Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 May 20.
Published in final edited form as: Neuron. 2020 Mar 11;106(4):637–648.e6. doi: 10.1016/j.neuron.2020.02.014

Hierarchical representations of aggression in a hypothalamic-midbrain circuit

Annegret L Falkner 1,2,*, Dongyu Wei 2, Anjeli Song 3, Li W Watsek 2, Irene Chen 2, Patricia Chen 1, James E Feng 2, Dayu Lin 2,4,5
PMCID: PMC7571490  NIHMSID: NIHMS1575430  PMID: 32164875

Summary

While the ventromedial hypothalamus, ventrolateral area (VMHvl) is now well established as a critical locus for the generation of conspecific aggression, its role is complex, with neurons responding during multiple phases of a social interaction. It has been previously unclear how the brain uses this complex multidimensional signal and coordinates a discrete action: the attack. Here, we find a hypothalamic-midbrain circuit that represents hierarchically organized signals during aggression. Optogenetic-assisted circuit mapping reveals a preferential projection from VMHvlvGlut2 to lPAGvGlut2 cells and inactivation of downstream lPAGvGlut2 populations results in aggression-specific deficits. lPAG neurons are selective for attack action and exhibit short-latency, time-locked spiking relative to the activity of jaw muscles for biting. Lastly, we find that this projection conveys male-biased signals from the VMHvl to downstream lPAGvGlut2 neurons that are sensitive to features of ongoing activity, suggesting that action-selectivity is generated by a combination of both pre and postsynaptic mechanisms.

Keywords: Aggression, circuit, hypothalamus, periaqueductal gray, social behavior, jaw muscle

Graphical Abstract

graphic file with name nihms-1575430-f0001.jpg

While stimulation of the hypothalamus has long been known to evoke attack, how these neurons organize aggressive action is unknown. Here, Falkner et al. describe an excitatory hypothalamic to midbrain circuit that transforms generalized social information into an action-specific neural code that is time locked to the bite during attack.

Introduction

The VMHvl has recently emerged as a critical hub for the integration of socially relevant information, exhibiting heterogeneous activity during the sensory, motor and motivational phases of aggression. Neurons in this small excitatory subregion of the hypothalamus not only respond during attack itself, but also show increased activity during sensory investigation of males and females and during the preparatory phase prior to attack(Falkner et al. 2016, 2014; Remedios et al. 2017; Lin et al. 2011; Hashikawa et al. 2017). In addition, suppression of VMHvl activity can have complex effects, decreasing not only the frequency of attack, but also investigatory, sexual, and aggression-seeking behaviors(Yang et al. 2013; Lee et al. 2014; Falkner et al. 2016). How then do neurons downstream of the VMHvl interpret this complex code to drive attack?

Outside of the hypothalamus, the periaqueductal gray receives the most prominent input from the VMHvl (Lo et al. 2019), and this area has long been known to play a role in the expression of various survival behaviors. The emerging role of the PAG in the expression of non-aggressive survival behaviors, such as stimulus-induced flight, appears to be that of a split-second action coordinator (Evans et al. 2018; Wang et al. 2019). We reasoned that a parallel circuit in the PAG might perform a similar function during conspecific attack.

In support of this, many decades of work across multiple species has firmly established a functional role of the PAG within the canonical “aggression circuit”. Stimulation experiments in cats and rats first demonstrated that electrical stimulation of the PAG was able to evoke elements of defensive aggression (Fernandez de Molina and Hunsperger 1962; Mos et al. 1982; T. R. Gregg and Siegel 2001), and that the effect of simulation can be facilitated through activation of glutamatergic synapses in the PAG (Siegel, Schubert, and Shaikh 1997). Destruction of this area through aspiration or through pharmacological means has been shown to dampen the behavioral effects of “upstream” stimulation, strongly indicating its role in the organization of aggressive action (Zalcman and Siegel 2006; Thomas R. Gregg and Siegel 2003; Schreiner and Kling 1953). Finally, a single paper reporting in vivo recordings in the PAG of cat shows a small number of neurons increasing during coordinated attack response (Adams 1968). While the functional role of the PAG in the aggression circuit has been firmly established, a systems level understanding of how this action coordination occurs in the circuit has been lacking.

The PAG is traditionally subdivided into columns that are molecularly and anatomically conserved across species (R. Bandler and Shipley 1994; Richard Bandler and Keay 1996), and have distinct roles in the coordination of survival behavior. Here we focused on the lateral PAG (lPAG) based on two criteria. First, this area is known to receive the densest projections from the VMHvl (Shimogawa, Sakuma, and Yamanouchi 2015), and second, previous studies have observed a high amount of immediate early gene expression following aggression (Lin et al. 2011). However, patterns of immediate early gene activation lack the temporal resolution to understand the precise interactions between the VMHvl and the PAG, and we do not yet understand the relationship between this putative action coordination area and its effectors. In this study, we use a combination of in vivo and ex vivo physiology, cell-type specific perturbation, and multi-site optical recordings to explore how this hypothalamic to midbrain circuit coordinates aggressive action.

Results

Since the PAG is a complex structure with several molecularly and anatomically distinct subregions that are known to receive hypothalamic input (Silva and McNaughton 2019), our first goal was to identify and map a functionally connected circuit using optogenetic-assisted circuit mapping. First we explored whether excitatory projection neurons from the VMHvl form functional connections with neurons in the lPAG. We targeted excitatory projection neurons by injecting a red-shifted cre-dependent excitatory opsin (AAV2.Syn.Flex.ChrimsonR.tdTomato) into the VMHvl of excitatory neurons (vGlut2-ires-cre mice) crossed with an Ai6 reporter (Fig. 1AB). We first confirmed that brief light pulses at the VMHvl were sufficient to reliably evoke action potentials (Fig. 1C). Then, we made coronal slices of the lPAG and performed voltage clamp recording from putative vGlut2+ and vGlut2- neurons (Fig. 1DE). We found that 50% (29/58) of identified vGlut2+ neurons exhibited short-latency excitatory post-synaptic currents (EPSCs) upon light delivery, while no single vGlut2- neurons (n =12) exhibited this excitatory response (Fig. 1FG). Bath application of tetrodotoxin (TTX) and 4-aminopyridine (4-AP) did not change the magnitude of light-evoked EPSCs, supporting the monosynaptic nature of the connection (Fig. 1HI). None of the recorded cells showed inhibitory postsynaptic currents (IPSCs) upon light delivery.

Figure 1.

Figure 1.

An excitatory circuit connect the VMHvl to jaw-projecting neurons in the lPAG (A) Viral strategy for targeting excitatory projections from VMHvl to lPAG. Slices were made of VMHvl and lPAG and whole cell recordings were performed. (B) Representative infrared differential interface contrast image (IR-DIC) from a recorded slice containing VMHvl (top, left) and lPAG (bottom, left). Yellow arrows indicate locations of recording pipette tips. Scale bar 500 μm. (right) Coronal section (right) showing expression of Chrimson-tdTomato (red) from a vGlut2 x Ai6 mouse. Scale bar 1mm. (C) Example trace showing current clamp recording of a VMHvlvGlut2 neuron expressing Chrimson-tdTomato. 605 nm light pulses (20 Hz, 20 ms for 500 ms, red ticks) reliably evoked time-locked spiking. Scale bars: 100 ms (horizontal) and 10 mV (vertical). (D) Histological image showing distribution of glutamatergic cells (green) and Chrimson-tdTomato expressing fibers from the VMHvl (red) in the PAG. Blue: Topro-3. Scale bar, 200 μm. (E) Enlarged views from (D) showing biocytin-filled vGlut2 negative (top row) and positive cells (bottom row) and their corresponding recording traces showing 1ms 605 nm light evoked EPSC (−70 mV) and IPSC (0 mV). Yellow arrows indicate the locations of the biocytin filled cells. Scale bar (left), 20 μm. Scale bars (right): 10 ms (horizontal) and 10 pA (vertical). (F) Stacked bar graphs showing the percentage of recorded PAG cells receiving EPSC during light stimulation. Two-tail Fisher’s test. *p = 0.0132. (G) Light-evoked EPSC amplitude (left) and latency (right) in PAG glutamatergic neurons (N=12). Error bars show means ± SEM. (H) Example traces showing 1ms 605 nm light evoked EPSC before (black) and after 1 μM TTX and 100 μM 4AP perfusion (red) in PAG glutamatergic neurons. Scale bars: 100 ms (horizontal) and 10 pA (vertical). (I) No change in light-evoked EPSC amplitude before and after 1 μM TTX and 100 μM 4AP perfusion in PAG glutamatergic neurons (n=6). Paired t-test. p > 0.05. (J) Schematic describing strategy for quantifying jaw-projecting lPAGvGlut2+ neurons. (K) 297/327 (90.8%) of PRV-614 labeled neurons in lPAG were vGlut2-Ai6 positive. (Example overlap shown in K, bottom right). (L) Strategy for functional targeting of jaw-projecting lPAG neurons (M) Example traces showing 1ms 605 nm light evoked EPSC in one PAGvGlut2 neuron. Scale bars: 10 ms (horizontal) and 10 pA (vertical). (N) Summary (left; n=27 cells, n=8 mice) and failure rate (right). Stacked bar graph showing 12 out of 23 PAG PRV+ vGlut2+ neurons received glutamatergic input from VMHvl. (O) Light-evoked EPSC amplitude and latency in PAG PRV+ vGlut2+ neurons (n=23). Error bars show means ± SEM.

We next hypothesized that a neural circuit that transforms a heterogeneous hypothalamic neural code into action would be linked to relevant musculature in as few synapses as possible. As a proxy for the implementation of aggressive action in mice, we focused on a single important effector for aggression: the jaw. By using intramuscular injections of a retrogradely transported pseudorabies virus (PRV) (Smith et al. 2000), neurons in the PAG have been previously identified having polysynaptic projections to jaw muscles critical for aggressive behavior (Fay and Norgren 1997). We injected the superficial masseter muscle of the jaw (the critical muscle for jaw closure) with a GFP-labeled PRV 152 and confirmed that the majority of labeled neurons were located in the lPAG (Figure S1). We found that PRV-labeled neurons in the lPAG were overwhelmingly glutamatergic (Fig 1JK). Using an Ai6 reporter line in combination with injections of a red-shifted PRV (PRV-614) in the jaw, we found that 297/327 red labeled PRV labeled neurons (90.8%) were labeled with GFP, indicating that these neurons are almost entirely excitatory (Fig. 1K).

