Abstract
The ventral pallidum (VP) is a central hub in the reward circuitry with diverse projections that have different behavioral roles attributed mostly to the connectivity with the downstream target. However, different VP projections may represent, as in the striatum, separate neuronal populations that differ in more than just connectivity. In this study, we performed in mice of both sexes a multimodal dissection of four major projections of the VP—to the lateral hypothalamus (VP→LH), ventral tegmental area (VP→VTA), lateral habenula (VP→LHb), and mediodorsal thalamus (VP→MDT)—with physiological, anatomical, genetic, and behavioral tools. We also tested for physiological differences between VP neurons receiving input from nucleus accumbens medium spiny neurons (MSNs) that express either the D1 (D1-MSNs) or the D2 (D2-MSNs) dopamine receptor. We show that each VP projection (1) when inhibited during a cocaine conditioned place preference (CPP) test affects performance differently, (2) receives a different pattern of inputs using rabies retrograde labeling, (3) shows differentially expressed genes using RNA sequencing, and (4) has projection-specific characteristics in excitability and synaptic input characteristics using whole-cell patch clamp. VP→LH and VP→VTA projections have different effects on CPP and show low overlap in circuit tracing experiments, as VP→VTA neurons receive more striatal input, while VP→LH neurons receive more olfactory input. Additionally, VP→VTA neurons are less excitable, while VP→LH neurons are more excitable than the average VP neuron, a difference driven mainly by D2-MSN-responding neurons. Thus, VP→VTA and VP→LH neurons may represent largely distinct populations of VP neurons.
Significance Statement
Many modern studies in neuroscience assign distinct behavioral roles to specific circuits. These behavioral roles are often attributed to the mere connectivity between two brain regions, although different projections may also differ in other aspects and may originate, like in the striatum, from largely separate neuronal populations. The ventral pallidum (VP) is a major structure of the reward system that sends projections to many different targets. In this work we provide for the first time a comprehensive and multimodal characterization of the four major outputs of the VP, with evidence for various differences between the projections. We also suggest that two of these projections, to the ventral tegmental area and the lateral hypothalamus, originate in largely separate neuronal populations.
Introduction
The ventral pallidum (VP) is a central hub in the basal ganglia and plays important roles in motivation, reward, and addiction (Smith et al., 2009; Root et al., 2015; Kupchik and Prasad, 2021). Despite its known general importance in reward-related behaviors, studies have attributed many different, sometimes contradicting, behavioral roles to the VP (Stratford et al., 1999; Mahler et al., 2014; Richard et al., 2016; Faget et al., 2018; Ottenheimer et al., 2018; Tooley et al., 2018; Gendelis et al., 2020; Heinsbroek et al., 2020; Levi et al., 2020). This suggests that the VP may not be a uniform structure but a complex collection of many neuronal subpopulations.
A characteristic of the VP that may underlie the complexity of this region is the numerous projections it sends to many targets within and outside the basal ganglia (Kupchik and Prasad, 2021). Thus, each projection may have behavioral functions that are different than those of other projections. For example, the VP projections to the lateral hypothalamus (LH), VP→LH, but not to the ventral tegmental area (VTA), VP→VTA, are important for renewal of alcohol drinking after abstinence (Prasad et al., 2020). The VP→VTA projections are also not involved in performance in the pavlovian-to-instrumental transfer task (Leung and Balleine, 2015) or in the tail suspension test (Knowland et al., 2017), while the VP projections to the mediodorsal thalamus (MDT), VP→MDT, or to the lateral habenula (LHb), VP→LHb, are implicated in these behaviors, respectively. Social behavior, on the other hand, requires the activity of the VP→VTA but not of the VP→LHb projection (Knowland et al., 2017).
It remains unknown whether different VP projections represent separate neuronal populations or rather all VP projections are similar in their cellular composition and their distinct functions depend on the downstream region they target.
The development of new technologies such as rabies-based tracing (Wickersham et al., 2007; Schwarz et al., 2015), single-cell RNA sequencing (Habib et al., 2017), and cell-specific optogenetics (Kim et al., 2017; Mardinly et al., 2018; Rost et al., 2022) or chemogenetics (Roth, 2016) has revealed in recent studies an increasing heterogeneity of neurons residing within the same brain region, defined by their connectivity patterns, gene expression, physiology, morphology, etc. (BRAIN Initiative Cell Census Network (BICCN), 2021; Wang et al., 2022). The high complexity of these methods has driven most studies to focus on one modality when analyzing the heterogeneity of neurons in a certain region. In particular, heterogeneity is often described either based on intracellular properties (gene expression, physiology, etc.) or on system-level connectivity, without integrating these two levels of complexity. As neurons are multimodal entities, there is a need for a multimodal interrogation of neuronal heterogeneity.
In this collaborative work, we integrate multilevel approaches to examine whether the different projections of the VP to the VTA, LH, MDT, and LHb are separate neuronal subpopulations based on their inputs, physiology, transcriptome, and behavioral relevance. Our data highlight the VP→VTA and VP→LH projections as highly distinct in these parameters, suggesting these two projections of the VP may be different subpopulations of VP neurons.
