A critical role for cortical amygdala circuitry in shaping social encounters

Aggression is an evolutionarily conserved behavior that controls social hierarchies and protects valuable resources like mates, food, and territory. In mice, aggressive behaviour can be broken down into an appetitive phase, which involves approach and investigation, and a consummatory phase, which involves biting, kicking, and wrestling. By performing an unsupervised weighted correlation network analysis on whole-brain c-Fos expression, we identified a cluster of brain regions including hypothalamic and amygdalar sub-regions and olfactory cortical regions highly co-activated in male, but not female aggressors (AGG). The posterolateral cortical amygdala (COApl), an extended olfactory structure, was found to be a hub region based on the number and strength of correlations with other regions in the cluster. Our data further show that estrogen receptor 1 (ESR1)-expressing cells in the COApl exhibit increased activity during attack behaviour, and during bouts of investigation which precede an attack, in male mice only. Chemogenetic or optogenetic inhibition of COApl ESR1 cells in AGG males reduces aggression and increases pro-social investigation without affecting social reward/reinforcement behavior. We further confirmed that COApl ESR1 projections to the ventrolateral portion of the ventromedial hypothalamus and central amygdala are necessary for these behaviours. Collectively, these data suggest that in aggressive males, COApl ESR1 cells respond specifically to social stimuli, thereby enhancing their salience and promoting attack behaviour.

part of the olfactory cortex, is a key node or "hub region" that is highly connected within the broader network of this cluster. We further investigated the mechanisms by which the COApl mediates aggressive social behavior using a variety of circuit techniques. We found that COApl ESR1 + neurons are highly active during bouts of social investigation which precede an attack, and that this region is necessary to make the transition from the appetitive phase (investigation) of aggressive interactions to the consummatory phase (attack).
The COApl is a key node in a male AGG network Following a RI test, we conducted a cleared whole-brain c-Fos mapping experiment using the iDISCO+ method combined with a network analysis commonly used in RNA-Seq studies to identify differentially activated regions between male and female AGG and NON mice (Fig. 1a). Prior to brain collection, male and female AGGs and NONs were exposed to an intruder mouse. We did not see a difference in total attack duration or investigation duration between either sex (Fig. 1b, c), but we did observe that males and females engaged in distinct actions which made up the total duration metrics as previously reported 3 (Extended Fig. 1a-f). We then assembled a single network for both AGG and NON mice (Fig. 1d). Each colour along the top and left side of the network indicates a module of highly clustered brain regions. Following the construction of each network, we performed a module preservation analysis using the AGG network as the reference network and the NON network as the test network. We found that the "pink" module was the least preserved module in the NON network ( Fig. 1e; Extended Fig. 2a-c). Inspection of this module showed that sub-regions of the amygdala and hypothalamus are highly interconnected in the AGG network. In the NON network however, sub-regions of the hypothalamus and amygdala are highly connected amongst themselves, but not between the amygdala and hypothalamus. Interestingly, male AGGs expressed this module, as determined by the eigenvalues, signi cantly more than female AGGs ( Fig. 1f). We also found that many of the top regions which showed a signi cant increase in c-Fos counts were different between male and female AGGs (Extended Fig. 2e, f). Of the 28 regions in the pink module, 11 regions differed in their total connectivity within the pink module between AGG and NON mice. Of the 4 regions which displayed higher connectivity in the AGG network, the COAp had the highest average connectivity (Fig. 1g). Further investigation of the COAp determined that male AGGs had more c-Fos + cells in this region compared to female AGGs and male NONs in the iDISCO+ data set (Fig. 1h). Since Clearmap cannot distinguish between the lateral and medial Cortical Amygdala, we manually cropped the lateral and medial portions of the COAp and found that male AGGs showed a signi cant increase in the number of cells expressing cfos in the lateral relative to the medial portion of the COAp (Extended Fig.  3A). Although cFos is also up in the COApm in AGGs relative to NONs, later functional validation studies with DREADDS shown in Extended Fig (5J-I) suggest only the COApl regulates aggression. When focusing on the COApl found that male AGGs showed a signi cant increase in the proportion of Esr1 cells which expressed c-Fos relative to female AGGs and male NONs (Fig. 1i-j). Detailed analyses of this cell population determined that the majority of Esr1 + cells were glutamatergic (~ 90%) and were equally expressed in the COApl in all four groups (Extended Fig. 3a-c). We also found that the majority of cells which expressed c-Fos also expressed Esr1 (~ 85%; Extended Fig. 3d).
