Abstract
A unique population of ventral tegmental area (VTA) neurons co-transmits glutamate and GABA. However, the circuit inputs to VTA VGluT2+VGaT+ neurons are unknown, limiting our understanding of their functional capabilities. By coupling monosynaptic rabies tracing with intersectional genetic targeting in male and female mice, we found that VTA VGluT2+VGaT+ neurons received diverse brainwide inputs. The largest numbers of monosynaptic inputs to VTA VGluT2+VGaT+ neurons were from superior colliculus (SC), lateral hypothalamus (LH), midbrain reticular nucleus, and periaqueductal gray, whereas the densest inputs relative to brain region volume were from the dorsal raphe nucleus, lateral habenula, and VTA. Based on these and prior data, we hypothesized that LH and SC inputs were from glutamatergic neurons. Optical activation of glutamatergic LH neurons activated VTA VGluT2+VGaT+ neurons regardless of stimulation frequency and resulted in flee-like ambulatory behavior. In contrast, optical activation of glutamatergic SC neurons activated VTA VGluT2+VGaT+ neurons for a brief period of time at high frequency and resulted in head rotation and arrested ambulatory behavior (freezing). Stimulation of glutamatergic LH neurons, but not glutamatergic SC neurons, was associated with VTA VGluT2+VGaT+ footshock-induced activity and inhibition of LH glutamatergic neurons disrupted VTA VGluT2+VGaT+ tailshock-induced activity. We interpret these results such that inputs to VTA VGluT2+VGaT+ neurons may integrate diverse signals related to the detection and processing of motivationally salient outcomes.
Significance Statement
Ventral tegmental area (VTA) glutamate neurons have roles in motivated behavior and unique neurotransmission capabilities. A specific VTA glutamate neuron subtype, those that co-transmit glutamate and GABA, have unique outcome-signaling properties compared with other VTA cell types. However, the circuits that regulate these neurons are unclear. We identified the whole-brain inputs to VTA glutamate and GABA co-transmitting neurons. We also identify two distinct glutamatergic inputs that activate VTA glutamate and GABA co-transmitting neurons and result in different behavioral repertoires suggestive of threat processing. Together, these results provide novel insights into the circuit and cell-type–specific influences on VTA glutamate and GABA co-transmitting neuronal activity as integrators of motivationally salient outcomes.
Introduction
Dopamine-, GABA-, and glutamate-releasing ventral tegmental area (VTA) neurons regulate rewarding and aversive experiences as well as the anticipation of these outcomes by learned predictors (Schultz, 1998; Lammel et al., 2011; Cohen et al., 2012; Tan et al., 2012; van Zessen et al., 2012; Ilango et al., 2014; Root et al., 2014a, 2018a; H. L. Wang et al., 2015; Qi et al., 2016; Yoo et al., 2016; Morales and Margolis, 2017). From their initial discovery (Yamaguchi et al., 2007), VTA glutamate-releasing neurons were highlighted for their molecular heterogeneity that reflects unique capabilities to release different combinations of neurotransmitters. In particular, subsets of VTA glutamate neurons release glutamate alone (requiring the vesicular glutamate transporter 2, VGluT2), glutamate with dopamine (requiring VGluT2 and tyrosine hydroxylase, TH), or glutamate with GABA (requiring VGluT2 and the vesicular GABA transporter, VGaT; Yamaguchi et al., 2007; Stuber et al., 2010; Yamaguchi et al., 2011; Tritsch et al., 2012; Root et al., 2014b, 2018b, 2020; Zhang et al., 2015). In contrast to other VTA cell types, we previously found in pavlovian reward and aversion tasks that VTA VGluT2+VGaT+ neurons showed a unique profile of neuronal activity consisting of activation by rewarding or aversive outcomes but not by learned predictors of those outcomes (Root et al., 2020).
The circuit inputs that regulate VTA VGluT2+VGaT+ neuronal signaling are unknown, limiting our understanding of how their unique outcome-signaling properties arise. To begin identifying the circuit inputs to VTA glutamate and GABA co-transmitting neurons, we first performed whole-brain monosynaptic rabies tracing of retrogradely labeled neurons from VTA VGluT2+VGaT+ neurons and utilized SHARCQ to register traced neurons to the Allen Brain Atlas (Lauridsen et al., 2022). Brainwide inputs largely arose from regions linked with threat, action selection, outcome valuation, and motor-related signaling. Within threat-related circuits, we identified that glutamatergic projections from the lateral hypothalamus (LH) and superior colliculus (SC) increased VTA VGluT2+VGaT+ neuronal activity in vivo. Based on our observations during stimulation of these neurons, we performed an initial exploration of stimulation-induced behaviors. Activation of glutamatergic LH and SC neurons resulted in distinct threat-related behavioral repertoires that were not correlated with VTA VGluT2+VGaT+ neuron activation, suggesting VGluT2+VGaT+ neurons are capable of signaling a generalized threat regardless of the motor responses that underlie them. However, glutamatergic LH activation of VTA VGluT2+VGaT+ neurons was selectively associated with VGluT2+VGaT+ footshock-related activity, suggesting this input provides threat-related information to these neurons. Together, these results provide novel insights into the cell-type–specific influences on VTA VGluT2+VGaT+ neuronal activity in the integration of motivationally salient outcome signaling.
Materials and Methods
Animals
VGluT2-IRES::Cre mice (Slc17a6tm2(cre)Lowl/J; Jax Stock 016963) were crossed with VGaT-2A::FlpO mice (Slc32a1tm2.1(flpo)Hze/J; Jax Stock 031331) at the University of Colorado Boulder to produce VGluT2::Cre/VGaT::FlpO offspring. Mice were maintained in a colony room with a 12 h light/dark cycle (lights on at 7:00 h) with ad libitum access to food and water. All animal procedures were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the University of Colorado Boulder Institutional Animal Care and Use Committee.
Stereotactic surgery
Male and female VGluT2::Cre/VGaT::FlpO mice (N = 8, 12–20 weeks of age) were anesthetized with 1–3% isoflurane and secured in a stereotactic frame (Kopf Instruments). INTRSECT vectors encoding optimized rabies glycoprotein (oG) and an avian cell-surface receptor tethered to mCherry (TVA-mCherry) were used for targeting VGluT2+VGaT+ neurons (Fenno et al., 2014, 2020; Hafner et al., 2019). AAV8-nEF-Con/Fon-TVA-mCherry (Addgene 131779, 5 × 1012 titer) and AAV8-EF1α-Con/Fon-oG (Addgene 131778, 5 × 1012 titer) were mixed in equal proportions (final titer, 2.5 × 1012 each) and injected into medial VTA (AP, −3.2 mm relative to the bregma; ML, 0.0 mm relative to the midline; DV, −4.4 mm from the skull surface; 500 nl). Three weeks later, mice were injected at the same coordinates with SAD-ΔG-EGFP(EnvA) (Salk Institute for Biological Studies, 400 nl). Total injection volume and flow rate (100 nl/min) were controlled with a microinjection pump (Micro4; World Precision Instruments). The syringe was left in place for an additional 10 min following injection to allow for virus diffusion, after which the syringe was slowly withdrawn to prevent spread by capillary action. Brains were harvested 1 week after rabies injection as described in Histology. Control mice (N = 4 VGluT2::Cre mice and N = 4 VGaT::Flp mice) were injected in VTA with the same Cre- and Flp-dependent TVA and oG AAVs, followed 3 weeks later by injection of SAD-ΔG-EGFP(EnvA) in the same site. Surgical, histological, and analytical procedures were identical to VGluT2::Cre/VGaT::FlpO mice.
