An integrative assay for measuring social aversion and motivation in freely behaving mice

Summary

Social aversion is a key feature of numerous mental health disorders, yet we lack adequate behavioral tools to interrogate social aversion in model systems. Here, we developed a behavioral task—selective access to unrestricted social interaction (SAUSI)—that integrates elements of social motivation, hesitancy, and free interaction to enable a multiplexed assessment of social aversion. Using SAUSI, we discovered that prolonged social isolation induces social aversion in mice—an effect largely driven by increases in social fear coupled with decreases in social motivation. Application of deep learning approaches revealed unique behavioral motifs underlying the socially aversive state produced by isolation, demonstrating the compatibility of modern computational pipelines with SAUSI. Last, we demonstrated that unique forms of social aversion can be induced by distinct stressors, highlighting the versatility of SAUSI. Our findings debut a fresh task for the behavioral toolbox—one that offers an integrative approach for assessing social aversion.

Graphical abstract

Highlights

• •SAUSI enables the multiplexed assessment of social aversion behaviors
• •SAUSI reveals that prolonged social isolation induces a state of social aversion
• •Unsupervised computational analyses identify distinct social states revealed by SAUSI
• •Distinct forms of social aversion are induced by social vs. non-social stressors

Motivation

An integrative assessment of social aversion behavior has been difficult to achieve due to the multi-faceted nature of social aversion and limitations in the ability of a single behavioral assay to capture multiple aspects of social behavior. Current behavioral assays used to assess social aversion either probe for changes in social motivation with restrained mice or allow for free interaction between mice without an element of social choice, but not both. To address this unmet need, we developed selective access to unrestricted social interaction, a behavioral assay capable of measuring social motivation, fear, and hesitancy in freely interacting mice.

Grammer et al. present SAUSI, a behavioral assay that measures social motivation, fear, and hesitancy in freely interacting mice. Using this assay, they show that prolonged social isolation induces social aversion and find distinct forms of social aversion induced by social vs. non-social stressors.

Introduction

A number of mental health disorders are known to include elements of social aversion, yet we lack the ability to adequately identify social aversion in model systems, limiting our capacity to interrogate the neurobiological mechanisms underlying this state.^1,2 While social aversion encompasses changes in social motivation, it also entails changes to a host of additional behaviors, such as social hesitancy and social fear. Currently, there is no behavioral assay that is able to test both social motivation (social choice, instrumental behavior-based access to social interaction, social approach, social hesitancy, etc.) and social fear (social freezing and reactivity) simultaneously. Various tasks have been developed to independently examine social motivation or social fear. Tasks involving operant access to a social conspecific have been used to index social reward and motivation.^{3,4,5,6,7,8,9,10,11} Other social motivation tasks, such as the three-chamber sociability assay,¹² can be used to measure approach behavior and interaction time^1,12,13; however, these assays often involve the confinement of the social conspecific mouse. In contrast, assays used to measure social aversion in response to perceived social threat allow mice to freely interact with each other but lack elements of social choice or motivation. These include the resident intruder (RI) assay, where a docile Balb/c intruder is placed into the home cage of the resident experimental mouse,¹⁴ and free social interaction, including aggressivity toward the intruder mouse, can be measured. Others include free interaction between two mice in a novel arena.¹⁵ While each of these assays is able to reveal behaviors critical to understanding the state of a rodent, none of them allows for the multiplexed, integrative examination of both social motivation and aversion within a single assay.¹⁶

Recent years have seen an explosion in computational tools and analysis pipelines for the estimation and analysis of complex animal behavior. For example, Social Leap Estimates Animal Poses (SLEAP),¹⁷ DeepLabCut,¹⁸ and Mouse Action Recognition System (MARS)¹⁹ are programs that use deep learning algorithms to track body key points of multiple animals and/or estimate animal pose. Large datasets generated by postural tracking software are then used for further supervised and/or unsupervised analyses. Behavioral analysis programs such as MARS¹⁹ and the behavioral classification system Simple Behavioral Analysis (SimBA)²⁰ allow behaviors initially identified by an experimenter to be automatically classified and quantified, reducing user error, bias, and discrepancies in scoring across—and even within—labs.^19,20 Unsupervised behavioral analysis pipelines such as Motion Mapper²¹ and MoSeq²² can computationally identify patterns of motion, detecting clusters composed of behaviors, poses, and movements that could not otherwise be detected with the human eye. Additional machine learning-based programs for behavioral analysis include Stytra,²³ Anipose,²⁴ Janelia Automatic Animal Behavior Annotator (JAABA),²⁵ DeepEthogram,²⁶ TRex,²⁷ Ctrax,²⁸ OptiMouse,²⁹ DeepPoseKit,³⁰ 3-Dimensional Aligned Neural Network for Computational Ethology (DANNCE),³¹ 3D virtual mouse,³² and more, highlighting the speed at which innovation in this space is occurring. These tools not only reduce bias and increase efficiency, but also allow us to visualize behavior in innovative and novel ways. Thus, the development of novel behavioral assays that lend themselves to these techniques is pivotal in a new age of validity, reproducibility, and translational meaning.³³

Here, we develop selective access to unrestricted social interaction (SAUSI), a single, integrative assay that allows for assessment of motivated social choice, hesitancy, decision-making, and free social interaction between pairs of mice and lends itself to modern machine learning and computational approaches. Because prolonged social isolation (SI) has been shown to reduce social motivation^13,34,35 and negatively impact other social behaviors,^15,36 we tested the ability of chronic SI to produce social aversion using SAUSI. We discovered that prolonged SI promotes a state of social aversion, characterized by multiple features, including social fear, social hesitancy, and reduced social motivation. We applied computational tools to SAUSI-generated behavioral data and identified isolation-specific behavioral motifs that significantly differed from those identified in group-housed (GH)control mice. We compared SAUSI with traditional assays conducted in the home cage or in a three-chamber apparatus and found that only SAUSI was able to unveil multiple components of social aversion. Finally, we demonstrate that SAUSI can be used to identify social aversion produced by a distinct experience, footshock stress, establishing the utility and broad applicability of this assay. These findings highlight SAUSI as an assay for the modern neuroscience toolkit, well equipped for the behavioral and computational analysis of complex social behaviors such as social aversion.

Results

Selective access to unrestricted social interaction: Design and development

We designed an arena that allows us to combine free, naturalistic interactions between an experimental mouse and a conspecific mouse in one compartment (social chamber) with free behavior of the experimental mouse in a separate compartment (home chamber). We further designed this arena to enable assessment of motivated social behavior, such that the experimental mouse can choose to access social interaction with the conspecific, or withdraw from interaction, via a tunnel that connects these two chambers (Figures 1A and 1B). Similar to the three-chamber task for sociability, the two chambers allow us to measure how much time a mouse prefers to spend in the social vs. the home chamber. However, instead of limiting social interaction by physically restricting the conspecific under a barrier, SAUSI allows for free interaction between the two mice when in the social chamber. Moreover, SAUSI employs an instrumental element to this social choice, by introducing a tunnel connecting the two chambers, where experimental mice must perform a tunnel crossing to gain access to social interaction. The narrow tunnel, which is compatible with hierarchy testing,³⁷ also allows for the measurement of precise time points in the mouse’s social decision-making as they approach the social chamber, and reveals hesitancy in mice when they shelter in the tunnel for discrete periods of time or decide to back out after initiating a crossover.

Figure 1.

Prolonged isolation promotes social aversion in SAUSI

(A) SAUSI apparatus side view (left) and top view (right). Numbers indicate the measurements (in centimeters) for each labeled arrow in the diagrams.