Materials and Methods
Experimental subjects
For electrophysiology, neural tracing, and behavioral experiments, subjects were naive C57bl6/J wild-type male and female mice that were ∼8 weeks old at the beginning of the experiments. A 12 h reversed light/dark cycle was maintained, with the lights turned off at 8:00 A.M. Experimental procedures were conducted during dark hours. Mice were group-housed and nesting/enrichment material was available. Experimentation began after a minimum of 7 d of acclimation to the animal facility. All experimental procedures were approved by the Authority for Biological and Biomedical Models in The Hebrew University of Jerusalem. For projection-specific gene expression analysis, adult (7–8 weeks old) male RiboTag (RT) mice (Rpl22tm1.1Psam/J) on a C57Bl/6 background were used. RT mice were given food and water ad libitum throughout the study. These studies were conducted in accordance with guidelines of the Institutional Animal Care and Use Committees at the University of Maryland School of Medicine.
Stereotaxic microinjections
Mice were anesthetized with isoflurane and fixed in a stereotaxic frame (Kopf, Model 940). Viruses or red RetroBeads™ (Lumafluor) were microinjected into one or more of the following structures—the MDT (coordinates in millimeters relative to the bregma: anterior/posterior (A/P) −1.5, medial/lateral (M/L) ±0.47, dorsal/ventral (D/V) −3.63), LHb (A/P −1.8, M/L ±0.7, D/V −2.91), LH (A/P −0.3, M/L ±0.9, D/V −5.5), VTA (A/P −2.9, M/L ±0.5, D/V −4.53), VP (A/P 0.7, M/L ±1.2, D/V −5.21), and nucleus accumbens (NAc; A/P 1.8, M/L ±1.0, D/V −4.6; see below for details on the injections in each experiment). Injections were performed bilaterally by drilling bilateral holes into the skull and then microinjecting the viral constructs [through a 33 ga NanoFil syringe (World Precision Instruments: 300 nl per hemisphere, 100 nl/min, needle retracted 5 min after injection terminated)] or RetroBeads [through a 30 ga syringe (Hamilton: 300 nl per hemisphere, 300 nl/min, needle retracted 5 min after injection terminated)] into the target structure.
Circuit-specific gene expression analysis was performed using retrograde AAV-Cre virus in combination with RT mice. Briefly, anesthetized RT mice were bilaterally injected with AAV5-Cre viruses (AAV sterotype 5 AAV5.hSyn.HI.eGFP-Cre.WPRE.SV40 (Addgene; #105540) at the following coordinates: VP (from the bregma; 10° angle, A/P +0.9, M/L ±2.2, D/V −5.3), MDT (from the bregma; 10° angle, A/P −0.8, M/L ±1.2, D/V −3.7), VTA (from the bregma; 7° angle, A/P −3.2, M/L ±1, D/V −4.6), LHb (from the bregma; 10° angle, A/P −1.2, M/L ±0.7, D/V −3.1), and LH (from the bregma; 10° angle, A/P −0.7, M/L ±1.4, D/V −5.2).
In all experiments, microinjection sites were verified visually using a Nikon Eclipse-NiE fluorescent microscope. Microinjections of fluorescent agents (RetroBeads for the electrophysiology experiments or retrograde viruses encoding a fluorescent marker for the double-labeling experiments) were verified by locating the center of the injection and examining the spread of the tracer. Microinjections of retrograde viruses that express Cre recombinase without a fluorescent marker (as in the behavioral, genetic, and rabies experiments) were verified by locating the needle tract. To estimate the spread of the virus in all these experiment, we injected, in separate mice, an identical virus but expressing eGFP instead of Cre (Fig. 5). Mice with injection center missing the center of the VP or in which >30% of the fluorescence was outside of the VP were excluded.
Double-labeling experiments
Each mouse was injected with two different retrograde tracers, retroAAV-GFP and retroAAV-RFP (ELSC viral core, The Hebrew University of Jerusalem). Each virus was injected bilaterally (300 nl per hemisphere) into one of the VP targets tested here. Two weeks after microinjections, mice were anesthetized with Pental (CTS Chemical Industries) and perfused with 4% paraformaldehyde (PFA). Sections of the VP (40 μm) were prepared using a sliding microtome (Leica, model SM2010R) and mounted on slides. The sections were scanned with Nikon ECLIPSE-NiE fluorescent microscope and scans were analyzed using ImageJ (National Institutes of Health).