Next, we tested whether this effector-projecting subpopulation in the IPAG receives functional input from excitatory neurons from the VMHvl. Three weeks after injecting the VMHvl of vGlut2 x Ai6 mice with Chrimson, we injected the jaw with a red-shifted PRV 614 (Fig. 1L). 96 hours following PRV injections, animals were sacrificed and lPAG slices were. We could evoke reliable EPSCs in slightly more than half of PRV+/vGlut2+ postsynaptic neurons in the lPAG (Fig. 1MN) by stimulating the excitatory projection from VMHvl. Intriguingly, the failure rate in our ability to evoke EPSCs following single 1ms pulse of light was non-zero, indicating that this input could not reliably drive activity in the jaw-projecting lPAG neurons (Fig. 1N, right). Overall, these data demonstrate a strongly preferential excitatory-to-excitatory circuit from glutamatergic hypothalamic neurons in the VMHvl to glutamatergic neurons in the lPAG, a subset of which project polysynaptically to aggression-relevant musculature.

We next tested the functional efficacy of this hypothalamic-midbrain projection. Attack of a castrated male has been shown to be reliably evoked through stimulation of the VMHvl cell bodies(Lin et al. 2011; Lee et al. 2014). We tested whether optogenetic activation of the VMHvl-PAG pathway could also reliably evoke attack. In some animals, we observed an upregulation of attack probability towards castrated males through this manipulation, but the ability to evoke attack was generally unreliable (Figure S2). Together, these data lend support to the hypothesis that the lPAG may organize specific motor aspects of attack, but may not be sufficient to generate the full suite of attack behaviors in a reliable manner similar to the VMHvl.

Chemogenetic inactivation of PAGvGlut2 leads to aggression-specific deficits

While inactivation of the VMHvl can result in a mixture of socially specific deficits (Yang et al. 2013; Lee et al. 2014; Falkner et al. 2016), we reasoned that if the lPAG was involved in translating VMHvl signals into an aggression motor-specific code, the effects may be action specific. To test this, we injected the lPAG of vGlut2-ires-cre males bilaterally with a cre-dependent inhibitory DREADD (AAV2-hSyn-DIO-hM4Di-mCherry) and tested whether inactivation of these neurons affected the expression of social and nonsocial behaviors following CNO or saline injection (Fig. 2AB). Here, we found that inactivation of the lPAGvGlut2 neurons following CNO i.p. injection resulted in aggression-specific deficits relative to saline injection (Fig. 2CF). Following CNO injections, mice spent less time attacking during interactions with males (Fig. 2C, Video S1), and this was entirely due to a reduction in the duration of attack episodes (Fig. 2D), rather than an increase in attack latency or a decrease in the number of attack episodes (Fig. 2EF). In contrast to attack, inactivation of the lPAG did not significantly affect the time spent investigating either males or females, or the time engaged in sexual behaviors, as quantified by the time spent mounting the female (Fig. G-I). We observed no significant differences in alternate measure of these behaviors, including mean behavior duration, latency, or number of episodes. We found no difference between saline and CNO injected animals in the latency to approach and consume palatable food (Fig. 2J), indicating that other behaviors requiring jaw closure were not affected.

Figure 2.

Figure 2.

Chemogenetic inactivation of lPAG results in aggression-specific deficits. (A) Viral strategy for reversible inactivation of lPAG. vGlut2-ires-cre animals were injected bilaterally with DIO-hM4D(Gi)-mCherry (N=8 animals) or DIO-mCherry (N=6 animals). After injection with either saline or CNO, animals were tested during interactions with males, females, then given access to palatable food. (B) Example histology showing expression of hM4Di-mCherry in the PAG. Scale bar 500μm. (C-F) Following CNO inactivation (left panels), percent time spent attacking was reduced relative to saline (C, p=0.016). This effect was due entirely to reduced attack duration following inactivation (D, p=0.004), and not to changes in attack latency (E, p=0.784) or to the number of attack episodes (F, p=0.772). No effects on attack were observed in control animals (right panels, C-F, p=0.854, p=0.236, p=0.480, p=0.854). (G-J) In contrast to attack, no effects on other social or jaw-dependent behaviors were observed. CNO Inactivation did not affect the perfect time spent investigating males (G, hM4Di: p=0.751 mCherry: p=0.239), time spent investigating females (H, hM4Di: p=0.270 mCherry: p=0.584), mounting females (I, hM4Di: p=0.332 mCherry: p=0.567), or the latency to approach and consume a yogurt pellet (J, hM4Di: p=0.998 mCherry: p=0.998), paired t-tests. We tracked the locations of the saline and CNO injected animals (K) during male-male (L) and male-female (M) interactions. No significant differences in the distribution of inter-animal distance (L-M left, p=0.243, p=0.154) or velocity (L-M right, p=0.999, 0.999) were observed between saline and CNO injection.

We used machine-vision tracking (DeepLabCut (Mathis et al. 2018)) to quantify the inter-animal proximity and movement velocity during interactions following saline and CNO administration (Fig. 2K). We found no significant difference in the overall distribution of inter-animal proximity or resident velocity in either male-male (Fig. 2L) or male-female interactions (Fig. 2M), indicating that inactivation of lPAGvGlut2did not result in non-specific motor or motivational deficits. Together, these cell-type specific inactivation data suggest a specific role for lPAGvGlut2 in the execution and maintained coordination of attack since.

PAG neurons encode a simplified action-selective code relative to VMHvl

The aggression-specific deficit during PAG inactivation indicates that the PAG’s role within the aggression circuit is uniquely action-selective. To confirm this, we recorded populations of single neurons in the lPAG during free social interactions and compared the response profiles to newly quantified measures of previously recorded neurons from the VMHvl(Falkner et al. 2014; Lin et al. 2011). Electrode tracks in all animals were confirmed to be located in the lPAG post hoc using DiI (Fig 3A). We examined the firing rates of single units aligned to attack, investigation of males, and investigation of females (Fig 3BE). Across the population (n=164 neurons, 6 animals), we observed that a subpopulation of neurons in the lPAG exhibited their peak firing aligned to the onset of attack while few neurons exhibited peak firing aligned to the onset of either investigation of male or female conspecifics (Fig 3CE).

Figure 3.

Figure 3.

Activity in the lPAG is more attack selective than the VMHvl. We performed chronic single unit recordings in the lPAG during social behaviors and compared these responses to activity in the VMHvl. (A) Histology showing example placement of electrode bundles (and one group of tetrodes (PAG3) in the lPAG and electrode track locations for all recording animals (N=6). Scale bar 1mm. (B) Example raster plot (top) and PETH (bottom) for activity of an example lPAG units aligned to attack (red), investigation of a male (blue), and investigation of a female (green) sorted by behavior. (C-E) Normalized responses of population (top) of recorded neurons sorted by peak of response aligned to attack (C, N = 159), investigation of male (D, N = 158), and investigation of female (E, N = 151). Bottom histograms show number of units with response peak above 95% CI in each bin. Dotted black lines (C-E) represent chance levels for each behavior. (F) Normalized population response mean ±SEM of VMHvl (top) and lPAG (bottom) during onsets of key behaviors interactions with males: attack (red) and investigate male (blue), and with females: investigate female (green). (N = 166,156 for VMHvl male attack and investigate, N=212 for female investigate, N = 159 for lPAG male attack and investigate, N=151 for female investigate). Comparison of responsivity of individual VMHvl neurons (G-H top, light gray) and lPAG neurons (G-H bottom, dark gray). VMHvl population is nonselective between attack and investigate male (G, top, p=0.806, N = 157), and selective for investigation of male compared to female (H, top, p=0.005, N=147), while lPAG is selective for attack relative to investigate male (G, bottom, p=3.4×10^−7, N=152), and nonselective for investigation of males and females (H, bottom, p=0.415, N=152). Tests in G-H performed using Wilcoxon signed-rank test. Pie chart insets displaying percentages of individually significant neurons (Bonferroni-corrected t-test) in VMHvl and lPAG show an increasing number of purely attack selective neurons in the lPAG relative to VMHvl and a decrease of investigation selective neurons in the lPAG. (I) Selectivity of population to attack compared to selectivity to investigate male shows that attack-shifted peak for lPAG population (dark gray) relative to VMHvl (light gray) shown using RI value. P=0.0001, Kolmogorev Smirnov test. (J) Attack responsive neurons in the VMHvl (light gray) have significantly increased activity prior to attack onset relative to attack responsive lPAG neurons (dark gray). N= 44 neurons in VMHvl N=46 neurons, lPAG, p = 0.0005, Bonferroni corrected unpaired ttest across all bins.