(B) SAUSI protocol. Experimental mice are habituated to the tunnel for 3 days (green). Conspecifics receive 1 day of tunnel deterrent training (yellow).

(C) Social hesitancy was increased in females by isolation, as indexed by reverse in tunnel behavior (interaction between the effects of housing condition and sex, F(1, 63) = 5.916, p = 0.018. No main effects of sex or housing condition were found. No significant differences were found for first tunnel sheltering or latency to enter the social chamber.

(D) Social fear was significantly increased by isolation. There was a significant interaction between housing and sex, F(1, 63) = 8.406, p < 0.01, as well as main effects of housing condition, F(1, 63) = 17.86, p < 0.0001, and sex, F(1, 63) = 8.748, p < 0.01, for social freezing. Social reactivity was also affected by isolation with main effects for housing condition, F(1, 63) = 5.889, p = 0.018, and sex, F(1, 63) = 4.823, p = 0.032, but no interaction (p > 0.9).

(E) Social motivation was negatively altered by isolation as indexed by significantly decreased prosocial initiations, with main effects of housing condition, F(1, 63) = 9.215, p < 0.01, and sex, F(1, 63) = 8.766, p < 0.01, but no interaction (p > 0.8). SI did not affect social chamber preference, although there was a significant decrease in preference for males, F(1, 63) = 5.378, p = 0.024.

(F) Social aversion, represented as the Z score average of all behaviors in (G), was significantly increased following SI, with a main effect of housing condition, F(1, 63) = 12.76, p < 0.001, and a significant interaction effect, F(1, 63) = 6.162, p = 0.016, but no main effect of sex (p > 0.9).

(G) Summary of SAUSI behaviors that combine to generate social aversion.

(H) Logistic regression analyses revealed factors most influential to the social aversion score and the accuracy of computationally determining original housing condition based on behavior, regardless of sex.

n = 18 females (circles), n = 16 males (triangles) per housing condition; for (C)–(G), two-way ANOVAs, followed by Fisher’s least significant distance (LSD) for post hoc comparisons. Error bars represent SEM. Effect of housing for males or females: ∗p < 0.05, ∗∗∗∗p < 0.0001. Effect of sex for GH or SI mice: ^#p < 0.05, ^##p < 0.01, ^####p < 0.0001.

See also Figures S1–S3 and Videos S2 and S3.

To ensure that conspecifics stay in the social chamber, they are prevented from crossing to the opposite home side by receiving deterrent training prior to the test day (Figures 1A and 1B). Footshock boards were designed as the primary form of deterrent to keep conspecifics from crossing to the opposite side.^38,39 BALB/c mice were used as conspecifics, as this strain has been shown to be relatively docile, removing potential confounds related to intruder aggression.⁴⁰ To ensure that conspecific mice do not cross over into the home chamber or block the tunnel entryway, we used an entry deterrent strategy, wherein a single, mild footshock was delivered via an electric board positioned directly in front of the tunnel if the mouse stood on the board (Figure 1A). Footshock boards were connected to grid scramblers (Med Associates), with mice completing the loop if at least one limb touched the positive and one limb touched the negative components of the footshock board (Figure S1). To determine the minimum shock intensity (mA) required to serve as a deterrent without significantly altering the conspecifics’ behavior, we conducted a behavior response curve experiment (Figure S2A). BALB/c mice were shocked at either 0 mA (controls), 0.1 mA., 0.2 mA, 0.3 mA, or 0.5 mA whenever they touched the footshock board across a 5-min training session. Mice were given one 5-min footshock deterrent training session/day for 2 days (with automated footshock delivery if they stepped on the footshock board) and then tested with a GH C57BL/6 experimental mouse on day 3 to assess social behavior and efficacy of the footshock training (Figure S2A). We found that 0.5- and 0.3-mA shocks, but not 0.1 mA or 0.2 mA, were sufficient to act as deterrents by significantly reducing the amount of total time on the footshock board (Figure S2B). However, 0.5 mA resulted in significant changes to the BALB/c’s behavior, such as increased freezing (Figure S2C). In contrast, 0.3 mA worked as a deterrent while not significantly impacting BALB/c behavior (Figure S2C). There were no changes in sniffing, thigmotaxis, or aggression across all groups. Locomotion was reduced by all levels of shock, even those that were not effective as a deterrent (Figure S2C). Critically, a 0.3-mA shock resulted in no changes to behavior of the experimental mouse when interacting with the mildly shocked conspecific (Figure S2D). Collectively, this design allows for the assessment of free social interaction, solitary behaviors, social motivation, social hesitancy, and decision-making.

Prolonged SI promotes social aversion

SI is known to promote social aversion.^{13,15,34,36,41} Yet, it has been difficult to test using rodents as a model system due to the lack of an appropriate assay that can measure multiple features of social aversion.¹ Using SAUSI, we tested whether prolonged SI results in increased social aversion in male and female mice. We found that isolation increased social hesitancy behaviors, including tunnel reversals (reversing out of the tunnel once the decision was made to cross over to the social chamber) (Figure 1C) and the latency to enter the social chamber (Figure 1C). These effects were more pronounced in females than males. No significant differences in tunnel sheltering during the first crossover were found (Figure 1C). We also found that prolonged SI produced an increase in social fear behaviors¹ such as social freezing (freezing in response to being sniffed by the conspecific) and social reactivity (jumping or darting in response to being sniffed) in both males and females (Figure 1D).^41,42,43 Isolated females showed higher rates of social freezing than isolated males (Figure 1D). Additionally, SI caused a reduction in prosocial initiation regardless of sex, while preference for the social chamber remained unaltered (Figure 1E). Interestingly, females spent longer in the social chamber than males overall. Finally, isolation increased the tonality of ultrasonic vocalizations (USVs) and decreased slope regardless of sex (Figure S3A). There was no effect of housing condition on the total number of USVs made; however, female pairs made significantly more calls than males (Figure S3A). Interestingly, aggression was also significantly increased following SI regardless of sex (Figure S3B), suggesting that isolation can induce aggression in a manner distinct from territoriality. In contrast, no changes were found with investigatory sniffing behavior (Figure S3C). Although isolation did not impact non-social freezing, isolated females showed heightened thigmotaxis during the baseline phase, indicating increased generalized anxiety (Figure S3D). Mobility during the baseline phase was not altered by isolation but was reduced in females compared with males (Figure S3E). Collectively, these results reveal that multiple social aversion behaviors are induced by SI and can be identified using SAUSI.

Next, we sought to harness the multiplexed nature of SAUSI and visualize the behavioral changes observed as a total cumulative, integrative social aversion score. We converted the datasets for each behavior in the categories of social fear, social hesitancy, and social motivation into Z scores and calculated the average Z score as an index of social aversion (Figure 1F). Because lower numerical social motivation scores (prosocial initiation and social chamber) indicate higher levels of social aversion, these scores were inverted (multiplied by −1) in the average Z score calculation. The social aversion Z score was significantly higher in socially isolated mice compared with GH controls, but only for females (Figure 1G), demonstrating that the effect of prolonged isolation to induce social aversion is predominantly driven by females. When we performed multi-logistic regression analyses on combined male and female data, we were able to determine whether a mouse was GH or isolated with 75.4% training accuracy (Figure 1H). The coefficients of this logistic regression equation reveal the behavioral features that most heavily drive our social aversion score (Figure 1H). The components that influenced the social aversion score most strongly were social freezing and social reactivity, indicating that social fear behaviors are the most important factors contributing to social aversion in isolated mice (Figure 1H).

