Investigative exploration and foraging leading to food consumption have vital importance, but are not well-understood. Since GABAergic inputs to the lateral and ventrolateral periaqueductal gray (l/vlPAG) control such behaviors, we dissected the role of vgat-expressing GABAergic l/vlPAG cells in exploration, foraging and hunting. Here, we show that vgat l/vlPAG cells encode approach to food and consumption of both live prey and non-prey foods. The activity of these cells is necessary and sufficient for inducing consumption, and vgat l/vlPAG activation produces exploratory foraging and compulsive eating without altering defensive behaviors. Moreover, l/vlPAG vgat cells are bidirectionally interconnected to several feeding, exploration and investigation nodes, including the zona incerta. Remarkably, the vgat l/vlPAG projection to the zona incerta bidirectionally controls approach towards food leading to consumption. These data indicate the PAG is not only a final downstream target of top-down exploration and foraging-related inputs, but that it also influences these behaviors through a bottom-up pathway.

Methods. All procedures conformed to guidelines established by the National Institutes of Health and have been approved by the University of California, Los Angeles Institutional Animal Care and Use Committee, protocols 2017-011 and 2017-075. All procedures conform to ARRIVE guidelines.

Mice. Vgat-Cre mice (Jackson Laboratory stock No. 028862) were used for all experiments. Male and female mice between 2 and 6 months of age were used in all experiments. Mice were maintained on a 12-hour reverse light-dark cycle with food and water ad libitum. All mice were handled for a minimum of 5 days prior to any behavioral task. VGAT-cre mice were used to direct the GCaMP6s expression in GABAergic neurons, since vgat, or vesicular GABA transporter, is a gene specifically expressed in GABAergic neurons. No animals were excluded. This study is reported in accordance with ARRIVE guidelines.

Viral Vectors. All vectors were purchased from Addgene, except for AAV5-hSyn-FLEX-TVA-P2A-eGFP-2A-oG and EnvA-dG-rabies-mCherry, which were purchased from the Salk Vector core. 

Surgeries. Ten-week-old mice were anesthetized with 1.5-3.0% isoflurane and affixed to a stereotaxic apparatus (Kopf Instruments). A scalpel was used to open an incision along the midline to expose the skull. After performing a craniotomy, 50 nL of virus was injected into the l/vlPAG  using a Hamilton 0.5 μL Neuros Model 7000.5 KH with a  beveled needle, and the bevel was placed to face medially (unilateral and counterbalanced for optogenetic activation and fiber photometry experiments, bilateral for inhibition experiments). The syringe was slowly retracted 12 minutes after the start of the infusion. For l/vlPAG, infusion location measured as anterior-posterior, medial-lateral, and dorso-ventral from bregma were -4.1 mm, ± 1.00 mm, -2.75 mm using a 15-degree angle. For optogenetic activation of vgat l/vlPAG cells, 50 nL of AAV5-Ef1a-DIO-eYFP or AAV5-EF1a-DIO-ChR2(H134R)-eYFP-WPRE-HGHpA of 2.3 x 1013 titer was delivered unilaterally to the l/vlPAG of vgat-cre mice. For optogenetic inhibition, 50 nL of AAV5-Ef1a-DIO-eYFP or AAV9-FLEX-Arch-GFP of 1.0 x 1013 titer was delivered unilaterally to the l/vlPAG of vgat-cre mice. Mice used in optogenetic experiments received a fiber optic cannula (0.22 NA, 200 mm diameter; Doric Lenses) 0.2 mm above viral infusion sites. For ZI optogenetic experiments, fiber optic devices location measured as anterior-posterior, medial-lateral, and dorso-ventral from bregma were -1.95 mm, ± 1.50 mm, -4.25 mm.