This selectivity for attack differs substantially from the response properties of the VMHvl. The VMHvl exhibits distinct activity peaks aligned to attack, investigation of males, and investigation of females (Fig 3F, top). In contrast, the population response at the lPAG reveals an increase only during attack (Fig 1F, bottom). To compare the responses of single neurons in these populations, each neuron was assigned a responsivity score for each behavior (relative to its response during nonsocial behaviors). We determined the number of Bonferroni-corrected, significant neurons for each behavior in both populations (Fig 3GH). We found that the population of lPAG neurons is significantly more attack-selective than investigate-selective, while the VMHvl population has equal responsivity for both behaviors (Fig 3G). In addition, we found that lPAG exhibited an increased number of uniquely attack responsive neurons, and showed a decrease of both co-selective and investigate-selective single units relative to the VMHvl (Fig 3G, p=0.0031, chi-square test). Consistent with this, we also observed that the lPAG population shows little selectivity for either male or female investigation (Fig 3H, p=0.0210). Overall, we found that the selectivity for attack was increased in the lPAG compared to the VMHvl and that the activity of attack selective neurons in the VMHvl increased earlier than attack selective neurons in the lPAG (Fig 3IJ).

lPAG neurons exhibit time-locked activity to jaw muscle activity

Given that a subpopulation of lPAGvGlut2 receives excitatory input from the VMHvl and projects polysynaptically to the jaw, we hypothesized that neurons in the lPAG may have a specific role in coordinating attack-relevant musculature. We developed a preparation to simultaneously record from neurons in the lPAG and EMG activity from the superficial masseter (EMGSM) while animals are attacking and performing other social interactions (Fig 4AB, Video S2, n=64 neurons in 4 animals). We used mutual information (MI) to quantify whether the EMGSM signal provides useful information in predicting the activity during male or female interactions. Mutual information provides a model-free method for capturing the amount of joint information between two signals, and increased mutual information indicates that one signal can be used to predict the other(Srivastava et al. 2017). We calculated MI for each neuron and associated EMGSM signal relative to a circularly permuted time shuffled control during interactions with males and females. We observed that across the population of recorded neurons, EMGSM increased the MI during male interactions but not during female sessions relative to shuffle control and also that MI provided by the EMGSM signal was significantly higher during male interactions compared to female interactions (Fig 4C), indicating that a subpopulation of neurons is modulated during attack-related jaw movement, but is not similarly activated by nonspecific jaw movements during female interactions.

Figure 4.

Figure 4.

lPAG spiking has precise temporal alignment with jaw muscle activity during aggression. (A) Example simultaneous recordings of jaw EMG and lPAG spiking during attack episodes, shown in red. (B) Example EMG (top) and activity from simultaneously lPAG neuron (bottom) during interaction with a male (C) Mutual information (MI) of lPAG spiking and EMG activity comparing during interactions with males and females (N=64 neurons). MI of activity during male interaction compared to time shuffled control (p=0.0214, paired t-test), MI of activity during female interactions to time shuffled control (p=0.1551, paired t-test), MI of activity during male and female interactions (p=0.0053, paired t-test). Example STEMG (D) and attack-aligned PETH (E) with precise temporal alignment to EMG. (F-G) STEMG (F) and attack aligned activity (G) of neurons with significant STEMG (red, top trace), and significant attack responsive neurons that are do not have significant STEMG (black, bottom trace), show distinct dynamics.

We hypothesized that the activity of jaw-activated neurons might have a tight temporal relationship to the muscle activity if it is involved in directly activating the muscle. We examined the relationship between individual spikes and the EMGSM signal. For each neuron, we generated a spike-triggered-EMG (STEMG) across an 800ms bin around each spike and used strict criteria (5 consecutive points must lie above 98% confidence interval) to determine whether each STEMG demonstrated a significant relationship with individual spikes(Davidson et al. 2007). We found that a subpopulation of recorded neurons (21.8%, 14/64 neurons) showed a significantly increased STEMG within 60ms of spikes during interactions with males (EMGsm+,Fig 4D,F). Most of these EMGsm+ neurons (12/14) were activated during attack (EMGsm+, Fig 4E,G). In contrast, neurons that were identified as being significantly activated during attack that were not overlapped with our EMG-identified subpopulation (EMGsm-/Atk+) showed distinctly different activity profiles in the STEMG and during attack (EMGsm-/Atk+). STEMG response profiles of EMGsm-/Atk+ neurons were diverse, but often exhibited suppression following PAG spiking (Fig 4F, black). Furthermore, activity of these EMGsm-/Atk+ neurons during attack showed that the responses were increased not just at the onset of attack, but showed activation that persisted through the attack (Fig 4G, black). These data are consistent with the hypothesis that attack-related neurons in the lPAG may activate multiple attack related muscles, including jaw opening musculature. A much smaller number of neurons (7.8%, 5/64 neurons) exhibited a time-locked relationship during female interactions, which suggests that a small minority of jaw responsive neurons may be recruited during other (nonaggressive) behaviors.

We closely examined the activity of these same neurons recorded simultaneously with EMG during other jaw-related behaviors including eating and grooming (Figure S3). While all three behaviors (attack, grooming, eating), produced significantly increased activity of the rectified EMG and can be observed in single behavioral trials, we observed an increase in the activity of the neural population only following attack and not in the other EMG-detected behaviors, providing further support to the specificity of this population in coordinating jaw related motor plans related to aggression.

Pathway-specific fiber photometry reveals male biased signal in the VMHvl-PAGvGlut2 projection

Our data demonstrates that neurons in the lPAG exhibit a greater degree of selectivity for aggressive action than the VMHvl. One simple mechanism by which this circuit could perform the transformation is if the projection from the VMHvl to PAG were a labeled line for either attack-specific or male-specific information. In order to explicitly test this, we targeted excitatory VMHvl-PAG projecting neurons by injecting the ipsilateral side of a vGlut2-ires-cre male mice with a retrogradely transported, cre-dependent, calcium indicator (HSV-Ef1a-LS1LGCaMP6f) into the lPAG. On the contralateral side, we injected a cre-dependent calcium indicator (AAV1.CAG.Flex.GCaMP6f.WPRE.SV40) in the VMHvl in order to compare the activity from the total VMHvl vGlut2+ population, not specified by projection. We positioned fibers over both the ipsi and contralateral VMHvl and used fiber photometry to simultaneously record respectively from VMHvl-PAGvGlut2 and the VMHvlvGlut2 populations during free social interactions (Fig 5AC).

Figure 5.

Figure 5.

Activity in lPAG-projecting VMHvl neurons conveys preferentially male-specific information (A) Experimental configuration of bilateral injection of GCaMP6f into vGlut2-ires-Cre mice for simultaneous fiber photometry recordings of VMHvl and VMHvl-PAG projection neurons. Ipsilateral injection targets VMHvlvGlut2 neurons and the contralateral injection targets VMHvl-PAGvGlut2 projection neurons. (B) Example histology of GCaMP-labeled VMHvlvGlut2 neurons (right) and VMHvl-lPAGvGlut2 neurons (left) and placement of fiber tracks. Scale bar 300μm. (C) Example simultaneous recording of VMHvl vGlut2 (black) and VMHvl-PAG vGlut2 (magenta) projection neurons during alternating interactions with males (blue) and females (red). (D) Population activity (mean +SEM for each animal, N=6 animals) of comparison between activity during male interaction and female interaction for VMHvl neurons (D left, p=0.0062) and for VMHvl-PAG projection neurons (D right, p=0.0002), shows that both populations exhibit increased activity to males. (E) Comparison of simultaneously recorded activity VMHvl vGlut2 and VMHvl-PAG vGlut2 neurons is not significantly different during male interactions (E left, p=0.3676), but activity during female interaction is reduced in VMHvl-PAG neurons (E right, p=0.044). (F) Distributions of the Pearson correlation between behavior-aligned activity in VMHvl and VMHvl-PAG. Correlation distributions are not significantly different between attack and investigation of male (F, left, p=0.358, unpaired t-test, N=83 attack trials, N=85 investigate male trials), but correlations are higher in both attack and investigate male than during investigation of females (F, middle p=0.00002; F, right p=0.004, unpaired t-test, N=180 investigate female trials). Solid lines represent distribution medians. (G) Correlation of simultaneously recorded VMHvlvGlut2 and VMHvl-PAGvGlut2 neurons is higher in male interactions than female interactions (p=0.0238). All tests (D-E,G) using paired t-test.

We found that both VMHvlvGlut2 and VMHvl-PAGvGlut2 populations showed a strong bias towards male compared to female mean activity (Fig 5D). However, we observed that activity during male interaction was not significantly different (Fig 5E), while during female interactions, VMHvl-PAG activation decreased relative to the overall VMHvl population response. In addition, to track how well the two signals match, we computed the correlation (Pearson r) between the simultaneously recorded activity of VMHvlvGlut2 and VMHvl-PAGvGlut2 populations during single behavior “trials” as animals were attacking or investigating males or females (Fig. 5FG). High correlations in these experiments indicate that VMHvl-PAGvGlut2 population sends a faithful copy of the VMHvlvGlut2 activity to the lPAG. We found that single trial correlations during male behaviors were similar (Fig. 5F, left). In contrast, correlations during investigation of females were significantly lower (Fig. 5F middle, right), indicating that the population is male selective Across all animals, we found that correlation coefficients during male interactions significantly increased during male interactions relative to female interactions (Fig 5G). Together, these data indicate that the VMHvl-PAGvGlut2 population behaves as a labeled line, selectively filtering activity during female interactions. We did not find evidence for an attack-specific labeled line (Figure 5F, Figure S4). To quantify this, we compared the PETHs of activity in the VMHvlvGlut2+ population and the VMHvl-PAGvGlut2+ projection during attack and investigation of males. Similar to the single unit data, both VMHvlvGlut2+ and VMHvl-PAGvGlut2+ activity also showed clear peaks aligned to both attack and investigation of males that were very strongly correlated during individual behavior episodes.

Together these data demonstrate that an excitatory projection from the VMHvl to the lPAG conveys male-biased information to excitatory downstream populations, behaving as a labeled line for aggression-relevant information. While this projection effectively filters female-evoked signals, it does not filter other male interactions such as investigation. This suggests that further downstream mechanisms are needed to transform VMHvl activity to an aggression-specific signal motor code. One possibility is that the PAG is sensitive to specific temporal features in its inputs prior attack-induced activation, including slower investigation-evoked increases that precede attack. To specifically test this, we performed simultaneous recordings of VMHvlvGlut2+ and PAGvGlut2 (Fig. 6AC) using multisite fiber photometry and computationally modeled the interactions between these two signals. Qualitatively, we observed that the during male interactions, PAGvGlut2 activity peaks often followed VMHvlvGlut2 activity peaks, while during female interactions, signal dynamics were more independent (Figure 6BC, Figure S4DF). In support of this, we found that in contrast to the correlation between VMHvlvGlut2 and VMHvl-PAGvGlut2, the trial-to-trial comparison of the VMHvlvGlut2+ and lPAGvGlut2 exhibited much higher correlations during attack than either investigation males or females (Fig. 6D). These lower correlations during investigation behaviors demonstrate more uncoupled signal between the two brain areas.

Figure 6.

Figure 6.

