Aversive stimuli drive hypothalamus-to-habenula excitation to promote escape behavior

A sudden aversive event produces escape behaviors, an innate response essential for survival in virtually all-animal species. Nuclei including the lateral habenula (LHb), the lateral hypothalamus (LH), and the midbrain are not only reciprocally connected, but also respond to negative events contributing to goal-directed behaviors. However, whether aversion encoding requires these neural circuits to ultimately prompt escape behaviors remains unclear. We observe that aversive stimuli, including foot-shocks, excite LHb neurons and promote escape behaviors in mice. The foot-shock-driven excitation within the LHb requires glutamatergic signaling from the LH, but not from the midbrain. This hypothalamic excitatory projection predominates over LHb neurons monosynaptically innervating aversion-encoding midbrain GABA cells. Finally, the selective chemogenetic silencing of the LH-to-LHb pathway impairs aversion-driven escape behaviors. These findings unveil a habenular neurocircuitry devoted to encode external threats and the consequent escape; a process that, if disrupted, may compromise the animal’s survival.

The lateral hypothalamus (LH) and the medial portion of the ventral tegmental area (mVTA), innervate the LHb (Herkenham and Nauta, 1977). Both the LH and the mVTA contain heterogeneous neuronal populations, but afferents to the LHb are mostly excitatory (Root et al., 2014b;Stamatakis et al., 2016). Notably, such LH and mVTA populations respond to negative events and contribute to shape aversive behaviors including escape (Morales and Margolis, 2017;Herkenham and Nauta, 1977;Matsumoto and Hikosaka, 2007;González et al., 2016;Wang et al., 2015). However, whether these projections instruct habenular neurons to process aversive stimuli and to ultimately orchestrate escape behaviors is unknown.
To test this hypothesis, we use a combination of electrophysiology with chemo-and optogenetics to show that the hypothalamic-to-habenula inputs, but not the mVTA-to-habenula projection, mediate foot-shock-driven glutamate-dependent excitation of the LHb. We demonstrate that such hypothalamic-habenular circuit underlies aversion-driven escape, unraveling a behavioral consequence of aversion processing within the LHb.

Results
Aversive stimuli lead to glutamate-driven excitation of lateral habenula To model aversive escape behaviors, we placed mice in a two-compartment chamber and delivered a series of unpredicted foot-shocks (30/session). Across trials, mice rapidly escaped to the compartment opposite to the one of shock delivery ( Figure 1A). To explore habenula's contribution to such behavioral response, we LHb-targeted a viral construct coding for the calcium sensor GCamp6f (rAAV-hSyn-GCamp6f-eGFP) ( Figure 1A). Positioning a multimode optical probe above the LHb allowed fluorescence transients detection, which represent bouts of neuronal activity (Cui et al., 2014). Foot-shocks (0.3 mA), and aversive air-puffs (painless stimulus), evoked escape behavior (i.e. phasic increase in locomotion) along with time-locked fluorescent transients within the LHb ( Figure 1A, Figure 1-figure supplement 1A-B). Notably, a progressive increase in locomotion in absence of foot-shock, artificially promoted using the rotarod, did not modify the fluorescence signal measured from the LHb (Figure 1-figure supplement 1C). Therefore, simultaneous monitoring of fluorescence and behavior suggests a relationship between LHb activity and escape (Lawson et al., 2014;Matsumoto and Hikosaka, 2007).
To examine aversive stimuli-mediated single-cell LHb dynamics, we analyzed single-unit activities in anesthetized mice. Foot-shocks (hindpaw delivery) led to fast (~60 ms) and phasic increase (~100 ms) in action potentials in~50% of neurons recorded across the LHb ( Figure