To further investigate COApl Esr1 neuron activity in vivo during aggressive encounters with an intruder mouse, we injected Cre-dependent adeno-associated virus (AAV)-DIO-GCaMP6s into the COApl of Esr1-Cre transgenic mice to label COApl Esr1+ neurons (Fig. 2a). We then measured uorescent Ca 2+ activity by bre photometry across three days of the RI test in male and female AGGs. Since the COApl receives input from the main olfactory bulb, we rst examined the response of COApl Esr1 neurons to three different odours (i.e., social, non-social, and predatory scents). We found that COApl Esr1 neurons displayed a signi cant increase in Ca 2+ activity during the investigation of soiled bedding from a cage of same-sex C57 mice but not when investigating fox urine or ethanol, which is consistent with previous literature 15 (Fig. 2b, c). We observed similar results in female mice, with a signi cant increase in Ca 2+ activity during the investigation of soiled bedding from a cage of same-sex mice (Extended Fig. 4a) but not during the investigation of fox urine or an ethanol swab (Extended Fig. 4b, c).
During the RI test, we observed that COApl Esr1 neurons displayed a signi cant increase in Ca 2+ activity during bouts of investigation in trials where an attack occurred relative to bouts of investigation during trials in which no attack occurred (Fig. 2d, e). In contrast, we observed the opposite effect in females: COApl Esr1 neurons displayed a signi cant increase in Ca 2+ activity during investigation trials in which no attack occured (Extended Fig. 4d). When examining individual investigation bouts during trials in which an attack occurred, we found that COApl Esr1 neurons increased Ca 2+ activity only during bouts of investigation which preceded an attack compared to bouts of investigation which occurred in isolation ( Fig. 2f, g). In females, we did not detect any differences in Ca 2+ activity in female AGGs between the two types of investigation (Extended Fig. 4e). Lastly, when examining attack bouts, we found that when attacks were preceded by a bout of investigation, there was no further increase in Ca 2+ activity during the attack. When an attack occurred spontaneously without prior investigation, we found that there was a signi cant increase in Ca 2+ activity during the attack in male (Fig. 2h, i), but not in female, AGGs (Extended Fig. 4f). In a separate cohort, we recorded Ca 2+ activity during social encounters across the four phases of the oestrous cycle but did not observe any differences in the signal during attack or investigatory behaviours as a function of cycle stage (Extended Fig. 4g-i).