SAUSI reveals the social state of an animal

Social behavior in rodents is highly complex and dynamic, continually shifting as animals engage in back-and-forth interactions. Recent advances in machine learning and computational approaches for the analysis of social behavior have enabled a deeper, broader, and less biased understanding of social behavior.^{33,44,45,46,47,48} To test whether SAUSI lends itself to next generation computational approaches for animal tracking, pose estimation, and analysis of behavior, we applied a variety of computational tools to videos obtained using SAUSI (behavior displayed in Figure 1).

Thirty-six high-speed videos were loaded into SLEAP, which allows for the tracking and pose estimation of multiple behaving mice.¹⁷ We tracked eight body points on each mouse (Figure 2A) and cleaned the data to include only two tracks (one track/animal). After initial tracking was performed, experimenters blind to experimental conditions manually corrected any switched tracks to maintain animal identity and interpolated the videos. Next, we extracted features from the data and tested whether these features could distinguish the social state of an animal (SI vs. GH condition). We used Motion Mapper,²¹ which enables the visualization of changes in feature space identified by stereotyped patterns of motion (Figure 2A). Using the Motion Mapper pipeline, x,y coordinates for each body point extracted from SLEAP were processed using a series of dimensionality reductions. The raw data were converted into angles, normalized, and analyzed using principal-component analyses (PCA) for the first dimensionality reduction. Next, we used UMAP to compute distances in high-dimensional space and neighboring points and embedded them on a new low-dimensional space, allowing for visualization of patterns of motion. Finally, we used watershed segmentation to separate the continuous stereotyped movements into groups (or regions), which were identified as collections of features or behavioral motifs (Figure 2A). Motion Mapper identified 10 distinct regions using data from all groups combined (Figure 2B). Occupation within each feature space was visualized for both isolated and GH mice (Figure 2C), and time spent engaging in each specified region was quantified, revealing increased time in regions 9 and 10 for isolated mice (Figure 2D).

Figure 2.

Deep learning approaches detect unique behavioral footprint for isolation-altered behavior in SAUSI

(A) Computational workflow for unsupervised behavioral classification.

(B) Map of behaviors that were generated using Motion Mapper.

(C) Heatmap distribution of time on the behavior map for GH and SI mice.

(D) Time spent occupying regions 9 and 10 were significantly increased in SI mice compared with GH controls.

(E) Multi-logistic regression performed on all behaviors identified using the Motion Mapper pipeline revealed feature importance as indexed by logistic regression coefficients displayed for each feature.

(F) Nose-to-nose distance extracted from SLEAP tracking output was not significantly different between GH and SI mice.

(G) Raster plots from a single animal in each group, depicting manually scored behavior (colored bars) and nose-to-nose distance (black) during SAUSI (n = 18 females per group; independent samples t tests, two-tailed).

Error bars represent SEM. ∗∗∗p < 0.001.

To test whether housing condition could be decoded based on Motion Mapper-generated features, we employed multi-logistic regression analyses and determined housing condition accuracy and importance of each feature (Figure 2E), allowing us to determine whether a mouse was GH or socially isolated with 75% training accuracy. The coefficients of this equation revealed regions 9 and 10 to be the most impactful in distinguishing between GH and SI mice. This demonstrates that the social state of the mouse (GH or SI) can be predicted from unsupervised, feature-based analysis extracted from postural tracking data.

Last, using data extracted from our SLEAP tracking, we assessed the distance between the nose of the experimental and conspecific mouse and found a numerical increase in distance for isolated mice (Figure 2F). When assessing how behaviors change in relation to nose-nose distance, we found a stark contrast between isolated mice and GH controls: SI mice display increased social hesitancy behaviors prior to engaging in social interaction, and, as mice gain proximity, these behaviors give way to social fear behaviors (Figure 2G). In conclusion, machine learning approaches revealed distinguishable GH and SI states, as well as dynamic changes to behavior depending on nose-nose distance, based on features extracted from behaving mice tracked during SAUSI. Although we expect this pipeline to be similarly applicable to other assays, we believe that the rich amount of data obtained from freely behaving mice in SAUSI allows for the ability to decode the social state of an animal.

SAUSI is compatible with perturbation and imaging approaches in tethered mice

Many neuroscience experiments employ tethering systems to allow for neural manipulations, recordings, or imaging during freely behaving animals. To address this, we adapted SAUSI to accommodate head mounts and tethering by developing an open-top tunnel design. This modification features a U-shaped channel in between the two outer chambers instead of an enclosed tunnel. We found this modification to be compatible with mice mounted with a miniature microendoscope (Inscopix) and wired to a DAQ box, as well as mice tethered for optogenetics experiments (Video S1). Next, we tested whether the presence of a tether affected a mouse’s ability to use the open-top tunnel or to engage in social aversion behaviors. We mounted a miniature microendoscope on male and female mice (n = 6) surgically injected with a virus encoding the calcium indicator GCaMP6f and implanted with a GRIN lens over the extended amygdala. Following recovery, mice were isolated (∼3 weeks) and run on SAUSI using a within-subjects design such that after habituation each mouse was tested twice—once under tethered conditions and once under untethered conditions (control), order counterbalanced (Figure 3A). We found that mice were able to use the tunnel with 3 days of habituation (Figure 3B), regardless of tethering. There were no differences in the number of tunnel crossings during the baseline phase of SAUSI testing days when mice were tethered (Figure 3B). Furthermore, there were no differences in social aversion behaviors between tethered and untethered conditions, and mice were able to display social aversion behaviors across all categories, including social hesitancy in the tunnel (Figures 3C–3E). Ultimately, these results suggest that tethering does not affect SAUSI behaviors.

Figure 3.

SAUSI is compatible for use with tethered mice

(A) Repeated measures experimental design. Mice were tethered for 3 days of SAUSI habituation. During the first SAUSI test, one-half of the mice were tested with a mounted microendoscope + tether and the other one-half served as controls and were tested untethered. On the second SAUSI test, the same mice were tested in the opposite condition (order counterbalanced).

(B) Tunnel use over the 3 SAUSI habituation days (all mice tethered). No differences in tunnel usage (number of tunnel crossings during the baseline phase) were found between tethered mice and controls.

(C) Tethering did not alter social fear behaviors.

(D) Tethering did not alter social hesitancy behaviors.

(E) No differences in social motivation were found between tethering and controls (n = 6 per group; dependent samples t tests, two-tailed).

See also Video S1.

SAUSI contrasts with traditional assays to reveal social aversion

To test whether SAUSI is unique in its ability to allow for the assessment of social aversion or whether social aversion could be similarly probed using other, well-established behavior assays, we compared behavior on SAUSI with behavior in the three-chamber social interaction assay^12,49 and the RI assay.^14,50 We used all females in these experiments because isolation-induced changes in social aversion with SAUSI were largely driven by females (Figure 1). In the three-chamber social interaction assay,^12,49 experimental mice were placed in the center chamber of a three-chamber arena. A sex- and age-matched conspecific was placed in one chamber under an overturned cup, while an inanimate object (e.g., plastic block) was placed under a cup in the opposite chamber. Given the lack of free social interaction due to the restrictive nature of the three-chamber design, we were unable to assess fear-specific behaviors (i.e., social freezing and social reactivity). However, we were able to see a reduction in social interaction preference and social approach preference (Figure 4A). There was no significant difference in social interaction latency in the three-chamber assay (Figure 4A), similar to what was observed in SAUSI (Figure S3A). In addition, mice showed minimal vocalizations in the three-chamber assay (Figure 4A), with no differences between GH and SI mice. There were no significant group differences in call slope or tonality (Figure S4A), in contrast with tonality and slope effects observed using SAUSI (Figure S3A). These data reveal the importance of the free interaction component present in SAUSI as social fear-related behaviors were largely absent in the three-chamber. Furthermore, limiting social interaction with physical restriction altered the number and quality of USVs in mice. These differences in social communication could impact the behavioral dynamics between two mice, further distancing them from a more naturalistic social setting.^51,52,53,54 However, it should be noted that the three-chamber assay is able to generate an isolation-induced reduction in social chamber preference that is not detectable in SAUSI (Figure 4E). This emphasizes that, while SAUSI is well suited to detect social aversion, there are nevertheless conditions under which other assays would be more appropriate (e.g., if attempting to look at social chamber preference).