For in vivo optical stimulation and photometry experiments, male and female VGluT2::Cre/VGaT::FlpO mice were injected with AAV8-EF1-Con/Fon-GCaMP6m (Addgene 137119, 8.5 × 1012 titer, 500 nl) into medial VTA (AP, −3.2; ML, 0.0; DV, −4.4 mm; 500 nl). In the same surgery, AAV8-nEF-Con/Foff-ChRmine-oScarlet (Addgene 137161, 1 × 1013 titer) was bilaterally injected into either LH (AP, −1.2; ML, ±1.2; DV, −5.0; 400 nl; N = 4 mice) or SC (AP, −3.5; ML, ±0.8; DV, −1.9 mm; 500 nl; N = 4 mice). All mice were implanted with a recording fiber cannula (400 μm core diameter, 0.66 NA; Doric Lenses) dorsal to medial VTA (AP, −3.2 mm relative to the bregma; ML, +1.0 mm relative to the midline at the skull surface, angled at 9.5°; DV, −4.2 mm from the skull surface). Recording fiber cannulae were angled to avoid damaging the aqueduct. Additionally, mice were implanted with optic fiber cannulae (200 μm core diameter, 0.37 NA; Doric Lenses) in either LH bilaterally [AP, −1.2; ML, −2.0 at 10° (left) and +2.9 at 20° (right); DV, −4.7 (left) and −4.9 (right)] or SC unilaterally (AP, −3.5; ML, −1.1 at 10°; DV, −1.73 mm). The LH stimulating implants in the left and right hemisphere were angled differently to better avoid the recording VTA implant on the skull. All implants were secured to the skull with screws and dental cement. Mice were allowed to recover for 3 weeks before experimentation.
For chemogenetic and photometry experiments, male and female VGluT2::Cre/VGaT::FlpO mice were injected with AAV8-hSyn-DIO-hM4Di-mCherry (Addgene 44362, 5 × 1012 titer, 400 nl) or AAV1-hSyn1-SIO-stGtACR2-FusionRed (Addgene 105677, 5 × 1012 titer, 400 nl) bilaterally into LH (AP, −1.2; ML, ±1.2; DV, −5.0 mm). In the same surgery, mice were injected with AAV8-EF1-Con/Fon-GCaMP6m and implanted with a recording fiber cannula into medial VTA as described above (400 μm core diameter, 0.66 NA). Mice were allowed to recover for 3 weeks before experimentation.
Pavlovian fear conditioning
Wild-type C57Bl6 mice and GCaMP mice were brought over 4 d to behavior chambers outfitted with rod flooring which was electrically connected to a shock generator (Med Associates). Mice were exposed to five auditory conditioned stimuli per day (CS+, 10 kHz tone, 30 s) that co-terminated with the delivery of footshock (US, 0.5 s, 0.5 mA, onset 29.5 s after the cue onset). Cues were presented on a variable interval 60 s schedule. On the fourth day, GCaMP6m signal was recorded. For wild-type mice, ANY-maze tracked the location of mice (30Hz) and quantified freezing 5 s before and during each conditioned stimulus. Freezing parameters in ANY-maze required at least 1 s of freezing to identify a freezing bout.
ChRmine stimulation
Three weeks after surgery, mice were restricted by the tail with gauze tape and placed in an enclosed Plexiglas box (Med Associates, MED-TLH-M). Tail restriction limited the mouse's actions so ChRmine-related photometry signals could be better interpreted. For frequency analysis, VTA GCaMP6m signals were recorded while optically stimulating ChRmine (589 nm, 10 ms pulse duration, 10–15 mW) at 5, 10, 20, and 40 Hz (Marshel et al., 2019). Five trials were delivered at each frequency. Each laser train was 1 s in duration, delivered on a 60 s intertrain interval and 5 min between frequencies. The frequency order was randomly assigned. After at least 3 d, mice were tail-restricted with a gauze tape in an unenclosed space allowing for rotational, ambulatory, or rearing movement that was behaviorally analyzed. ChRmine was activated at the maximum GCaMP frequency response for each input (20 Hz in LH and 40 Hz in SC, 10–15 mW) using 589 nm light at 10 ms pulse duration and 1 s train duration. Ten trains were delivered with 1 min intertrain intervals. Video recordings were made concurrently with ChRmine stimulation [Tucker-Davis Technologies (TDT), 30 Hz].
Chemogenetic inhibition
Three weeks after surgery, mice were given 0.1 mg/kg clozapine hydrochloride (Cayman Chemical 25779) intraperitoneally. Mice were placed in an enclosed Plexiglas box and restricted by the tail with a gauze tape. Copper electrodes with electrode paste were affixed to the tail. Ten minutes after clozapine injection, VTA GCaMP6m recordings began. Mice were administered 10 tailshocks (0.3 mA, 0.5 s duration) 1 min apart.
GCaMP recordings
GCaMP6m was excited at 465 and 405 nm (isosbestic control) with amplitude-modulated signals from two light-emitting diodes reflected off dichroic mirrors and coupled into an optic fiber. Signals from GCaMP and the isosbestic control channel were returned through the same optic fiber and acquired with a femtowatt photoreceiver (Newport), digitized at 1 kHz, and recorded by a real-time signal processor (TDT). Analysis of the recorded calcium signal was performed using custom-written MATLAB scripts. Signals (465 and 405 nm) were downsampled (10 times) and perievent time histograms (PETHs) were created trial-by-trial between −10 and +30 s surrounding each optical stimulation train onset and −10 and +10 s surrounding conditioned stimulus (CS+), footshock, or tailshock onsets. For each trial, data was detrended by regressing the isosbestic control signal (405 nm) on the GCaMP signal (465 nm) and then generating a predicted 405 nm signal using the linear model generated during the regression. The predicted 405 nm channel was subtracted from the 465 nm signal to remove movement, photobleaching, and fiber bending artifacts (dF). Baseline-normalized maximum Z-scores (normalized dF) were taken from −6 to −3 s prior to LH train, CS+, footshock, or tailshock onset, and maximum event Z-scores were taken from 0 to 2 s following LH train, SC train, CS+, footshock, or tailshock onset (McGovern et al., 2021, 2023). Due to the shorter activation profile following SC stimulation, PETHs were created trial-by-trial between −5 and +5 s surrounding each train onset, and baseline-normalized maximum Z-score was taken from −3 to 0 s prior to the SC train onset. The area under the curve (AUC) was collected between 0 and 2 s following the stimulation train onset. Because LH and SC neurons differed in their stimulation-induced activity profiles, the timepoint at which the normalized signal returned to half of the maximum observed value was calculated (half maximum time). AUC was collected between the stimulation onset and half maximum time (half maximum AUC; McGovern et al., 2024).
Video scoring
Behavioral optical stimulation experiments were video recorded at 30 Hz (TDT). For simultaneous optical stimulation and photometry experiments, foot treading, rearing, freezing, and turning behavior was timestamped in OpenScope (TDT). Foot treading was defined as more than one full gait cycle with all four paws. Rearing was defined as both forepaws lifting from the ground. Supported rearing was defined as rearing while supported against the wall with the forepaws. Freezing was marked as the complete cessation of movement except that which is necessary for breathing. For SC VGluT2 ChRmine experiments, head turning was quantified by subtracting the absolute angle of the head at the freeze onset from the absolute head angle immediately before the stimulation onset. Angles were determined by placing a protractor parallel to the base of the apparatus with the origin positioned at the midpoint between the animal's scapulae and extending a straight line to the bridge of the nose. Body turn calculations were performed the same way but with the origin of the protractor at the midpoint between the animal's hindlegs and with the line extending to the midpoint of the scapulae. All angles are presented in absolute values. Supplemental movies were created with Canva.
Histology
Mice were anesthetized with isoflurane and perfused transcardially with 0.1 M phosphate buffer followed by 4% (w/v) paraformaldehyde in 0.1 M phosphate buffer, pH 7.3. Brains were extracted, postfixed for 2 h in the same fixative, and cryoprotected in 18% sucrose in phosphate buffer at 4°C. For the tracing experiment, coronal sections of the whole brain (30 μm) were taken on a cryostat, mounted to gelatin-coated slides, and coverslipped with DAPI-containing ProLong Diamond medium. Sections were imaged on a PerkinElmer Opera Phenix High-Content Screening System at 20× air objective magnification. Fluorescence images in the DAPI, EGFP, and mCherry channels were taken in a single plane, in which somata were in focus. For stimulation and photometry experiments, coronal sections containing VTA and either LH or SC (30 μm) were taken, mounted to gelatin-coated slides, and imaged for optical fiber placement and expression of oScarlet, mCherry, FusionRed, and GCaMP6m on a ZEISS Axioscope.