For photometry recordings, 50 nL of AAV9.Syn.Flex.GCaMP6s.WPRE.SV40 of 2.3 x 10*12 titer was injected into the l/vlPAG of vgat-cre mice and an optical fiber was implanted (0.48 NA, 400 mm diameter; Neurophotometrics) 0.2 mm above the injection site or above ZI at the same dorso-ventral coordinate aforementioned. Dental cement (The Bosworth Company, Skokie, IL, USA) was used to securely attach the fiber optic cannula to the skull. Half the mice in each cage were randomly assigned to YFP/GFP control or ChR2/Arch groups. Only mice with opsin expression restricted to the intended targets were used for behavioral assays.

For miniaturized microscope surgeries, after performing a craniotomy, 50 nl of virus was injected into the l/vlPAG (coordinates in mm, from skull surface): −4.10 anteromedial, ±1.00 lateral, −2.75 depth, 15-degree angle. Five days after virus injection, the animals underwent a second surgery in which two skull screws were inserted and a microendoscope was implanted above the injection site. A 0.5 mm diameter, ~4-mm-long gradient refractive index (GRIN) lens (Inscopix, Palo Alto, CA) was implanted above the l/vlPAG (−2.35 mm ventral to the skull surface) (Reis et al, 2021). The lens was fixed to the skull with cyanoacrylate glue and adhesive cement (Metabond; Parkell, Edgewood, NY, USA). The exposed end of the GRIN lens was protected with transparent Kwik-seal glue and animals were returned to a clean cage. Two weeks later, a small aluminum base plate was cemented onto the animal’s head on top of the previously formed dental cement. Animals were provided with analgesic and anti-inflammatory (carprofen).

For rabies virus tracing, we injected 50 nL of AAV5-hSyn-FLEX-TVA-P2A-GFP-2A-oG virus of titer 8.4 x 1012 in the l/vlPAG of vgat cre mice. Three weeks later, 200 nL of EnvA-dG-rabies-mcherry of 2.17 x 108 titer was injected in the same location. Mice were perfused 5 days after the injection of the second virus. Images were acquired with a Zeiss confocal microscope (AXIO Observer.Z1/ 7), using a x 10 objective. We counted manually along every slice using Zen Image Processing software the number of starter cells within the l/vlPAG around the injection site (mCherry and GFP double-positive cells). To analyze monosynaptic inputs of vgat+ l/vlPAG neurons we quantified the number of mCherry+ cells from a series of input brain areas with ImageJ software. We use the following nomenclature for the nomenclature for the brain areas evaluated: medial preoptic area (MPA), bed nucleus of the stria terminalis (BST, including all subdivisions), nucleus of the horizontal limb of the diagonal band (HDB), anterior hypothalamus (AH, including the entire anterior hypothalamic area and the latero-anterior hypothalamic nucleus), lateral hypothalamic area (LHa, including the parasubthalamic nucleus), central nucleus of the amygdala (CeA), zona incerta (ZI, we used the posterior hypothalamus to separate the most anterior portion - ZIa, and the most posterior portion - ZIp), ventromedial hypothalamic nucleus (VMH), dorsomedial hypothalamus (DMH), posterior hypothalamic nucleus (PH), premammillary nucleus (PMd), substantia nigra (SN, SNr - reticular part, SNc - compact part, and SNl - latera partl) and cuneiform nucleus (Cun). For counting cells in the MPA, we collected coronal sections from bregma +0.62 mm to bregma −0.22 mm. For counting cells in the BST, we collected coronal sections from bregma +0.38 mm to bregma −0.10 mm. For counting cells in the HDB, we collected coronal sections from bregma +0.62 mm to bregma −0.10 mm. For counting cells in the AH, we collected coronal sections from bregma −0.58 mm to bregma −1.06 mm. For counting cells in the LHa, we collected coronal sections from bregma −0.46 mm to bregma −2.30 mm. For counting cells in the CeA, we collected coronal sections from bregma −1.06 mm to bregma −1.70 mm. For counting cells in the ZIa, we collected coronal sections from bregma −1.06 mm to bregma −1.70 mm. For counting cells in the ZIp, we collected coronal sections from bregma −1.82 mm to bregma −2.70 mm.  For counting cells in the VMH, we collected coronal sections from bregma −1.22 mm to bregma −1.94 mm. For counting cells in the DMH, we collected coronal sections from bregma −1.58 mm to bregma −2.06 mm. For counting cells in the PH, we collected coronal sections from bregma −1.82 mm to bregma −2.54 mm. For counting cells in the PMd, we collected coronal sections from bregma −2.46 mm to bregma −2.70 mm. For counting cells in the SN, we collected coronal sections from bregma −2.70 mm to bregma −3.64 mm.5.20 mm. For each mouse all quantifications were performed on 40 μm coronal sections stained with DAPI. The outlines of these brain areas were according to the Paxinos Mouse Brain Atlas.