Simultaneous recordings of VMHvlvGlut2 and lPAGvGlut2 reveal lPAG activity is preferentially coupled to VMHvl during attack and is preferentially sensitive to VMHvl activity history during male interactions (A) Experimental configuration of simultaneous recordings during interactions with males (B) and females (C). (D) Distributions of the Pearson correlation between behavior-aligned activity in VMHvlvGlut2 and lPAGvGlut2. Correlation distributions are significantly higher for attack than investigation of males or females (D, left, p=0.00003, unpaired t-test, D, right, p=0.00006, N=112 attack trials, N=73 investigate male trials, N=174 investigate female trials), but correlation distributions are similarly low for investigation of either males or females (D, middle p=0.240). Solid lines represent distribution medians. (E) Cross correlation of simultaneously recorded signals during male interactions (blue), female interactions (red) and no-interaction baseline (black). (F) Comparison of summed cross correlation in pre epoch (−10s to 0s) and post epoch (0s to 10s) for male, female, and no-interaction baseline shows significant asymmetry only during male interaction (N=7 animals, male p=0.0174*, female p=0.9333, baseline p=0.6148). (G) PAG vGlut2 activity was fit with a linear model using a variable amount of preceding VMHvlvGlut2 signal as the regressors. (H) Fit percent (middle) and time (right) associated with best fit models of cross-validated data using a time-varying input from VMHvl. Dotted lines represent data from time shuffled controls. (I) Controls for time-varying regression model. Comparison of forward (VMHvl → lPAG) and reverse (lPAG → VMHvl) regressive models. Model fit percentages for fits during male (I, left) and female (I, right) interactions. Black traces (I) show reverse models. During male interactions (blue), fits are significantly increased in forward model relative to reverse model (*p<0.05, paired t-test for each time bin). (J) Conceptual model of pathway selectivity of male-responsive information.

To quantify the time-dependence of the interactions between the VMHvlvGlut2+ and lPAGvGlut2 activity, we computed the cross correlation of the simultaneously recorded signals separately for social behaviors and a baseline no-interaction epoch. We found that the cross correlation during male interactions was strongly and asymmetrically skewed across a multi-second timescale, indicating that increases in VMHvl signals “lead” PAG signals during interactions with males, but not females (Fig. 6EF). The integral of the cross correlation during the “pre” epoch (VMHvl leads PAG) relative to the “post” epoch (PAG leads VMHvl) significantly increased during male interactions across the population of recorded animals (n=7, p=0.0174, paired t-test) and not significantly different during either the interactions with females or no-interaction baseline period (p=0.9333, p=0.6148, paired t-test).

The slow temporal dynamics of the cross correlation during male interactions suggests that PAG activity is influenced by VMHvl activity stretching backwards in time. To specifically quantify this temporal relationship between the VMHvl and lPAG, we modeled the ongoing PAG activity using a time-varying linear regressive model (Fig. 6G). We iteratively fit PAG activity with the activity of the VMHvl (VMHvl → lPAG) with a family of models with a variable number of regressors representing increasing numbers of previous time bins, used Akaike Information Criteria (AIC) to select the best model within each family, and cross validated this model on a separate data set for each recorded animal. We performed this model selection separately for male and female interactions and across a variety of bin sizes, ranging from 50 ms to 1 s. We found that cross validated fits for the best fit model were significantly better during male but not for female interactions relative to a time shuffled control. Importantly, this effect was consistent across a range of bin sizes (Fig. 6H), indicating that this variable made little difference in the model fit. In addition, we extracted the model order (number of time bins) for the best fit model for each animal for male and female interactions for each bin size. We found that the elapsed time associated with these best fit models was significantly longer for male interactions than female interactions (Fig. 6H, right), indicating that during male interactions, PAG signals are influenced by VMHvl signals stretching farther back in time.

On a conceptual level, this suggests that activity in the VMHvl increases prior to activation of the PAG, an effect that is consistent the action specificity of the PAG and the fact that activity in the VMHvl appears to increase prior to lPAG (Fig. 3J). As a control for the signal kinetics, we also fit the reverse circuit model, (lPAG→ VMHvl) and found that both the fits, and model order are significantly decreased relative with the forward circuit model during male (Fig. 6I). Together, these results suggest a model by which during interactions with males, but not during interactions with females, the PAG receives VMHvl inputs conveying information about the sensory properties of the stimulus obtained during investigation, and this information is integrated over several seconds in order to coordinate attack behavior (Fig. 6J).

Discussion

These findings provide the first direct evidence that the role of the lPAG in the aggression circuit is to transform the complex sensory-motor and motivational signals of the hypothalamus into aggressive action-specific code. Since neural signals are time locked to jaw movement during aggression, we propose that the role of the lPAG in aggression is to coordinate effector-specific musculature, including the jaw. We build on decades of elegant functional manipulation experiments to provide a systems-level description of the hierarchical representations of aggression in the hypothalamus and midbrain, with downstream lPAG neurons representing a “simplified” action-selective neural code relative to the VMHvl. These results add to a growing literature that position the PAG as a critical coordinator of survival actions that are determined by combinations of noisy sensory and state-dependent inputs(Koutsikou, Apps, and Lumb 2017).

Since PAG neurons are active later and more acutely during attack behavior than its hypothalamic inputs from the VMHvl, it may serve as a behavioral initiation threshold, as has been suggested for other survival behaviors. We do not mean to suggest that the lPAG is “for” aggression: the PAG (and in particular hypothalamic to PAG pathways) have been implicated in many sensory-driven behaviors including but not limited to threat responsivity(Wang, Chen, and Lin 2015; Evans et al. 2018), prey capture(Li et al. 2018), itch(Gao et al. 2018), oro-motor coordination(Stanek et al. 2014), and social avoidance following defeat(Franklin et al. 2017). Additionally, the PAG has a well-documented role across species in the generation of vocalization, a behavior that also requires the integrations of social-sensory signals and the coordination of facial and laryngeal musculature(Kittelberger, Land, and Bass 2006; Gert Holstege 2014; Tschida et al. 2019). Here we add to this literature by characterizing the neural code during conspecific attack and further hypothesize that the PAG is capable of orchestrating many complex survival behaviors by coordinating output to relevant muscles through sex-selective processing of specific temporal features in its inputs (Fang et al. 2018).

How does this transformation occur? Here we show that the lPAG provides a functional link between excitatory activity in the VMHvl and a jaw-projecting excitatory population. This PAG-projecting VMHvl population is a “partial” labeled line, filtering out some aggression-irrelevant signals, but is not a “pure” action selective signal, implying that further downstream mechanisms are needed to generate the action-selective code. One simple hypothesis that that the PAG simply represents the accumulation bound from VMHvl input: When VMHvl activity is high, this drives activity in the PAG when a particular threshold is reached. While simple and compelling, this does not account for the fact that at the level of the VMHvl population, activity during male investigation and attack are indistinguishable, activity from both behaviors is relayed to the PAG, yet the PAG is activated only during attack. A second possibility that that if input is noisy, high fidelity spikes from the VMHvl occur only at or around the initiation of attack, activating post-synaptic PAG neurons, similar to the effects observed in a circuit for defensive escape (Evans et al. 2018). We observed that the failure rate for the VMHvl-lPAG synapse here is relatively low (~10%, Fig 1). Although lower than the failure rate observed in the escape circuit (~30%), it is non-zero, indicating that spike fidelity may still play a role in driving this circuit.

Another possibility is that transformation to an action-selective code may occur at the local level within the lPAG. Mixed sensory and action signals may target a subpopulation of lPAG neurons that respond during both sensory and action phases of aggression, and then local attack-selective populations project to an effector-specific output lPAG population. A subpopulation of single neurons within that population responds to sensory variables, demonstrating significantly response heterogeneity within this circuit (Figure 3G). Second, a subpopulation of these lPAGvGlut2 PRV+ neurons do not receive direct input (Figure 1N), suggesting that further processing may occur at the local level. Further experiments using high density probes to perform in vivo recordings from a large population of neurons simultaneously will likely yield important insights about the local lPAG dynamics during the aggression transformation.

Finally, the lPAG is only one of many outputs from the VMHvl: Recent reports show at least 25 outputs from the esr1+ VMHvl subpopulation alone (Lo et al. 2019). Therefore, the lPAG may represent only one piece of the VMHvl output circuit, and may be involved in solely in coordinating effector-specific motor components of aggression. The lPAG may generate action specificity is by requiring coincident input from other nodes in the aggression circuit. Unlike stimulation-evoked attack from the VMHvl, which occurs in a time-locked and robust manner by recruiting a host of arousal and endocrinological mechanisms in addition to putative motor control populations, we found that our ability to evoke attack through the VMHvl-PAG pathway was generally unreliable. We reasoned that a more nuanced “on-manifold” activation pattern may be required to drive complex action. One likely candidate region (though there are many) that may provide coincident input is the dorsal medial PAG, which also receives excitatory input from the VMHvl and projects monosynaptically to the lPAG (Jansen et al. 1998). This projection could potentially behave as a second information loop in the hypothalamic-midbrain hierarchy.

Lastly, our functional inactivation experiments that specifically target lPAGvglut2 neurons show a highly specific effect on the duration of attack (as opposed to the latency or number of behavioral episodes). Since attack latency is often used as a proxy for “escalated” or pathological aggression (Miczek, de Boer, and Haller 2013), this suggests that the animal’s overall state is not affected through inactivation, and that effects may be more motor in nature. This strongly indicates that the role of the lPAG in the aggression circuit is to coordinate the activity of multiple muscle groups, rather than having than a more general effect on aggressive state. We liken the relationship between the hypothalamus and the PAG to the execution of a piano sonata, where the activity in the VMHvl provides the song, and the lPAG represents the keys on the piano. The hypothalamus is able to specify the key of the song, set the mood, and determine which chords get played, while the PAG population determines how keys are pressed and depressed.

STAR METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

  • Further information about resources, reagents used, and requests for code should be directed to and will be fulfilled by the Lead Contact, Annegret Falkner (afalkner@princeton.edu).