Atlas registration and neuron quantification
Somata of EGFP-positive neurons were manually counted in FIJI using the Multi-point tool. Files of counted neurons’ X,Y positions within the image were exported using the Measure function. Somata of EGFP-mCherry coexpressing neurons were manually counted in Photoshop using the Count tool. Files of counted neurons’ X,Y positions were exported using the custom-written get_xy_photoshop.js JavaScript script (available at https://www.root-lab.org). Histological images and files of counted neurons’ locations were registered to the Allen Mouse Brain Atlas using the open-source software tool SHARCQ (https://github.com/wildrootlab/SHARCQ; Lauridsen et al., 2022). To avoid oversampling, we compared the coronal sections within 0.1 mm anteroposterior position of each other, and we removed the section with lower tissue integrity or imaging quality from analysis. SHARCQ analysis with the Allen Atlas CCFv3 (Q. Wang et al., 2020) localized 13.2% (8,756) of input neurons and 26.7% (606) of starter neurons simply to the hypothalamus, midbrain, pallidum, or pons with no further subregion specification. The registered coordinates of those neurons were analyzed in the Kim Atlas (Chon et al., 2019) to further classify neurons into more specific regions. Including both Allen Atlas and Kim Atlas data, neurons were categorized into 1,387 distinct regions which were merged into 556 regions in post-SHARCQ analysis.
Based on manual inspection, starter neurons designated to cerebellar peduncles, crossed tectospinal pathway, dorsal tegmental decussation, Edinger–Westphal nucleus, epithalamus related, extrapyramidal fiber systems, hypothalamic lateral zone, hypothalamus related, interpeduncular fossa, mammillary peduncle, mammillotegmental tract, medial forebrain bundle, medial lemniscus, midbrain, midbrain motor related, midbrain raphe nuclei, midbrain reticular nucleus, nucleus of Darkschewitsch, oculomotor nerve, oculomotor nucleus, periaqueductal gray, posterior hypothalamic nucleus, red nucleus, rubrospinal tract, substantia nigra compact part, tectospinal pathway, and ventral tegmental decussation were individually inspected and determined to belong to VTA. Likewise, starter neurons designated to lateral mammillary nucleus, mammillary body, mammillary related, and medial mammillary nucleus were determined to belong to the supramammillary nucleus. Input neurons localized to white matter tracts which are completely encapsulated by the VTA (dorsal tegmental decussation, mammillary peduncle, mammillotegmental tract, and ventral tegmental decussation) were moved post hoc to the VTA category. Neurons localized to the superior cerebellar peduncles posterior to −4.80 mm relative to the bregma were moved post hoc to the parabrachial nucleus.
For density analysis, brain region volumes were calculated by summing the voxels in each region and converting to cubic millimeter according to the voxel size of the atlas annotation (10 µm). The density of input neurons per brain region was calculated in neurons/mm3. Regions with fewer than 10 neurons on average were excluded from further density analysis.
Rendering
The open-source Python package brainrender (Claudi et al., 2021) was used to generate 3D whole-brain renders and coronal slices of the Allen Mouse Brain Atlas with neuron locations acquired in SHARCQ. Anatomical coordinates of starter or input neurons were combined across eight animals for all plots. In 3D coronal slices, select brain structures were outlined for anatomical reference.
Density heatmaps of starter and input regions were generated using the Matplotlib Python library and Seaborn KDE analysis according to the process described by the Wang Lab (Takatoh et al., 2021). Anatomical coordinates of starter or input neurons in brain regions of interest were isolated. For bilateral structures, neurons of each hemisphere were plotted separately to prevent proximity algorithms from misrepresenting density across the midline. Density plots were overlaid onto 3D coronal atlas slice renders generated in brainrender. Plots were aligned in Adobe Photoshop using axes as guides, which were removed from the final image. Automatically generated scale bars indicate the color-coded density in neurons/mm3. Bilateral structures have discrete scale bars for each hemisphere. Parameters, code, and documentation are available at www.root-lab.org.
Statistical testing
Statistical testing was performed in SPSS or GraphPad Prism. For LH VGluT2 or SC VGluT2 ChRmine stimulation experiments, a within-subject ANOVA compared stimulation epoch (baseline, stimulation) and frequency (5, 10, 20, 40 Hz). Main effects were followed up by Sidak-adjusted pairwise comparisons. A repeated-measure ANOVA or paired t test compared the duration of behavioral reactions or freezing during baseline and stimulation. Simple linear regressions were used to correlate within-subject GCaMP activity (maximum Z-score or AUC from 0 to 2 s) and parameters of the behavioral responses (reaction/freeze duration, latency to head turn, head turn angle, or latency to freeze). For SC VGluT2 ChRmine stimulation experiments, a paired t test compared the average proportion of trials resulting in turn-and-freeze behavior in stimulation epochs versus interstimulation intervals. Within subjects, the normality of head turn and body turn angle distributions were tested using a Shapiro–Wilk test. A nonparametric Kruskal–Wallis test compared the latencies of head turn, peak GCaMP value, and freezing. The main effect was followed up by Dunn's multiple comparison test. To examine CS+ and footshock-related activity, a within-subject ANOVA compared the maximum baseline Z-score (−6 to −3 s from the CS+) with the maximum Z-score following the CS+ onset (0 to 2 s) and the maximum Z-score following the footshock onset (0 to 2 s). A main effect was followed up by contrast tests comparing the baseline with CS+ and the baseline with footshock (McGovern et al., 2021). For cue-induced freezing behavior in wild-type mice, a paired t test compared the percentage of freezing 5 s before each cue and the initial 5 s of each cue. Pearson's correlation was used to correlate VTA VGluT2+VGaT+ maximum Z-score (0 to 2 s) following 40 Hz stimulation of each pathway with footshock-induced maximum Z-score (0 to 2 s). For chemogenetic inhibition experiments, a repeated-measure ANOVA compared maximum Z-score GCaMP activity (0 to 2 s) across the shock epoch (baseline, shock) and treatment group (hM4Di, stGtACR2). For all tests, alpha was set to 0.05.
Results
We aimed to identify circuits that contribute to VTA VGluT2+VGaT+ neuronal signaling. To accomplish this, VGluT2::Cre/VGaT::Flp mice were injected in VTA with a mix of rabies helper viruses dependent on the expression of both Cre and Flp which encode a cell-surface avian receptor tethered to mCherry (TVA-mCherry) and optimized mammalian rabies glycoprotein (oG) (Watabe-Uchida et al., 2012; Kim et al., 2016; Hafner et al., 2019). Three weeks later, mice were injected in VTA with an avian-enveloped, glycoprotein-deleted rabies virus encoding EGFP (Fig. 1A). After 1 additional week, brains were extracted, sectioned, imaged, counted, and registered to the Allen Atlas by SHARCQ (Lauridsen et al., 2022; Fig. 1B). The 2,273 putative starter neurons were identified by co-expression of TVA-mCherry and rabies-EGFP (284.12 ± 47.51 neurons/mouse). The vast majority of putative starter neurons were found within the VTA (85.96 ± 2.50%), and a smaller subset were found within the supramammillary nucleus (14.04 ± 2.50%), consistent with the localization of midbrain VGluT2+VGaT+ neurons (Root et al., 2018b; Fig. 1C–J). VTA starter neurons were highly dense within the interfascicular nucleus but also showed high density in the rostral linear nucleus and other portions of the VTA (Fig. 1H).
