Capturing and Manipulating Activated Neuronal Ensembles with CANE Delineates A Hypothalamic Social Fear Circuit

The design of the CANE technology

To achieve both anatomical and temporal specificity in capturing behaviorally activated Fos+ neurons, we conceived a lock and key strategy requiring an engineered mouse and pseudotyped viruses (Figure 1). We developed a knock-in mouse (Fos-2A-dsTVA) in which activity-induced Fos expression triggers co-expression of a destabilized TVA (dsTVA) receptor (Figure 1). TVA is an avian specific receptor that is not present in the mouse genome (Barnard et al., 2006; Barnard and Young, 2003), and 2A is a self-processing peptide that enables Fos and dsTVA to be co-translated (de Felipe et al., 2006). To construct dsTVA, the c-terminus of the TVA receptor is fused with a “PEST” sequence that targets the protein for degradation (Li et al., 1998). See Figure S1 and Experimental Procedures for detailed description. This knock-in line is designed with the expectation that in the freely behaving Fos-2A-dsTVA mouse (hereafter referred to as Fos^TVA), activated neurons will transiently express dsTVA with a time course similar to that of Fos itself. TVA is an exquisitely specific receptor for the envelope protein of the avian sarcoma and leukosis virus (EnvA) (Barnard and Young, 2003). Therefore, EnvA-coated (pseudotyped) viruses can be introduced into the brain of Fos^TVA mice within a short time window that parallels the expression of dsTVA/Fos to specifically infect those recently activated neurons (Figure 1). More specifically, lentiviruses (LVs) and deficient monosynaptic rabies virus (RVs) can be pseudoptyped with the EnvA envelope protein (EnvA-LV, or EnvA-RV, respectively) to exploit this transient susceptibility of activated neurons to viral infection (Figure 1). Because LVs are non-toxic and enable stable transgene expression, they are ideal for introducing and persistently expressing desired transgenes in transiently Fos+ neurons (Figure 1), while the monosynaptic RVs (Callaway, 2008; Wickersham et al., 2007) can be used for transsynaptic tracing of presynaptic inputs onto Fos+ neurons. We name this technology CANE for capturing activated neuronal ensembles.

Rationale to use the hypothalamic neurons activated by the resident-intruder behavior as a platform for testing the CANE

CANE is designed to use engineered viruses to infect activated neurons in the Fos^TVA mice. We need to demonstrate that this method is specific and efficient. To do this, we will need to use CANE to capture neurons activated by the first behavior and drive expression of a transgene such as a fluorescent protein in these neurons. After allowing more than 10 days to 3 weeks for sufficient expression of the transgene, mice will be asked to perform the behavior a second time to induce Fos expression (Figure 2A, scheme). Showing that captured neurons following one behavioral epoch overlap highly with Fos+ neurons following a second behavior would provide a stringent validation of this technology’s efficiency and specificity.

However, rarely had any previous studies examined whether the same populations of neurons that express Fos following a brief, one-time behavior were re-activated and re-express Fos weeks later by the “same” behavior. Furthermore, no behaviors performed at different times can be identical. Thus, to minimize the inevitable behavioral variability, we tested CANE using innate behaviors that are thought to involve genetically hard-wired hypothalamic neurons and are therefore more stereotyped (Dulac and Kimchi, 2007; Manoli et al., 2013). In the rodent male-male resident-intruder assay, an aggressive male resident attacks an intruder evoking a highly stereotyped behavior in the intruder which includes retreating to a corner and avoiding the resident (see Movie S1).

Previous studies showed the presence of Fos+ neurons in the ventromedial hypothalamus ventrolateral division, VMHvl, in both the intruder and the resident after a brief encounter (Kollack-Walker et al., 1997; Lin et al., 2011; Silva et al., 2013). We also observed that brief (10 minutes or less) male-male encounters, in which the intruder was attacked 10–15 times by the resident, consistently induced Fos expression in similar numbers of neurons in all intruders examined (262 ± 20 Fos+ neurons; n = 9 intruders). We called these neurons social fear neurons (SFNs; Figure S2). Similarly, the encounters consistently induced Fos expression in 138 ± 14 VMHvl neurons in resident mice (n = 9 residents, aggression neurons) (Figure S2). The consistent numbers of Fos+ SFNs and aggression neurons induced by a highly stereotyped behavior suggest that these neurons are likely re-activated by repeated social fear or aggression behavior, therefore ideal for testing the CANE technology.