Figure 4.

Traditional assays fail to detect full suite of social aversion behaviors

(A) During the three-chamber test, isolated mice showed a reduction in social interaction preference (mouse preference score = $\frac{mouseinteractiontime}{mouse+objectinteractiontime}$) and social approach preference (approach ratio = $\frac{\#mouseapproaches}{\#objectapproaches}$), but no significant changes in latency to approach the mouse cup. Vocalizations were unaltered by isolation. (n = 21 mice per group; independent samples t tests, two-tailed).

(B) During home cage RI testing, isolated mice did not display social freezing, but did engage in higher amounts of sniffing, aggression, and vocalizing (n = 8 mice per group; independent samples t tests, 2-tailed).

(C) During RI-novel cage, no significant differences in behavior were found (n = 8 mice per group; independent samples t tests, two-tailed).

(D) During RI-no bedding, isolated mice displayed social freezing behavior but no increases in sniffing or aggression (n = 8 mice per group; independent samples t tests, two-tailed).

(E) Summary table summarizing effects of isolation in SAUSI, RI home cage (H.C.), and three chamber.

n.a., not applicable.

Error bars represent SEM. ∗p < 0.05, ∗∗p < 0.01.

See also Figure S4.

Next, we examined whether we could identify isolation-induced social aversion in the RI assay, which takes place in the home cage of the experimental mouse.^14,50 Experimental resident mice (either GH or SI) were tested with a 3-min baseline (no intruder) followed by a 10-min test phase, during which a novel, age- and sex-matched intruder mouse from a more docile strain (BALB/c) was introduced into the cage and the mice were allowed to freely interact. We found that resident mice (regardless of housing condition) spent virtually no time engaging in social fear behaviors, despite being free to demonstrate these behaviors (Figures 4B and S4A). We also found no changes in prosocial initiation (Figure S4B). Instead, we found a significant increase in sniffing behavior and aggression for SI vs. GH mice (Figure 4B). There was also a significant increase in the number of USVs produced by isolated mice (Figure 4B); however, changes to slope or tonality were not significant (Figure S4B). As with the three-chamber assay, these results support the idea that SAUSI is unique in its ability to reveal social aversion, while at the same time highlighting the utility of an assay such as RI in cases where investigators are interested in assessing territorial aggression.

We hypothesized that the failure to find any SI-induced changes in social fear behaviors during the RI assay may be due to competition with territorial behaviors (e.g., sniffing and aggression),⁵⁵ as well as an inability to escape from the environment.⁵⁶ To determine whether territorial behavior competes with social aversion behavior, we next performed RI in a novel cage (with new bedding), as RI aggression has been shown to be highly context dependent.^14,56 We found that, in a novel cage environment, isolation did not produce increases in social freezing, sniffing, or aggression (Figures 4C and S4C). To test whether this could be due to the presence of bedding material, as opposed to plastic flooring (as used in SAUSI), we tested animals on the RI assay in a novel cage without bedding. Our RI-novel-no-bedding experiment revealed significant isolation-induced social freezing behavior (Figure 4D), but no other significant differences were found (Figures 4D and S4D). This indicates that the absence of bedding may be a driving force of social freezing.

SAUSI probes for social aversion across distinct forms of stress

To assess whether SAUSI is generalizable—capable of detecting social aversion induced by other forms of stress—we tested the impact of a non-social stressor, footshock, to induce social aversion in SAUSI (Figure 5A). Male and female mice were subject to a series of random footshocks in a novel context or simply exposed to the context for the same amount of time (controls). All shocked mice acquired fear to the shocked context, as previously shown⁵⁷ (Figure 5B). When tested for social aversion using SAUSI, we found that shocked mice show significantly increased social aversion when compared with unshocked controls (Figure 5C). Further analyses of social aversion sub-categories revealed that social hesitancy was significantly increased in shocked mice, specifically in the latency to enter the social chamber and time spent reversing in the tunnel (Figure 5D). Surprisingly, and in contrast with SI, shocked mice did not display increased social fear behaviors when compared with controls, with almost none of the mice showing social freezing or shock-induced differences in social reactivity (Figure 5E). Interestingly, males were overall more socially reactive compared with females (Figure 5E), revealing sex differences in social aversion that may be stressor specific. Finally, social motivation was significantly reduced in shocked mice, including a reduced number of prosocial initiations and percentage of time spent in the social chamber, effects primarily driven by males (Figure 5F). While a decrease in baseline mobility was found in mice that received a shock, there were no changes to USVs, aggression, sniffing, or generalized anxiety (Figures S5A–S5E). Shocked females displayed increased USV tonality, a greater number of USVs, more aggression, and more non-social freezing and thigmotaxis compared with males (Figures S5A–S5D). Overall, our results indicate that footshock-induced social aversion behaviors are largely driven by changes in social motivation and social hesitancy, which are more prominent in males then females. This differs from isolation-induced social aversion, which was driven by social fear and more pronounced in females then males. These data reveal nuanced differences in social aversion and motivation behaviors that differ across distinct forms of stress, highlighting the importance of the nature of stressor and the general utility of SAUSI.

Figure 5.

Footshock stress induces a distinct form of social aversion

(A) Experimental design.

(B) Total percent freezing during footshock exposure. Mice that received footshock (shock) froze for significantly more time than controls (CNTL), with a main effect of shock, F(1,28) = 191.8, p < 0.0001, but no effect of sex (p > 0.5) or interaction (p > 0.3).

(C) In SAUSI, social aversion was significantly increased following footshock compared with controls, with a main effect of shock, F(1,27) = 13.49, p = 0.001, but no significant effect of sex (p > 0.5) or interaction (p = 0.088).

(D) Social hesitancy behaviors—latency to enter the social chamber and reversing in the tunnel—were significantly increased by footshock stress. Latency to enter the social chamber revealed a significant main effect of shock, F(1,27) = 6.743, p = 0.015, and shock × sex interaction, F(1,27) = 5.214, p = 0.031, but no effect of sex (p = 0.093). Reverse in tunnel had a significant effect of shock, F(1,27) = 9.912, p < 0.01, but no effect of sex (p > 0.5) or interaction (p > 0.3). There were no significant differences in first tunnel sheltering.

(E) Footshock stress did not impact social fear behaviors. No effects were found with social freezing. Social reactivity revealed a main effect of sex, F(1,27) = 12.02, p < 0.01, but no effects of shock (p > 0.9) or interaction (p > 0.3).