Experimental design and VTA VGluT2+VGaT+ starter neurons. A, Schematic of viral injections for retrograde tracing in VGluT2::Cre/VGaT::Flp mice. B, Timeline of viral injections, tissue processing, and analysis. C–E, Example fluorescence images of viral injections in the VTA. C, VGluT2+VGaT+ neurons expressing the helper viruses. D, Neurons expressing the modified rabies virus. E, Neurons with mCherry-EGFP overlap (yellow triangle) coexpressed the helper viruses and rabies virus and comprised the population of starter neurons. Neurons with only mCherry expression (red triangle) did not receive the rabies virus and thus were not included in the starter neuron population. Neurons with EGFP only (green triangle) expressed the rabies virus and comprised the population of monosynaptic input neurons. F, Example histological section (top) registered to the Allen Mouse Brain Atlas using SHARCQ (bottom). G, Average proportion of starter neurons localized to the VTA and SuM. Dots represent individual animals; N = 8. H, Density heatmap of all starter neurons from −2.97 to −3.27 mm relative to the bregma (N = 8 animals). I, Coronal, sagittal, and horizontal 3D representations of all starter neurons (N = 8 animals). J, 3D coronal slice representations of starter neurons from (left to right) −2.37 to −2.67, −2.67 to −2.97, −2.97 to −3.27, −3.27 to −3.57, and −3.57 to −3.87 mm relative to the bregma (N = 8 animals). Abbreviations: SuM, supramammillary nucleus; VTA, ventral tegmental area.
To identify the intersectional specificity of the starter neurons, we first injected mice expressing Cre under the control of only the VGluT2 promoter (VGluT2::Cre) or mice expressing Flp under the control of only the VGaT promoter (VGaT::Flp) in VTA with Cre- and Flp-dependent AAVs encoding TVA-mCherry and oG. After 3 weeks, mice were injected in VTA with rabies-EGFP and following an additional week examined for brainwide TVA-mCherry and rabies-EGFP expression (Fig. 2A). In VGluT2::Cre mice, 11.25 ± 3.61 VTA neurons expressed rabies-EGFP, and in VGaT::Flp mice, 3.25 ± 1.31 VTA neurons expressed rabies-EGFP (Fig. 2B). In both cases, we observed no mCherry expression nor were retrogradely labeled rabies-EGFP neurons identified outside the injection site (Fig. 2C,D). We interpret these data to indicate that rabies-EGFP neurons outside the VTA in VGluT2::Cre/VGaT::Flp mice were retrogradely labeled from Cre and Flp coexpressing VTA neurons. However, a small number of rabies-EGFP neurons locally within the VTA may result from recombinase-independent labeling.
Intersectional control experiment. A, Schematic of viral injections for retrograde tracing in VGluT2::Cre or VGaT::Flp mice. B, Example fluorescence image of viral injection in the VTA. C, Number of EGFP-expressing VTA neurons in VGluT2::Cre/VGaT::Flp, VGluT2::Cre, or VGaT::Flp mice. D, Number of EGFP and mCherry coexpressing VTA neurons in VGluT2::Cre/VGaT::Flp, VGluT2::Cre, or VGaT::Flp mice.
Using monosynaptic rabies viral tracing of inputs to VGluT2+VGaT+ neurons, 66,465 input neurons (defined by EGFP expression without mCherry co-expression) were found to synapse onto VGluT2+VGaT+ neurons (8,308 ± 384.52 inputs per mouse). SHARCQ registration of these inputs to the Allen Atlas revealed high percentages of inputs in SC (7.69 ± 0.53%), midbrain reticular nucleus (6.24 ± 0.42%), periaqueductal gray (6.14 ± 0.32%), locally within the VTA (5.32 ± 0.29%), and LH (4.03 ± 0.37%) (Fig. 3A–C). Considering the input number relative to region volume, input neurons were densest in the dorsal raphe nucleus (1,232.08 ± 161.26 neurons/mm3), lateral habenula (980.51 ± 114.79 neurons/mm3), locally within the VTA (713.74 ± 50.99 neurons/mm3), and supraoculomotor periaqueductal gray (480.86 ± 96.85 neurons/mm3; Fig. 3D,E).
Brainwide monosynaptic inputs to VTA VGluT2+VGaT+ neurons. A, Average proportion of input neurons. All regions accounting for at least 0.5% of inputs are depicted. Dots represent individual animals; N = 8. B, Coronal, sagittal, and horizontal 3D representations of all input neurons (N = 8 animals). C, 3D coronal slice representations of input neurons from (left to right) 1.40 to 1.20, 0.20 to 0, −1.60 to −1.80, −3.15 to −3.35, −3.45 to −3.65, −4.30 to −4.50, and −5.8 to −6.0 mm relative to the bregma (N = 8 animals). D, Average density of input neurons. All regions with at least 10 neurons and a density of at least 100 neurons/mm3 are depicted. Dots represent individual animals; N = 8. E. Density heatmaps of input neurons in the DRN (top, −4.20 to −4.40 mm relative to the bregma), LHb (middle, −1.60 to −1.80 mm relative to the bregma), and PAG (bottom, −3.80 to −4.00 mm relative to the bregma). The bilateral LHb has a scale bar for each hemisphere. Abbreviations: DN, dentate nucleus; DRN, dorsal raphe nucleus; IP, interposed nucleus; LH, lateral hypothalamus; LHb, lateral habenula; MRN, midbrain reticular nucleus; NAcc, nucleus accumbens; PAG, periaqueductal gray; SC, superior colliculus; SNR, substantia nigra reticulata; VP/SI, ventral pallidum/substantia innominata; VTA, ventral tegmental nucleus.
Cortical inputs accounted for a substantially smaller proportion of total inputs than subcortical inputs (2.92% of total, 243 ± 28.22 cortical inputs per mouse). Nonetheless, the primary cortical inputs arose from the secondary motor cortex (17.51 ± 1.57%), orbital cortex (15.03 ± 3.79%), and insular cortex (14.90 ± 2.06%; Fig. 4). Smaller proportions of inputs arose from more sensory-related regions including somatosensory, piriform, gustatory, and visceral cortical regions.
Cortical monosynaptic inputs to VTA VGluT2+VGaT+ neurons. A, Average proportion of input neurons in cortical regions. Interior pie chart illustrates broad cortical categories. Exterior pie chart illustrates subregions of cortical categories. B, 3D coronal slice representations of input neurons (N = 8 animals) from (left to right) 3.00 to 2.80, 2.50 to 2.30, 2.00 to 1.80, 1.50 to 1.30, 1.00 to 0.80, and 0.50 to 0.30 mm relative to the bregma. Abbreviations: ACA, anterior cingulate cortex; AI, agranular insular cortex; AON, anterior olfactory nucleus; M1, primary motor cortex; M2, secondary motor cortex; OFC, olfactory cortex; PIR, piriform area; S1, primary somatosensory cortex.