dsTVA tracks Fos expression pattern in Fos^TVA mice

For CANE to work, the expression of dsTVA needs to mimic that of Fos. We used male Fos^TVA mice as intruders, where they were attacked 10–15 times over 10 minutes or less. The mice were returned to their home cage and then killed either immediately (0 hour) or 1, 2, or 6 hours later (Figures 2B and 2C). Immediately following the social fear experience (0 hour), minimal levels of Fos+ or TVA+ signals were observed in VMH of the intruder, indicating that background expression of TVA is very low (Figure 2C). At 1-hour after the encounter, moderate Fos and TVA expression was detected in a column of cells in VMHvl, as well as in a few cells scattered in VMHdm-c (Figure 2C). Both Fos and TVA expression reached peak levels 2 hrs after the encounter (68 ± 2 % Fos+ neurons express TVA, n = 4). The fewer number of TVA+ cells compared to Fos+ neurons was likely due to the detection limitation of TVA-antibody. By 6 hours, both Fos and TVA expression levels were diminished. Most importantly, TVA staining always co-localized with strong Fos expressing neurons, with the anti-TVA signal in the cell membrane surrounding the Fos+ nucleus (Figure 2C, inset). Our analyses establish that TVA expression mimics the spatiotemporal pattern of Fos expression in Fos^TVA mice, which is a critical criterion necessary for selectively tagging neurons that transiently express Fos.

Developing a pseudotyped lentivirus (CANE-LV-Cre) that selectively infects Fos+ neurons

To efficiently and selectively infect neurons that express Fos/TVA, we generated a lenti-hSyn-Cre (synapsin promoter driving Cre recombinase expression) construct and packaged pseudotyped lentiviruses (LVs) with either the wild type EnvA coat protein, or with several mutated forms of the EnvA (EnvA^M4, EnvA^M5, and EnvA^M21) protein that have different degrees of reduced binding affinities to TVA (Rong et al., 1997) (see Experimental Procedures). We generated multiple mutated-EnvA proteins for pseudotyping LVs because the extremely high affinity of the wild type EnvA protein for the TVA receptor may result in background infection of neurons even if they express Fos/TVA at very low levels; whereas mutated-EnvA with reduced affinity will result in the infection of only those neurons with high-level TVA/Fos.

To test these different pseudotyped viruses, we crossed Fos^TVA mice with Rosa-flox-STOP-tdTomato (Ai14) reporter mice (Madisen et al., 2010). Fos^TVA; Ai14 mice were subjected to a brief intruder social fear experience (experimental group) or placed in a novel cage (control group), then injected 1–2 hours later with 0.5 μl of pseudotyped LV-Cre into VMHvl. 10–14 days later, we examined Cre-induced tomato expression. Notably, only the intruder mice injected with the mutated EnvA^M21 coated LVs exhibited highly selective infection of VMHvl neurons (putative SFNs), with no background infection in the control mice (Figure S3) (for more details, see later). In contrast, the wild type EnvA, as well as EnvA^M4 and EnvA^M5 coated LVs all exhibited varying degrees of background infection even in control mice likely due to high affinity binding of the viral coat to neurons with low basal levels of Fos/TVA expression (Figure S3). Therefore, all subsequent experiments were performed with EnvA^M21 coated LVs. We name the EnvA^M21 coated LV as CANE-LV.

CANE-LV-Cre selectively captures hypothalamic social fear neurons (SFNs)

We next determined whether CANE captured social fear neurons (SFNs) in VHMvl overlapped with Fos+ SFNs induced by a second intruder experience. We injected CANE-LV-Cre after the first intruder experience in Fos^TVA; Ai14 mice, and 10–14 days later, the intruder mice were attacked by the same aggressive resident for inducing Fos expression (Figure 3A). Despite the fact that no two resident-intruder encounters could be perfectly identical, we found extensive overlap between CANE+ (red) and Fos+ neurons (green) (Figure 3C). Again, there was minimal background infection in control mice (Figure 3B). With injection of 0.5 μl of CANE-LV-Cre (5 × 10⁶ ifu/ml), we captured 131 ± 8 neurons (CANE+, about 50% of the number of Fos+ neurons), and 63 ± 3% of captured neurons were Fos+ (n = 9, Figure 3G). Thus, the second encounter reactivated many (> 60%) of the same cells excited by the first social fear behavior. Concentrated higher titer CANE-LVs (1 × 10⁸ ifu/ml) could capture more neurons (200~240 neurons, see later result in Figure 7B). Additionally, we tested the capturing efficiency at different time points (Figure S4). The best capturing result was obtained with injection at 1.5–2.0h after the behavior, while injections at 30min (earlier) or 6.5h (later) had significantly lower capturing efficiency. These results further confirm that timed injection of CANE-viruses can be used to selectively, permanently, and relatively efficiently capture neurons that transiently express Fos after a brief behavior.