Whole-brain mapping of inputs to VP projections
To label monosynaptic inputs to specific VP projections, we used the “Testing the Relationship between Input and Output” (TRIO) technique (Schwarz et al., 2015). Mice were first injected with a retrograde virus expressing Cre recombinase (retroAAV-Cre; ELSC viral core, The Hebrew University of Jerusalem) into one of the four targets of the VP examined here (300 nl). We also injected these mice with a cocktail of two helper viruses expressing the TVA receptor (AAV-FLEx-TVA-mCherry) and the rabies glycoprotein (AAV-FLEx-RG; total of 500 nl, 250 nl per virus) into the VP. Two weeks after these injections, we injected mice with the modified rabies virus (500 nl) into the VP. We allowed the virus to spread for 7 d before anesthetizing and perfusing the mice with PFA as described above. Brains were then postfixed with 4% PFA overnight at 4°C. Tissue clearing and staining were performed according to the published protocol (Renier et al., 2014). Briefly, brains were dehydrated, bleached to reduce background autofluorescence, rehydrated, and then incubated with rabbit anti-RFP antibody (1:1000, Rockland, catalog #600-401-379) and chicken anti-GFP antibody (1:2000, Aves Labs, catalog #GFP1010) for 5 d, followed by washing and incubation with Cy3-conjugated anti-rabbit secondary antibody (1:500, Jackson ImmunoResearch, catalog #711-165-152) and Cy5-conjugated anti-chicken secondary antibody (1:400, Jackson ImmunoResearch, catalog #703-175-155) for 5 d. Following additional washing, brains were dehydrated and cleared in methanol and dichloromethane, and then refractive index matching occurred in dibenzyl ether. Brains were imaged at 4× magnification, 0.35 numerical aperture, in the vertical orientation on a LaVision Ultramicroscope II light sheet microscope with a near-isotropic xyz resolution of 1.6 µm × 1.6 µm × 2 µm. Images were acquired with a 488 nm laser for an autofluorescence reference channel and a 561 nm and 642 nm laser for a secondary antibody fluorescence. The ClearMap Python package (www.github.com/christophkirst/clearmap) was used for cell detection and registration of cell coordinates onto the Allen Brain Atlas. Data analysis occurred on an Antec P110 Silent Workstation running Ubuntu 20.04LTS, and followed steps outlined in ClearMap documentation. Briefly, data were downsampled, and the Elastix toolbox (http://elastix.lumc.nl/; Klein et al., 2010; Shamonin et al., 2013) was used to perform automated 3D affine and B-spline transformation to register the autofluorescence signal channel to the 25 µm resolution Allen Brain Atlas reference and to correct for any motion between imaging of the autofluorescence and signal channel images. Elastix's “pixel classification” machine learning workflow was used to teach a model to segment input neurons from background, which later was used in Headless mode in ClearMap to detect input neurons. The Transformix module of the Elastix toolbox was then used to apply transformation vectors from the registration step to cellular coordinates, and cell counts for each region were calculated.
RNA sequencing and bioinformatics
Three weeks following viral injections, VP tissue was collected from RT mice infused with retrograde Cre virus (AAV5-Cre). Projection-specific RNA isolation was performed using polyribosome immunoprecipitation as described previously (Chandra et al., 2015; Engeln et al., 2021). Briefly, pooled tissue from RT mice with virally mediated Cre expression in VP projection neurons (n = 5 mice per sample) or in VP was homogenized, and 800 μl of the supernatant was incubated in HA-coupled magnetic beads (Invitrogen, 100.03D; Covance, MMS-101R) overnight at 4°C. Magnetic beads were washed using high salt buffer. Following TRK lysis buffer, RNA was extracted with the RNeasy Micro kit (Qiagen: 74004). Libraries were prepared from 10 ng of RNA using the NEBNext Ultra kit (New England Biolabs) and sequenced on an Illumina HiSeq 4000 with a 75 bp paired-end read. Four biological replicate samples per projection group type were used RNA sequencing as performed in previous studies (Engeln et al., 2022; Morais-Silva et al., 2023). RNA-sequencing data are available through the Gene Expression Omnibus database (accession number: GSE218580).
A total of 75–110 million reads were obtained for each sample. Reads were aligned to the mouse genome using TopHat, and the number of reads that aligned to the predicted coding regions was determined using HTSeq. Significant differential expression was assessed using limma by comparing VP projection neuron samples with the total VP. Genes with the absolute value of log fold change ≥0.3 and an uncorrected p-value <0.05 in the pairwise comparisons were used for downstream analysis (Fig. 6-1). Cytoscape 3.7.2 software was used for downstream analysis: transcriptional regulator networks were identified using the iRegulon app, and Gene Ontology (GO) functional enrichment analysis was performed using the BiNGO app (Fig. 6-2). From these lists, top GO terms were selected based off of the highest −log10(adjusted-p-value), and top upstream regulators were selected based on the number of predicted target differentially expressed genes (DEGs).