LH has previously been shown to be a primary input to VTA VGluT2+ neurons (Faget et al., 2016; An et al., 2021; Simon et al., 2024). Channelrhodopsin2 activation of LH VGluT2+ inputs to VTA increases c-Fos expression in VTA VGluT2+VGaT− neurons but not VGluT2+VGaT+ neurons (Barbano et al., 2020). Nevertheless, given the strong input from LH VGluT2 neurons to VTA VGluT2 neurons (Barbano et al., 2020), we hypothesized that the large number and density of LH neurons that synapse on VTA VGluT2+VGaT+ neurons (Fig. 5A,B) were glutamatergic. Furthermore, we hypothesized that activation of LH VGluT2 neurons would increase the activity of VTA VGluT2+VGaT+ neurons on a phasic timescale but may not be sufficient to increase c-Fos expression. To test this hypothesis, we injected VGluT2::Cre/VGaT::Flp mice in LH with a Cre-dependent vector that expressed the red-shifted opsin ChRmine in LH VGluT2+ neurons and in VTA with a Cre- and Flp-dependent vector that expressed GCaMP6m in VTA VGluT2+VGaT+ neurons (Figs. 5C, 6A–D). ChRmine activation of LH VGluT2 neurons robustly increased VTA VGluT2+VGaT+ neuron activity (main effect of stimulation over the baseline, F(1,3) = 11.657; p = 0.042), which was invariant to stimulation frequency (F(3,9) = 1.485; p = 0.283; Fig. 5D). AUC from 0 to 2 s following the stimulation onset was 19.71 ± 3.32. Time to half maximum was 1.42 ± 0.19 s, and AUC over half maximum time was 18.16 ± 3.71. At the stimulation onset, we noticed that mice frequently attempted to rear or tread within the enclosed environment used for frequency testing. To more closely examine the behaviors induced by LH VGluT2 neuron stimulation and determine if this behavioral response is related to VTA VGluT2+VGaT+ neuron activity, we stimulated LH VGluT2 neurons while mice were tail-restricted in an unenclosed environment. We chose 20 Hz stimulation because this frequency of activating LH VGluT2 neurons causes defensive behavior that depends on VTA VGluT2+ neurons (Barbano et al., 2020). In the unenclosed environment, stimulation of LH VGluT2 neurons at 20 Hz resulted in increased supported rearing (0.59 ± 0.18 s) and paw treading (0.82 ± 0.30 s) behavior in tail-restricted mice (Fig. 5E,F; Movie 1). Reaction durations were greater following stimulation than at the baseline (F(1,3) = 15.330; p = 0.0020), but did not differ between rearing and foot treading (F(1,3) = 0.662; p = 0.43). Stimulation of LH VGluT2 neurons at 20 Hz in the unenclosed environment also increased VTA VGluT2+VGaT+ neuronal activity (Fig. 5G). However, there was no significant correlation between LH VGluT2 ChRmine-induced reaction durations and VTA VGluT2+VGaT+ neuronal activity (maximum Z-score or AUC).
LH VGluT2+ neurons form a functional connection with VTA VGluT2+VGaT+ neurons and cause ambulatory behavior. A, Monosynaptic input neurons in LH. B, Bilateral density heatmap of input neurons in the LH from −1.60 to −1.80 mm relative to the bregma. The left scale bar corresponds to the left hemisphere; the right scale bar corresponds to the right hemisphere. C, D, Mice were first optically stimulated in an enclosed space. C, Schematic of viral injections and bilateral optical fiber implants for simultaneous optical stimulation of LH VGluT2+ neurons and fiber photometry of VTA VGluT2+VGaT+ neurons. D, Normalized averages (left) of GCaMP response to different frequencies of ChRmine laser stimulation (N = 4 animals). Average Z-scores (right) at stimulation were significantly higher than the baseline, invariant to stimulation frequency. Dots represent individual animals; N = 4. E–I, Mice were then optically stimulated in an unenclosed space while tail-restricted. E, Depiction of supported rearing and foot treading behavior elicited by ChRmine stimulation (wall not shown). F, Supported rearing and foot treading were significantly higher immediately following ChRmine stimulation compared with before stimulation. Dots represent individual animals; N = 4. G, Normalized average GCaMP response to 20 Hz ChRmine stimulation over 10 trials (N = 4 animals).
Histological localization of fiber optics targeting LH or SC VGluT2 neurons and VTA VGluT2+VGaT+ neurons. A, Example bilateral optical fiber placement and viral expression of AAV8-nEF-Con/Foff-ChRmine-oScarlet. B, Anatomical locations of all bilateral optical fiber cannulae (N = 4 animals). C, Example recording fiber placement and viral expression of AAV8-EF1-Con/Fon-GCaMP6m. D, Anatomical locations of all medial recording fiber cannulae (N = 4 animals). E, Example unilateral optical fiber placement and viral expression of AAV8-Nef-Con/Foff-ChRmine-oScarlet. F, Anatomical locations of all unilateral optical fiber cannulae (N = 4 animals). G, Example recording fiber placement and viral expression of AAV8-EF1-Con/Fon-GCaMP6m. H, Anatomical locations of all medial recording fiber cannulae (N = 4 animals). Scale bars in A and C, 100 µm.
Another major input to VTA VGluT2+ neurons is from the SC (An et al., 2021), which accounted for the largest proportion of inputs to VGluT2+VGaT+ neurons (Figs. 2A, 7A). SC inputs were densest in the intermediate and ventral layers of SC (Fig. 7B), suggesting SC inputs were from glutamatergic neurons (Masullo et al., 2019; Liu et al., 2022). To test this, we injected VGluT2::Cre/VGaT::Flp mice in SC with a Cre-dependent vector that expressed the red-shifted opsin ChRmine in SC VGluT2 neurons and injected in VTA with a Cre- and Flp-dependent vector that expressed GCaMP6m in VTA VGluT2+VGaT+ neurons (Figs. 7C, 6E–H). In contrast to LH VGluT2 inputs, ChRmine activation of SC VGluT2 neurons resulted in frequency-dependent signaling in VTA VGluT2+VGaT+ neurons; F(3,9) = 3.8723; p = 0.0497. ChRmine activation of SC VGluT2 neurons significantly increased VTA VGluT2+VGaT+ neuronal activity at 10, 20, and 40 Hz, but not 5 Hz (Sidak-adjusted pairwise comparisons 5 Hz, p = 0.4181; 10 Hz, p = 0.0307; 20 Hz, p = 0.0009; 40 Hz, p = 0.0034; Fig. 7D). AUC from 0 to 2 s following the stimulation onset was −2.90 ± 1.13. Half maximum time was 0.66 ± 0.061 s; and half maximum AUC was 0.84 ± 0.22. We noticed that stimulation of SC VGluT2 neurons often resulted in a rapid head movement and body posture change followed by prolonged freezing behavior within the enclosed testing environment. To more closely examine the behaviors induced by SC VGluT2 stimulation and determine if this behavioral response was related to VTA VGluT2+VGaT+ neuron activity, we stimulated SC VGluT2 neurons at the most reliable frequency that activated VTA VGluT2+VGaT+ neurons (40 Hz), while mice were tail-restricted in an unenclosed environment. In the unenclosed environment, SC VGluT2 neuron stimulation at 40 Hz caused a constellation of behaviors (Fig. 7E; Movie 2) and again activated VTA VGluT2+VGaT+ neuronal activity (Fig. 7F). ChRmine stimulation resulted in head turning followed immediately by freezing, which was rarely observed during interstimulation intervals; t(3) = 5.278; p = 0.0133 (Fig. 7G). Following ChRmine stimulation, mice turned their heads an average of 48.45° ± 5.07° from their starting position to their freezing position. All subjects except one (subject A518, Shapiro–Wilk test; W = 0.76; p = 0.0042) had head turn angles normally distributed around their mean turn angle. Body turns were observed with less vigor and consistency compared with head turns. The mean angle of body turn following SC VGluT2 ChRmine stimulation was 25.78° ± 4.92°. Two of four subjects had body turn angles that departed from a normal distribution around their mean angle (subject A518, Shapiro–Wilk test; W = 0.68; p = 0.0009; subject A525, W = 0.69; p = 0.0012; Fig. 5H). A Kruskal–Wallis test yielded overall differences in the mean latencies to head turn (0.26 ± 0.06 s after the SC VGluT2 stimulation onset), freezing behavior (0.56 ± 0.05 s), and maximum VTA VGluT2+VGaT+GCaMP activity (0.50 ± 0.02 s; H = 29.30; p < 0.0001). Dunn's multiple comparison tests found that this main effect was explained by significant differences between head turn latency and freeze latency (p < 0.0001) and between head turn latency and time to peak GCaMP activity (p < 0.0001), but not between freeze latency and time to peak GCaMP (p = 1.0; Fig. 7I). Irrespective to pairing with head turning, freezing behavior was rarely observed in the interstimulation intervals. The duration of the first continuous freezing bout following SC VGluT2 ChRmine stimulation (4.50 ± 0.33 s) was significantly longer than all freezing in interstimulation intervals (0.33 ± 0.22 s); t(3) = 18.44; p = 0.0003 (Fig. 7J). Subject A518 displayed moderate negative correlations between head turn angle and maximum GCaMP value (R2 = 0.508; p = 0.0207), head turn angle and GCaMP AUC (R2 = 0.610; p = 0.0076), and freeze duration and maximum GCaMP value (R2 = 0.450; p = 0.0339). On the whole, however, there was no significant correlation between SC VGluT2 ChRmine-induced behaviors (latency to head turn, head turn angle, latency to freeze, freeze duration) and VTA VGluT2+VGaT+ neuronal activity (maximum Z-score or AUC).