CANE capturing reveals that social fear and social aggression activate largely distinct ensembles in VMHvl

The CANE method allowed us to resolve whether SFNs are similar or distinct from the aggression-activated VMHvl neurons in the same animal. To this end, we first tested whether CANE-LV can also be used to efficiently and selectively capture aggression-activated neurons (AGNs) in VMHvl. As mentioned above, in our animal housing conditions, the total number of Fos+ AGNs (average 138 ± 14 per animal in VMHvl, n = 9) was considerably fewer than that of Fos+ SFNs (262 ± 20, n=9). We co-injected CANE-LV-Cre together with a Cre-dependent AAV-Flex-GFP into the VMHvl of Fos^TVA aggressive resident mice after they attacked the intruder mice. Then, 10–14 days later, the resident mice were exposed to another intruder in their home cage, which elicited another round of attacks by the resident. On average, 55 ± 4% CANE+ AGNs in the resident’s VMHvl re-expressed Fos, consistent with the idea that aggression repeatedly activates a similar population of VMHvl neurons (n=6, Figures 3E and 3G).

In a separate group of mice, we modified this two-bout paradigm so that male mice first behaved as aggressive residents, but then were introduced as intruders to experience social fear (1^st encounter = aggression; 2^nd encounter = social fear). Here, co-injection of CANE-LV-Cre and AAV-Flex-GFP was used to label AGNs following the first encounter, and anti-Fos staining was used to visualize SFNs following the second encounter. Only 19 ± 2% of CANE+ AGNs co-expressed Fos (SFNs) (n=5, Figures 3F and 3G). We also performed the converse paradigm: 1^st social fear-2^nd aggression. Again, the CANE+ SFNs were largely non-overlapping with Fos+ AGNs (10 ± 1% CANE+ neurons were Fos+, n = 9) (Figures 3D and 3G). In summary, we show repeated bouts of social aggression or repeated bouts of social fear weeks apart reactivate many of the same VMHvl neurons, which indicates that CANE is both efficient and selective. However, we see limited overlap after either a bout of social aggression followed by social fear, or a bout of social fear followed by social aggression, indicating that these opposite behaviors activate largely distinct populations of neurons in the VMHvl.

We also examined a couple molecular markers known to be expressed in VMH neurons, namely estrogen receptor 1 (Esr1, a presumed marker for AGNs in VMHvl), and steroidogenic factor 1 (SF1, a marker for predator fear neurons in the dorso-medial part of VMH, VMHdm-c) (Kunwar et al., 2015; Lee et al., 2014; Silva et al., 2013; Wang et al., 2015). Using two-color immunofluorescence or two-color in situ hybridization, we found that 65 ± 2% of Fos+ AGNs co-expressed Esr1 (n=4); whereas only 39 ± 4% Fos+ SFNs co-expressed Esr1, (n=4, Figures S5A and S5B). In both cases, about 43% of Esr1+ neurons were Fos+, further suggesting that Esr1 is not an ideal marker to distinguish these two populations of VMHvl neurons (Figure S5E). Only 4 ± 1% of Fos+ SFNs co-expressed SF1, confirming that SFNs are largely distinct from predator fear neurons (n=4, Figure S5C). Finally, all Fos+ SFNs express vGlut2 and therefore are likely to be excitatory neurons (n=4, Figure S5D).

Optogenetic activation of SFNs using CANE induces social fear behaviors in resident mice

All previous electrical or optogenetic stimulation experiments in various mammalian species showed that activation of VMHvl elicits aggressive behavior (Fuchs et al., 1985; Kruk et al., 1983; Lammers et al., 1988; Lee et al., 2014; Lin et al., 2011). Here we found that less than 3 percent of VMHvl neurons (9442 ± 277 neurons total; n = 4) are Fos+ SFNs. We therefore wondered whether activating this small number of CANE captured SFNs would be sufficient to evoke fear-like behaviors in non-fearful situations. To test this idea, we first co-injected CANE-LV-Cre with Cre-dependent AAV-Flex-ChR2-EYFP (Zhang et al., 2010) unilaterally into the VMHvl of Fos^TVA mice after a brief social fear experience (Figures 4A), resulting in the expression of ChR2 selectively in SFNs in VMHvl (Figure 4E). Photo stimulation-dependent activation of ChR2-expressing SFNs was confirmed using whole-cell patch-clamp recording in acute hypothalamic slices in vitro (Figures 4B–4D, 100% spike fidelity is achieved up to 10 Hz, 2 ms or 20 ms light pulses). Mice unilaterally co-injected with CANE-LV-Cre and AAV-Flex-GFP were used as controls for in vivo optogenetic experiments. All mice were tested in their home cages under infrared light and were exposed to either an intruder male or an intruder female on different days, each for ~20 min. Animals were photo-stimulated 5–6 times (10 Hz, 20 ms pulse-width, 2–3 mw/mm², 30 s duration for each stimulation).

