched sibling fish, which were not isolated. To quantify social preference, we calculated a Visual Preference Index (VPI) that compares the amount of time fish spend in the chamber nearest the conspecifics versus the opposite chamber where they are visually isolated from social cues (see Methods). Full social isolation (Fi) caused a significant decrease in social preference relative to normally raised sibling controls (C) ( Figure 1A, left and middle panel: C vs Fi, p= 8.3 e -8 , Mann-Whitney). Specifically, there was an increase in the number of individuals that had a large negative VPI. We, therefore, decided to divide the fish into three groups: a) anti-social (-S) fish with VPIs below -0.5; b) pro-social (+S) fish with VPIs above +0.5; c) non-social (ns) fish with -0.5 < VPI < +0.5. Fish that underwent Partial isolation (Pi), i.e. isolated from social cues for 48 hrs immediately prior to the behavioural experiment, exhibited an intermediate, yet highly significant, change in social preference (Fig.1A, right panel, C vs Pi, p=2.5e -8 , Mann-Whitney).
As previously reported (Zellner, Padnos, Hunter, MacPhail, & Padilla, 2011), we found that fish raised in full isolation were significantly less active than their normally raised siblings during the acclimation session ( Figure. 1B, C vs Fi, p=9.0e -6 , C vs Pi, p=2.8e-9 Mann-Whitney). However, when we compared the swimming activity of anti-social isolated and anti-social control fish, there were no significant differences ( Figure 1C; C (-S) vs Fi (-S), p=0.48, Mann-Whitney). We then compared the swimming behaviour of isolated and normally raised anti-social (-S) fish in more detail and noticed that fish from both groups exhibited prolonged periods of quiescence (freezing) ( Figure 1C left, C (-S) vs Fi (-S), p=0.48, Mann-Whitney). This freezing behaviour is consistent with similar anxiety behaviours observed in many species, and previously reported in adult zebrafish exposed to stressors (Giacomini et al., 2015;S Shams, Seguin, Facciol, Chatterjee, & Gerlai, 2017). The pro-social fully isolated fish, Fi (+S), instead, showed a different behavioural phenotype compared to the pro-social normally raised fish, C (+S), exhibiting an increase in freezing behaviour especially when observing conspecifics ( Figure 1C right, Figure 1D, and Supp. Movie 1).
Anxiety-like behavioural responses have been shown to increase following periods of social isolation (S Shams et al., 2017) and they are also known to vary across individuals in a population (Egan et al., 2009). The behavioural similarities between anti-social fish isolated and anti-social fish found in the normal population led us to hypothesize that maybe isolation predisposes fish to a high-anxiety state. If this is the case, neural activity of anti-social isolated and normally raised fish should be similar when presented with social cues. To test this hypothesis, we performed whole-brain two-photon imaging of c-fos expression, an immediate early gene whose expression is associated with increased neural activity (Herrera & Robertson, 1996), in juvenile brains following testing in the social preference assay. Dissected brains were imaged with the dorsal surface down (bottom-up) to achieve clear views of the ventral brain structures that have been previously implicated in the social brain network (Figure 2A, see Methods). Volumes of 1.5 mm x 1.5 mm x 700 µm, with a voxel size of 1x1x3 µm (2048x2048 pixels), were acquired from 135 zebrafish brains across all experimental groups and registered to a reference brain (Marquart et al., 2017).
These c-fos whole-brain functional maps allowed us to compare the neural activity patterns of different test groups. For example, normally reared pro-social fish, C (+S) and a control group of sibling fish that were placed in the behavioural assay for 30 minutes without any social cue (no social-cue controls: (nsc)). The resulting average difference stacks (C (+S) vs. C (nsc)) were used to identify changes in neural activity associated with exposure to a visual social cue ( Figure 2A).