SC VGluT2+ neurons form a functional connection with VTA VGluT2+VGaT+ neurons and reduce ambulatory behavior. A, Monosynaptic input neurons in the SC. B, Bilateral density heatmap of input neurons in the SC from −3.50 to −3.70 mm relative to the bregma. The left scale bar corresponds to the left hemisphere; the right scale bar corresponds to the right hemisphere. C, D, Mice were first optically stimulated in an enclosed space. C, Schematic of viral injections and unilateral optical fiber implant for simultaneous optical stimulation of SC VGluT2+ neurons and fiber photometry of VTA VGluT2+VGaT+ neurons. D, Normalized averages (left) of GCaMP response to different frequencies of ChRmine laser stimulation (N = 4 animals). Average Z-scores (right) depended on the interaction of frequency and stimulation condition. Dots represent individual animals; N = 4. E–J, Mice were then optically stimulated in an unenclosed space while tail-restricted. E, Depiction of head turning, freezing, and body turning behavior elicited by ChRmine stimulation. F, Normalized average GCaMP response to 40 Hz ChRmine stimulation over 10 trials (N = 4 animals). G, Proportion of stimulation trials and interstimulation intervals that resulted in head turning with freezing. Dots represent individual animals; N = 4. Head turning and freezing behavior was significantly more likely after stimulation compared with during interstimulation intervals. H, Average angle of head turn and body turn from starting position to freezing position following ChRmine stimulation for each test subject. Dots represent individual stimulation trials. All subjects except A518 had head turn angles normally distributed around their mean turn angle. Subjects A517 and A519 had body turn angles normally distributed around their mean turn angle, while A518 and A525 had body turn angle distributions that departed from normality. I, Box-and-whisker plot of head turn, peak GCaMP, and freeze latencies following ChRmine stimulation. Dots represent individual stimulation trials. Head turn latency was significantly shorter than peak GCaMP latency and freeze latency. Peak GCaMP and freeze latency did not differ. J, Average time spent freezing immediately following ChRmine stimulation and during interstimulation intervals, irrespective of pairing with head turning. The duration of continuous freezing bouts following stimulation (4.50 ± 0.33 s) was significantly longer than freezing in interstimulation intervals (0.33 ± 0.22 s). Dots represent individual animals; N = 4.
Unlike other VTA neurons, VTA VGluT2+VGaT+ neurons were recently classified by their sensitivity to outcomes but not the learned predictors of those outcomes (Root et al., 2020). VGluT2::Cre/VGaT::Flp GCaMP mice that were previously stimulated in LH or SC VGluT2 neurons were trained to associate an auditory CS+ with footshock over 3 d, and GCaMP recordings of VTA VGluT2+VGaT+ neurons were conducted on a final day of CS+ footshock pairings (Fig. 8A). In a separate group of wild-type mice, this training resulted in significantly increased CS+-induced freezing over baseline precue freezing levels; t(9) = −2.379; p = 0.041 (Fig. 8B). VTA VGluT2+VGaT+ neuronal activity differed between the baseline, cue, and shock epochs with a main effect of time; F(2,14) = 7.673; p = 0.027. Pairwise comparisons showed that footshock significantly increased VTA VGluT2+VGaT+ neuronal activity over the baseline; F(1,7) = 7.684; p = 0.028; but there was no difference in activity between the baseline and CS+; F(1,7) = 0.205; p = 0.664 (Fig. 8C–E). Based on escape-related and freezing-related behaviors that can result from footshock exposure in rodents, we assessed whether a relationship existed between LH VGluT2 or SC VGluT2 stimulation-induced VTA VGluT2+VGaT+ neuronal activity and footshock-related activity of VGluT2+VGaT+ neurons. We compared 40 Hz stimulation because both LH VGluT2 and SC VGluT2 stimulation significantly maximally increased VGluT2+VGaT+ neuronal activity at this frequency. While SC VGluT2 neuron stimulation was not correlated with VGluT2+VGaT+ footshock-induced neuronal activity (r = 0.448; p = 0.552), LH VGluT2 neuron stimulation was highly correlated with VGluT2+VGaT+ footshock-induced neuronal activity (r = 0.983; p = 0.017; Fig. 8F). To further assess this relationship, we tested the role of LH VGluT2 neurons in generating the shock-induced response of VTA VGluT2+VGaT+ neurons. VGluT2::Cre/VGaT::Flp mice were injected in VTA with Cre- and Flp-dependent GCaMP6m and in LH with Cre-dependent inhibitory DREADD-mCherry (hM4Di) or soma-tagged GtACR2-FusionRed as a control (Fig. 8G–I). We chemogenetically inhibited LH VGluT2-DREADD neurons with a behaviorally subthreshold dose of 0.1 mg/kg clozapine (Gomez et al., 2017) during tailshock and measured the resultant GCaMP activity in VTA VGluT2+VGaT+ neurons. In control mice, tailshock significantly increased VTA VGluT2+VGaT+ neuron activity but in hM4Di DREADD mice tailshock did not increase VTA VGluT2+VGaT+ neuronal activity (interaction of epoch and treatment, F(1,7) = 16.23; p = 0.0012; Fig. 8J–K). Therefore, tailshock-induced activation of VTA VGluT2+VGaT+ neurons depends on LH VGluT2 neuronal activity. Together, these results support that VTA VGluT2+VGaT+ neurons signal outcomes and suggest that LH VGluT2 neurons play a role in VTA VGluT2+VGaT+ signaling of shock-related outcomes.
VTA VGluT2+VGaT+ neurons signal footshock outcomes that are correlated with LH VGluT2+ stimulation-induced VGluT2+VGaT+ activity. A, Schematic of viral injection and optical fiber implant for fiber photometry of VTA VGluT2+VGaT+ neurons from mice expressing ChRmine in either LH VGluT2 neurons or SC VGluT2 neurons. B, Wild-type mice show increased freezing behavior in response to a footshock-predicting CS+ compared with prior to CS+ presentation. Dots represent individual animals; N = 10. C, Normalized average maximum GCaMP response at the baseline, CS+, and footshock. Maximum GCaMP response was significantly higher at footshock compared with the baseline and CS+. Dots represent individual animals; N = 8. D, Normalized average GCaMP response at the CS+ onset (N = 8 animals). E, Normalized average GCaMP response at the footshock onset (N = 8 animals). F, VTA VGluT2+VGaT+ footshock-induced neuronal activity correlated with 40 Hz ChRmine stimulation-induced neuronal activity from LH VGluT2 neurons but not SC VGluT2 neurons. Dots represent individual animals; N = 8. G, Schematic for fiber photometry of VTA VGluT2+VGaT+ neurons in mice expressing inhibitory DREADD (hM4Di) or soma-tagged GtACR2 controls. Clozapine chemogenetically inhibits hM4Di but not stGtACR2 neurons. H, Example hM4Di expression in LH (left) and GCaMP expression in VTA (right). Scale bars, 200 µm. I, Cartoon of VTA optic fiber placements for hM4Di (red) and stGtACR2 (black) mice. J, Normalized average GCaMP response to clozapine and 0.3 mA tailshock (0.5 s duration). K, Tailshock increases GCaMP activity significantly above the baseline in control mice but not in hM4Di mice. Dots represent individual animals; N = 9.