Perfusion and histological verification. Mice were anesthetized with Fatal-Plus and transcardially perfused with PBS, followed by a solution of 4% paraformaldehyde. Extracted brains were stored for 12 hours at 4°C in 4% paraformaldehyde. Brains were then placed in sucrose for a minimum of 24 hours. Brains were sectioned in the coronal plane in a cryostat, washed in phosphate buffered saline and mounted on glass slides using PVA-DABCO. Images were acquired using a Keyence BZ-X fluorescence microscope with a 4x, 10x or 20x air objective.

Optogenetic light delivery. For ChR2 mice, blue light was generated by a 473 nm laser (Dragon Lasers, Changchun Jilin, China) at 4.5 mW unless otherwise indicated. Green light was generated by a 532 nm laser (Dragon Lasers), and bilaterally delivered to mice at 7 mW. Mice were food-deprived for 18 hours prior to inhibition experiments. The power of the light was adjusted to the indicated intensity at the tip of the optic fiber before the experiments. A Master-8 pulse generator (A.M.P.I., Jerusalem, Israel) was used to drive the blue laser pulses of 5 milliseconds at  20 Hz. This stimulation pattern was used for all ChR2 experiments. The laser output was delivered to the animal via an optical fiber (200 μm core, 0.22 numerical aperture, Doric Lenses, Canada) coupled with the fiberoptic implanted on the animals through a zirconia sleeve.

Behavioral protocols. Preparation of the behavioral tests. On the day of the behavioral test, the animals were transferred to the testing room and habituated to the room conditions for 2h before the experiments started. All behavioral tests were conducted during the dark phase of the circadian period. For optogenetic experiments, the behavioral measures were taken as the difference between laser off and on epochs for each mouse and then averaged.

Walnut assay. To investigate behavioral and neural correlates of appetitive and consummatory responses related to feeding, we used walnuts as a high palatable and caloric food resource. Mice were allowed to eat walnuts in their home cages 2 days prior to the test. They were habituated to the test chamber (dimension: 13 x 13 x 25 cm) for 5 minutes on the day prior to the test. On the test day, mice were introduced to the chamber in the presence of a walnut and allowed to eat it. Each photostimulation consisted of 2-min epochs and presented once, totalling 4-min exposure. Epoch presentation (off or on first) was counterbalanced across mice. Walnut weight was measured before and after each epoch.

Cricket assay. Mice were trained to hunt crickets during 2 days before the start of experiments. Each mouse was allowed to kill up to 3 crickets on each day. Mice were acclimated to the experimental chamber (dimension: 33 cm diameter, 22 cm walls) for 10 minutes on the day prior to the test. On the test day, mice were allowed to hunt a single cricket per epoch. During “on'' trials, light was delivered in 1-min epochs.The same procedure was conducted during the “off” trials but no light was delivered. Each mouse was submitted to three to five epochs per light condition. For behavioral scoring, latency to first cricket attack was defined as the latency to start the attack and bite towards the cricket. Latency to successful predation was considered the latency to kill or remove a limb of the cricket. Probability of successful predation was calculated using a binary score (1 for a successful trial and 0 for an unsuccessful trial) if the mouse killed or removed a limb off the cricket during the 60s epoch.