Manipulating COApl Esr1 neurons alters consummatory drive in male AGGs
The results discussed above suggest that COApl Esr1 neurons are highly active during the transition from the appetitive phase of aggressive social interactions to the consummatory phase. To test the hypothesis that this cell population mediates this transition, we utilized viral vectors expressing designer receptors exclusively activated by designer drugs (DREADDs) to bidirectionally manipulate the activity of COApl Esr1 neurons. ~3-4 weeks before the RI and social self-administration (SA) tests, we injected AAV-DIO-hM3Dq, AAV-DIO-hM4Di or AAV-DIO-mCherry viruses into the COApl of 8-week-old Esr1-Cre mice (Fig. 3a, b). We found that inhibition of COApl Esr1 neurons led to a striking reduction of total attack duration and a concomitant increase in total investigation. Conversely, we found that exciting COApl Esr1 neurons led to a signi cant increase in total attack duration without a decrease in total investigation (Fig. 3c, d). We also injected AAV-DIO-hM4Di into the medial portion of the COAp and did not see any changes in attack behavior ( Supplementary Fig. 5m), which is consistent with a previous lesion study in rats 16 . When examining distinct aggressive actions, we found that the number of bites during RI was affected most by the inhibition or excitation of COApl Esr1 neurons (Extended Fig. 5a-c). Furthermore, we found that mice injected with the hM4Di virus showed a signi cant increase in anogenital investigation and a signi cant decrease in the number of active withdrawals (Extended Fig. 5d-f). Since COApl Esr1 neurons receive input from the main olfactory bulb [17][18][19] , we ran two control experiments to ensure that the observed alterations in social behaviour were not due to anosmia. We rst ran a hidden food test to assess general olfactory processing 20 . In the test, mice are food restricted overnight and must nd a piece of food buried under bedding by using their sense of smell. Inhibition of COApl Esr1 neurons did not affect the ability to locate the food (Fig. 3e). Next, we tested the ability of male AGGs to differentiate between males and females 21 . Male mice were placed in an open eld arena with a male and female mouse under mesh cups on opposite sides of the arena. We computed a discrimination ratio in which the difference in time spent investigating the male and the female was divided by the total time spent investigation both mice. A negative score indicates more time spent investigating the male and a positive score indicates more time spent investigating the female. We found that inhibiting COApl Esr1 neurons did not affect the males' preference for investigating the female mice (Fig 3f). To further support the idea that COApl Esr1 neurons participate in mediating the transition from the appetitive to consummatory phase of aggression, we examined the relationship between the change in attack and investigation duration due to manipulations of COApl Esr1 neurons. In mice injected with hM4Di, we found a signi cant negative correlation between the change in attack duration and the change in investigation duration (Fig 3g). There was a trend in mice injected with hM3Dq for a negative correlation between the increase in attack duration and decrease in investigation duration ( Fig. 3h) with no effect observed in mice injected with the mCherry virus (Fig 3i).
Lastly, we tested whether inhibition of COApl Esr1 neurons alters the motivation to engage in attack behaviour by utilizing the aggression self-administration paradigm 22 . We found that both the mCherry and hM4Di groups learned the task equally well in terms of rewarded trials and the occurrence of attacks (Extended Fig. 5g-i). Following the acquisition of the task, mice were tested using a counterbalanced injection regimen (i.e., CNO/saline). We did not observe any difference in lever presses (Fig. 3j) between the two groups on CNO, suggesting that COApl Esr1 cells do not affect the reinforcing properties of social interaction (Fig. 3k). However, we observed a striking decrease in the percentage of rewarded trials which led to an attack (Fig. 3l). We also tested mice on a progressive schedule and did not nd a difference in the maximum number of lever presses made for access to an intruder, suggesting that the COApl Esr1 cells do not regulate the motivation to work for access to an intruder. Lastly, manipulations of COApl Esr1 neurons in female AGGs during the RI test did not affect attack (Extended Fig. 6a-c) or investigation behaviours (Extended Fig. 6d-f). Collectively, these results suggest that the inhibition of COApl Esr1 neurons selectively alter the consummatory drive in social interactions in male AGGs.