In male-male interactions, the resident male normally investigated and sometimes attacked the intruder male, both of which are considered as forms of “active social contact” here. Unilateral photo-activation of ChR2-expressing SFNs caused the resident male to cease ongoing interactions or attacks, and retreat to a corner or against a wall of the cage where they remained largely stationary (i.e., stationary cornering behavior) (Movies S2). By contrast, illuminating GFP-expressing SFNs in control mice had negligible effects on the resident’s ongoing investigative or aggressive behaviors (Movie S3, Figure 4N). Post hoc histology revealed that the average duration of stationary cornering behavior positively correlated with the number of ChR2-captured SFNs (R² = 0.83) (Figure 4H), whereas the time spent in active social contact inversely correlated with the number of ChR2 cells (R² = 0.62) (Figure 4J). The population size effect on fear-like behavior in male-male interactions could also be modeled by a sigmoid fit (R² = 0.88) (Figure 4L), suggesting a possible threshold effect. We performed cluster analysis to separate the experimental mice into two groups: a Lo-ChR2 group (n ≤ 137 of ChR2+ neurons, n = 6 mice) and a Hi-ChR2 group (n > 137 of ChR2+ neurons, n = 6 mice). The light-induced stationary cornering behavior and reduced social interactions were displayed by the Hi-ChR2 but not the Lo-ChR2 group (Figures 4M and 4N, 14.47 ± 1.32, 2.38 ± 0.75, or 1.29 ± 0.43 seconds stationary time; and 5.09 ± 0.82, 8.89 ± 1.22, or 9.86 ± 1.57 seconds interaction time for Hi-ChR2, Lo-ChR2, and GFP groups, respectively). These results indicate that activating ~140 or more VMHvl SFNs (on one side of the brain) is sufficient to elicit fear-like behaviors in resident male mice presented with a male intruder.

In male-female interactions, the resident male normally spent most of its time chasing and investigating the female. Photo-activation of SFNs also resulted in a population size dependent stationary cornering behavior in males presented with females (Figure 4I and Movie S4, linear regression analysis, R² = 0.84). Again, the Hi-ChR2 group showed a longer duration of cornering behavior during photo-stimulation than either the Lo-ChR2 or GFP control groups (9.42 ± 1.27, 1.91 ± 0.80, or 1.84 ± 0.67 seconds stationary time for Hi-ChR2, Lo-ChR2, and GFP groups, respectively, Figure 4M). However, photostimulation only moderately reduced the active contact with females (Figures 4K and 4N). Thus, unilateral activation of SFNs in male mice elicited milder fear-like behaviors toward females than toward males.

We also measured the latency from light stimulation to the onset of stationary cornering behavior in both behavior paradigms, and the Hi-ChR2 group showed significantly shorter latency to exhibit cornering behavior in male-male encounters compared with Lo-ChR2 and GFP-expressing controls (11.03 ± 1.32 s, 25.74 ± 1.15 s, or 26.69 ± 1.39 s latency for Hi-ChR2, Lo-ChR2, and GFP groups, respectively, Figure 4O) and a moderate latency to cornering when interacting with females (15.20 ± 1.51 s, 24.12 ± 1.52 s, or 25.63 ± 1.71 s latency for Hi-ChR2, Lo-ChR2, and GFP groups, respectively, Figure 4O). Finally, optogenetic activation of SFNs never elicited any aggressive behavior in any situations, further indicating that SFNs are distinct from neurons eliciting aggression in the VMHvl.

Optogenetic activation of SFNs using CANE induces social avoidance in a three-chamber social preference test

To further demonstrate that CANE-mediated activation of SFNs have a causal effect on social interaction behaviors, we employed the three-chamber social interaction test, which is a widely used assay to examine social preference or avoidance (Figures 5A and 5B). We unilaterally injected expressed ChR2-YFP or GFP (controls) in SFNs using CANE. Behavioral results from mice with more than 140 CANE-captured SFNs, determined by post hoc analyses, were quantified. In the three-chamber test, a non-aggressive male or a female mouse was contained in an inverted wire cup in the “social” chamber (Figure 5A). The wire containment reduces the potential for territorial disputes in the male-male interactions. Representative spatial heat maps showed the locations of the free-moving experimental mice in these different conditions (Figures 5C and S6A). When the contained mouse was a male, optical stimulation of SFNs expressing ChR2, but not GFP, caused the animal to avoid the social chamber and stay in the non-social chamber (Figure 5C). Quantitative analysis showed a drastic reduction in the time spent in the social zone or in social contact with the male (Figures 5D and 5E). Milder reductions in social interactions were observed when a female mouse was placed in the social chamber (Figures S6B and S6C). Thus, the three-chamber tests independently confirmed that activating SFNs in the VMHvl induces fear-like social avoidance towards a non-aggressive male.