Empty chambers. Mice were allowed to acclimate to the chamber (dimension: 13 x 13 x 25 cm) for 2 minutes before the start of the experiments. In a first experiment mice were allowed to explore the empty chamber and the frequency of rearing behavior was measured. In a following experiment, to investigate context-specific actions, a wired mesh (8 cm wide situated 7 cm above the floor) was added at one of the walls of the chamber allowing the occurrence of climbing behavior. In both assays, blue light was delivered in alternating 1-min epochs for four epochs, totalling 4-min exposure. The frequency of rearing behavior and the latency were quantified during each epoch and analyzed from videos recorded from a side camera. These data are plotted in Fig. 4b-c.

Holeboard task. Exploratory behavior was analyzed using a modified version of the hole-board apparatus, consisting of a test box (46 cm × 46 cm × 36 cm) and a hole-board frame with 25 holes in a grid-pattern (1.7 cm diameter, 9 cm apart), placed 4 cm above the floor of the testing box. Odor-impregnated bedding from cages of the same sex, which is a strong exploratory motivator, was placed below the hole-board frame. Mice were exposed only once and alternating 1-min epochs for six epochs totalling 6-min exposure. As a measure of exploration, the number of head-dips (hole visits) was measured by visual inspection of an experimenter blind to treatment. These data are plotted in Fig. 4d.

Real-time place preference. Mice were placed in a two-chamber context (20 × 42 × 27 cm) for 10 minutes to freely explore the environment. Both chambers are identical. During the next 10 minutes blue light was delivered to the l/vlPAG of vgat-cre mice expressing either ChR2 or YFP when they entered one of the chambers. Laser stimulation was only delivered during exploration of the stimulation chamber. 

Following ball. The experiment was performed similarly as previously described7. Mice were habituated to a rectangular chamber (46 cm × 46 cm × 36 cm) for 10 min on the day prior to the test. The contours of the letters “B” and “G” were marked on the bottom of the chamber. During the test day, after an acclimation period of 10 min, with mice at the left corner of the chamber, a table tennis ball connected to a stick was gently placed on the start point of the letter “B”, and photostimulation started. Each trial consisted in moving the ball so the mouse could follow the “BG” tracks. Speed was adjusted to prevent the mouse from biting the ball. The trial was terminated after the ball completed the “G” track, and photostimulation was discontinued. Video recordings were made of each mouse performing 5 trials for each epoch condition. The tracks of the mice and objects, and the distance between them were analyzed using custom Matlab code (MathWorks, USA).

Climbing assay. To test how context-specific actions of vgat l/vlPAG stimulation are selected when food is available in the environment, mice were exposed to an arena (dimension: 33 cm diameter, 27 cm walls) in which  walls were covered with a wired mesh. In this context, the appetitive stimuli was available on the walls 16 cm above the floor. Thus, mice were able to climb the walls and approach the food. For each mouse, the 'latency to bite walnut' and 'time in walnut corner' were taken as the average across all off and all on trials alternating 1-min epochs for five epochs. 

Object versus walnut preference assay. The mice were habituated to the arena (dimension: 24 cm x 24 cm x 24 cm) on the day prior to the experiment. Unreachable object and a walnut of similar size were hung on the opposite corner of the arena 14 cm above the floor. Blue light was delivered in alternating 3-min epochs for three epochs, totalling 9-min exposure. Mouse behavior at each corner was analyzed from videos recorded from a camera situated at the top of the arena.