After determining the effect of inhibiting COApl Esr1 neurons on aggressive behaviour in males, we sought to investigate how inhibiting this cell population alters the connectivity of the pink module from the AGG network. To this end, we injected 8-week-old male AGGs with AAV-DIO-hM4Di or AAV-DIO-mCherry viruses. Following recovery from surgery, all mice were given an injection of CNO prior to the last RI test and were perfused 90 minutes later (Fig. 4a). As expected, inhibiting COApl Esr1 cells decreased attack duration and increased investigation duration (Fig. 4b, c; Extended Fig. 7a-f). Brains were perfused, cleared, imaged, and registered to the Allen brain atlas. We then constructed a network for each group of mice. For visualization, the clusters were organized according to the original AGG network to ensure that the same brain regions would make up similar clusters in the hM4Di and mCherry networks ( Fig. 4d-g). As in the modules from Fig. 1, visual inspection indicated a high degree of interconnectedness between amygdalar and hypothalamic subregions under control mCherry conditions. Following inhibition of COApl Esr1 cells, we observed high levels of interconnectedness within, but not between, amygdalar and hypothalamic subregions, which is consistent with data from NONs in Fig 1. We also found that inhibiting COApl Esr1 cells with hM4Di signi cantly decreased expression of the pink module relative to the mCherry group ( Fig.  4d). We next determined if the hM4Di and mCherry networks differed in how well they preserved the pink module in the original AGG network. To this end, we performed a module preservation analysis separately for each network with the original AGG network serving as the reference network. Using the same metrics from Fig.1, we found that the hM4Di network showed a signi cant decrease in these metrics compared to the mCherry (Fig. 4e), suggesting that inhibiting COApl Esr1 neurons diminished the preservation of the pink module found in the original network. Of the remaining 7 clusters, the hM4Di group only showed a signi cant decrease in the 'red' cluster ( Supplementary Fig. 7g). Interestingly, this cluster contains regions which COApl Esr1 neurons project to, including the bed nucleus of the stria terminalis (BNST), the anteroventral periventricular nucleus (AVPV), and the substantia innominata (SI).
When examining c-Fos counts in individual regions of the pink network, we found that 89% of regions showed a signi cant reduction in c-Fos cells following inhibition with hM4Di ( Fig. 4h and Supplementary  Fig. 7h).

Inhibiting of COApl Esr1 neuron outputs alters aggressive social behaviour
We next sought to understand which regions within the pink network downstream of COApl Esr1 neurons contributed to the behavioural effects of local manipulation (Fig. 5a). To rst identify downstream regions, we injected AAV-DIO-eYFP in COApl Esr1 neurons to visualize axonal terminals projecting from COApl Esr1 cells. We also injected HSV-1 (H129 TK-TT) to con rm that visualized axonal terminals formed monosynaptic connections with downstream regions of interest (Fig. 5b). Interestingly, many of these regions were previously identi ed to promote aggressive behaviour, including the ventrolateral portion of the ventromedial hypothalamus (VMHvl), the medial amygdala (MEA), the central amygdala (CEA), and the BNST. Furthermore, we found that this cell population also projected to numerous regions involved in olfaction, such as the piriform cortex (PIR), endopiriform cortex (EP), piriform amygdala area (PAA), postpiriform transition area (TR), and the anterior olfactory nucleus (AON). Interestingly, of the 28 regions in the pink module, COApl Esr1 neurons formed monosynaptic connections with 13 of them, indicating that many of these regions are both functionally and physically connected. In addition, of these 13 regions, 12 of them showed a signi cant decrease in c-Fos following inhibition of COApl Esr1 DREADDs (Fig. 4h). Based on this information, we chose to target the VMHvl (Fig. 5c), which is necessary for aggressive behaviour, and the central amygdala (Fig. 5d), which has been shown to modulate jaw movements during biting, a key component of aggressive behaviour. To evaluate whether these circuits alter aggressive behavior, we injected AAV-DIO-NpHR into COApl Esr1 neurons of 8-week-old male AGGs and implanted ferrules into either the VMHvl or the CEA (Fig. 5e). Four weeks later, mice were tested for two days in the RI test in which the order of testing (laser on/off) was counterbalanced. We found that inhibiting the COApl-CEA pathway resulted in a signi cant decrease in total attack duration without affecting social investigation (Fig. 5f, g). In addition, we found that inhibiting the COApl-VMHvl pathway produced a strong reduction in attack duration, and concomitantly increased social investigation, mirroring the effects of locally inhibiting COApl Esr1 neurons.