CANE-mediated bilateral silencing of SFNs reduces social fear

We next asked whether silencing SFNs using CANE would reduce social fear. We expressed the neuronal silencer, tetanus toxin light chain (TeLC), in SFNs by co-injecting CANE-LV-Cre and AAV-Flex-TeLC-GFP (or AAV-Flex-GFP as controls) bilaterally into the VMHvl after a social fear experience (Figures 6A and 6B, a total of 397 ± 28 neurons from both sides were infected, n=8 mice). It was shown previously that immediately after being defeated by aggressive residents, the intruder mice often exhibit social withdrawal behaviors (Silva et al., 2013), a sign of fear for conspecifics. We therefore subjected control mice or mice with TeLC expressed in SFNs (SFNs-silenced group) to a very brief social fear experience during which they were attacked 3–5 times, after which they were returned to their home cage (Figure 6C). Three minutes later, a non-aggressive male mouse was introduced to the home cage (Figure 6C). Control mice avoided the intruder nearly every time the intruder approached, and often maintained a constant distance from the intruder (Figures 6D–6F). By contrast, the SFNs-silenced mice displayed normal resident behaviors, including close investigations of the intruder through sniffing the intruder’s snout and anogenital regions (Figures 6G–6I). Two SFNs-silenced mice even attacked the intruder. Thus silencing SFNs reduced the expression of social withdrawal in mice after they had recently been attacked by an aggressive male, or reduced the generalization of social fear toward a non-aggressive conspecific male.

Mapping the axonal projection targets of CANE-labeled VMHvl social fear neurons

The activation and silencing results described above indicated that CANE can be used to establish the causal function of captured neurons. To understand how such a small number of SFNs (~260 on one side of VMHvl) is both necessary and sufficient to regulate social fear, we decided to elucidate their functional connectivity. We first used CANE to identify the output targets of SFNs. High titer CANE-LV-Cre (1×10⁸ ifu/ml) and AAV-Flex-GFP were co-injected into the VMHvl of Fos^TVA mice to label SFNs with GFP (Figure 7A). 3 weeks later, the mice were subjected to another social fear experience to re-induce Fos expression (61 ± 4% of CANE-labeled VMHvl neurons were also Fos+, n = 3, again confirming specificity of labeling) (Figures 7B–7D). These neurons projected axons widely to regions of the limbic system (Figures 7E–7R). To measure SFN projection patterns in each region, we quantified both the “total axon projections” (averaged and normalized total pixel numbers of GFP-labeled axons, n = 3) and the “density of innervation” (total pixel numbers of GFP-labeled axons divided by the area of the nuclei and averaged across animals, n = 3) using a method similar to that described previously (Oh et al., 2014). This analysis identified 18 brain regions that are targets of VMHvl SFNs, including bed nucleus of the stria terminalis (BNST), medial and lateral preoptic area/nucleus (MPA and LPO), anterior hypothalamic nucleus (AHA) and paraventricular hypothalamic nucleus (PVH), limbic cortex regions (PrL, IL and Cg1), nucleus accumbens (Nac), lateral septum (LS), paraventricular thalamic nucleus (PVT), central and medial amygdala (CeA and MeA), the dorsal and ventral hypothalamic premammillary nucleus (PMd and PMv), dorsomedial and lateral periaqueductal grey (PAG_DM and PAG_L) and in the most lateral region of the midbrain reticular formation (MRN_L) (Figures 7E–7R). The majority of projections were ipsilateral with small numbers of axons innervating the contralateral side (Figures 7P–7R). A schematic summary of the SFN projection pattern is shown in Figure 8T (green lines with arrows). Note that all of the SFN target also contain Fos+ neurons induced by the same intruder experience (Figure 7, red signals), indicating that SFNs project to regions that are also activated by social fear.