Hypothalamic-habenular projections guide escape behaviors in mice
If LH inputs instruct LHb-to-midbrain projections to orchestrate aversion-driven escape behaviors, our prediction is that impairing their function may be detrimental for animal's ability to cope with a   threat. To test this, we targeted LH neurons with a Cre-dependent DREADDi construct (rAAV-EF1a-DIO-DREADDi-mCherry). We next infused the canine-derived Cav2-Cre vector within the LHb allowing for retrograde Cre-transport and subsequent DREADDi expression to selectively silence LHfiLHb projections in mice ( Figure 4A). Control (CTRL; rAAV-EF1a-td-Tomato) animals, after CNO injection, successfully shuttled to the compartment opposite to the one of shock delivery with short latencies across trials. Instead, mice expressing DREADDi in LHfiLHb escaped with higher latencies, along with increased shuttling failure rate ( Figure 4B). In contrast, no difference in latencies or failures was detected when silencing VTAfiLHb projections consistent with the lack of contribution to foot-shock mediated LHb excitation ( Next, we investigated the role of LHfiLHb in a more naturalistic setting, mimicking the attack of a predator with a projected shadow to model innate escape (Kunwar et al., 2015). When exposing CTRL mice, after CNO injection, to such paradigm, they rapidly escaped from the arena center to a 'safe' area (nest) ( Figure 4C and Video 1). In contrast LHfiLHb DREADDi mice had significant larger escape latencies ( Figure 4C and Video 2). Altogether, these data attribute to the LHfiLHb projection the crucial role in guiding escape, an evolutionary conserved innate response to a threat.

Discussion
The functional connectivity allowing hypothalamic circuits to orchestrate defensive and escape behaviors remained to date modestly described (Kunwar et al., 2015;Stamatakis et al., 2016;Mongeau et al., 2003). Our data fit the LHb within this neurocircuitry inferring causality between LHfiLHb excitation and the behavioral processing of aversive stimuli. Nevertheless, how sensory information reaches the LH, and how precise aversion-encoding hypothalamic cell-types control LHb remain to be established (Herrera et al., 2016;González et al., 2016). Aversive-driven escape is in equilibrium with freezing behaviors (innate and learned) that, at least partly, rely on periaqueductal gray and the amygdala (Fadok et al., 2017;Tovote et al., 2016;Wei et al., 2015). This raises interest to decipher whether an anatomical and functional connectivity exists between these latter nuclei and LHfiLHb circuits, and whether they encode negative stimuli in synergy or if devoted to specific behavioral aspects (i.e. flight, fight or freezing).
We found that shock-driven LHb excitation and escape behavior are independent of the VTAfiLHb projection. However, we report, consistent with published data, that VTAfiLHb terminals activation, drives aversive behaviors highlighting that cautious interpretation should be given to motivated behaviors generated by optogenetic manipulation of habenular circuits (Root et al., 2014a;Shabel et al., 2012;Stamatakis et al., 2016;Yoo et al., 2016).
Our data suggest a relevant control by LHfiLHb neurons onto midbrain GABA neurons, which in turn may lead to aversion-driven dopamine inhibition (Ungless et al., 2004). However, we also find that LHfiLHb neurons directly synapse onto 5HT and DA neurons. Notably, while habenula lesions reduced reward omission-driven inhibition of DA neurons, it left instead intact the inhibitory response to aversive stimuli (Tian and Uchida, 2015). Altogether, this heightens the need of understanding the computational properties and behavioral relevance of cell-type specific LHb projections onto monoaminergic systems (Lammel et al., 2012;Tan et al., 2012;van Zessen et al., 2012;Schweimer and Ungless, 2010).  In conclusion, our data describe a neural circuit instrumental for escaping a threat, and open indepth investigation of cell specificities and synaptic adaptations of the LHfiLHb projection to better decipher LHb processing in both health and disease states.

Experimental subjects
All in vivo and ex vivo procedures were performed on C57Bl/6J mice (males) wild-type or slc30a1-Cre (VGat-Cre), Pitx3-Cre and Slc6a4-Cre (Sert-Cre) mice aged 4-12 weeks. Mice were used in accordance with the guidelines of the Ministry of Agriculture and Forestry for animal handling and