Discussion
Using an unsupervised approach to analyzing whole-brain c-Fos patterns, we identi ed the COApl as a key region controlling the transition from social investigation to aggressive behaviour in male, but not in female, mice. Interestingly, many of the regions highly connected to the COApl have previously been identi ed as part of the "core aggression circuit" such as the MEA, VMH, and the MPOA. Given that these regions are clustered together with many other regions whose function have not been examined in the context of aggression, it is possible that the core aggression circuit is larger than previously considered and requires further investigation. Interestingly, we nd that inhibition of the COApl resulted in a signi cant decrease in c-Fos expressing cells in 79% of the regions within the AGG network (pink module). This is of importance when considering the causal nature of focal manipulations. Understanding the relationship between the manipulated region and other downstream regions is paramount to understanding the relationship between brain-wide activity and behaviour in our studies. 23,24 .
Based on our current understanding of the circuitry mediating aggression, one might have predicted that a loss of function manipulation in the COApl would result in decreased aggression due to the lack of social approach 10,25 . This expectation is perhaps due to previous studies demonstrating that loss of function manipulations in the main olfactory bulb, which projects to the COApl, reduce social investigation and aggression 26,27 . However, our ndings that COApl Esr1 neurons are more active during RI trials in which an attack occurs, and that this population is speci cally active during investigation bouts which precede an attack, suggests that this region is not simply relaying olfactory 'cues' to downstream regions. This is further supported by our ndings which demonstrate that inhibiting COApl Esr1 neurons increasesinvestigation rather than decreasing it without resulting in anosmia. In addition, our ndings that COApl Esr1 neurons respond to odours from conspeci cs, but not from predators, suggests that they are not acting as a simple "threat detector", which is in line with a previous study demonstrating that the COApl is not involved in fear responses to predator odours 19 . While it was previously reported that the COApl is required for the attraction to non-social odourants such as 2-phenylethanol, this was not celltype speci c 18 . Our data support the idea that COApl Esr1 neurons are exclusively activated by odours emitted from conspeci cs and therefore shapes the type of social encounter that ensues.
One potential explanation of our ndings is that COApl Esr1 neurons are involved in the motivation to attack the intruder thereby regulating the rewarding/reinforcing properties of aggression 28 . In support of this explanation, inhibiting COApl Esr1 neurons with hM4Di robustly decreased the percentage of rewarded trials which led to an attack in an aggression self-administration task. However, inhibiting COApl Esr1 neurons did not change the number of lever presses and mice in both groups worked equally hard for access to a social target, suggesting that their motivation to interact with the intruder-albeit in a less aggressive and more pro-social way-remained intact. Though limited, some previously published studies of other nuclei that make up the pallial amygdala (basolateral, basomedial, cortical and posterior amygdala) support its role in rodent social behavior [29][30][31] . For example, inhibition of basolateral amygdala (BLA) inputs to the ventral hippocampus and the medial prefrontal cortex increases a liative social behaviour in mice, though aggression was not assayed in these studies. Conversely, manipulating the posterior amygdala alters aggressive behaviour without affecting pro-social investigation 32,33 . PAllila amygdala subcircuits have also been implicated in social behavior in human and non-human primate studies. These studies have focused largely on the BLA and lateral amygdala given their extensive connections with the visual system 34,35 , which primates are highly reliant upon during social interactions. Some studies have demonstrated that neurons in the BLA re when primates make eye contact with one another, and bilateral lesions often reduce fear of others and limits aggression. However, some contradictory results have also been reported, perhaps due to their use of non-speci c excitotoxic lesions (reviewed in 36 ). A recent study using the GABA agonist muscimol to reversibly inhibit the BLA, increased a liative behaviors between pairs of rhesus monkeys, while activating the BLA resulted in social withdrawals without affecting anxiety 37 , mirroring our ndings in mice. Furthermore, single unit recordings in humans indicate that neurons in the amygdala are sensitive to the subjective perception of emotions shown in facial stimuli, as opposed to just the features of stimulus 38 . This suggests that the amygdala of different individuals can respond differently to the same sensory cues depending on the behavioural state/belief of the individual. Collectively, these ndings have led to the interpretation that the amygdala plays a key role in social perception 39,40 . Thus, it is tempting to speculate that COApl plays a role in the perception of the intruder which is affected by the "aggressive state" of the resident. Internal states have previously been shown to bias perception towards speci c actions in other areas of rodent research. For instance, a state of hunger has been shown to enhance the responsivity of the insular cortex and lateral amygdala to sensory cues related to food 41,42 . Additionally, the auditory cortex of mothers is more responsive to pup cries than virgin mice 12,43 . Our research thus provides an important foundation for understanding the neural mechanisms underlying social perception and action selection. Future studies designed to delineate the mechanisms by which internal states and perceptual processes interact to execute distinct behaviours during dyadic social interactions will thus be of high importance.