  • This study did not generate new unique reagents.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals

Experimental mice for recording, perturbation and behavior were sexually experienced, wild-type male C57BL/6N (12–24 weeks, Charles River), wild-type male Swiss Webster (12–24 weeks, Taconic), or vGlut2-ires-Cre mice. Sexually naïve vGlut2 x Ai6 mice were used for slice physiology and tracing experiments. Intruders used for social interaction tests were group housed, sexually inexperienced BALB/c males, castrated group housed BALB/c males, or group housed C57BL/6 females (both 10–30 weeks). Mice were maintained on a reversed 12-h light/dark cycle (dark cycle starts at noon) and given food and water ad libitum. All experiments were performed in the dark cycle of the animals. All procedures were approved by the IACUC of NYULMC and Princeton University in compliance with the NIH guidelines for the care and use of laboratory animals.

METHOD DETAILS

In vitro electrophysiological recordings

vGlut2:Cre x Ai6 mice were injected with 100 nL rAAV2.syn.flex.ChrimsonR.tdT into VMHvl. Three weeks after virus injection, acute horizontal brain slices of VMHvl and PAG (275 μm in thickness) were collected using standard methods (Fang et al. 2018). After being anesthetized by isoflurane inhalation, the mice were perfused by ice-cold choline based cutting solution containing (in mM) 25 NaHCO3,25 glucose, 1.25 NaH2PO4, 7 MgCl2, 2.5 KCl, 0.5 CaCl2, 110 choline chloride, 11.6 ascorbic acid, and 3.1 pyruvic acid. The slices were collected in the same cutting solution using a Leica VT1200s vibratome, incubated for 20 min in oxygenated artificial cerebrospinal fluid (ACSF) solution (in mM: 125 NaCl, 2.5 KCl, 1.25 NaH2PO4, 25 NaHCO3, 1 MgCl2, 2 CaCl2 and 11 glucose) (osmolality, 295 mmol/kg) at 32–34°C and then maintained at room temperature until use. Standard whole cell recordings were performed with MultiClamp 700B amplifier (Molecular Devices) and Clampex 11.0 software (Axon Instruments). Membrane currents were low-pass filtered at 2 kHz and digitized at 10 kHz with Digidata 1550B (Axon Instruments). Electrode resistances were 2–4 M , and most neurons had series resistance from 4 to 15 M . Glutamatergic (green fluorescent) or GABAergic (non-fluorescent) cells as well as Chrimson-tdTomato expressed VMHvl cells were identified with an Olympus 40 x water-immersion objective with GFP and TXRED filters. The slices were superfused with ACSF warmed to 32– 34°C and bubbled with 95% O 2 and 5% CO2. The intracellular solutions for voltage clamp recording contained (in mM) 135 CsMeSO3, 10 HEPES, 1 EGTA, 3.3 QX-314 (Cl– salt), 4 Mg-ATP, 0.3 Na-GTP, and 8 Na2-Phosphocreatine (osmolality, 295 mmol/kg; pH 7.3 adjusted with CsOH), and for current clamp recording contained (in mM) 130 K MeSO3, 5 KCl, 0.5 EGTA, 20 HEPES, 1.8 MgCl2, 0.1 CaCl2, 4 Na2-ATP, and 0.2 Na-GTP (osmolality, 295 mmol/kg; pH 7.3 adjusted with KOH). To activate Chrimson-expressing VMHvl glutamatergic neurons and Chrimson-expression axons in PAG, brief pulses of full field illumination (20 ms for VMHvl during current clamp recording and 1 ms duration for PAG during voltage clamp recording) were delivered onto the recorded neuron with 605 nm LED light (pE-300white; CoolLED) at 35 s intervals. Voltage clamp recording was conducted for PAG neurons, and the membrane voltage was held at −70 mV for EPSC recording, and at 0 mV for IPSC recording. Current clamp recording was conducted in VMHvl glutamatergic neurons expressing ChrimsontdTomato, where the neurons were maintained at resting potential and spiking activity was detected with or without red light pulses (20 ms, 20 Hz for 500 ms). For recordings of cells labeled with PRV, PRV-614 was injected to jaw muscle three weeks after rAAV2.syn.flex.ChrimsonR.tdT injection into VMHvl of vGlut2:Cre x Ai6 mice. 96 hours after PRV injection, acute horizontal brain slices of PAG were collected for whole cell recordings.

Behavior analysis and tracking

All freely moving behaviors were recorded using top and side GigE cameras using StreamPix (Norpix) or Spinview (FLIR) and all videos were manually annotated for pre-identified behaviors and tracked for positional and velocity information using custom Matlab Software (https://github.com/pdollar/toolbox), the Behavioral Observation Research Interaction Software (BORIS, https://www.boris.unito.it/) and DeepLabCut(Mathis et al. 2018). Behaviors were manually classified as previously described(Falkner et al. 2014); individual behaviors included attack, investigation of males, investigation of females, mounting, eating, and grooming. All social interactions were variants of the “resident intruder assay” (5–10 minutes of free interaction with male or female intruder), or alternating interactions with males and females (1 min each) separated by 1 min (Fig 5) in the home cage of the experimental animal.

PRV injections

The right masseter was exposed and PRV-152 or PRV-614 (kind gift from Lynn Enquist), was injected at 5 separate locations, 250uL per injection along the A-P axis of the muscle. The skin was sutured closed and PRV was incubated for 96–112 hours prior to sacrifice. For vGlut2 Ai6 overlap experiments, PRV labeled neurons in PAG were counted on sequential 50uM thick sections through the PAG. A total of 9 sections were counted across 3 animals.

Optogenetic activation

For optogenetic activation experiments we injected wildtype males with a 2:1 mixture of 100–150nl of AAV2-CMV-Cre (2 × 1012 PFU/ml) and AAV2.EF1a-DIO-hChR2(H134R)-EYFP (1 × 1012 PFU/ml, both from UNC vector core) ipsilaterally in the VMHvl (−1.82A/P, 0.72M/L, −5.8D/V) and a 230- m multimode optic fiber (Thorlabs) was positioned 0.5mm over the injection site and over the terminal field in the lPAG targeting the middle third of the column (coordinates: −4.29 A-P, +/−0.45 M-L, −1.9 D-V). vGlut2-ires-cre males were injected with 130nl of AAV-EF1a-double floxed-hChR2(H134R)-mCherry-WPRE-HGHpA (Addgene 20297, 1.8 × 10^13, Lot V31400). Optogenetic testing on the cell bodies was performed 3 weeks following viral injection. The other end of the optic fiber was connected to a 473-nm laser (Shanghai Dream Lasers or Opto Engine) controlled by computer-programmed TTL pulses. For optogenetic activation, animals were screened for ‘functional’ injection sites using a resident-intruder test to determine whether stimulation of each injection site was sufficient to evoke attack of a castrated Balb/c male. High frequency (10–20 Hz, 10-ms pulses, 0.5–1.5 mW) stimulation was delivered through the optic fiber for 20–30 s, with a minimum duration of 30 s between stimulation trials. Stimulated animals received 5–10 stimulation trials during each test session for each injection site (~15 min). Consistent with our previous report, light activation of functionally defined sites caused animal to orient toward, approach, investigate, and eventually attack the intruder. If animals showed robust attack during this test (attack within 30s of stimulation on, on >=50% of trials), they were used for the terminal stimulation experiments.

Chemogenetic Inactivation

For reversible inactivation with DREADD gi, we stereotactically injected vGlut2-ire-cre mice with 100nl of AAV2-hSyn-DIO-hM4Di-mChery (Addgene, 44362-AAAV2, titer: 1.4 × 10^13 pfu/ml, Lot v30430) bilaterally in the lPAG at 3 sites along the A-P column of the PAG (coordinates: −3.93 A-P, +/−0.30 M-L, −2.3 D-V;−4.29 A-P, +/−0.45 M-L, −2.3 D-V; −4.75 A-P, +/−0.45 M-L, −2.3 DV;).

Control animals were injected with AAV2/Ef1a-DIO-mCherry (UNC GTC Vector Core, titer: 5.3×10^12 pfu/ml, Lot: AV4375H). Three weeks following injections animals were tested for aggression on three consecutive days. If levels of aggression had stabilized, animals were injected with either 1mg/kg CNO or the equivalent amount of saline i,p., then after 30 minutes, were presented with a male for 5 minutes followed by a 2 minute break, then a female for 5 minutes followed by a two minute break. Animals were then given a yogurt chip and allowed to approach and consume the food. Behaviors were compared between the saline and CNO injection days for both groups of animals.

Extracellular recording of freely moving mice

Methods for physiological recording in freely moving animals were described previously(Lin et al. 2011; Falkner et al. 2014). Custom-built 16-channel (or 14 channel with EMG) tungsten electrode bundles or groups of tetrodes were attached to a moveable microdrive and implanted over the lPAG. After allowing 2 weeks for recovery, we connected the implanted electrode to a 16-channel headstage. Signals were streamed into a commercial acquisition system through a torqueless, feedback-controlled commutator (Tucker Davis Technology) and band-pass filtered between 100 and 5,000 Hz. Digital infrared videos of animal behavior from both side- and top-view cameras were simultaneously recorded at 640×480 pixel resolution at 25 frames per second (Streampix, Norpix). Video frame acquisition was triggered by a TTL pulse from the acquisition system to achieve synchronization between the video and the electrophysiological recording. Spikes were sorted manually using commercial software (OfflineSorter, Plexon) based on principal component analysis. Unit isolation was verified using autocorrelation histograms. To consider the recorded cell as a single unit, cells had to have a signal/noise ratio >2; spike shape had to be stable throughout the recording; and the percentage of spikes occurring with inter-spike intervals (ISIs) <3 ms (the typical refractory period for a neuron) in a continuous recording sequence had to be <0.1%. We checked for redundancies within days by examining the cross correlations of co-recorded neurons and checked for redundancies across days by comparing waveforms and temporal response profiles. After the first recording, the implanted electrode was slowly moved down in 40-μm increments. The placement of the electrode was examined histologically with the aid of DiI coated on the electrodes. Animals were excluded if electrodes were not confined to the PAG. Recordings of the VMHvl (Fig 3FJ) were performed previously using identical methods(Falkner et al. 2014; Lin et al. 2011) and reanalyzed here for direct comparison to PAG neurons.