Slice preparation for whole-cell recordings
As described before (Gendelis et al., 2020; Inbar et al., 2020; Levi et al., 2020), mice were anesthetized with 150 mg/kg ketamine HCl and then decapitated. Coronal slices (200 µm) of the VP were prepared (VT1200S Leica vibratome) and moved to vials containing the following artificial cerebrospinal fluid (in mM): 126 NaCl, 1.4 NaH2PO4, 25 NaHCO3, 11 glucose, 1.2 MgCl2, 2.4 CaCl2, 2.5 KCl, 2.0 Na pyruvate, 0.4 ascorbic acid, bubbled with 95% O2 and 5% CO2 and a mixture of 5 mM kynurenic acid and 100 mM MK-801. Slices were stored in room temperature (22–24°C) until recording.
In vitro whole-cell recording
Recordings were performed at 32°C (TC-344B, Warner Instrument Corporation). VP neurons were visualized using an Olympus BX51WI microscope and recorded by using glass pipettes (1.3–2.5 MΩ, World Precision Instruments) filled with the following internal solution (in mM): 68 KCl, 65 d-gluconic acid potassium salt, 7.5 HEPES potassium, 1 EGTA, 1.25 MgCl2, 10 NaCl, 2.0 MgATP, and 0.4 NaGTP), pH 7.2–7.3, 275 mOsm. MultiClamp 700B (Axon Instruments) was used to record both membrane and action potentials and postsynaptic currents (PSCs) in whole-cell configuration. Excitability and passive membrane properties were measured in current-clamp mode, while synaptic activity was measured in voltage-clamp mode at a holding potential of −80 mV. Recordings were acquired at 10 kHz and filtered at 2 kHz using the AxoGraph X software (AxoGraph Scientific). To evoke inhibitory PSCs (IPSCs) from NAc terminals expressing ChR2, we used a 470 nm LED light source (Mightex Systems; 0.1–1 ms in duration) directed at the slice through the x60 objective. The stimulation intensity was set to evoke 50% of maximal IPSC at −80 mV. Recordings were collected every 10 s. Series resistance (Rs) measured with a −2 mV hyperpolarizing step (10 ms) given with each stimulus and holding current were always monitored online. Recordings with unstable Rs or when Rs exceeded 20 MΩ were aborted.
Current-clamp experiments
After penetrating the neuron, we switched to current-clamp mode and recorded the resting membrane potential of the neurons and spontaneous action potentials for 60 s. Cells with unstable membrane potential were discarded. Action potentials were later detected by their waveform using the AxoGraph software, baselined and analyzed. Potentials of <20 mV in amplitude were discarded. We then applied the current step protocol—five 500-ms-long depolarization current steps ranging from 0 pA to +80 pA (20 pA intervals) were applied—the inter-step interval was 3 s. The five-step protocol was repeated five times with 3 s between repetitions. Baseline membrane potential was adjusted to be ∼−50 mV in all neurons by injecting a constant current.
Cocaine conditioned place preference (CPP)
Mice were first injected with retroAAV-Cre in one of the four VP targets examined here and with an AAV encoding for the Gi-coupled designer receptor exclusively activated by designer drug (DREADD), hM4Di, in a Cre-dependent manner (AAV-DIO-hM4Di-mCherry), or a sham virus (AAV2-EF1a-DIO-EYFP-WPRE-pA) into the VP. After 2 weeks of acclimation to the reverse light cycle and recovery from surgery, the mice were trained in the unbiased cocaine CPP paradigm as described previously (Inbar et al., 2020; Fig. 1). On the first day, the mice were allowed to freely explore both sides of a 30 cm × 30 cm arena, divided into two, each side with a different wall pattern and floor texture. On the following days, the experimental mice received one daily injection of either cocaine (15 mg/kg, i.p.) or saline such that cocaine was always given in one side of the box (the “cocaine-paired” side) and saline in the other. Cocaine-paired sides were counterbalanced for pattern and side. Cocaine/saline injections alternated daily until each mouse received four of each. Then, the mice were left in their cages for 14 d (abstinence) before being tested for their preference of the cocaine-paired side. On the test day, the mice first received an injection of the DREADD ligand clozapine-N-oxide (CNO, 3 mg/kg, i.p.), were left in their home cages for 30 min and then put in the CPP box for 15 min. During the test, the mice were allowed to move freely between the two sides of the arena. Movement was tracked using the MediaRecorder (Noldus) and analyzed using the OptiMouse software (Ben-Shaul, 2017), and CPP scores were calculated off-line as the ratio between the difference in time spent between the cocaine-paired and unpaired sides and the total time [CPP score = (time in paired zone − time in unpaired zone) / (time in paired zone + time in unpaired zone)].
Inhibition of different VP projections affects cocaine preference after abstinence in different manners. A, Viral injection strategy. A retrograde AAV expressing Cre (retroAAV-Cre) was injected into one of the four targets of the VP tested here and an AAV encoding Cre dependently for the inhibitory DREADD hM4Di (AAV-DIO-hM4Di-eYFP) or a control sham virus (AAV-DIO-eYFP) were injected into the VP. B, Cocaine CPP paradigm. After habituating to the CPP box, mice received daily alternating injections of cocaine (15 mg/kg i.p.) in one side of the box or saline in the other side of the box (4 injections of each). Mice then underwent 14 d of abstinence in their home cages. On the test day, all mice first received an injection of CNO (3 mg/kg i.p.) 30 min prior to the beginning of the test and then were left in the CPP box for 15 min, during which their movement was recorded. A CPP score was calculated using the following equation:
Statistics and data analysis
All statistical analyses were performed using the GraphPad Prism 9.2 (GraphPad Software). Parametric tests were used unless stated otherwise with p values <0.05 considered as significant. All tests are indicated in the figure legends. Data in graphs represent mean ± SEM.