Stereotactic injections
For surgery, all mice were anaesthetized with ketamine (100 mg/kg)/xylazine (10 mg/kg) (Sigma-Aldrich, France). Viral injections were performed using a glass pipette mounted on a stereotactic frame (Kopf, France). Volumes ranged between 200 and 400 nl per side, infused at a rate of 100-150 nl/min. The injection pipette was withdrawn from the brain 10 min after the infusion.
C57B6J mice (4-7 weeks) were injected in the Lateral hypothalamus (LH, À1.35 mm AP, 0.9 mm ML, À5.2 mm DV) or in the medial portion of the ventral tegmental area (mVTA, À2.6 mm AP, 0.0 mm ML, À4.8 mm DV). Animals were allowed to recover for a minimum of three weeks before recordings were performed.
For in vivo optogenetic experiments we injected mice in the LH or mVTA with rAAV2.1-hSyn-CoChr-eGFP (University of North Carolina, US) or rAAV2.1-CAG-dt-Tomato as a control virus. A single fiber optic was placed above and medially to the LHb (À1.4 mm AP, 0.2 mm ML, 2.8 mm DV) and then cemented on the mouse skull (Superbond resin cements, Sun medical, Japan).
In a separate set of experiments mice were injected unilaterally in the LHb (À1.4 mm AP, 0.45 ML, 3.1 mm DV) with the genetically encoded calcium indicator GCamp6f (rAAV2.1-hSyn-GCamp6f-eGFP; Pennsylvania University, US). A single fiber probe/optical fiber was placed and fixed 0.5 mm above the injection site. Only mice expressing GCamp6f-eGFP were used for such experiments.
For behavioral studies, C57B6J mice were infused with rAAV8-hSyn-DIO-HM4Di-mCherry virus (University of Pennsylvania, US) in the LH or mVTA. After a delay period of 3-5 weeks, mice were injected in the LHb (À1.45 mm AP, 0.45 mm ML, À3 mm DV) with retrograde Cav2-Cre virus (titer:~2,5 Â 10 12 pp/ml, IGMM CNRS, France). The injection sites were examined for all experiments and only data from animals with correct injections were included in the analysis.