LH VGluT2+ neuronal activation causes escape-like behavior. VGluT2::Cre/VGaT::Flp mice with ChRmine expression and bilateral optic fiber cannulae in LH. While tail-restricted in an unenclosed environment, a 1 s train of 20 Hz ChRmine stimulation induced foot treading (view from below) and supported rearing (view from the side) onto the hind legs. Responses began shortly after the laser onset and continued for an average of 1.74 ± 0.14 s. [View online]
SC VGluT2+ neuronal activation causes behavior resembling threat detection and avoidance. VGluT2::Cre/VGaT::Flp mouse with ChRmine expression and an optic fiber cannula in SC, view from above. While tail-restricted in an unenclosed environment, a 1 s train of 40 Hz ChRmine stimulation induced head orientation followed by a prolonged freezing bout. Head turn began immediately after the laser onset. Mice remained frozen for an average of 4.52 ± 1.68 s. [View online]
Discussion
It is well established that subsets of VTA neurons are capable of releasing one or more neurotransmitters (Stuber et al., 2010; Tritsch et al., 2012; Root et al., 2014b; Zhang et al., 2015). Based on the capability of VTA neurons to release one or more neurotransmitters, we hypothesized that the circuits and behavioral functions of VTA neurons depend on multiple genetic characteristics of neurotransmitter release. In support, VTA neurons that release glutamate without GABA (VGluT2+VGaT-), release GABA without glutamate (VGluT2-VGaT+), release both glutamate and GABA (VGluT2+VGaT+), or release dopamine (TH+) have distinct and partially overlapping axonal targets as well as unique neuronal activation profiles in response to rewarding or aversive stimuli and their learned predictors (Root et al., 2020). Here, we focused on the population of VTA neurons that release both glutamate and GABA (VGluT2+VGaT+ neurons) and are activated by rewarding or aversive outcomes (Root et al., 2020). We aimed to identify the circuit inputs that regulate VTA VGluT2+VGaT+ neuronal signaling and their contribution to behavior, which are currently unknown.
Monosynaptic rabies viral tracing with helper viruses dependent on Cre and Flp recombinases was used to identify brainwide inputs to VTA VGluT2+VGaT+ neurons. To determine the intersectional specificity of the rabies helper viruses, we injected VGluT2::Cre mice and VGaT::Flp mice with the same Cre- and Flp-dependent AAVs encoding TVA-mCherry and oG, followed by rabies-EGFP 3 weeks later. While we did not observe mCherry-labeled neurons or retrogradely labeled EGFP neurons throughout the brain, we observed EGFP-labeled neurons locally within the injection site. The number of locally labeled EGFP neurons was small, ∼2.55% of starter neurons in VGluT2::Cre/VGaT::Flp mice. Local EGFP labeling may have been due to leaky recombinase-independent expression of TVA and the phenomenon termed “invisible TVA” (Hafner et al., 2019). In the original description of intersectional rabies virus tracing, the authors report similar levels of leaky TVA expression in mice expressing only one of the two required recombinases, with an average of 10.0 and 4.0 rabies-EGFP cells in Cre-only mice and Flp-only mice, respectively. Furthermore, starter cells at the injection site may be “invisible” by expressing enough TVA-mCherry to allow the rabies-EGFP virus to transduce the cell, but not enough to express detectable levels of mCherry fluorophore. These cells would therefore appear to express only EGFP and would be presumed input neurons rather than starter neurons. The lack of rabies-EGFP cells outside the injection site supports leaky and invisible TVA but not leaky oG, since oG would allow rabies virus to spread transsynaptically from the starter cells, which was not observed in our single recombinase controls. Taken together, these results indicate minor off-target rabies-EGFP labeling of neurons likely expressing only Cre or only Flp in the injection site due to leaky, invisible TVA. Accordingly, the number of local VTA input neurons reported here may represent an overestimation of 1.51%. However, this off-target labeling at the injection site did not spread transsynaptically, due to the intersectional specificity of oG. This supports the assumption that retrogradely labeled input neurons counted outside the VTA correspond to on-target VGluT2+VGaT+ starter neurons.
Using monosynaptic rabies viral tracing of inputs to VGluT2+VGaT+ neurons, we found that most synaptic inputs arose from subcortical structures implicated in motor actions, outcome valuation, and threat responding, such as SC, LH, lateral habenula, periaqueductal gray, dorsal raphe, substantia innominata/ventral pallidum, locally within the VTA, and parabrachial nucleus (Matsumoto and Hikosaka, 2009; Bromberg-Martin et al., 2010; Hikosaka, 2010; Thevarajah et al., 2010; Benarroch, 2012; Cohen et al., 2012; Lecca et al., 2014; Root et al., 2015, 2020; Bonnavion et al., 2016; Stuber and Wise, 2016; Basso and May, 2017; Zahm and Root, 2017; Palmiter, 2018; Silva and McNaughton, 2019; Hoang and Sharpe, 2021). These circuits likely contribute to the high sensitivity of VTA VGluT2+VGaT+ neurons toward stressful stimuli that result in social and exploratory deficits in mice (McGovern et al., 2024). On the whole, brainwide inputs to VTA VGluT2+VGaT+ neurons were similar to those found examining all VTA VGluT2+ neuron inputs (Faget et al., 2016; An et al., 2021). Highly dense inputs arose from lateral habenula, dorsal raphe, locally within the VTA, and subdivisions of the periaqueductal gray. The highest density of inputs was defined by the Allen Atlas as the Edinger–Westphal nucleus. However, this labeled region best corresponds to subdivisions of the VTA involving the linear portions of the rostral linear nucleus and caudal linear nucleus that contain large numbers of VGluT2-expressing neurons (Root et al., 2018b). Further, because Edinger–Westphal nucleus neurons release peptides, but not neurotransmitters such as glutamate or GABA (Priest et al., 2023), it is unlikely that rabies transsynaptic tracing would label Edinger–Westphal nucleus neurons that use paracrine signaling. Thus, we consider the dense inputs within the labeled Edinger–Westphal nucleus as local VTA inputs, which were large in number and density.
Cortically, there were significantly fewer inputs to VTA VGluT2+VGaT+ neurons than subcortical inputs, which is consistent with inputs to other VTA cell types (Watabe-Uchida et al., 2012; Faget et al., 2016; An et al., 2021). More than half of cortical inputs were from the primary and secondary motor cortex, orbital cortex, and agranular insular cortex, which are implicated in action selection, outcome valuation, decision-making, and fear memory (Schoenbaum et al., 2011; Sul et al., 2011; Shi et al., 2020; Knudsen and Wallis, 2022). Integration of M2 inputs, the largest cortical input region onto VTA VGluT2+VGaT+ neurons, could represent a proprioceptive feedback circuit to ensure correct responding to relevant outcomes.
Based on our prior examinations of glutamate or GABA inputs to VTA VGluT2 neurons (McGovern et al., 2021) and the location of lateral hypothalamic and SC inputs to VTA VGluT2+VGaT+ neurons, we hypothesized these pathways were glutamatergic. Optogenetic stimulation of the LH VGluT2 neuron→VTA pathway results in real-time place avoidance and reduction of accumbal dopamine (Nieh et al., 2016). Further, fleeing behavior that is induced by a visual threat depends on a LH VGluT2 neuron→VTA VGluT2 neuron pathway (Barbano et al., 2020, 2024). We found that LH VGluT2 neurons robustly increased VTA VGluT2+VGaT+ neuronal activity across stimulation frequency parameters. LH VGluT2 stimulation also resulted in rearing and paw treading movements. All rearing behavior was supported rearing in which the mouse placed its front paws against the wall of the tail-restraint device. Unsupported rearing has been implicated in states of moderate anxiety (Sturman et al., 2018), but internal states related to supported rearing are less clear. We interpreted the supported rearing and paw treading movements as attempted flee or escape behavior. However, these behavioral reactions induced by LH VGluT2 stimulation were not correlated with the activation of VTA VGluT2+VGaT+ neurons by LH VGluT2 neurons. There are two important limitations of our excitatory optogenetic stimulation experiments. First, our readout of LH or SC VGluT2 neuron influence on VTA VGluT2+VGaT+ neurons was by changes in intracellular calcium (GCaMP) at the population level rather than changes in receptor-specific currents or firing patterns from single neurons. Second, LH VGluT2 neuron stimulation was at the soma and not pathway-specific. Although our rabies tracing experiments demonstrate the existence of synapses from LH to VTA VGluT2+VGaT+ neurons, these limitations reduce our ability to unequivocally conclude that LH VGluT2 neurons provide synapses to VTA VGluT2+VGaT+ that regulate behavior. Further, it is likely that LH VGluT2 neuron efferents that feedforward to VTA VGluT2+VGaT+ neurons influenced our GCaMP responses. Some possibilities include local connections from VTA VGluT2+VGaT− neurons that are innervated by LH VGluT2 neurons (Barbano et al., 2020) as well as LH VGluT2 neuron inputs to the lateral habenula (Lecca et al., 2017), which was among the densest inputs to VTA VGluT2+VGaT+ neurons.