Results
Different roles for VP projections in cocaine CPP
The VP is central in processing cocaine reward, and our previous data show synaptic plasticity in the VP following cocaine CPP and abstinence (Inbar et al., 2020, 2022; Levi et al., 2020). However, it is not known whether some VP projections are more relevant than others for cocaine preference after abstinence. To examine this, we inhibited each of the four VP projections we chose to examine here—VP→MDT, VP→LHb, VP→LH, and VP→VTA—during the preference test performed after cocaine CPP and 14 d of abstinence. To achieve this, we injected a retrograde AAV encoding for Cre recombinase (retroAAV-Cre) in the target area to express Cre in the specific VP projection neurons and expressed the inhibitory version of the DREADDs in a Cre-dependent manner in the VP (AAV-DIO-hM4Di-GFP; control mice were injected with a sham virus, AAV-DIO-eYFP) (Fig. 1A; for accuracy of injections, see Fig. 5). We then trained mice to associate one side of the CPP box with cocaine using the cocaine CPP paradigm (Fig. 1B; see Materials and Methods) and then allowed them to undergo abstinence in their home cages for 14 d as we did before (Inbar et al., 2020, 2022; Levi et al., 2020). Thirty minutes before the CPP test, we injected all mice with CNO (3 mg/kg, i.p.) and then tested their preference to the cocaine-paired side.
Our data show that inhibiting the different VP projections had different effects on the preference of the cocaine-paired side [Fig. 1C–H; one-way ANOVA, main projection effect, F(3,25) = 5.81, p = 0.004]. In particular, the average CPP score when inhibiting the VP→LH projection (0.34 ± 0.19) was significantly lower than the CPP score when inhibiting the VP→VTA [0.63 ± 0.10; Tukey's multiple comparisons test, q(25) = 5.3, p = 0.005] or VP→MDT [0.61 ± 0.15; Tukey's multiple comparisons test, q(25) = 4.7, p = 0.01] projections. Comparison of the CPP scores to that of the control sham virus group (average CPP score of 0.51) revealed that inhibiting the VP→LH projection significantly reduced the CPP score [Dunnett's multiple comparison test, q(40) = 2.6, p = 0.04]. Inhibiting the other VP projections did not change the CPP score significantly compared with the sham group. Inhibiting the different projections did not change the overall locomotion of the mice compared with each other or to the sham group [Fig. 1I; one-way ANOVA, F(4,39) = 0.4, p = 0.83]. Thus, our data demonstrate that different VP projections have different roles in renewal of cocaine CPP after abstinence and highlight the different and maybe opposite roles of the VP→LH projection compared with the VP→VTA and VP→MDT projections in cocaine preference.
Partial overlap between VP projection neurons
Our finding here (Fig. 1) and previous studies (Leung and Balleine, 2015; Prasad et al., 2020) showing that different VP projections have different behavioral roles raises the possibility that these different projections represent separate groups of neurons. To test the degree of overlap between the four VP projections examined here, we performed six experiments examining the overlap between all possible pairs of the four targets of the VP. For each pair of combination, we injected one retrograde virus expressing GFP (retroAAV-GFP) into one of the targets and another retrograde virus expressing RFP (retroAAV-RFP) into the other target (Fig. 2A,B; for accuracy of injections, see Fig. 5). Then we examined the proportion of the double-labeled yellow neurons out of all labeled neurons in the VP (Fig. 2C). Our data show that overlap between projections ranged between 7 and 20%, with the VP→LH projections showing the lowest overlap with other projections, particularly with VP→VTA projections (7%).
Partial overlap between ventral pallidal projections to the MDT, LHb, LH, and VTA. A, Experimental setup. We injected each mouse with two retrograde viruses expressing either RFP (retroAAV-RFP) or GFP (retroAAV-GFP), each in a different target of the VP. Thus, we could identify neurons either those projecting to target or projecting to both (yellow). B, Sample images of labeled VP→MDT neurons, VP→LH neurons, and double-labeled neurons projecting to both targets. C, Pie charts depicting for each pair of VP targets the proportion of double-labeled VP neurons projecting to both targets out of all labeled neurons (arranged from lowest to highest proportion of double-labeled neurons). The proportion of VP neurons projecting to both targets tested was in the range of 7–20%. Note that VP→LH neurons showed the lowest proportions of overlap, especially with VP→VTA neurons. Three hemispheres were tested for each of the six pairs of projections. D, Experimental setup of the optogenetic study. Mice were injected with retroAAV-Cre in the LH and AAV-DIO-ChR2 in the VP to express ChR2 in VP→LH projections. Neurons from all four targets of the VP were patched, and the infected terminals were activated optogenetically to identify responding neurons in all four targets. E, Percentage of neurons in each region (out of all patched neurons) receiving synaptic input from axons of VP→LH neurons.