We first investigated the regions activated in pro-social normally raised fish ( Figure 2B C (+S)) that have been previously reported to be social brain areas (O'Connell & Hofmann, 2011). Coronal sections are shown for two different areas: caudal hypothalamus and the preoptic area. Significant activation in a region of the dorsal caudal hypothalamus was a prominent signature of C (+S) fish ( Figure 2B and 2D, C(+S) vs C (nsc) p=0.007, Mann-Whitney). In contrast, anti-social, C (-S), fish showed a large decrease in activation in an adjacent, non-overlapping area of the ventral caudal hypothalamus ( Figure 2B and 2D, C (-S) vs C (nsc) p=0.003, Mann-Whitney test). This functional division in the hypothalamus seems to mirror the distribution of dopaminergic and serotoninergic markers (Filippi, Mahler, Schweitzer, & Driever, 2010;Lillesaar, 2011) (Figure 2C), suggesting that this area could be responsible for the changes of serotonin and dopamine levels widely reported in fish viewing social cues (Huang et al., 2015;Soaleha Shams, Amlani, Buske, Chatterjee, & Gerlai, 2018). In the preoptic area, we observed only a small dorsal (PM) increase and ventrolateral (Pa) decrease in activity, which was significant only in the Pa of anti-social control fish (C(-S) vs C (nsc) p=0.003 Mann-Whitney, Figure 2B and 2D). These results are consistent with brain areas that have been previously identified as involved in social behaviour in a number of species (O'Connell & Hofmann, 2011).
We were now able to ask whether neural activity was similar in anti-social isolated and anti-social normally raised fish following the presentation of a social cue. Contrary to our hypothesis, c-fos functional maps of anti-social fully isolated fish (Fi (-S), Figure 2B) revealed a completely different activity profile than their anti-social normally raised siblings (C (-S), Figure 2B). The ventral caudal hypothalamus of Fi (-S) fish was not inactivated, and the preoptic was strongly activated in both dorsal (PM) and ventral (Pa) regions, but significantly only in the PM ( Figure 2B and 2D, Fi (-S) vs Fi (nsc) p=0.006 PM; p=0.07 Pa, Mann-Whitney). Similar functional activity changes were observed in fish exposed to a brief isolation of only 48 hours prior to testing. (Pi (-S), Figure 2B), albeit less strong (Pi (-S) vs Pi (nsc) p=0.04 Pa; p=0.04 PM, Mann-Whitney, Figure 2D). The altered functional brain maps of isolated fish provide strong evidence that social isolation, while reducing social preference behaviour, does not do so by the same mechanism that produces social aversion in normally reared fish.
We were also interested in understanding why social isolation promotes social aversion instead of increasing the drive towards social cues. An important clue was found in the pattern of brain changes that were unique to isolated fish. When we compared activity levels in the functional brain maps of fully isolated fish (Fi (nsc); Figure 3A) which had never been exposed to social cues to normally raised controls, we found a significant increase in areas associated with visual processing (the optic tectum (McDowell, Dixon, Houchins, & Bilotta, 2004)) and stress responses (the posterior tuberal nucleus, PTN (Ziv et al., 2013).
In pro-social control fish (C (+S), viewing a social cue causes a significant increase of neuronal activity in visual brain areas like the tectum ( Figure 3B top; C (+S) vs C (nsc) p=0.004 Mann-Whitney). However, in fully isolated fish, there was already an increase in neuronal activity in the tectum in the absence of social cues (Fi (nsc)). This suggests that isolation increases visual sensitivity, as previously reported in humans (J T Cacioppo, Cacioppo, Capitanio, & Cole, 2015); (Figure 3B = Fi (nsc) vs C (nsc) p=0.0005 Mann-Whitney). This increased sensitivity of fully isolated fish (Fi (nsc)) was not found in partially isolated fish, Pi (nsc), suggesting that the visual sensitization effects of isolation are cumulative. In addition, increased tectal activity was also found in both full and partially isolated anti-social fish that were exposed to social cues (Fi(-S) and Pi(-S), even though these fish largely avoided the chamber with visual access to conspecifics (Fi (-S) vs C (nsc) p=0.016, and Pi (-S) vs C (nsc) p=0.0007, Mann-Whitney, Figure 3B).