Developing CANE-rabies virus for transsynaptic tracing of presynaptic inputs for VMHvl social fear neurons

We next wanted to establish CANE-based method to map presynaptic inputs onto captured neurons in combination with the monosynaptic rabies virus (RV) strategy (Wickersham et al., 2007). To achieve this, we first needed to produce a pseudotyped deficient RV expressing mCherry in place of endogenous glycoprotein (rabies-G) with the mutated low-affinity EnvA^M21 envelope (CANE-RV-mCherry), and to demonstrate that it can specifically infect Fos+ neurons. To test viral specificity, we injected CANE-RV-mCherry into the VMHvl of Fos^TVA mice after a social fear experience and 5 days later the mice were again used as intruders to induce Fos (Figure 8A). 61 ± 6% of CANE-RV-mCherry captured neurons were also Fos+ (Figures 8B–8E). Furthermore, no long-range infection outside of VMHvl was detected, indicating that dsTVA was not sufficiently expressed in axon terminals from neurons projecting into VMHvl to allow for CANE-RV infection (Figure 8B and data not shown). These results showed that CANE-RV selectively infects Fos+ neurons only at cell bodies. As a control experiment, we injected CANE-RV-mCherry into the VMHvl of Fos^TVA mice housed in their home cage. Only minimal background infection was observed without social behavior (Figure S7A).

Having demonstrated the specificity of CANE-RV-mCherry to infect Fos+ neurons, we next performed experiments to trace sources of synaptic inputs onto SFNs. We first co-injected CANE-LV-Cre and the helper virus AAV-SynP-DIO-TVA-EGFP-RG (Kohara et al., 2014) into the VMHvl after a social fear experience, which enabled Cre-dependent expression of the TVA receptor, the rabies-G and GFP in SFNs (Figures 8F and 8G). Three weeks later, CANE-RV-mCherry was injected into the VMHvl. GFP+/mCherry+ double positive neurons represented the “starter” SFNs that expressed the rabies-G and could complement the deficient CANE-RV to enable its transsynaptic spreading (Figures 8F–8I). Such “starter” neurons were observed only in VMHvl (Figure 8I).

Neurons that were only mCherry+ outside of the VMHvl, representing putative monosynaptic afferents to SFNs, were observed in multiple structures within the ventral forebrain including nucleus accumbens (Nac), lateral septum (LS), medial and lateral preoptic area (MPA/MPN, LPO), the bed nucleus of the stria terminalis (BNST), as well as in several hypothalamic nuclei including anterior- (AHA), paraventricular- (PVH), paraventricular thalamic nucleus (PVT), and the dorsal and ventral premammillary- (PMd, PMv) nuclei (Figure 8J–Q). The labeling results (normalized by dividing numbers of labeled presynaptic neurons by the number of starter neurons and averaged across animals, n= 3) (Figure 8S) are schematically summarized in Figure 8T (red lines; regions with > 0.25 average labeled neurons per starter neuron are shown). Interestingly, all regions that provide input onto SFNs also receive input from SFNs, as revealed in the projection studies described above (Figure 8T). These findings indicate that SFNs in VMHvl are reciprocally connected with multiple brain regions (orange-colored regions in Figure 8T) that together drive social fear behaviors (see Discussion). Notably, no transsynaptically labeled neurons were observed in medial amygdala (MeA) (Figure 8R), which is known to contain neurons that are selectively activated by male or female mice, or by predators (Bergan et al., 2014; Choi et al., 2005; Gross and Canteras, 2012; Perez-Gomez et al., 2015; Petrovich et al., 2001; Unger et al., 2015) and that project axons to VMHvl. This result suggests that inputs from MeA to VMHvl terminate on neurons other than SFNs, although we cannot rule out the possibility that the lack of MeA labeling was caused by differences in viral neurotropism.

As a further control experiment, we injected CANE-LV-Cre and AAV-SynP-DIO-TVA-EGFP-RG into VMHvl of Fos^TVA mice housed in their home cage without social behavior. Three weeks later, CANE-RV-mCherry was injected into the same region (with only home cage experience). Again, only minimal background infection of both viruses was observed in the VMHvl in these controls (Figure S7B).

CANE is a generally applicable tool to label activated/Fos+ neurons

To further demonstrate the specificity and general applicability of this technology, we used CANE to capture activated neurons in two other behavioral paradigms. First, we used CANE to label barrel cortical neurons activated by whisker dependent tactile exploration. Fos^TVA; Ai14 mice with only a single whisker (C2 whisker) remained, or with all-but-C2 whiskers intact were introduced in a novel environment in the dark. 2 hours after the exploration, CANE-LV-Cre was injected into the barrel cortex in these mice to infect activated neurons (Figure 9A). 10 days later, mice were subjected to another 2-hour environmental exploration to re-induce Fos expression. As expected, in the single C2 whisker only mice, the majority of Fos+ neurons were observed in layer 4 of the C2 barrel column, as were CANE+ neurons (Figure 9B). Importantly, 80 ± 5% of CANE+ neurons re-expressed Fos (n=3). In the all-but-C2 mice, C2 column had the least number of CANE+ (or Fos+) neurons with the majority of CANE+ neurons observed in layer 4 of neighboring barrel columns surrounding C2 (Figure 9B). Note that CANE-LV is a lentiviral vector that has a diffusion range of ~3–4 barrel columns. These results strongly suggested that sensory-experience activated cortical neurons can be selectively captured by CANE.