Optogenetic stimulation of SC VGluT2 neurons significantly increased VTA VGluT2+VGaT+ neuron activity at 10, 20, and 40 Hz frequencies but not 5 Hz. Previous research has shown that optogenetic stimulation of the SC VGluT2→VTA pathway activates VTA dopamine neurons as well as VTA GABA neurons within the medial VTA where VGluT2+VGaT+ neurons are located, but not laterally where VGluT2-VGaT+ neurons predominate (Root et al., 2018b, 2020; Solie et al., 2022). We found that SC VGluT2 stimulation resulted in head turning followed by freezing, which resembled orientation toward and avoidance of detection by a threat (Dean et al., 1989; Bradley, 2009). In support of the role of the SC VGluT2→VTA pathway in signaling orientation toward threatening stimuli, the densest SC inputs were consistent with the location of glutamatergic Pitx2 SC neurons which drive head orientation (Masullo et al., 2019). However, head turns occurred before the peak activation of VTA VGluT2+VGaT+ neurons, and the activity of VGluT2+VGaT+ neurons did not correlate with movement parameters of the head turn, indicating VTA VGluT2+VGaT+ neurons are likely unnecessary for generating this head orientation response. Likewise, although the peak activation of VTA VGluT2+VGaT+ neurons was concurrent with freezing behavior, there was no correlation between VGluT2+VGaT+ activity and freezing behavior. Together with the results of LH VGluT2 neuronal stimulation, we found no evidence that the outcome signaling of VTA VGluT2+VGaT+ neurons is directly related to movement kinematics during outcomes. Experimental limitations on VTA readout (GCaMP) and somatic stimulation that was not pathway-specific limit our ability to unequivocally conclude that LH- or SC-induced behavior changes are completely uncorrelated from VTA VGluT2+VGaT+ neuron activity. However, because stimulation of both LH VGluT2 and SC VGluT2 neurons activated VTA VGluT2+VGaT+ neurons and LH VGluT2 neurons caused ambulatory escape-like behavior while SC VGluT2 neurons caused halted ambulatory behavior (freezing), a common feature of VTA VGluT2+VGaT+ neuronal activation between stimulation sites is a behavioral response consistent with the general interpretation of a threat.
We previously found that VTA VGluT2+VGaT+ neurons are highly sensitive to uncontrollable tailshocks that induce social avoidance, exaggerated fear, and reduced exploratory behavior (McGovern et al., 2024). Here we identify that chemogenetic inhibition of LH VGluT2 neurons regulate the tailshock-induced activation of VTA VGluT2+VGaT+ neurons. Though tailshock-induced activity of VTA VGluT2+VGaT+ neurons was dependent on LH VGluT2 neurons, because our chemogenetic inhibition was not pathway-specific, we cannot unequivocally conclude that a LH VGluT2 pathway to VTA controlled this effect alone. In addition to a direct pathway, one possibility is the robust LH VGluT2 innervation of LHb (Stamatakis et al., 2016; Rossi et al., 2021), of which LHb was a large and dense input to VTA VGluT2+VGaT+ neurons. Future research will be necessary to identify polysynaptic interactions between LH VGluT2 neurons, lateral habenula neurons, and VTA VGluT2+VGaT+ neurons as they relate to aversive experiences. In addition, because SC VGluT2 neurons were not chemogenetically examined, the role of these neurons in aversion or other behaviorally relevant signaling pattern of VTA VGluT2+VGaT+ neurons remains unknown.
In conclusion, these results provide novel insights into the cell-type–specific influences on VTA VGluT2+VGaT+ neuronal activity. We found that VTA VGluT2+VGaT+ neurons received large numbers of monosynaptic inputs from SC, LH, midbrain reticular nucleus, and periaqueductal gray, whereas the densest inputs were from dorsal raphe nucleus, LHb, and VTA. We also found that stimulation of LH VGluT2 and SC VGluT2 neurons each activated VTA VGluT2+VGaT+ neurons in vivo. LH stimulation more reliably resulted in activation of VGluT2+VGaT+ neurons across frequencies compared with SC stimulation. Our initial examination of these pathways revealed that LH VGluT2 and SC VGluT2 stimulation induced opposite ambulatory behavioral responses, and while some threat-response behaviors were concurrent with VGluT2+VGaT+ neuronal activity, no stimulation-induced behavioral kinematics correlated with VTA VGluT2+VGaT+ neuronal activity. Instead, we found that LH VGluT2 stimulation-induced activation of VTA VGluT2+VGaT+ neurons was correlated with footshock-induced activation of VTA VGluT2+VGaT+ neuronal activity, while SC VGluT2 stimulation did not correlate with footshock-induced activity. Further examination identified that tailshock-induced activation of VTA VGluT2+VGaT+ neurons was dependent on LH VGluT2 neuronal activity, which may result from direct or polysynaptic connections. From these data, we hypothesize that LH VGluT2 neurons mediate the sensitivity of VTA VGluT2+VGaT+ neurons to aversive outcomes but not in generating immediate aversion-related kinematics. The immediate threat-related kinematics may result from other projections of LH VGluT2 or SC VGluT2 neurons, whereas a generalized threat signal is integrated onto VTA VGluT2+VGaT+ neurons regardless of the motor behaviors that underlie them. VTA VGluT2+VGaT+ neurons likely relay signals related to aversive outcomes to LHb, its primary output (Root et al., 2014b, 2018b, 2020). However, it should be noted that VTA VGluT2+VGaT+ neurons are also activated by rewarding outcomes (Root et al., 2020), which suggests a wider role in outcome detection and salience processing that likely arises from regions outside of the LH and SC (McGovern et al., 2024).
Footnotes
This research was supported by the Webb-Waring Biomedical Research Award from the Boettcher Foundation (D.H.R), R01 DA047443 (D.H.R), F31 MH125569 (D.J.M), F31 MH132322 (A.L), a 2020 NARSAD Young Investigator Grant from the Brain and Behavior Research Foundation (D.H.R), The Curci Foundation (E.D.P), and The University of Colorado Boulder. The imaging work was performed at the BioFrontiers Institute Advanced Light Microscopy Core (RRID: SCR_018302). Laser scanning confocal microscopy was performed on a Nikon A1R microscope supported by National Institute of Standards and Technology-CU Cooperative Agreement award number 70NANB15H226. The PerkinElmer Opera Phenix is supported by National Institutes of Health Grant MH125569, 1S10OD025072, and MH132322. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Prism, MATLAB, brainrender, and BioRender.com were used to make figures and schematics. We thank Nicholas Steinmetz and Jun Takatoh for their help in density analyses.
The authors declare no competing financial interests.
L.E.F.’s current address: Department of Neuroscience, The University of Texas at Austin, Austin, Texas 78712, USA; Department of Psychiatry & Behavioral Sciences, Dell Medical School, The University of Texas at Austin, Austin, TX, 78712, USA.
- Correspondence should be addressed to David H. Root at David.Root{at}Colorado.edu.