The low level of overlap between VP→LH and VP→VTA neurons supports a previous study showing different behavioral roles for these projections (Prasad et al., 2020) but contradicts another study using anterograde tracing (Tripathi et al., 2013) that showed that most of the VP axons, regardless of their final target, pass through the LH. Since our data may be biased by the efficiency of retrograde viral transport or microinjection localization in the two injection sites, we took a second approach to examine the level of overlap between VP→LH neurons and the other projections. In this approach (Fig. 2D), we injected the LH with retroAAV-Cre and the VP with a virus expressing the optogenetic sodium channel channelrhodopsin 2 (ChR2) in a Cre-dependent manner (AAV-DIO-ChR2). Thus, only VP→LH neurons are expected to express ChR2. We then used patch-clamp electrophysiology and optogenetics to activate VP→LH axons and detect the synaptic input in each of the four targets of the VP we examine here. Our data show (Fig. 2E) that two out of five neurons (40%) in the VTA responded to the activation of VP→LH neurons, while only one out of five MDT neurons and no LHb neurons responded. For positive control, five out six LH neurons responded to VP→LH activation. Thus, the overlap between VP projections may be higher than that observed when using two retrograde tracers. Nevertheless, VP→LH neurons may have less overlap with other projections than the average VP projection neuron.
Analysis of the localization of the specific VP projection neurons in the VP revealed that these neurons are intermingled throughout the VP, and neither projection shows a unique localization in the VP (Fig. 3).
Localization of VP projection neurons in the VP. Projections to the MDT, LHb, LH, and VTA were labeled retrogradely as described in Figure 2. A, Location of projection neurons inside the VP from 1.10 anterior to the bregma to −0.34 mm posterior to the bregma. AC, anterior commissure; ACA, anterior commissure, anterior limb; ACP, anterior commissure, posterior limb; BST, bed nucleus of the stria terminalis; HDB, horizontal diagonal band; IC, internal capsule; IPAC, interstitial nucleus of the posterior limb of the anterior commissure; MCPO, medial preoptic area; NAcC, nucleus accumbens core; NAcS, nucleus accumbens shell; SI, substantia innominata; VP, ventral pallidum. B, Cell count for each projection across the VP sections shown in A. A two-way ANOVA analysis showed no main projection effect [F(3,120) = 5.04 × 10−18, p > 0.99]. Three hemispheres were tested for each projection.
Different input patterns for different VP projections
The four VP projections we test here differ, by definition, in the target they project to. However, they may also receive their inputs from different sources, thus generating parallel segregated circuits. To test whether the VP projections differ in their input patterns, we used the TRIO method (Schwarz et al., 2015), which allows labeling of monosynaptic inputs to a specific projection neuron (Fig. 4A). We first injected a retrograde virus encoding Cre recombinase (retroAAV-Cre) into either of the four VP targets (i.e., VTA, LH, MDT, or LHb) and then viruses that encode for the TVA receptor and the rabies glycoprotein in a Cre-dependent manner into the VP. Thus, only VP neurons projecting to the site injected with the retroAAV-Cre would express the TVA receptor and the rabies glycoprotein (“starter cells,” Fig. 5). Lastly, we injected the modified rabies virus (Osakada and Callaway, 2013) into the VP and tested neuron labeling using the iDISCO + method (Renier et al., 2014) and light sheet fluorescence microscopy (Fig. 4A,B).
Distinct monosynaptic input patterns between different ventral pallidal projections. A, To visualize the monosynaptic inputs to each ventral pallidum projection, we injected a retrograde virus expressing Cre (retroAAV-Cre) into one of the targets of the VP and the rabies helper viruses expressing in a Cre-dependent manner the avian tumor receptor A (AAV-FLEx-TVA) and the rabies glycoprotein (AAV-FLEx-G) into the VP. Fourteen days later, we injected the pseudotyped rabies virus (RVdG) into the VP to label all monosynaptic inputs to a certain VP projection. We then cleared the brains with iDISCO+ and imaged the cleared brains with light sheet fluorescence microscopy (LSFM). B, Top: LSFM image of labeled neurons in the NAc. Bottom: magnification of the neuron marked by a dotted square in the top image. C, Proportions of the major inputs to the four VP projections. Colored ring surrounding pie charts indicates grouping of subparts of the same brain region into the regions used in panel D. Although the four VP projections receive input from the same brain regions, the proportions of each input vary between the projections. D, The proportions of specific inputs between the four VP projections are significantly different. The VP→LH projections show several differences from the VP→VTA projections, including stronger olfactory input but weaker striatal and amygdalar inputs. *p < 0.05; **p < 0.01; ***p < 0.001. Three hemispheres were tested for each projection.