(F) Social motivation scores including the number of prosocial initiations and time spent in the social chamber were decreased after footshock. For prosocial initiation, there was a main effect of shock, F(1,27) = 8.079,p < 0.01, but no effect of sex (p > 0.9) or interaction (p > 0.5). Social chamber preference had a main effect of shock, F(1,27) = 9.657,p < 0.01, but no effect of sex (p > 0.8) or interaction (p = 0.063).

n = 8 females (circles), n = 8 males (triangles) per group. A two-way ANOVA was performed to analyze the effect of footshock and sex. Fisher’s LSD was used for post-hoc comparisons. Error bars represent SEM. Effect of footshock for males or females: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Effect of sex for control or shocked mice: ^##p < 0.01.

See also Figure S5.

Discussion

SAUSI represents an integrative assay that can be used to assess social motivation, hesitancy, decision making, and in-depth free social interaction in mice. SAUSI achieves this goal by providing mice with the choice to access a social chamber, while also allowing for free social interaction following that choice. Using this assay, we characterize social aversion behavior—identifying critical features that drive this state: social fear, social motivation, and social hesitancy. Importantly, SAUSI is compatible with USV recordings to monitor changes in vocal communication; lends itself to computational, machine vision-based approaches; and can be used in tethered mice.

Our results demonstrate that the ability to reveal isolation-induced changes in behavior are largely assay and context dependent. Certain behavioral assays or environmental contexts reveal territorial-driven changes to behavior following SI, while others are more sensitive to social motivation changes. Our results indicate that it is critical to choose an assay that is best suited for the behavior of interest, as competing behaviors (e.g., aggression) or other environmental factors (e.g., bedding) may obscure or dominate findings. Here, we demonstrate that SAUSI is uniquely poised to reveal the influence of SI on social aversion.

While we expected isolation to induce aggression in our RI assay,^13,34,35,36 we were excited to find SI also induced aggression in SAUSI. These suggest that isolation-induced aggression may be offensive in nature. While distinctions between offensive and defensive aggression have been described extensively,^14,58,59,60 as indicated by posture, advancement, and location of attack or submissive posture and flight behaviors, respectively, the circumstances and emotional states that motivate particular classes of attack remain relatively unknown. Recent studies suggest that there is a rewarding component for the aggressor, which could provide motivation seek aggressive encounters.^{5,61,62,63,64} Other theories suggest the establishment of dominance as key.¹⁴ However, SAUSI reveals that isolation-induced aggression is likely intrinsically motivated (proactive or offensive) in nature.^{5,61,62,63,64} This is demonstrated by the fact that isolated mice repeatedly choose to enter the social chamber in order to engage in immediate and repeated attacks without provocation or reciprocation. This characteristic of isolation-induced aggression is something that one could not have determined with the classic RI assay, as animals are forced to be in the same chamber and thus may be more restricted in their choice of whether to engage in attack. Indeed, it can often difficult to determine whether an attack is offensive or defensive in the RI assay. Interestingly, although footshock stress induced social anxiety, it did not cause an increase in aggression in SAUSI. This finding is consistent with the idea that footshock-induced aggression⁶⁵ is more reactive or defensive in nature.^{5,61,62,63,64} Finally, in contrast with the repeated training required to assess aggressive motivation,^5,66 SAUSI provides the ability to assess motivation for aggression in a more naturalistic setting, without training and with free interaction.

One intriguing discovery we made is that SAUSI allows for a number of behaviors to be displayed along a continuum, mapped with regard to conspecific proximity. This finding is consistent with the theory that animals display a series of stereotypical and species-specific defensive behaviors across a predatory imminence continuum.^{67,68,69,70,71} Similar to the series of stereotyped behaviors an animal produces in response to distinct stages of imminence with regards to a predator, we find that animals produce a series of distinct behaviors depending on conspecific imminence in SAUSI. When the distance between the experimental mouse and the conspecific mouse is large, animals display increased hesitancy behaviors. As experimental mice gain proximity to the conspecific, they demonstrate increased hesitation to approach the social chamber or shelter for long periods in the tunnel. Finally, once mice enter the social chamber and begin to interact with the conspecific, they engage in social freezing and heightened social reactivity in response to being investigated. This continuum reveals the emergence of social aversion^67,68,69,70 and highlights SAUSI’s unique design and ability to distinguish between different phases of behavior with respect to conspecific imminence.

In the current study, behavioral changes induced by SI and footshock stress were found to be sex dependent. The stronger isolation-induced social aversion displayed by female mice is consistent with human literature demonstrating that women are impacted by stress and social anxiety at higher rates than men.^72,73 Some evidence suggests that social anxiety has altered clinical representation based on sex differences.^73,74,75 While social stress has been linked to the development of social anxiety in men and women alike,⁷⁶ the unique, sex-specific impacts that SI may have on social anxiety are unknown. Interestingly and in direct contrast with what we found with isolation, footshock stress induced stronger social aversion in male mice compared to female mice, suggesting that while both males and females are able to display social aversion, the factors driving social aversion for each sex are likely distinct. This evidence suggests the importance of studying both male and female subjects in social aversion research and highlights the interesting possibility that diverging states can emerge from the same stressor,^{13,77,78,79,80} or that the same state can emerge from distinct stressors. These differing outcomes provide additional motivation to understand the neurobiology of isolation and the manifestation of distinct behaviors that arise from this state.

Using new behavioral tools to investigate social aversion is critical to understanding the neurobiology underlying this state. Moreover, social anxiety disorder (SAD), the fear of social situations, affects approximately 25 million adults in the United States alone,^81,82 and it is estimated that more than 5 million adults in the United States have been diagnosed with autism spectrum disorder (ASD),⁸³ another disorder characterized by disturbances to social connection. Each year, ASD- and SAD-related disorders have steadily risen,^84,85 particularly in the wake of the growing epidemic of loneliness,⁸⁶ suggesting that far more people are currently impacted. Despite the prevalence of SAD and ASD, our understanding of their underlying neurobiology is limited, hindered by a lack of readouts for behaviors induced by these disorders, such as social aversion. Thus, our findings not only debut SAUSI as a multiplexed assay with which to assess social aversion but also set the stage for critical discoveries in our understanding of the neurobiology of social anxiety and associated disorders.

Limitations of the study

Considerations for tethered-animal experiments

As with other assays, the use of a tether introduces complexity to the task. Our open-top tunnel allowed mice to move freely across chambers, with minimal cord snagging. In vivo imaging quality was unaffected by cord catching, but these instances should be monitored and optimized for each setup. Intervention by an experimenter to free a caught cord can be reduced by having the appropriate amount of slack in the cord (not too long) and a well-serviced cord with no obstructions (no tape or bends on the cord). We recommend the use of a commutator to ensure a greater range of motion for the mouse. Additional days of habituation are recommended for mice with heavy head mounts to ensure their ability to maintain an upright head position.

Alternatives to footshock-based deterrence

Although 0.3 mA of shock delivered to conspecifics resulted in no changes to social behaviors, footshock in general is a stressful experience. Thus, it is important to include appropriate controls in each experiment. We attempted two other BALB/c deterrent methods: experimenter handling and auditory deterrent. Handling deterrent consisted of removing conspecifics out of the tunnel by their tail prior to and during behavioral testing. This method was only somewhat effective and gave rise to issues with conspecifics sitting in front of the tunnel entrance and blocking it. Handling-deterred mice were consistently interested in crossing over, requiring repeated experimenter intervention. Auditory deterrent training consisted of playing a 2 kHZ white noise at 50% amplification. White noise was delivered in response to conspecifics entering the space directly in front of the tunnel and during crossing attempts. Auditory-deterred mice primarily froze in place, blocking the tunnel entrance while the tone was played. Additionally, this noise was disruptive to experimental mice. Experimenters may attempt to identify other deterrent approaches but should ensure that none of these leave the conspecific blocking entrance to the tunnel, as this impacts behavior of the experimental mouse and their ability to freely cross over into the social chamber.