Compulsive eating. The mice were habituated to an experimental chamber (dimension: 12 x 18 x 16 cm, with 50% of the floor area covered by a shock grid) for 10 minutes and food deprived for 18 hours on the day prior to the test. On the test day, a walnut was introduced into the environment on top of the grid and the amount of food eaten was measured during a 5 min baseline period (shock and light were off). Mice could only have access to the walnut by entering the grid area. After baseline, a shock generator was turned on and was only turned off at the end of the test. Shocks (0.08 mA, 0.1 seconds, 1Hz) were strong enough to decrease the amount of food consumed, but weak enough not to completely abolish the behaviors similarly as previously described31. Each shock period (laser off or laser on) lasted 5 min. Walnut weight was measured before and after each period (baseline, shock-laser off, and shock-laser on).

Elevated plus maze. The arms of the elevated plus maze (EPM) were 30 x 7 cm. The height of the closed arm walls were 20 cm high. The maze is elevated 65 cm from the floor and is placed in the center of the behavior room away from other stimuli. Mice were placed in the center of EPM facing a closed arm. For optogenetic experiments, light was delivered in alternating 2-min epochs for five epochs, totalling 10-min exposure.

Rat assay. We used a long rectangular chamber (70 x 25 x 30 cm) as previously described17. Mice were acclimated to this environment for at least two days for 10 min each day. During rat exposure, a live rat is restrained to one end of the chamber using a harness attached to a cable with one end taped to the chamber wall. For the optogenetic experiments light alternated off or on in 3-min epochs for 4 epochs, totalling in a 12-min trial.

U-shaped maze with walnut and object. The mice were habituated to an U-shaped maze enclosed arena (dimension: 24 x 20 x 13 cm) for 10min on the day before the fiber photometry experiment. Each corridor (8 cm wide) was separated by an 18 cm wall, and had a window situated on its end. Then, a plastic cap (3 cm diameter) or a piece of walnut was positioned at the end and outside of the window of each corridor. The interaction with the object or the food was limited by a wired mesh positioned at the window. Mouse approaches to the windows were recorded from a camera situated at the top of the maze. The mean df/F activity during approach to object or walnut was used for the quantitative analyses.

Aggression. Mice were placed in a small chamber of dimensions 29 x 19 x 13 cm for 8 minutes in the presence of a novel conspecific of the same sex. Baseline was recorded for 5 minutes and optogenetic stimulation was performed alternating 1-min epochs for three epochs. 

Fiber photometry. Photometry was performed as described in detail previously (Kim et al., 2016). Briefly, we used a 405-nm LED and a 470-nm LED (Thorlabs, M405F1 and M470F1) for the Ca2+-dependent and Ca2+-independent isosbestic control measurements. The two LEDs were band-pass filtered (Thorlabs, FB410-10 and FB470-10) and then combined with a 425-nm long-pass dichroic mirror (Thorlabs, DMLP425R) and coupled into the microscope using a 495-nm long-pass dichroic mirror (Semrock, FF495-Di02-25 3 36). Mice were connected with a branched patch cord (400 mm, Doric Lenses, Quebec, Canada) using a zirconia sleeve to the optical system. The signal was captured at 20 Hz (alternating 405-nm LED and 470-nm LED). To correct for signal artifacts of a non-biological origin (i.e., photo-bleaching and movement artifacts), custom Matlab scripts leveraged the reference signal (405-nm), unaffected by calcium saturation, to isolate and remove these effects from the calcium signal (470-nm). Correlation of df/F activity and threat proximity during EPM test and Rat assay were calculated as previously described18. 

Miniscope video capture. All videos were recorded at 30 frames/sec using a Logitech HD C310 webcam and custom-built head-mounted UCLA miniscope. Open-source UCLA Miniscope software and hardware (http://miniscope.org/) were used to capture and synchronize the neural and behavioral videos 17.