Electrophysiology Analysis

Spikes in single neurons were convolved with a 25 ms Gaussian for presentation (Fig 3D). Responsivity index (RI) for each behavior (Fig 3ij) was computed as

RI=ActivityBehaviourActivityBaselineNonsocialActivityBehaviour+ActivityBaselineNonsocial

Where Activitybehavior is defined as the mean activity across all episodes of a particular behavior (e.g. attack) and ActivityBaselineNonsocial is defined as the mean activity across all episodes designated as non-social within the given social interaction. Within-neuron significance was determined using a paired t-test for each neuron (behavior vs. nonsocial) compared to a Bonferroni-corrected threshold for each tested population.

EMG implantation and recording

To perform simultaneous recordings of PAG neurons and jaw muscle activity, we implanted animals with chronic EMG electrodes in the right masseter superficial muscles of the jaw. Electrodes were constructed using a pair of 0.001 inch flexible multi-strand stainless steel wires (A-M Systems, No. 793200) with the insulation removed from a 0.5-mm segment of each wire such that pairs of electrodes recorded signals from separate but nearby areas of the same muscle. Electrode wires were threaded through the muscle during a surgical procedure and anchored with a knot on the outside of the muscle. EMG wires were then threaded under the skin to the base of the skull where they were attached to ground electrodes. EMG wire output was relayed through a preamplifier and commutator to the digitizer with a sampling rate of 3,000 Hz (Tucker Davis Technology). Signals were processed by taking the difference from the pair of electrodes, and this differential signal was low pass filtered at 300 Hz.

EMG analysis

Mutual Information

Mutual information was computed between simultaneously recorded PAG activity and jaw EMG. EMG signals were rectified, low pass filtered at 20Hz, and downsampled to 1kHz. Spike trains were converted into a continuous instantaneous firing rate (IFR) with the same number of points as the downsampled EMG signals. For each pair of recorded PAG instantaneous firing rate and EMG signal, the continuous signals were discretized and MI was computed according to the definition(Shannon and Weaver 1964; Schilling, n.d.; Timme and Lapish 2018):

MI(X;Y)=x,yPXY(x,y)logPXY(x,y)PX(x)PY(y)=EPXYlogPXYPXPY

Where x and y represent the instantaneous firing rate and EMG signal respectively.

For each signal pair, the MI was compared relative to the mean of ten iterations of a circularly permuted time shuffled control.

Spike Triggered EMG

STEMGs were computed on rectified EMG signals by averaging the EMG signal in an 800ms window around each PAG spike recorded during interactions with a male or with a female. STEMGs were computed separately for each 10s increment during male and female interactions, smoothed with a 5ms moving average, then normalized by the number of spikes. To correct for drift due to volleys of successive spikes, STEMGs were baseline corrected by subtracting a baseline 100ms boxcar filtered version of the STEMG. To determine whether STEMGs contained significant peaks, we set strict criteria: STEMGs had to have a minimum of 5 consecutive points that crossed above the 98% confidence interval within 60ms of the 0 (the spike onset).

Fiber photometry recordings

A rig for performing simultaneous fiber photometry recordings from 2 locations was constructed following basic specifications previously described with a few modifications. For simultaneous VMH and VMHvl-PAG recordings, we injected vGlut2-ires-cre males with 100–160nl of AAV1.CAG.Flex.GCaMP6f.WPRE.SV40 (Lot CS0956, CS0845, CS0224WL, Upenn, final titer:: 9.3 × 10^12 PFU/ml) ipsilaterally into the VMHvl, and 240nl of HSV-Ef1a-LS1L-GCaMP6f (MIT vector core,Lot RN506, final titer: 1.0 × 10^9 PFU/ml) contralaterally in the PAG. For simultaneous recordings of the VMHvl and lPAG, we injected 80–120nl of AAV2/1 CAG::Flex-GCaMP6f-WPRE-SV40 ipsilaterally into the VMHvl, and 160–240nl of AAV2/1 CAG::Flex-GCaMP6f-WPRE-SV40 in the lPAG. VIruses were injected using the following coordinates: VMHvl (−1.82A/P, 0.72M/L, −5.8D/V), lPAG(−3.64A/P, 0.5M/L-2.4D/V).

A 400-μm optic fiber (Thorlabs, BFH48–400) housed in a metal ferrule (Thorlabs, SFLC440–10) was implanted 0.4 mm above each injection site, except for the HSV-Ef1a-LS1L-GCaMP6f injection, where the fiber was placed over the VMHvl. After three weeks of viral incubation and before recording, a matching optic fiber was connected to the each implanted fiber using a ferrule sleeve. A 400-Hz sinusoidal blue LED light (30–50 W) (LED light: M470F1; LED driver: LEDD1B; both from Thorlabs) was bandpass filtered (passing band: 472 ± 15 nm, Semrock, FF02–472/30–25) and delivered to the brain to excite GCaMP6. The emission light then traveled through the same optic fiber, was bandpass filtered (passing band: 534 ± 25 nm, Semrock, FF01–535/50), detected by a femtowatt silicon photoreceiver (Newport, 2151) and recorded using a real-time processor (RZ5, TDT). The envelope of the 400-Hz signals that reflects the intensity of the GCaMP6 signals were extracted in real-time using a custom TDT program. Baseline adjusted fluorescence signals were regressed using a 30s spline approximation.

Histology and imaging

Animals were deeply anaesthetized using 0.5 ml of a ketamine-xylazine cocktail (10 mg/ml ketamine and 5 mg/ml xylazine) and transcardially perfused with phosphate buffered saline (PBS) followed by cold 4% paraformaldehyde in PBS. Brains were immersed overnight in a 20% sucrose solution, embedded with cutting medium (Tissue-Tek) and sectioned using a cryostat (Leica). Standard immunohistochemistry procedures were followed to stain 30-μm coronal brain sections for all mice. DAPI (1:20,000, Life Technologies, catalog number D21490, widely validated) was used to assess electrode track for physiology and fiber track for photometry. We acquired 2.5x or 5x, fluorescent images to determine cannula or electrode placements amd 20x fluorescent images to determine viral expression. We used 10× fluorescent images to count PRV+. Cell counting in synaptophysin-injected animals was done manually using ImageJ on 30-μm sections separated by 60 μm that had observed GFP label in the lPAG.

Time-varying linear regression model

We modeled the population response of the PAG during male and female interactions for each animal by fitting the PAG response during each interaction including 10s prior to the introduction and 10s following the removal of the animals with a series of autoregressive models using time-varying lengths of simultaneously recorded VMHvl signal as the variable regressors using the form:

PAG(t)=β0+β1VMH(t)+β2VMH(t1)+β3VMH(t2)+βnVMH(tn)

where t represents the current time bin and the other regressors represent variable amounts of elapsed time. PAG and VMHvl signals were binned in either 50ms, 100ms, 250ms, 500ms, and 1s bins and models were fit independently for each of these bin sizes. For each VMHvl and PAG signal, data was halved and the first half was used to fit and the second half was used to cross validate, using least squares. Model order was set to maximum value of 10s of elapsed time, set independently for each bin size. For each family of models fit to the data, the best order model was determined using AIC (Akaike Information Criteria) on the fit to the cross validated data. A timeshifted null model was fit by circularly permuting the input data (VMHvl) signal and fitting the unpermuted PAG signal, and fit and model order were re-fit for the timeshifted data. Model order for the selected model was converted to elapsed time for each binsize. Fits were performed separately for male and female interactions. Additionally, we tested the alternative hypothesis that PAG signals influence VMHvl signal by fitting the “reverse” model of the form:

VMH(t)=β0+β1PAG(t)+β2PAG(t1)+β3PAG(t2)+βnPAG(tn)

A similar model family was fit for the reverse model, and the model order (and elapsed time) was determined for each model independently for each bin size.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analysis was performed using Matlab. Parametric tests, including Student’s t-test, paired t-test, and two-sample t test were used if distributions passed Kolmogorov–Smirnov tests for normality. For within-neuron tests of firing rate significance, a non-parametric Wilcoxon signed rank test was used since spike rates were often low and not normally distributed. Repeated tests of significance were corrected with a strict Bonferroni correction. For all statistical tests, significance was measured against an alpha value of 0.05 unless otherwise stated. All error bars show s.e.m. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications. Data collection and analysis were not performed blind to the conditions of the experiments. Statistical detail can be found in the figure legends for individual figures and in some cases, reported in the results.

Statistical analyses used in each figure are listed below.

Figure 1. (I,Q) Paired t-test

Figure 2. (C-J, L-M) Paired t-test

Figure 3. (B) Single neuron activity average using 25ms Gaussian smoothing (C-E) Top panels are the z-scored responses of individual neurons aligned to each behavior, and histograms represent the number of neurons whose peak response lies at that bin relative to behavior onset. Dotted line represents chance level (number of bins/number of neurons). (G-H) Wilcoxon signed rank test for comparison of population response. Percentages of neurons in pie charts computed using within-neuron significance test across one or both behaviors, with Bonferroni correction. (I) Kolmogorov-Smirnov test (J) Unpaired t-test across all bins using Bonferroni corrected threshold across all bins.

Figure 4. (C) Kolmogorov-Smirnov test between cumulative distributions. (F-G) Significant EMGsm+ neurons see methods.

Figure 5. (D,E) Fisher’s test, (D-G,) paired t-test, (D-E,G) Error bars show +/−SEM, comparisons of means with paired t-test.

Figure 6. (D) paired t-test (E) Error bars show +/−SEM (F) Wilcoxon signed rank test of male, female, and baseline activity. (H-I) Paired t-test between forward and null model at each bin for fit, Paired t-test between male and female interactions for model time.

DATA AND CODE AVAILABILITY

All data and code are available upon reasonable request.

Supplementary Material

1
2

Supplementary Figure 1. (C) Paired t-test.

Supplementary Figure 2. (D-E) Paired t-test.

Supplementary Figure 3. (C,H,M) Wilcoxon signed rank test on mean EMG 5s pre compared to 5s post behavior. (E,J,O) Wilcoxon signed rank test on mean activity 5s pre compared to 5s post behavior.