A similar cumulative response was observed in the posterior tuberal nucleus of fully isolated fish. Full isolation caused a significant increase in PTN activity in absence of social cues (Fi (nsc) vs C (nsc) p=0.015, Mann-Whitney, Figure 3B bottom), and in anti-social fish (Fi(-S). During partial isolation, PTN activity was slightly increase and variable in absence of social cues (Pi(nsc), but significantly increased as well in antisocial fish (Pi(-S)). In humans, loneliness is associated with stronger activation of the visual cortex in response to negative social images (John T. Cacioppo, Norris, Decety, Monteleone, & Nusbaum, 2009) and it has been shown to increase attention to negative social stimuli (e.g. rejection, social threats, and exclusion). We therefore propose that social isolation in fish initially activates neuronal and neuroendocrine responses that promote anxiety-like behaviors, such as increased vigilance for social threats, hostility and social withdrawal, as observed in humans (John T. Cacioppo et al., 2009).
To test whether reducing anxiety could reverse the anti-social behaviour observed in isolated zebrafish, we acutely treated control and partially isolated fish with Buspirone, a drug that has been shown to reduce anxiety in humans, mice, and zebrafish (Bencan, Sledge, & Levin, 2009;Lalonde & Strazielle, 2009;Lau, Mathur, Gould, & Guo, 2011;Patel & Hillard, 2006) Buspirone, an agonist of the auto-inhibitory 5HT1A receptors, has been shown to enhance social interaction of rats (File & Seth, 2003;Gould et al., 2011) and sociability of zebrafish (Barba-Escobedo & Gould, 2012), and social phobia in humans (Schneier et al., 1993;van Vliet, den Boer, Westenberg, & Pian, 1997) Its ability to counter the effects of social isolation in zebrafish has not been investigated.
We first tested the effects of acute exposure to Buspirone in control fish, and, as expected, we observed a small significant increase in social preference relative to untreated controls (Supp. Figure 1A-B: C (no drug) vs C (30 µM) p=0.01 Mann-Whitney). We then treated partially isolated fish with 30 µM (Supp. Figure 1A-B, n=46 fish) and 50 µM (Supp Figure 1A-B, n=72 fish) Buspirone. Remarkably, the acute drug treatment was sufficient in both concentrations to reverse the anti-social phenotype caused by isolation ( Figure 1A and 4A; Pi vs Pi (Buspirone 30 µM and 50 µM combined), p= 2.56 e-05 Mann-Whitney).
When we compared the time course of this phenotype reversal by computing the VPIs for each minute throughout the 15 minutes of the behavioural experiment (Figure 4 B), we found that the Buspirone treated isolated fish, while initially anti-social, would rapidly recover normal social preference behaviour within the first 5 minutes of exposure to social cues (VPI at 0 and 3 minutes between Pi and Pi(Buspirone) p= 0.01, and p=0.09 Mann-Whitney). In cotrast, the VPIs of untreated isolated fish remained significantly lower than controls throughout the entire session. Buspirone's impact on the rate of recovery of social preference suggests it may be reducing anxiety by regulating circuit plasticity, perhaps by promoting down-regulation of the hypersensitivity acquired during the isolation period.
In summary, our results demonstrate that lonely fish, which have been isolated from social cues and show anti-social behaviour, have a completely different brain activity compared to loner anti-social fish found in the normal population. In addition, the functional changes caused by social deprivation are consistent with an increase in anxiety state found in humans, and could be reversed with an existing anxiolytic drug that acts on the monoaminergic system. Zebrafish, thus, will provide a powerful new tool for studying the impact of loneliness (isolation) on brain function and exploring different strategies for reducing, or even reversing, its effects.