Generalizability of SAUSI

Adolescent SI and footshock stress are two behavioral manipulations that we have shown can induce social aversion behaviors. Other behavioral and/or genetic manipulations (i.e., different mouse strains, genetic mouse lines for autism and social anxiety, or other forms of psychosocial stress) should be performed and mice tested on SAUSI to determine the generalizability of the assay to other forms of social aversion. Slight variations to protocol (e.g., the number of habituation training sessions or the mA of footshock used for deterrence) should be expected; hence, pilot experiments are critical when using SAUSI in novel contexts.

Resource availability

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Moriel Zelikowsky (moriel.zelikowsky@neuro.utah.edu).

Materials availability

All items described here are commercially available or open source available at Zenodo: https://doi.org/10.5281/zenodo.15610884.

Data and code availability

• •All data reported in this paper will be shared by the lead contact upon request.
• •All original code has been deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884 and is publicly available as of the date of publication.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank the University of Utah Machine Shop for working with us and our design to help build the SAUSI apparatus; Adilenne Maese for technical assistance; Dr. Sewon Park for Matlab code; Alan Mo for GitHub management; Dr. Neda Nategh for discussions about model fitting; and Drs. Benjamin Dykstra and Kanishk Jain for addressing Motion Mapper queries. This work was supported by the National Science Foundation Graduate Research Fellowship Program (J.G.), a travel scholarship to the Short Course on the Application of Machine learning for Automated Quantification of Behavior, Jackson Laboratory, Maine (J.G.), a NIMH R01 MH132822 (M.Z.), a Klingenstein-Simons Early Investigator Award (M.Z.), a Whitehall Fellowship (M.Z.), a Sloan Fellowship (M.Z.), and a McKnight Scholars Award (M.Z.).

Author contributions

Conceptualization, J.G. and M.Z.; investigation, J.G., R.V., and A.B.; writing, J.G. and M.Z.; funding, J.G. and M.Z.; resources, M.Z.; supervision: M.Z.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Experimental models: Organisms/strains
C57BL/6J mice	Charles River Laboratories	027 https://emodels.criver.com/search?species=Mouse
BALB/c mice	Charles River Laboratories	028 https://emodels.criver.com/search?species=Mouse
Software and algorithms
SLEAP	Pereira et al.,17 This paper	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
deepSqueak	Coffey et al.,87 This paper	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
Motionmapper	Berman et al.,21 This paper	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
Regression analysis	This paper	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
Raster plots	This paper	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
Other
Basler GigE camera acA1300-60gc	Noldus	https://my.noldus.com/shared/knowledge-base/answer/256
Microphone	Avisoft Bioacoustics	https://avisoft.com/ultrasound-microphones/cm16-cmpa/#40011
Behavioral apparatuses	This paper - CAD files (machine shop)	deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884
overhead pole for camera and lighting mount	Fudesy Photo Video Studio 10Ft Heavy Duty Adjustable Backdrop Stand (Amazon)	N/A
GLFERA gooseneck lamps with clamping base	Amazon	N/A
Joydeco Blackout Curtains	Amazon	N/A
Anti-static spray - Plant Therapy Unscented Natural Wrinkle Releaser	Amazon	N/A
metallic tape for foot shock boards (GENNEL 5 mm × 20 m Conductive Cloth Fabric Adhesive Tape)	Amazon	N/A
Buttons for manual foot shock board operation (Starelo 19mm 3/4″ Momentary Push Button Switch – push & hold-ON; release-OFF)	Amazon	N/A
Alligator clip wires for wiring foot shock boards (Electronix Express 2 Wire 30 Ft Retractable Test Leads, 18 Gauge Wire with Alligator Clips)	Amazon	N/A
Foot shock board power supply	Med Associates Fear Chamber and Aversive Stimulator/Scrambler	https://med-associates.com/product/aversive-stimulator-scrambler-module/

Experimental model and study participant details

One hundred and twenty-two male or female C57Bl/6N mice and ninety-six male or female BALB/c mice were housed in cages (Thoren type 9 https://thoren.com/?page_id=71) with Teklad pelleted bedding (https://www.inotivco.com/7084-pelleted-paper-contact-bedding) on a ventilated rack in a 12 h light/dark cycle with unrestricted access to food (Teklad Global Soy Protein-Free Extruded Rodent Diet 2920X) and water. All mice were between 10 and 17 weeks old at the time of behavioral testing. All animal work was reviewed for compliance and approved by the Institutional Animal Care and Use Committee of The University of Utah (IACUC protocol #00001504).

Method details

Design and construction of SAUSI

The SAUSI apparatus was designed in CAD and built by the machine shop at the University of Utah. The floors were made of white High Density PolyEthylene (HDPE) plastic, the tunnel (tube) and walls were made of transparent acrylic plastic. The tunnel block (5.3 × 2.5 × 6.3 cm) was made of metal (673 g), and the U-channel tunnel block was 3D printed. Transparent, plastic wall extenders (33 cm tall) were fitted to the exterior walls of the SAUSI box and secured with clips to ensure that mice could not jump out of the apparatus.

Footshock boards were made by hand using metallic tape (optionally on a nonconductive backing (such as scotch tape) for portability). These were connected via alligator clip wires to the floor of Med Associates fear chamber boxes. During the conspecific tunnel deterrent session (Figures 1, S1, and S2), the footshock boards are wired directly to the fear chambers for automated shocking. During the SAUSI test day, this connection is mediated by off-on-off buttons for manual shock administration. Footshock boards were cleaned with 70% ethanol, dried, and reused multiple times.

Black-out curtains were hung surrounding the SAUSI setup to prevent preferences due to differences in visual stimuli across the room. They also improved the even distribution of lighting and increased sound dampening. Scent-free, anti-static spray was sprayed on the footshock boards and curtains at the beginning of each day of the SAUSI protocol. This reduced the chances of accidental shock due to static.

Room conditions

A bar was hung above the apparatus for mounting two yellow-light lamps facing away from the mice and the single top view camera. Two side view cameras were also mounted to each side of the SAUSI box with permanent fixtures for consistency of recording quality. The Avisoft microphone (for USV capture) was clamped to the exterior middle chamber wall (Figure 1).

Lighting: 2 lamps facing upward, yellow hue, lux = 22–23 per chamber.

Temperature: 70–75 F in behavior suite.

Handling: plastic beakers (4″ diameter, 5″ tall) were used to transport mice from homecages into SAUSI.

On the first day of the SAUSI protocol, mice were weighed. Each experimental mouse was paired with a lower-weight BALB/c (2 weeks younger) that was sex matched. Each BALB/c was tested with both a GH and SI mouse (once each) in counterbalanced order. The hind region of BALB/c mice was colored with Stoelting animal markers (https://stoeltingco.com/Neuroscience/Animal-Markers∼9769) to be identified throughout training and testing.

Mice

All mice were purchased from Charles River Laboratories. The strain of experimental mice was C57BL/6N. The strain of conspecific mice was BALB/c. Experimental mice for isolation arrived at the University of Utah vivarium facilities at 5–6 weeks old or 8 weeks old for footshock experiments. All mice were housed on a ventilated rack with dark, opaque plastic dividers stationed in between cages. These features were present to reduce the visual, auditory, and olfactory stimuli of surrounding cages for isolation.