Supplementary Figure 4. (A-B,D-E) Activity shown is z-scored across the whole interaction trace. (C,F) Activity for each individual behavior is baseline subtracted using a 1s bin 5s prior to interaction. Individual behaviors compared using studenťs ttest, and comparison between behaviors using an unpaired t-test.

3

Supplementary Video 1, Related to Figure 2. DREADD Gi-mediated Inactivation of lPAG results in aggression-specific deficits. Example intermale aggression following saline injection, compared with behavior following CNO mediated inactivation. During inactivation, intermale aggression is decreased while investigatory behaviors and other social behaviors are unchanged.

Download video file (120.1MB, mp4)
4

Supplementary Video 2, Related to Figure 4. Example of single lPAG neuron recorded simultaneously with EMG in the superficial masseter muscle of the jaw with corresponding behavior. lPAG neuron (bottom trace) shows time-locked spiking with EMG peaks (top trace).

Download video file (643.8KB, mp4)

Key Resource Table

REAGENT or RESOURCE SOURCE IDENTIFIER
Bacterial and Virus Strains
AAV2-hSyn-DIO-hM4Di-mChery Gift from Bryan Roth Addgene viral prep # 44362-AAV2
AAV2.EF1a-DIO-hChR2(H134R)-EYFP Gift from Karl Deisseroth Addgene viral prep: 20298 Lot# v52304
AAV2-EF1a-double floxed-hChR2(H134R)-mCherry-WPRE-HGHpA Gift from Karl Deisseroth Addgene viral prep # 20297-AAV1) Lot# AV4375H
AAV2/Ef1a-DIO-mCherry Gift from Bryan Roth UNC GTC Vector Core Prep
rAAV2.syn.flex.ChrimsonR.tdT Gift from Ed Boyden Klapoetke et al. 2014 UNC Vector Core prep: Lot # AV6555B
Pseudorabies virus (PRV) 152 Gift from Lynn Enquist Smith et al., 2000 N/A
Pseudorabies virus (PRV) 614 Gift from Lynn Enquist Smith et al., 2000 N/A
AAV1.CAG.Flex.GCaMP6f.WPRE.SV40 Gift from Douglas Kim and GENIE project Chen et al., 2013 UPenn Vector Core: Lot # CS0956, CS0845, CS0224WL
HSV-Ef1a-LS1L-GCaMP6f MIT Vector Core Lot RN506
Chemicals, Peptides, and Recombinant Proteins
tetrodotoxin Alomone Labs T-550; CAS: 18660-81-6
4-aminopyridine Sigma-Aldrich 275875; CAS: 504-24-5
QX-314 chloride Tocris 2313; CAS: 5369-03-9
DAPI Life Technologies D21490
CNO Sigma Cat#C0832
Mounting Medium Vectashield Cat#H1000
Experimental Models: Organisms/Strains
Swiss Webster Wildtype mouse Taconic Tac: SW
BALB/c wildtype mouse Taconic Balb/cAnNTac
Mouse: Slc17a6tm2(cre)Lowl (vglut2-ires-cre) The Jackson Laboratory JAX: 028863
Mouse: Gt(ROSA)26Sortm6(CAG-ZsGreen1)Hze (Ai6 RCL-ZsGreen) The Jackson Laboratory JAX: 007906
Software and Algorithms
DeepLabCut Mathis et al., 2018 https://github.com/AlexEMG/DeepLabCut
BORIS Friard and Gamba, 2016 https://www.boris.unito.it/
CalTech behaviorAnnotator Lin et al., 2011 https://github.com/pdollar/toolbox
Offline Sorter Plexon https://plexon.com/products/offline-sorter/
StreamPix 5 NorPix https://www.norpix.com/products/streampix/streampix.php
SpinView FLIR https://www.flir.com/products/spinnaker-sdk/
ClampEx 11.0 Digidata Axon Instruments https://www.moleculardevices.com/products/axon-patch-clamp-system/acquisition-and-analysis-software/pclamp-software-suite#RelatedProducts
Matlab (Custom Code) Available upon reasonable request
Other
Nanojector
400 μm multimode optic fibers World Precision Instruments Cat# Nanoliter 2000
230 μm multimode optic fibers Thorlabs Cat# BFH48–400
LED light Thorlabs Cat# TS1450308
LED driver Thorlabs Cat# M470F1
Bandpass filter Thorlabs Cat# LEDD1B
Adjustable zooming lens Semrock Cat# FF02–472/30–25, FF01–535/505
Femtowatt silicon photoreceiver Thorlabs, Edmund Optics Cat# SM1NR01, #62–561
Real-time processor RP5 and RX8 Newport Cat# 2151
13 mm tungsten microwires TDT RP5/RX8
Multi stranded wire for EMG California Fine Wire Cat# 100211
omnetics nano-connector A-M Systems Cat# 793200
Feedback-controlled commutator Omnetics Cat# A79014–001
16-channel preamplifier TDT Cat# ACO32
TDT Cat# RA16PA

Highlights:

  • VMHvlvGlut2 neurons target lPAGvGlut2 neurons that project polysynaptically to the jaw.

  • Inactivating lPAGvGlut2 neurons results in aggressive action-specific deficits.

  • Single unit lPAG activity is action-specific and time locked to EMG-detected biting.

  • VMHvl-lPAG projection relays male-biased signals to generate action selectivity.

Acknowledgements

The authors wish to thank Lynn Enquist for the gift of the PRV 152 and 614, A. Chow, N. Cuvelier,, C. Heins and K. Liu for assistance with video annotation, T. Akay for EMG guidance, A. Person, K. Hashikawa, M. Halassa, and M. Long, for helpful discussions, A. Hu and S. Oline for technical support, and D. Blackman for graphics. This work was supported by the Brain & Behavior Research Foundation (ALF), the Alfred P. Sloan Foundation (ALF), NIMH (R00MH109674) (ALF), Irma T. Hirschl Trust (DL), and NIH R01MH101377, R21MH105774 and 1U19NS107616–01(DL).

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Declaration of interest

The authors declare no competing interests.

Hierarchical representations of aggression in a hypothalamic-midbrain circuit

Falkner et al.