Next, we performed experiments that allowed us to directly compare CANE-labeling with Fos-tTA based labeling. Tayler et al. (Tayler et al., 2013) used Fos-tTA system to induce the stable H2B-GFP expression in neurons activated by contextual fear conditioning in basolateral amygdala, dentate gyrus and other regions. 2 weeks later, they re-exposed labeled mice to the same context again to examine what percentage of H2B-GFP+ neurons were reactivated (using anti-Fos staining) by the remote memory. They discovered that Fos-tTA/H2B-GFP labeled ~2.5% of basolateral amygdala neurons, and ~4% dentate gyrus neurons upon fear conditioning training. Importantly, only ~10% of H2B-GFP labeled amygdala neurons re-expressed Fos+, and only ~2% of labeled dentate neurons were Fos+ upon re-exposure to the context. Results from this previous study indicate that remote memory only reactivate a very small number of neurons that were active during memory encoding. We followed the contextual fear conditioning paradigm (Maren, 2001; Maren et al., 2013) (Figure S8A) but used CANE to capture activated neurons. 2 weeks later, mice were placed back to the conditioning box to induce Fos. We found that CANE labeled 6.9 ± 0.9% of amygdala, and 8.7 ± 0.7% of dentate neurons (Figures S8B-S8D), and that 17.6 ± 1.1% of CANE-captured amygdala, and 3.3 ± 0.4% of captured dentate neurons re-expressed Fos by the remote memory retrieval task (Figures S8B, S8C and S8E). Thus, CANE captured moderately higher numbers of neurons and revealed moderately higher number of re-activated neurons, but overall our results are comparable to those obtained by Tayler et al., further validating the general applicability of CANE technology.

The aggression locus also contains a largely separable circuit regulating social fear

Prior conclusions that VMHvl is a locus for aggression were based on the fact that ablating progesterone receptor expressing neurons in VMHvl significantly reduced male aggression (Yang et al 2013), while electrically activating VMHvl or optogenetically stimulating Esr1+ VMHvl neurons elicits aggression or mounting behaviors (Fuchs et al., 1985; Kruk et al., 1983; Lammers et al., 1988; Lee et al., 2014; Lin et al., 2011). Here, using CANE-capturing and Fos staining in four different behavioral combinations (fear-fear, aggression-aggression, fear-aggression, aggression-fear), we showed that the social aggression and social fear ensembles in VMHvl are composed of largely non-overlapping pools of neurons. However, there is a 10~20% overlap between social fear and aggression VMHvl neurons. We speculate that the small overlap may represent shared neurons between the two behaviors that encode social cues; or alternatively, these shared neurons might be a potential substrate for fear induced aggression.

Although the numbers of SFNs in VMHvl labeled with CANE was relatively small (~200 Fos+ neurons), they exerted pronounced behavioral effects when synchronously activated with light. Our connectivity mapping study revealed that many of the regions targeted by VMHvl SFNs provide inputs back onto SFNs. It is likely that synchronously activating a small number of ChR2-epxressing SFNs in the presence of otherwise non-threatening social cues may recruit fear-associated neurons in multiple downstream targets of SFNs, some of which could in turn recurrently excite both tagged and untagged SFNs in the VMHvl. Through these recurrent interconnections, activity in SFNs may be sufficient to recruit and activate many parts of the social fear circuitry, exceeding the threshold required to trigger the behavioral expression of fear. The input-output map of VMHvl SFNs provides a blueprint for future studies that seek to understand how each of the nodes in the distributed recurrently connected circuit contributes to social fear. Future experiments are needed to test whether the identified hypothalamic SFNs are also involved in other types of anxiety and fear behaviors.