SAUSI protocol

The SAUSI apparatus contains two outer rooms (“chambers”), one empty (“home”) and one containing a conspecific (“social”). These rooms are connected by a narrow tunnel with a miniature footshock board on either side, which deter conspecifics from crossing over, and allows for observable decision making in the experimental mice.

To habituate mice to the apparatus, tunnel, and handling, experimental mice are placed in the arena for 5 min each on days 1 and 2. The setup for habituation was identical to the test day to habituate the mice to the environment (lighting, cameras, microphones, foot shock boards present but off). During this assisted habituation, experimenters gently guided the mouse through the tunnel every 30 s. Time restarts after each time the mouse crosses through the tunnel. For the very first entrance on the first day, mice were held by the tail with all 4 paws on the ground with their face directed near the entrance of the tunnel (Video S2). From there, experimenters waited until the mouse enters the tunnel. Once all four paws are in, the mouse is blocked from reversing out of the tunnel³⁷ (Video S3). After the first entrance, “corralling” the mouse close to the tunnel entrance without touching it was sufficient for them to enter the tunnel. On the third day, mice were placed in the apparatus and monitored to ensure that the mice can use the tunnel at will (inspired by the hierarchy tube test habituation protocol³⁷). At the end of each session, the experimental mouse was taken out of the SAUSI apparatus with a plastic beaker for transportation. The side that mice start on was alternated for each day of habituation.

The conspecifics undergo a 5-min foot shock deterrent session on day 3 (the day prior to SAUSI testing) to prevent them from crossing to the other side and blocking the tunnel entryway. Two conspecifics were trained to deter at once (place one on each side). The footshock boards were set to automatic for the entirety of the 5-min session so that any time the conspecific steps on the foot shock board they receive a shock. In the rare occurrence that the conspecific entered the tunnel during deterrent training, that mouse was excluded from use on test day. Conspecifics were reused without additional deterrent training for a minimum of 2 days after original deterrent session. In our footshock-deterrent shock response curve experiment (Figure S2), female pairs were used to establish the optimal conspecific footshock of 0.3mA (minimal needed to generate tunnel entry deterrence while leaving behavior of the conspecific as well as experimental mouse unchanged).

On day 4, the social behavior test consisted of a 3-min habituation period for the experimental mouse (where no other mice or stimuli were present in the chamber) followed by a 10-min test phase with the conspecific present (Figure 1). To begin the baseline phase, the tunnel was blocked off in the starting chamber. The experimental mouse was placed in one chamber (alternating between left and right for each trial) and the block was removed to signify the start of the 3 min. Once three minutes passed, the experimenter waits until the mouse is in the “home” chamber and blocks entry to the tunnel using the tunnel block. The conspecific is then placed in the opposite chamber (“social”) and the block is removed to signify the start of the 10-min test phase. During the social test, the footshock boards are controlled manually by a button and can be used in the rare case that a conspecific steps on the foot shock board while the experimental mouse is not also on it. Note, if urine is present on the footshock board, it will not function properly, and should be quickly cleaned with a paper towel damp with ethanol (see above). To transport mice to and from the SAUSI apparatus, the tunnel is first blocked off and a plastic beaker (4″ diameter, 5″ tall) is used to transition the mouse back to its home cage.

For isolation, half of the mice were placed in social isolation, while the other half remained in their original group-housed cages. Isolation began at 6 weeks old and lasted for a total of 4 weeks (tested at 10 weeks old). Conspecifics were group-housed with 4 mice per cage and were tested at 8 weeks old (2 weeks younger than experimental mice). Each conspecific was paired with a GH and an SI mouse during the test day, counterbalanced for order.

SAUSI with tethered mice

To test whether SAUSI could be used with tethered mice and a modified, open-top tunnel (Figure 3), we used our Inscopix nVue microendoscope system to test tethered mice being imaged during SAUSI (https://inscopix.com/nvue-system/ Inscopix product code 1000-005596). Male (n = 3) and Female (n = 3) C57 BL/6 mice underwent stereotaxic surgery at 6 weeks old to implant a ProView Integrated GRIN Lens 0.6 mm × 7.3 mm (Inscopix product code 1050–004413) with an integrated base plate over the Bed Nucleus of the Stria Terminalis (lens lowered to AP +0.5, ML + - 0.85, DV -4.0 from bregma). Mice were also injected with a virus encoding the calcium indicator GCaMP6f (AAV5-Syn-GCaMP6f-WPRE, Addgene). Mice were given 7 weeks to recover and isolated as adults at 13 weeks old for a duration of 3 weeks before behavioral testing.

For behavioral testing, the tether-compatible open-top tunnel was used throughout (see above). Mice were attached to a miniature microendoscope which was wired to a commutator (Inscopix product code 1000–005088) connected to a DAQ, which allowed mice to move freely without cord entanglement. Video was captured using the nVision system (https://inscopix.com/nvision-system/). Untethered mice (controls) were handled similarly to tethered mice (scruffed and microendoscope attachment screwed on and off) prior to testing.

All mice were habituated with the tether throughout the 3 SAUSI habituation days. For SAUSI test days, we used a within-subjects design, where ½ the mice were tested as untethered controls (no microendoscope/tether attached) and the other ½ were tested with the attached microendoscope/tether. The same mice were retested the following day under the opposite condition (counterbalanced for test day and sex). All SAUSI protocols were identical to previous testing except mice were handled by tail rather than in transport beakers and the experimenter was present in the room throughout each session every day to ensure there were no problems with the cord.

Videos were converted to.mp4 files using the Inscopix Data Processing Software. Videos were tracked and hand scored using Ethovision. Data were graphed and analyzed using GraphPad Prism.

Behavioral paradigms

3-Chamber assay

Lighting, video, and audio recordings (as well as all other factors including curtains, wall extenders, anti-static spray, cup handling) were set up identically to SAUSI. The temperature was 70–75 F in the behavior suite, with air circulation on. The shell of the apparatus used for SAUSI was used for 3-chamber testing, with the tunnel replaced by removable doors and with no foot shock boards present. Mesh cups used to restrain conspecifics (https://www.noldus.com/applications/sociability-cage) were placed in the top left or bottom right of the respective chamber and weighted down. Videos were recorded with a camera mounted overhead using MediaRecorder (Noldus) and scored using automated tracking in EthoVision (Noldus). USVs were captured using Avisoft and were extracted and analyzed using DeepSqueak.

The 3-chamber task began with a 5-min baseline phase, where experimental mice were placed in the middle chamber and allowed to explore all 3 empty chambers. This was followed by the test phase (10 min) where animals could choose to explore an inanimate object (black plastic block) under a cup in one chamber or a conspecific mouse under a cup in the opposite chamber (“social” chamber).

The social chamber was counterbalanced for each trial. Each conspecific was used twice: once with a GH mouse and once with an SI mouse, counterbalanced for order. The apparatus, cups, and doors were cleaned with 70% ethanol between each trial.

Prior to testing, half of the mice were placed in social isolation, while the other half remained in their original group-housed cages. Isolation began at 6–14 weeks old and lasted for a total of 3–4 weeks. Conspecifics were group-housed with 4 mice per cage.