Injection sites and labeling of inputs to VP projections using the TRIO method. To label the monosynaptic inputs to each of the four VP projections we tested here, we injected a retrograde AAV expressing Cre into the MDT, LHb, LH, or VTA (left column; here we depict injection of retroAAV-GFP pseudocolored to cyan to demonstrate injection accuracy and viral spread) and viruses expressing in a Cre-dependent manner the TVA receptor linked to mCherry and the rabies glycoprotein RG into the VP. We also injected the VP with the modified rabies virus (RVdG) expressing GFP. Neurons in the VP that were infected with both the TVA receptor (red) and the RVdG (green) were considered as starter cells (yellow; second column from left, arrows point to starter cells). Rabies-labeled neurons in the BNST, LPO, and NDB are shown in the three rightmost columns.
Our data show that although all four VP projections receive input from largely the same sources, they differ in the relative contribution of each input (Fig. 4C,D, Table 1). The main inputs to all VP projections originated in the hypothalamus, striatum, amygdala, and olfactory areas, with lesser contribution from the midbrain, hippocampus, and cortex.
Number of rabies-labeled neurons in main inputs to each VP projection
Some inputs were more pronounced on specific projections [Fig. 4D; two-way ANOVA, main effect for input source, F(9,80) = 23.8, p < 0.0001; interaction (input source × VP projection) effect, F(27,80) = 3.42, p < 0.0001]. For example, hypothalamic inputs were less dominant in VP→LHb neurons [Tukey's multiple comparison test, q(80) = 4.0, p = 0.03 compared with VP→LH, q(80) = 4.3, p = 0.02 compared with VP→MDT], while olfactory inputs were significantly stronger on VP→LH projections [Tukey's multiple comparison test, q(80) = 5.3, p = 0.002 compared with VP→LHb, q(80) = 7.2, p < 0.0001 compared with VP→VTA, q(80) = 5.6, p = 0.001 compared with VP→MDT], and striatal inputs (mainly from the NAc) were specifically enriched in VP→VTA projections [Tukey's multiple comparison test, q(80) = 4.3, p = 0.02 compared with VP→LH, q(80) = 3.9, p = 0.03 compared with VP→MDT]. Interestingly, the VP→LH and VP→VTA projections, which show opposite influence on cocaine preference after abstinence (Fig. 1C) and low overlap (Fig. 2C), also differ in their inputs, the VP→LH projections receiving more olfactory and less striatal and amygdalar inputs compared with VP→VTA projections.
Distinct molecular signatures within VP projection neurons
Our data so far support the possibility that VP projections, at least those to the LH and the VTA, may originate in largely distinct sets of neurons. We thus aimed to examine whether these sets of VP projection neurons exhibit distinct profiles at the molecular level. Although studies have identified the neurochemical composition of VP projection neurons using in situ hybridization (Faget et al., 2018; Tooley et al., 2018; Vachez et al., 2021), we profiled global translatome-wide signatures within VP projection neurons. This allows us to identify patterns and novel genes enriched within VP projection neurons. To this end, gene expression profiles of VP projection neurons were assessed using RNA sequencing (Fig. 6A). Retrograde Cre virus (AAV5-Cre) was infused either into the VTA, LHb, LH, MDT, or VP. This resulted in Cre-dependent expression of HA-tagged ribosomal protein within upstream VP neurons. Infusion of AAV5-Cre directly into the VP served as a comparison of nonpathway-specific VP (global VP) gene expression. VP tissue was collected from mice, and RT methods were used to isolate pathway-specific ribosome-associated mRNA from VP neurons. RNA-sequencing libraries were constructed from each VP projection group and followed by translatome profiling (Fig. 6A).
VP projection neurons have distinct molecular signatures. A, Retrograde Cre (AAV5-Cre) was infused into one of the four downstream VP regions (MDT, LHb, LH, VTA) or VP within RT male mice (n = 16 mice per projection). HA-immunoprecipitation procedures were employed to isolate RNA from distinct VP projection neurons or all VP neurons. RNA-sequencing libraries were generated and analyzed from these samples to characterize baseline gene expression profiles within VP neurons projecting to the MDT, Lab, LH, and VTA. B, DEG analysis was generated by comparing gene expression within specific VP projection neuron types relative to the global VP gene expression (uncorrected p < 0.05). Transcriptional patterns within each VP projection neuron type are shown in a heat map. Full list of DEGs is given in Figure 6-1 and the associated GO terms in Figure 6-2. C, RRHO analysis comparing gene expression in a threshold-free manner suggests VP→MDT/VP→LH and VP→MDT/VP→LHb projection neurons have the highest concordance of gene expression patterns.
Figure 6-1
Differentially-expressed genes within each type of VP projection neurons. Download Figure 6-1, XLSX file.
Figure 6-2
Gene Ontology terms associated with upregulated DEGs within each type of VP projection neuron. Download Figure 6-2, XLSX file.