Real-time place aversion and optogenetics
Four weeks following surgery, mice implanted with optical fibers above the LHb were placed in a custom-made behavioral arena with two compartments with different visual cues and wall texture for 15 min. The counterbalanced side of the chamber was defined as the light-paired stimulation side. At the beginning of the session, the mouse was placed in the non-light paired side of the chamber. Every time the mouse crossed to the light-paired side a 20 Hz constant laser stimulation (473 nm,~10 mW) was delivered. Light stimulation was interrupted when the mouse exit the light-paired side. Time spent on the stimulation-paired side and velocity were recorded via a digital camera interfaced with Any-maze software (Stoelting, Ireland).
Airpuffs experiments. Two fiber-coupled LEDs (Thorlabs M470F3 and M4051FP1) provided excitation light at different wavelengths: 470 nm to excite the calcium-dependent range of GCaMP6f, and 405 nm, the isosbestic excitation wavelength of GCaMP6f. Square pulses of 470 nm and 405 nm excitation light were alternated at 20 Hz, allowing near-simultaneous recording of, respectively, calcium-related fluorescence changes and a reference signal reporting calcium-independent fluorescence changes (e.g. movement artefacts; Kim et al., 2016). The light paths of the two excitation wavelengths were aligned using a custom-built fiber-coupled filter cube containing a dichroic mirror (Thorlabs DMLP425R), preceded by a collimator (350-700 nm; Thorlabs F240FC-A) and GFP excitation filter (Thorlabs MF469-35) for the 470 light path, and a collimator (395-415 nm; Thorlabs F671FC-405) for the 405 light path. Output excitation light was fiber-coupled to a Doric Fluorescence Mini Cube (FMC3_E(460-490)_F(500-550)), with the excitation filter removed. A single multimode patch cable was used for excitation and emission inside the brain (Thorlabs, 200 mm core, 0.39 NA), connected to the Doric cube using a metal FC connector and plugged into the fiber-optic implant using a metal ferrule-to-ferrule connector. Emitted fluorescence (~525 nm, regardless of excitation light) was filtered at the Doric cube, and then detected and amplified by a fiber-coupled photodetector (Thorlabs PDF10A). The light power of the 470 nm and 405 nm excitation wavelengths were set separately according to the signal seen in each mouse, and measured after the mouse was unplugged (range: 40-100 mW). The analogue signal from the photodetector was sent to an ADC input of a CED Micro1401, and signals were recorded using Spike2. Offline, 470-and 405excited signals were separated according to the time of excitation light, and DF/F was calculated for each, according to (F-F0)/F0, where F0 was defined as the median of a 10 s baseline period (for the air puff experiment, the first 10 s of each trial). DF/F 405 was subtracted from DF/F 470 , to remove any non-calcium-related fluorescence changes from the signal.
Puffs of compressed air lasting approximately 1 s were delivered to the base of the tail, every 60 s. This was done 5-7 times per mouse, and was aversive (mice moved quickly away) but not painful.
Foot-shocks experiments: Footshocks (3 s) were delivered in a chamber provided with an electrified grid-floor with an intensity of 0.3mA. Each mouse tested received 5-10 shocks in total with an inter-shock interval of 60 s. The mouse immediately reacted to the shock escaping in the other side of the chamber. A subset of animals was tested in the rotarod while monitoring locomotion. RPM increased continuously from 3 to 30 in bouts of 3 RPMs. The experiments were replicated two to three times in the laboratory.
Experiments were sampled on-and off-line by a computer connected to CED Power 1401 laboratory interface (Cambridge Electronic Design, Cambridge, UK) running the Spike2 software (Cambridge Electronic Design). Single units were isolated and the spontaneous activity was recorded for 5 min before triggering the shock protocol.
Spontaneous firing rate, percent of spikes in bursts and coefficient of variation (CV = standard deviation of interspike intervals/mean interspike interval; a measure of firing regularity) were determined. The criteria to identify a burst were made by a qualitative analysis per each neuron of the interspike interval histogram of 200 s of recordings. We defined the initiation of a burst as at least 2 action potentials occurred in an interval <10 ms. The burst was considered finished when the interval between the last 2 action potentials was >20 ms. Additionally, autocorrelograms were generated using a 10 ms bin width for intervals up to 2 s, to qualitatively classify neurons as firing in the regular, irregular or burst firing mode. Autocorrelograms showing three or more regularly occurring peaks were characteristic of the regular firing pattern. An initial trough that rose smoothly to a steady state was classified as irregular firing pattern, whereas an initial peak, followed by decay to a steady state, was indicating a burst pattern the bursting mode (elsewhere (aptic physiology with Prof. Fernando Valenzuela at the University of New Mexico, of the project (Lecca et al., 2011).
After recording baseline activity, each cell was tested for its response to repetitive (each 5 s) shocks (0.5 s, 3mA) delivered to the hind paw contralateral to the recording side with a spike2 automatic program. PSTHs and raster plots were built from 20 to 30 shocks and displayed using 10 ms bin width. A cell was considered excited when the mean number of action potentials/bin (bin length = 10 ms) in at least one of the four epochs (50 ms per epoch) after the shock inset was 2 times the Standard Deviation (SD) higher than baseline levels (the average number of action potentials/bin in the 2 s period before the shock). The onset of the response was calculated as the first of at least 2 consecutive bin higher than the 2SD of the averaged baseline. The duration of the response was calculated from the latency to the first of at least 5 consecutive bins not different than the baseline +2 SD. The magnitude of the response was obtained subtracting the baseline firing rate to the firing during the duration of the shock response.
Graphical representation of the foot-shocks responses were obtain as follow: for each cell excited by foot-shocks we normalized the PSTH subtracting the averaged baseline (2 s prior the shock, bin 10 ms) to every bin divided the number of sweeps. Then we implemented all the cells in a single histogram graph, reporting the mean ± the s.E.M.
For the recordings in the lateral hypothalamus (AP, À1.0 -À1.4 mm; L, 0.8-1.0 mm) we lowered the recording pipette at the following depth: V, from -4.7 to -5.2 mm. In this case to select LHbprojecting LH neurons, we tested each cell for its response to the ChR2-activating light. Only optoidentified neurons were considered for the analysis. Firing rate, coefficient of variation (%) as well as the response to foot-shocks were assessed using the same criteria for the LHb neurons.
A double barrel pipette assembly (injection tip,<50 mm in diameter attached~100 mm beyond the recording tip) was used for recording LHb spike activity with simultaneous local microinjection of drugs. The injection pipette was filled with one of the following: a mixture of the specific NMDA antagonist amino-5-phosphonopentanoic acid (AP-5; 100 mm) and the specific non-NMDA 2,3-Dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide (NBQX, 100 mM), or phosphate buffered saline. CNO (100 mM) was used for local chemogenetic silencing. Drugs were microinjected into the LHb, using brief pulses of pneumatic pressure (40 psi, 40 ms, Picospritzer, IM-300, Narishige, Japan). In all experiments a total volume of 60 nl was infused over 30 s for each injection. Two injections maximum per animal were given at an interval >30 min.
At the end of each experiment, the electrode placement was determined with an iontophoretic deposit of pontamine sky blue dye (À80 mA, continuous current for 5 min). Brains were then rapidly removed and fixed in 4% paraformaldehyde solution. The position of the electrodes was microscopically identified on serial sections (60 mm). Only recordings in the correct area were considered for analysis.