Resident Intruder

Lighting was provided by two lamps facing upward with yellow hue, lux = 14–15. Temperature was 70–75 F in behavior the suite. Videos were recorded from an overhead camera (43 cm from the cage floor) using MediaRecorder and manually scored in EthoVision. USVs were captured using Avisoft (microphone 38.1 cm from the cage floor) and were analyzed with DeepSqueak. A custom crafted, plexi-glass test apparatus, designed for homecage test assays such that a homecage can be slotted into the apparatus with the top off and the apparatus surrounds each cage extending it vertically and contains a ledge to prevent mice from exiting the cage during testing. (deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884).

The RI tests began with a 3-min baseline phase, where the experimental mouse was recorded in the cage on its own. This was followed by a 10-min test phase, in which a conspecific mouse was placed in the cage with the experimental mouse.

Cages used (Thoren type 9- https://thoren.com/?page_id=71) were 19.56 × 30.91 × 13.34 (cm.) with a floor area of 435.7 (Sq. cm.). Bedding that mice were continuously housed in (Teklad pelleted bedding - https://www.inotivco.com/7084-pelleted-paper-contact-bedding) or new bedding of the same type were used for RI-home cage and RI-novel bedding. Nestlets were removed for visibility throughout all testing. Food and water were not accessible during the baseline and test phases of the task.

Group-housed mice that were not actively being tested were placed together in a clean holding cage. Each group-housed mouse shared an “untested” holding cage with their cage mates prior to testing. After being tested, they were placed in another clean “tested” holding cage so that no untested mice were exposed to tested mice.

For isolation, half of the mice were placed in social isolation, while the other half remained in their original group-housed cages. Isolation began at 6 weeks old and lasted for a total of 4 weeks (tested at ∼10 weeks old). Conspecifics were group-housed with 4 mice per cage and were tested at 8 weeks old (2 weeks younger than experimental mice). Each conspecific was paired with a GH and an SI mouse during the test day, counterbalanced for order.

Footshock stress paradigm

All mice (shock and controls) were isolated for 7 days prior to the SAUSI test day to prevent within-cage aggression following shock, and to avoid the threshold for chronic social isolation (2+ weeks).³⁴ On the third day of the SAUSI timeline, after receiving the unassisted tunnel habituation, each mouse was placed in a Med Associates fear chamber. Mice in the shock group received 10, 1-s, 1-mA pseudorandomized foot shocks over a 60 min period.⁵⁷ Context only controls were placed in the Med Associates fear chambers as well, but received no shocks over the same 60-min total duration.

Conspecifics were group-housed with 4 mice per cage and were tested at 8 weeks old. Each conspecific was paired with a CNTL and a Shock mouse during the test day, counterbalanced for order.

Quantification and statistical analysis

Data are presented as mean ± Standard Error of Mean (SEM) with individual values displayed. At least 2–5 replications were conducted in each main figure. Prism (GraphPad) was used for statistical analyses. Student’s t test, ANOVAs, and RM ANOVAs were used followed by posthoc tests (as indicated in figure legends were used), controlling for repeated testing. The cut-off for statistical significance was set at alpha<0.05, two-tailed.

Acoustic analysis

The Avisoft Bioacoustics recording system (https://avisoft.com/) was used to collect vocalization data during SAUSI testing. Specifically, we used the condenser ultrasound microphone CM16/CMPA (part# 40011) with the Ultrasound Recording Interface (UltraSoundGate 416H, part # 34164) and recorded with the Avisoft-RECORDER software (https://avisoft.com/recorder/).

DeepSqueak was then used to analyze Ultrasonic Vocalizations (USVs) collected from Avisoft. In DeepSqueak, the Mouse Detector YOLO R2.mat neural network was used with the following settings: Total analysis length = 0, Frequency cut off high = 250, Frequency cutoff low = 25, Score threshold = 0. DeepSqueak gathers information (referred to as call statistics)⁸⁷ about each individual USV detected, including call length, frequency metrics, slope, sinuosity, mean power, and tonality (where values near one have lower signal to noise ratio). Definitions for each call statistic can be found on the DeepSqueak GitHub page (https://github.com/DrCoffey/DeepSqueak/wiki/export-to-excel).

After obtaining.csv files containing the features for each USV in each animal, we used a python script (deposited at Zenodo: https://doi.org/10.5281/zenodo.15610884) to retrieve summary data for all mice (total # USVs, average, and median values for each feature). These summary data were graphed and analyzed using GraphPad Prism.

Behavior analysis

Behavior	Event Type	Description	Equation
Face Sniff	Start-stop	Sniff in front of the ears
Body Sniff	Start-stop	Sniff behind ears (including ears) and anterior to anogenital/tail regions
Anogenital Sniff	Start-stop	Sniff anogenital region (below the tail)
Tail Sniff	Start-stop	Sniff tail only
All Sniff	Start-stop	All sniffing combined	All sniff = Face + body + anogenital + tail sniff
Conspecific sniffa	Point event	Any new sniff bout by the conspecific
First Tunnel Sheltering	Start-stop	In tunnel before first time entering the social chamber (can be multiple times before entering). must have all 4 paws in tunnel.
Reverse in Tunnel	Start-stop	Backing up in the tunnel after all 4 paws entered.
Social Freeze	Start-stop	freezing in response to being sniffed by the conspecific. Can be during or just after being sniffed by the conspecific (within 1 s).
Social Reactivity	Point event	experimental mouse reacts when conspecific sniffs it (within 1 s of the end of a sniffing bout from the conspecific). Reacting can include retreating, darting, vigilance, flinching, or jumping.	$\frac{socialreactivitybouts}{conspecificsniff}\ast 100\%$
Social Initiation	Point event	Any time the c57 approaches the BALB/c to sniff and closes the gap between them/comes from more than one body length distance away. The BALB/c does not close the gap.
Social Chamber Preference	Start-stop	Preference to spend time in the social chamber.	$\frac{socialchambertime\left(s\right)}{totaltime\left(s\right)}\ast 100\%$
Groom	Start-stop	Self grooming
Non-social freeze	Start-stop	Freezing not in response to being sniffed by the conspecific.
aggression	Start-stop	Attack, biting, mounting
jump	Point event	Attempt to exit the apparatus. All 4 paws leave the ground.

^aall behaviors are referencing behaviors produced by the experimental mouse, except for conspecific sniff, which measures the bouts of sniffing performed by the BALB/c mouse.

Videos were recorded in MediaRecorder. The top view perspective was used for automated analysis latency and time spent in each chamber with EthoVision⁸⁸ as well as postural data tracking in SLEAP.¹⁷ Side view perspectives were used for hand scoring behaviors in EthoVision.

DeepSqueak was used to analyze all audio recording data from avisoft

SLEAP¹⁷ was used for postural tracking. All videos used for tracking were obtained from the top view camera and were trimmed to only include the test phase. These videos were used for training in SLEAP with >600 frames manually labeled by an observer blind to experimental conditions. The “export labels package” feature from the SLEAP user interface was used to export labeled frames, which were uploaded in google colab, where training was performed (Google Colab file accessible using the “train on google colab” feature in the SLEAP user interface). Bottom-Up processing was used for training with default settings. For inference, tracker was set to “flow”, clean_instance_count was set to 2, track_window was set to 30, and all remaining settings were left as default. Google Colab was used for training and inference due to high computational needs.

The Motion Mapper pipeline²¹ was used for unsupervised behavioral analysis. (modified code available here at Zenodo: https://doi.org/10.5281/zenodo.15610884).

Python and MATLAB were both used to generate nose-to-nose distance calculations, behavior raster plots, and all regression data (code available here at Zenodo: https://doi.org/10.5281/zenodo.15610884).

GraphPad PRISM was used to generate all statistics and graphs comparing GH and SI conditions.

Published: July 17, 2025