Animals
Wild-type Swiss-Webster (SW) mice (male and female, 12-15 weeks; Charles River) and Esr1-Cre mice (017911, B6N.129S6(Cg)-Esr1 tm1.1(cre)And /J; Jackson Laboratory) were crossed with SW females to obtain F1 mice used as experimental mice. Intruders for the resident-intruder (RI) test were 8-12-week-old male or female C57/BL6J mice (Jackson Laboratory). All delivered mice were allowed 1 week of acclimation to the housing facility prior to any experimental protocol. At 8 weeks, mice were separated from their littermates and paired with a member of the opposite sex for sexual experience. Females were paired with castrated SW males (8-12 weeks; Charles River) to prevent pregnancy. All mice were maintained on a 12/12 h light/dark cycle (07:00-19:00) with ad libitum access to food and water. Housing and experimental rooms were maintained at 20-22 °C and 40-60% humidity. Experiments were conducted during the light phase. Procedures were performed in accordance with the National Institutes of Health Guide for Care and approved by the Use of Laboratory Animals and the Icahn School of Medicine at Mount Sinai Institutional Animal Care and Use Committee.

RI Test
Mice were screened using protocols adapted from previous studies 3, 44 . Brie y, cage tops were removed and replaced with Plexiglas covers to monitor trials. A novel C57BL/6J mouse matching the sex of the resident was introduced into each cage and mice were allowed to freely interact for 5 min. After 5 min elapsed, intruder mice were returned to their home cages and, in the case of female resident-intruder trials, cohabiting male mice were returned to their home cages. All videos were recorded for later analysis. Resident behaviors were manually annotated using Observer XT 11.5 (Noldus Interactive Technologies).

Aggression Self-Administration (SA) Test
Aggression SA testing was conducted as previously described 45 . Brie y, SW resident mice were placed in standard Med Associates operant chambers and underwent 8 trials per day for 16 days in which they could press a lever (FR1) in order to receive access to a subordinate same-sex mouse through a guillotine door into their operant chamber. A house light illuminated the chambers during trials, and an inactive lever was extended at all times. At the beginning of a trial, the house light turned on, and a is lever extended 10 s later. If the resident pressed for a reward, a 2 s tone played and a guillotine door next to the active lever opened for 12 s. A same-sex C57BL/6J intruder manually pushed into the chamber, and the intruder was removed at the end of the trial. If residents did not press, the active lever retracted after 60 s. All chambers also had recessed food pellet receptacles with beam-break registered entries ports. We conducted all Progressive Ratio (PR) tests using the same parameters we used for self-administration training, except for the trial design. Following reward obtention (number of required active lever presses for a given trial), the automatic guillotine door was opened for 10 s and the same-sex intruder (C57BL/6J) was manually pushed inside the main operant chamber. Immediately after the automatic door closed, the active/inactive levers were re-extended. The intruder mouse was removed from the chamber following an attack or after 30 s had elapsed. The PR session was terminated if no reinforced responses occurred after a total duration of 20 min had elapsed. Throughout the sessions, for each trial, the number of responses required to activate the guillotine door and gain access to the intruder was incremented following an exponential progression (R = 5 x e0.12*P -5, where P is the previous ratio). The breakpoint value corresponds to the total number of rewards (number access to the intruder during the whole session). Esr1-cre mice transfected in the COApl using AAV9-DIO-hM4Di-mCherry or AAV9-DIO-mCherry (see titer, surgeries information and coordinates below) were administered either saline (SAL) or Clozapine-N-Oxide (CNO), counterbalanced over the two days of the PR.