Animals
AB strain zebrafish maintenance and breeding was performed at 28.5C with a 14h:10h light-dark cycle. Isolated fish were housed in custom chambers (length=15 cm, width=5 cm, height= 10 cm) made of opaque white acrylic with translucent lids, either from fertilization (full isolation) or for 48 hours prior to the behavioural experiment (partial isolation). All experiments were performed according to protocols approved by local ethical committee (AWERB Bloomsbury Campus UCL) and the UK Home Office.

Behavioural assay and analysis
Experimental details and image acquisition were performed as described previously (Dreosti et al., 2015) and the source code can be found at http://www.dreo-sci.com/resources/. The visual preference index (VPI) was calculated by subtracting the number of frames in which the fish was located on the side of the arena nearest the social stimulus (Social cue (SC) side in Figure 1B) by the number of frames located on the opposite side of the arena (nsc (No SC) side). This difference was then divided by the total number of frames recorded [VPI = (SC -No SC)/Total frames].

Imaging and registration
A custom built two-photon microscope (INSS) was used for image acquisition of whole-brain in situs. Both DAPI and Cy3 Images were collected with a 10x objective (Olympus, W Plan-Apochromat 10x/0.5 M27 75mm) using a "Chameleon" titanium-sapphire laser tuned to 1030 nm (Coherent Inc., Santa Clara, CA, U.S.A.) and controlled using custom written software in LabView.. Registration  The registered image stacks were then normalised to adjust for intensity variations between imaging sessions. caused by a variety of sources (staining efficiency, laser power fluctuations, light detector sensitivity, etc.). Normalization was accomplished by computing an intensity histogram for each fish brain's volume (with 10000 discrete intensity bins spanning the range -4000.0 to 70000.0) for all 512*512*273 voxels. The minimum value bin (with at least 100 voxels) was used as the bias offset, and subtracted from all voxel values. The mode value, minus the bias, provided a robust estimate of the background/baseline fluorescence and was thus used to normalize voxel values for the entire volume. Therefore, after normalization, intensity value of 1 reflected the background level, 2 reports fluorescence level that is twice the background, and so on. Histogram normalization was performed for each individual fish brain volume prior to any region or voxel-based analysis. Figure 2B and 3A Reconstruction of cross section images were obtained by using the Fiji "Volume viewer" plugin. Schematics of cross-and horizontal-section were obtained by using the "Neuroanatomy of the zebrafish brain" Figure 2D and 3B Percentages of c-fos activation were calculated for each of the 6 different areas highlighted in Figure 2B and 3A, using custom written Python functions, in the following way. A 3D mask for each area was generated by using the "Segmentation Editor" plugin Fiji (https://imagej.net/Segmentation_Editor). C-fos percentage values for each condition (C (+S), C (-S), Fi (-S), Pi (-S)) were obtained by subtracting and then dividing each c-fos average value of the mask by the basal c-fos average value calculated in control fish No Social Cue.

Statistics
Statistical analysis was performed using Python scipy stats libraries. Since VPI, percent time moving, and c-fos activity distributions were generally not normally distributed, we used the non-parametric Mann-Whitney U-test of independent samples for hypothesis testing throughout the manuscript.

Drug treatment
Juvenile fish were treated with 30µM or 50µM Buspirone (Buspirone HCl, Sigma) for 15 minutes prior the experiment. After washing, fish were run through the behavioural assay. Each fish was used only once.

Data availability
All the images, video, protocols, analysis scripts, and data that support the findings of this study are available either from this website (http://www.dreo-sci.com/resources/) or from the corresponding author upon request.    Fig. 3A (green) for each experiment condition: normally raised controls without social cue C (nsc); pro-social controls C (+S); anti-social controls C (-S); anti-social fully isolated Fi (-S); fully isolated without social cue Fi (nsc); anti-social partially isolated Pi (-S); partially isolated without social cue Pi (nsc).