Behavioral paradigms
All behavioral tests were conducted during the light phase (7:00-19:00). Mice were habituated to the experimental room light level (35 lux) in their home cage (5 mice/cage) for at least 1 hr prior the testing. Animals were used for maximum two behavioral paradigms compatible one another, with behavioral paradigms repeated at least twice. Animals were randomly assigned to the experimental groups. Operators were blind to the experimental group during the scoring. All mice used for DREADDi silencing experiments received a single injection of CNO (1 mg/kg i.p.)~20 min prior the test.

Hot plate test
A standard hot plate (Biosed, Chaville, France), adjusted to 52˚C, was used to assess motor reactions in response to noxious stimuli. Mice were confined on the plate by a Plexiglas cylinder (diameter 19 cm, height 26 cm). The latency to a hind paw response (licking or shaking) or jumping was taken as the nociceptive threshold.

Locomotor activity
To assess the locomotor activity we tested mice in an open field arena. Mice were placed in the center of a plastic box (50 cm x 50 cm x 45 cm) in a room with dim light. Following a 5 min habituation period, the animal's behavior was videotaped and subsequently analyzed (Any-maze, France).

Shuttle box test
A shuttle box (13 cm Â 18 cm Â 30 cm) was equipped with an electrified grid floor and a door separating the two compartments. The test session consisted of 30 trials of escapable foot-shocks (10 s at 0.1-0.3 mA) separated by an interval of 30 s. The shock ended when animals shuttled to the opposite compartment. Failure was defined as the absence of shuttling to the other compartment within the 10 s of shock delivery. The time employed by the mouse to shuttle in the other compartment during the shock (latency) was also calculated (Lecca et al., 2016).

Looming visual stimulus test
Mice were tested for behavior in a looming visual stimulus test, as described elsewhere (Yilmaz and Meister, 2013). Animals were placed in an open-top Plexiglas box (50 Â 50 Â 45 cm). A triangular shaped nest (20 Â 12 cm) was placed in one corner. Recordings were performed under illumination provided by the projector screen placed above the arena. After 10 min of habituation, a looming stimulus was presented from the screen when an animal was in the center. The stimulus of 0.5 s duration was repeated 10 times with an interstimulus interval of 0.5 s. The latency to escape and the freezing time after escaping was calculated for each mouse. The analysis was performed off-line (Anymaze, Ireland).

Statistical analysis
Online/offline analyses were performed using Spike2 (Cambridge Electronic Design) IGOR-6 (Wavemetrics, US) and Prism (Graphpad, US). Data distribution was systematically tested with D'Agostino Pearson and Shapiro-Wilk normality tests. Depending on the distribution, parametric or not parametric test were used. Single data points are always plotted. Electrophysiological and behavioral experiments were replicated at least three times within the laboratory. Sample size was preestimated from previously published research and from pilot experiments performed in the laboratory. Compiled data are expressed as mean ± S.E.M. Significance was set at p<0.05 using two-sided unpaired t-test, Kolmogorov-Smirnov test, one or two-way ANOVA with multiple comparison when applicable. The use of the paired t-test and two way ANOVA for repeated measured were stated in the legend figure text. The Chi-Square test was used when required.