Hidden Food Test
Mice were habituated to and allowed to consume a palatable food (i.e., Reese's mini peanut butter cup) two days prior to being tested. Mice were then food-restricted the night before the test. Mice were tested under red light and placed in a clean mouse cage. The food was placed 8 cm deep in the bedding of the cage and the time to nd the food was recorded with a stop-watch.

Sex Discrimination Test
Male and female C57/BL6J mice were used as target mice and placed on opposite sides of an open eld enclosure in mesh cups. Test mice were placed in the open eld and allowed to explore the arena for ve minutes under red light. Mice were recorded with an overhead Ethovision camera and time spent interacting with each mouse was scored o ine manually. To calculate the discrimination ratio, the total time spent interacting with both mice was divided by the difference between time spent with the female and male. A positive discrimination ratio indicates more time spent with the female, a negative ratio indicates more time spent with the male.
Perfusion and brain tissue processing For iDISCO+, mice were injected with 10% chloral hydrate and perfused transcardially with ice-cold 1× PBS (pH 7.4), followed by xation with cold 4% paraformaldehyde in 1× PBS. Brains were post-xed for 12 h in the same xative at 4 °C. For FISH, brains were rapidly removed and ash-frozen in −30 °C isopentane for 5-10 s then kept at −80 °C until sectioning. Slices were sectioned at 15 μm using a cryostat (Leica). Animals injected with H129ΔTK-TT were perfused 48 h after injection.
In Situ Hybridization and Imaging . Reagent Alt-A was used for the fourth ampli cation step, with Channel 1 at 488 nm, Channel 2 at 550 nm and Channel 3 at 647 nm. Finally, slides were treated for 30 s with DAPI and immediately coverslipped with Eco-Mount. All slices were imaged using a Zeiss LSM 780 confocal microscope. Cells from all ISH images were counted blindly across groups. Cells with at least 5 puncta for each probe were considered to be positive for the probe of interest. iDISCO+ staining, imaging and ClearMap analysis The iDISCO+ staining protocol was modi ed from http://www.idisco.info. Fixed whole brains were incubated with the primary Fos antibody (no. 226003, 1:1,000, Synaptic Systems) and secondary donkey anti-rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 647 (no. A-31573, 1:1,000, Thermo Fisher Scienti c) for 7 days each. A LaVision lightsheet microscope with zoom body was used for half-brain sagittal scanning, with dynamic focus and a step size of 4 μm. Cleared brains were processed as previously described using ClearMap 13 . Fos + cells were quanti ed using the cell detection module, with cell detection parameters optimized and validated based on the intensity and shape parameters of the signal. The auto uorescence channel was aligned to the Allen Institute's Common Coordinate Framework using the Elastix toolbox. To compare cell counts between AGG and NON animals in both sexes and hM4Di vs. mCherry, a negative binomial regression was applied using the glm.nb function from the MASS package in R. Group classi cations were dummy coded (0 for the AGG group and 1 for the NON group). The maximum-likelihood coe cients α and β were determined through iterative reweighted least squares. A signi cant β means that group status is related to cell count number at the speci ed region of interest. Z-scores correspond to this β coe cient, normalized by its sample standard deviation. P values were corrected for multiple comparisons using the Benjamini-Hochberg procedure to decrease false discovery rate. Q-values below 0.05 were considered signi cant. The open-source Bonsai programme was used to control the system; 470 and 415 nm LED lights were used for GCaMP6s signal and auto uorescence measurement. Light at the bre tip ranged from 40 to 80 μW and was constant across trials over testing days. Simultaneous recording of 40 fps from both 470 and 415 nm channels was achieved phase to phase and visualized via Bonsai. Three weeks after virus injection and ferrule implantation, mice were tested in the RI test and were presented different odours in a counterbalanced manner. Once connected to the apparatus, mice were allowed to rest and habituate for 3-5 min before starting. For the RI test and odour presentation, we recorded Ca 2+ uorescence during 2 min of baseline activity without an intruder, followed by 5 min of intruder or odor exposure. The 415 nm channel served as the control channel and was subtracted from the GCaMP6s channel to eliminate auto uorescence, bleaching and motion effects. Change in uorescence (ΔF/F) was computed by subtracting the average value during the nal minute of baseline recording and then dividing the resulting value by the averaged value during the nal minute of baseline. This value was then z-scored by subtracting the average ΔF/F from each value and dividing by the standard deviation. Behavioural data were manually annotated in Ethovision and time stamps were aligned with uorescence recording.