Miniscope postprocessing. The open-source UCLA miniscope analysis package (https://github.com/daharoni/Miniscope_Analysis) was used to motion correct the miniscope videos32. They were then temporally downsampled by a factor of four and spatially downsampled by a factor of two. The cell activity and footprints were extracted using the open-source package Constrained Nonnegative Matrix Factorization for microEndoscopic data (CNMF-E; https://github.com/zhoupc/CNMF_E) 33. Only cells whose variance was greater than or equal to 10% of the maximum variance among non-outliers were used in the analysis17. Neurons were coregistered across sessions using the open-source probabilistic modeling package CellReg (https://github.com/zivlab/CellReg) 34.   

Behavioral quantification. To extract the pose of freely-behaving animals in the described assays, we implemented DeepLabCut35, an open-source convolutional neural network-based toolbox, to identify mouse nose, ear and tailbase xy-coordinates in each recorded video frame, as well as, when applicable, cricket xy-coordinates. These coordinates were then used to calculate speed and position at each time point.

'Cricket approach' was defined as epochs for which the mouse approached the initial cricket position by >10 cm at a speed >3cm/s.

'Rearing' was defined as epochs for which the mouse extended its body up the enclosure wall of either the walnut or cricket assay. This was automatically detected as samples when the mouse head position exceeded a minimum distance from either (1) the center of the circular cricket assay enclosure or (2) a minimum vertical threshold for the walnut eating assay.

The DeepLabCut coordinates were not conducive to automatic classification of 'cricket eating', 'walnut approach', or 'walnut eating' behaviors. These behaviors were therefore scored manually by the experimenters.

Principal component analysis and silhouette score. For individual cricket or walnut sessions, the multidimensional neural activity from each session was reduced to its top three principal components using Matlab function 'pca'. For combined cricket/walnut session analysis, the session pairs were first coregistered using CellReg34, and the cell activity of putative neurons identified across sessions was first concatenated. The samples that corresponded with 'approach', 'eat', or 'rear' were plotted in 3-dimensional space. The silhouette score was calculated across these three behavioral clusters using the Matlab function 'silhouette'. The Mahalanobis distance between cricket/walnut 'approach' and cricket/walnut 'eat' was calculated using Matlab function 'mahal'. The pairwise Euclidean distance between the dimensionally-reduced 'approach' and 'eat' samples and their mean  for concatenated cricket/walnut sessions was calculated using the Matlab function 'pdist'.

Behavior decoding. Discrete classification of behaviors was performed using multinomial logistic regression. Time-points  following  approach or eat  by  2  s  were selected,  and  a  matched  number  of  non-behavior  time-points were randomly selected for training and validation. Each time point was treated as an individual data point. Training and validation were performed using five fold cross- validation, with a minimum of 10 seconds between training and validation sets. As equal numbers of behavior and non-behavior samples were used to build the training and validation sets, chance accuracy was 50%. Sessions with less than five behaviors  were  excluded  from  the  analysis.

Behavior cell classification. We  used  a  Generalized Linear Model (GLM)  to  identify  cells  that  showed  increased  calcium  activity  during  each behavior. We fit this model to each cell’s activity, with behavior indices as the predictor variable and behavior coefficients as the measure of fit. Behavior onset times were then  randomized  100  times  and  a  bootstrap  distribution  built  from  the  resulting  GLM  coefficients.  A cell was considered a behavior-categorized cell if its coefficient was either greater than (+ cell) or less than (- cell)  95% of the bootstrap coefficient values.

Mean peak amplitude. To calculate the mean peak amplitude for a cell on a given session, the local maxima were identified from the raw calcium activity (Matlab function ‘findpeaks’), and these maxima amplitudes were averaged.

Statistics. Nonparametric Wilcoxon signed-rank or rank-sum tests were used, unless otherwise stated using MATLAB. Two-tailed tests were used throughout with α=0.05. Asterisks in plots indicate the p values. Standard error of the mean was plotted in each figure as an estimate of variation. Multiple comparisons were adjusted with the false discovery rate method. All error bars in plots represent the standard error of the mean.

All files can be opened with Matlab 2021. Octave is an open-source alternative.