References

  1. Adams DB 1968. “Cells Related to Fighting Behavior Recorded from Midbrain Central Gray Neuropil of Cat.” Science 159 (3817): 894–96. [DOI] [PubMed] [Google Scholar]
  2. Bandler Richard, and Keay Kevin A.. 1996. “Chapter 17 Columnar Organization in the Midbrain Periaqueductal Gray and the Integration of Emotional Expression.” In Progress in Brain Research, edited by Holstege G, Bandler R, and Saper CB, 107:285–300. Elsevier. [DOI] [PubMed] [Google Scholar]
  3. Bandler R, and Shipley MT. 1994. “Columnar Organization in the Midbrain Periaqueductal Gray: Modules for Emotional Expression?” Trends in Neurosciences 17 (9): 379–89. [DOI] [PubMed] [Google Scholar]
  4. Chen Tsai-Wen, Wardill Trevor J., Sun Yi, Pulver Stefan R., Renninger Sabine L., Baohan Amy, Schreiter Eric R., et al. 2013. “Ultrasensitive Fluorescent Proteins for Imaging Neuronal Activity.” Nature 499 (7458): 295–300. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Davidson Adam G., Ryan O’Dell Vanessa Chan, and Schieber Marc H.. 2007. “Comparing Effects in Spike-Triggered Averages of Rectified EMG across Different Behaviors.” Journal of Neuroscience Methods 163 (2): 283–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Evans Dominic A., Stempel A. Vanessa, Vale Ruben, Ruehle Sabine, Lefler Yaara, and Branco Tiago. 2018. “A Synaptic Threshold Mechanism for Computing Escape Decisions.” Nature 558 (7711): 590–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Falkner Annegret L., Dollar Piotr, Perona Pietro, Anderson David J., and Lin Dayu. 2014. “Decoding Ventromedial Hypothalamic Neural Activity during Male Mouse Aggression.” The Journal of Neuroscience: The Official Journal of the Society for Neuroscience 34 (17): 5971–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Falkner Annegret L., Grosenick Logan, Davidson Thomas J., Deisseroth Karl, and Lin Dayu. 2016. “Hypothalamic Control of Male Aggression-Seeking Behavior.” Nature Neuroscience 19 (4): 596–604. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Fang Yi-Ya, Yamaguchi Takashi, Song Soomin C., Tritsch Nicolas X., and Lin Dayu. 2018. “A Hypothalamic Midbrain Pathway Essential for Driving Maternal Behaviors.” Neuron 98 (1): 192–207.e10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Fay RA, and Norgren R. 1997. “Identification of Rat Brainstem Multisynaptic Connections to the Oral Motor Nuclei Using Pseudorabies Virus. I. Masticatory Muscle Motor Systems.” Brain Research. Brain Research Reviews 25 (3): 255–75. [DOI] [PubMed] [Google Scholar]
  11. Fernandez de Molina A, and Hunsperger RW. 1962. “Organization of the Subcortical System Governing Defence and Flight Reactions in the Cat.” The Journal of Physiology 160 (February): 200–213. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Franklin Tamara B., Silva Bianca A., Perova Zinaida, Marrone Livia, Masferrer Maria E., Zhan Yang, Kaplan Angie, et al. 2017. “Prefrontal Cortical Control of a Brainstem Social Behavior Circuit.” Nature Neuroscience 20 (2): 260–70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Friard Olivier, and Gamba Marco. 2016. “BORIS: A Free, Versatile Open-Source Event-Logging Software for Video/audio Coding and Live Observations.” Edited by Richard Fitzjohn. Methods in Ecology and Evolution / British Ecological Society 7 (11): 1325–30. [Google Scholar]
  14. Gao Zheng-Run, Chen Wen-Zhen, Liu Ming-Zhe, Chen Xiao-Jun, Wan Li, Zhang Xin-Yan, Yuan Lei, et al. 2018. “Tac1-Expressing Neurons in the Periaqueductal Gray Facilitate the Itch-Scratching Cycle via Descending Regulation.” Neuron, December 10.1016/j.neuron.2018.11.010. [DOI] [PubMed] [Google Scholar]
  15. Gregg Thomas R., and Siegel Allan. 2003. “Differential Effects of NK1 Receptors in the Midbrain Periaqueductal Gray upon Defensive Rage and Predatory Attack in the Cat.” Brain Research 994 (1): 55–66. [DOI] [PubMed] [Google Scholar]
  16. Gregg TR, and Siegel A. 2001. “Brain Structures and Neurotransmitters Regulating Aggression in Cats: Implications for Human Aggression.” Progress in Neuro-Psychopharmacology & Biological Psychiatry 25 (1): 91–140. [DOI] [PubMed] [Google Scholar]
  17. Hashikawa Koichi, Hashikawa Yoshiko, Tremblay Robin, Zhang Jiaxing, Feng James E., Sabol Alexander, Piper Walter T., Lee Hyosang, Rudy Bernardo, and Lin Dayu. 2017. “Esr1+ Cells in the Ventromedial Hypothalamus Control Female Aggression.” Nature Neuroscience 20 (11): 1580–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Holstege Gert. 2014. “The Periaqueductal Gray Controls Brainstem Emotional Motor Systems Including Respiration.” Progress in Brain Research 209: 379–405. [DOI] [PubMed] [Google Scholar]
  19. Jansen AS, Farkas E, Mac Sams J, and Loewy AD. 1998. “Local Connections between the Columns of the Periaqueductal Gray Matter: A Case for Intrinsic Neuromodulation.” Brain Research 784 (1–2): 329–36. [DOI] [PubMed] [Google Scholar]
  20. Kittelberger J. Matthew, Land Bruce R., and Bass Andrew H.. 2006. “Midbrain Periaqueductal Gray and Vocal Patterning in a Teleost Fish.” Journal of Neurophysiology 96 (1): 71–85. [DOI] [PubMed] [Google Scholar]
  21. Klapoetke Nathan C., Murata Yasunobu, Sung Soo Kim Stefan R. Pulver, Amanda Birdsey-Benson Yong Ku Cho, Morimoto Tania K., et al. 2014. “Independent Optical Excitation of Distinct Neural Populations.” Nature Methods 11 (3): 338–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Koutsikou Stella, Apps Richard, and Lumb Bridget M.. 2017. “Top down Control of Spinal Sensorimotor Circuits Essential for Survival.” The Journal of Physiology 595 (13): 4151–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Lee Hyosang, Kim Dong-Wook, Remedios Ryan, Anthony Todd E., Chang Angela, Madisen Linda, Zeng Hongkui, and Anderson David J.. 2014. “Scalable Control of Mounting and Attack by Esr1+ Neurons in the Ventromedial Hypothalamus.” Nature 509 (7502): 627–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lin Dayu, Boyle Maureen P., Dollar Piotr, Lee Hyosang, Lein ES, Perona Pietro, and Anderson David J.. 2011. “Functional Identification of an Aggression Locus in the Mouse Hypothalamus.” Nature 470 (7333): 221–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Li Yi, Zeng Jiawei, Zhang Juen, Yue Chenyu, Zhong Weixin, Liu Zhixiang, Feng Qiru, and Luo Minmin. 2018. “Hypothalamic Circuits for Predation and Evasion.” Neuron 97 (4): 911–24.e5. [DOI] [PubMed] [Google Scholar]
  26. Lo Liching, Yao Shenqin, Kim Dong-Wook, Cetin Ali, Harris Julie, Zeng Hongkui, Anderson David J., and Weissbourd Brandon. 2019. “Connectional Architecture of a Mouse Hypothalamic Circuit Node Controlling Social Behavior.” Proceedings of the National Academy of Sciences of the United States of America 116 (15): 7503–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Mathis Alexander, Mamidanna Pranav, Cury Kevin M., Abe Taiga, Murthy Venkatesh N., Mathis Mackenzie Weygandt, and Bethge Matthias. 2018. “DeepLabCut: Markerless Pose Estimation of User-Defined Body Parts with Deep Learning.” Nature Neuroscience 21 (9): 1281–89. [DOI] [PubMed] [Google Scholar]
  28. Miczek Klaus A., de Boer Sietse F., and Jozsef Haller. 2013. “Excessive Aggression as Model of Violence: A Critical Evaluation of Current Preclinical Methods.” Psychopharmacology 226 (3): 445–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Mos J, Kruk MR, Van der Poel AM, and Meelis W. 1982. “Aggressive Behavior Induced by Electrical Stimulation in the Midbrain Central Gray of Male Rats.” Aggressive Behavior 8 (3): 261–84. [Google Scholar]
  30. Remedios Ryan, Kennedy Ann, Zelikowsky Moriel, Grewe Benjamin F., Schnitzer Mark J., and Anderson David J.. 2017. “Social Behaviour Shapes Hypothalamic Neural Ensemble Representations of Conspecific Sex.” Nature 550 (7676): 388–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Schilling Donald L. n.d. “WILEY SERIES IN TELECOMMUNICATIONS.” http://www.cs-114.org/wp-content/uploads/2015/01/Elements_of_Information_Theory_Elements.pdf.
  32. Schreiner L, and Kling A. 1953. “Behavioral Changes Following Rhinencephalic Injury in Cat.” Journal of Neurophysiology 16 (6): 643–59. [DOI] [PubMed] [Google Scholar]
  33. Shannon Claude Elwood, and Weaver Warren. 1964. The Mathematical Theory of Communication. Urbana: University of Illinois Press. [Google Scholar]
  34. Shimogawa Yuji, Sakuma Yasuo, and Yamanouchi Korehito. 2015. “Efferent and Afferent Connections of the Ventromedial Hypothalamic Nucleus Determined by Neural Tracer Analysis: Implications for Lordosis Regulation in Female Rats.” Neuroscience Research 91 (February): 19–33. [DOI] [PubMed] [Google Scholar]
  35. Siegel A, Schubert KL, and Shaikh MB. 1997. “Neurotransmitters Regulating Defensive Rage Behavior in the Cat.” Neuroscience and Biobehavioral Reviews 21 (6): 733–42. [DOI] [PubMed] [Google Scholar]
  36. Silva Carlos, and Neil McNaughton. 2019. “Are Periaqueductal Gray and Dorsal Raphe the Foundation of Appetitive and Aversive Control? A Comprehensive Review.” Progress in Neurobiology 177 (June): 33–72. [DOI] [PubMed] [Google Scholar]
  37. Smith BN, Banfield BW, Smeraski CA, Wilcox CL, Dudek FE, Enquist LW, and Pickard GE. 2000. “Pseudorabies Virus Expressing Enhanced Green Fluorescent Protein: A Tool for in Vitro Electrophysiological Analysis of Transsynaptically Labeled Neurons in Identified Central Nervous System Circuits.” Proceedings of the National Academy of Sciences of the United States of America 97 (16): 9264–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Srivastava Kyle H., Holmes Caroline M., Vellema Michiel, Pack Andrea R., Elemans Coen P. H., Nemenman Ilya, and Sober Samuel J.. 2017. “Motor Control by Precisely Timed Spike Patterns.” Proceedings of the National Academy of Sciences of the United States of America 114 (5): 1171–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Stanek Edward 4th, Cheng Steven, Takatoh Jun, Han Bao-Xia, and Wang Fan. 2014. “Monosynaptic Premotor Circuit Tracing Reveals Neural Substrates for Oro-Motor Coordination.” eLife 3 (April): e02511. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Timme Nicholas M., and Lapish Christopher. 2018. “A Tutorial for Information Theory in Neuroscience.” eNeuro 5 (3). 10.1523/ENEURO.0052-18.2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Tschida Katherine, Michael Valerie, Takatoh Jun, Han Bao-Xia, Zhao Shengli, Sakurai Katsuyasu, Mooney Richard, and Wang Fan. 2019. “A Specialized Neural Circuit Gates Social Vocalizations in the Mouse.” Neuron 103 (3): 459–72.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Wang Li, Chen Irene Z., and Lin Dayu. 2015. “Collateral Pathways from the Ventromedial Hypothalamus Mediate Defensive Behaviors.” Neuron 85 (6): 1344–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Yang Cindy F., Chiang Michael C., Gray Daniel C., Prabhakaran Mahalakshmi, Alvarado Maricruz, Juntti Scott A., Unger Elizabeth K., Wells James A., and Shah Nirao M.. 2013. “Sexually Dimorphic Neurons in the Ventromedial Hypothalamus Govern Mating in Both Sexes and Aggression in Males.” Cell 153 (4): 896–909. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Zalcman Steven S., and Siegel Allan. 2006. “The Neurobiology of Aggression and Rage: Role of Cytokines.” Brain, Behavior, and Immunity 20 (6): 507–14. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
2

Supplementary Figure 1. (C) Paired t-test.

Supplementary Figure 2. (D-E) Paired t-test.

Supplementary Figure 3. (C,H,M) Wilcoxon signed rank test on mean EMG 5s pre compared to 5s post behavior. (E,J,O) Wilcoxon signed rank test on mean activity 5s pre compared to 5s post behavior.

Supplementary Figure 4. (A-B,D-E) Activity shown is z-scored across the whole interaction trace. (C,F) Activity for each individual behavior is baseline subtracted using a 1s bin 5s prior to interaction. Individual behaviors compared using studenťs ttest, and comparison between behaviors using an unpaired t-test.

3

Supplementary Video 1, Related to Figure 2. DREADD Gi-mediated Inactivation of lPAG results in aggression-specific deficits. Example intermale aggression following saline injection, compared with behavior following CNO mediated inactivation. During inactivation, intermale aggression is decreased while investigatory behaviors and other social behaviors are unchanged.

Download video file (120.1MB, mp4)
4

Supplementary Video 2, Related to Figure 4. Example of single lPAG neuron recorded simultaneously with EMG in the superficial masseter muscle of the jaw with corresponding behavior. lPAG neuron (bottom trace) shows time-locked spiking with EMG peaks (top trace).

Download video file (643.8KB, mp4)

Data Availability Statement

All data and code are available upon reasonable request.

RESOURCES