Weighted correlation network analysis
For network construction, we followed previously published guidelines 14 . Brie y, a n x n adjacency matrix was created which encodes the connection strengths between pairs of nodes (brain regions), using the power adjacency function: A ij = |s ij | β . A ij refers the adjacency matrix and s ij refers to the correlation between regions i and j raised to the power of β. The correlation is raised to the power of β in order to reduce the in uence of potentially specious correlations. To measure the similarity between the nodes for clustering the topological overlap measure (TOM) was used: refers to the TOM matrix, In words, this is the sum of the product of the shared connections between regions i and j plus the direct connection between regions i and j. This value is divided by the denominator to achieve a value between 0 and 1. In order to compare the expression of these modules between sexes within a network, a singular value decomposition (SVD) was computed on each module. The SVD is de ned as: X n is the TOM matrix for the n th module in the network, U and V T are matrices of orthonormal vectors, and S is a diagonal matrix of eigenvalues that denote how much variation is accounted for by each column of U and V.

Statistical analysis
All statistical tests and associated information are reported in the Figure legends. All t-tests and two-way ANOVAs were performed using GraphPad Prism software (GraphPad Software Inc.). Two-way repeatedmeasures ANOVA analysis was followed by Šídák's multiple-comparisons test for post-hoc analysis. For comparisons between groups for region by region iDISCO+ analysis, P values were corrected for multiple comparisons using the Benjamini-Hochberg procedure to reduce false discovery rate. Q values below 0.05 were considered signi cant. Figure 1 The COApl is a key node in a differentially connected module in aggressive males. Male NON), 6 (Female NON) * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.001) See regions table for abbreviations .

Figure 2
In vivo activity of COApl Esr1 cells during distinct bouts of investigation in male AGGs a, Experimental timeline. b, Peri-event plot of COApl Esr1 activity two seconds before and two seconds after the investigation of an odour; c, Quanti cation of activity before and after the event. There was a main effect of cue (F (2,16) = 5.407, p = 0.0161) and a cue x time interaction (F (2,16) = 11.91, p = 0.0007). Post-hoc comparisons revealed an increase in COApl Esr1 activity only during the investigation of soiled bedding (p < 0.0001) d, f, h, peri-event plot of COApl Esr1 activity four seconds before and four seconds after performing the behaviour of interest; e, g, i, quanti cation of activity before and after the event.
When combining all bouts of investigation on days in which an attack did not occur compared to bouts of investigation on a day in which an attack occurred, there was a signi cant increase in activity during  better preserved the connectivity and density of the pink module than the hM4Di network (t (9) = 3.983, p = 0.0032) when separately analyzed. Each pair of points connected by a line are a single preservation metric from each network. n = 19. (HM4di), 21 (mCherry) f, g, Whole brain c-Fos network in mice injected with mCherry (MC) or hM4Di (H4) and network plot of the pink module from the WT AGG network.