CANE as a general tool for studying spatially intermixed but functionally distinct ensembles

Similar to VMHvl, many brain regions contain Fos+ neurons that encode distinct or even opposite functions (Xiu et al., 2014). The CANE method is likely to be especially well suited to dissect such intertwined functional sub-circuits, especially when other molecular markers for distinguishing these circuits are unavailable (see Supplemental Table S1 for detailed comparison of CANE with other immediate early genes based methods). First, the Fos^TVA mouse for the CANE method is a knock-in line that maintains endogenous Fos expression, thus this method can in principle be applied to any brain regions with sufficient Fos expression. Second, CANE has a better temporal precision for labeling recently activated neurons without the ambiguity caused by the slow metabolism of doxycycline or tamoxifen. The CANE virus is injected into the desired brain region at the time when Fos-expression reaches peak levels, therefore the time window of infection is precisely controlled by the experimenter and can completely follow endogenous Fos expression. Third, the precise tagging allows for specific transsynaptic tracing of presynaptic inputs to functionally defined neurons in a heterogeneous brain region. Fourth, CANE is highly versatile. It can be used to label neurons in the same nucleus but activated by different stimuli or behavioral conditions by sequential labeling with CANE viruses expressing different fluorescent proteins or transgenes (an example of two-color sequential labeling of aggression versus fear neurons is shown in Figure S9). Of course CANE also has its own limitations: it cannot be used for systematic capturing of activated neurons across the entire brain because it requires stereotaxic surgical injection of the CANE viruses. Further, since anesthesia is needed for surgery, using CANE to capture neurons involved in processes affected by anesthesia requires additional steps: a cannula needs to be implanted into the desired brain region first, followed by the behavior episode to induce Fos, and followed by CANE virus delivery without additional surgery and anesthesia. A full comparison of advantages and limitations between different activity-dependent tagging methods is listed in Supplemental Table S1. In conclusion, CANE is a widely applicable and powerful tool to study specific functions and connectivity of transiently activated neuronal ensembles in many brain regions and in diverse behaviors and sensory processes.

Generating Fos^TVA knock-in mice

Fos^TVA knock-in mice were generated by inserting a 2A sequence, followed by a destabilized nuclear CFP (nCFP fusing a PEST sequence), the second 2A sequence, followed by destabilized TVA (TVA fusing a PEST sequence), bGH polyA, and a Frt-flanked PGK-neomycin resistance cassette immediately before the stop codon of Fos coding sequence by homologous recombination. All experiments were conducted according to protocols approved by The Duke University Institutional Animal Care and Use Committee. The Fos^TVA mice are available from Jackson Laboratory, stock No. 027831.

Production of CANE-lentiviruses (CANE-LV) and CANE-rabies viruses (CANE-RV)

EnvA^M21 has three (R213A, R223A, R224A) mutations in the extracellular domain of the EnvA protein. CANE-LV-Cre viruses are produced and packages using HEK293T cells by co-transfecting a plasmid encoding the EnvA^M21-VSVG fusion envelope protein, the pLenti-hSyn-Cre-WPRE plasmid, and the psPAX2 plasmid into these cells. CANE-RV-mCherry viruses are produced by pseudotyping the SAD-DeltaG-mCherry rabies virus (Osakada and Callaway, 2013) using a cell lines that produce the EnvA^M21(extracellular and transmembrane)-rabiesG (cytoplasmic region of SAD-B19 rabies glycoprotein) fusion envelope.

Virus injection for capturing social fear neurons (SFNs)

Adult male Fos^TVA mice at ages more than 8 weeks were singly housed for at least one day, and then subjected to the intruder social fear experience. Briefly, for capturing SFNs using CANE-LV or CANE-RV, they were introduced into a singly housed aggressive ICR male mouse for up to 10 min to be attacked 10~15 times, and returned into their home cage. One hour later, mice were anesthetized and undergo stereotaxic surgery for virus injection.

Optogenetic activation of captured SFNs in the VMHvl

SFNs in the VMHvl were captured using virus mixture of CANE-LV-Cre and AAV-EF1α-Flex-ChR2-EYFP or AAV-CAG-Flex-GFP in Fos^TVA mice. After more than 7 days of recovery periods, virus injected mice were implanted with a custom-made optic. 3 to 4 weeks later, the animals were subjected to behavioral testing in their home cage. Blue (473 nm) light was delivered with 20 ms pulses at 10 Hz, and the final output powers ranging from 2 to 3 mWmm⁻².

Behavioral effects of silencing SFNs

Fos^TVA mice in which either TeLC or GFP were expressed bilaterally in SFNs were subjected resident-intruder test for 2 min and were attacked 3~5 times by an aggressive resident. Then mice were returned to their own home cage and waited for 3 min, followed by social interaction test by introducing a non-aggressive adult male C57BL6 mouse for 5 min.

For more detailed full description of all experimental procedures, please refer to the Supplemental Experimental Procedures.