First, pairwise comparisons of gene expression patterns from VP→VTA, VP→LHb, VP→LH, and VP→MDT were compared with gene expression within nonprojection-specific VP neurons (Fig. 6B). The following number of DEGs was detected in each VP projection neuron: VP→MDT, 4,318; VP→VTA, 1,982; VP→LHb, 1,709, and VP→LH, 1,308. To further characterize global gene expression patterns, we performed rank–rank hypergeometric overlap (RRHO) analysis (Plaisier et al., 2010), which compares gene expression between two lists in a threshold-free manner. Heat maps display overlap at those points, determined by relative effect sizes in differential gene expression using −log10(p-value). Comparisons between gene expression lists of all six possible pairs of VP projections were performed (Fig. 6C). High concordance is seen within genes upregulated within VP→LH versus VP→MDT in relation to VP global neurons [peak: −log10(p-value), 541] and VP→LHb versus VP→MDT [peak: −log10(p-value), 334]. High concordance is also observed within downregulated genes in VP→VTA versus VP→LH neurons [peak: −log10(p-value), 393]. These data overall suggest that VP→MDT has the highest number of upregulated genes and exhibits the highest overlap of upregulated gene expression patterns among other VP projection neuron types.
To identify cell-type-specific molecules and processes, additional analysis was performed on genes upregulated (i.e., enriched) within the VP projections. VP→MDT neurons had the highest number of upregulated DEGs (708) in comparison with other VP neuron types (VP→VTA, 489; VP→LH, 454; VP→LHb, 402). Upregulated genes were largely distinct from one another (Fig. 7A), with only 66 upregulated genes shared among all VP neuron types. Consistent with the patterns detected from RRHO analysis, VP→MDT and VP→LH projection neurons shared the highest number of upregulated genes (267 total), and VP→LH and VP→LHb have little gene overlap (133 total).
GO analysis identifies biological processes and molecular functions enriched within VP projection neurons. A, Genes upregulated within each VP projection neuron type were compared in Venn diagrams. The VP→MDT projection neuron has the highest number of enriched genes, with VP→MDT and VP→LH sharing the highest number of enriched genes. B, Top GO terms enriched in upregulated genes from each VP projection neuron. Note that all projections are enriched with GABA-related genes but only some with glutamate or dopamine-related genes. bind., binding; biosyn., biosynthetic; cytosk., cytoskeletal; DAR, dopamine receptor; GABAR, GABA receptor; GABARA, GABA-A receptor; GluR, glutamate receptor; org., organization; path., pathway; proc., processes; prot., protein; reg., regulation; secr., secretion; sig., signaling; syn., synaptic. C, Expression of GABAergic and glutamatergic cell-type marker genes from each VP projection neuron.
To characterize the biological processes and cellular functions associated with genes enriched within each VP neuronal type, GO analysis was performed (Fig. 7B). In this analysis, we focused on the neurotransmitters that the VP projections release and respond to. For example, while VP neurons releasing glutamate were shown to generate aversion through their projection to the LHb (Faget et al., 2018), it is not known whether the VP→LHb projection is particularly rich with glutamatergic neurons. The GO analysis provides evidence for distinct afferent and efferent neurotransmitter release among VP projection types. VP→VTA neurons have genes enriched for regulation of glutamate secretion and regulation of hormone secretion, whereas VP→LHb neurons are enriched for dopamine biosynthetic and metabolic processes. GO term analysis also suggested that VP→LH neurons have a higher proportion of glutamatergic neurons, with enrichment for genes associated with regulation of synaptic glutamate transmission. In terms of afferent inputs, all VP projection neuron types showed enrichment for genes related to GABA receptor activity, indicating GABAergic input to all projection neuron types. However, only DEGs from VP→VTA and VP→LHb projection neurons showed enrichment for dopamine receptor activity, with dopamine receptor signaling pathway and dopamine-binding GO terms, respectively. Similarly, only VP→VTA neurons had enrichment of glutamate receptor activity, suggesting stronger glutamatergic input to these projections. Expression of marker genes for known VP cell types was assessed in each projection type (Fig. 7C). Enrichment of markers for GABAergic cell types (Slc32a1, GAD2) was seen in VP→LH and VP→MDT projection neurons, suggesting these projections are engaged more than others in GABA neurotransmission. The marker gene for the VP glutamatergic neuron Slc17a6 (encoding for vGluT2) was particularly enriched in VP→LH neurons, corroborating our GO term analysis above suggesting high proportion of glutamatergic neurons in this projection. All projections were depleted of the glutamatergic marker Slc17a7, encoding for vGluT1. Proenkephalin, a gene marker for a subset of GABAergic VP cell types was found highly enriched only in VP→VTA projection neurons. Overall, the enriched genes and biological processes within each projection neuron type provide evidence for circuit specificity within VP projection neurons.
Genes critical for ion transport enriched within VP→LH and VP→VTA neurons
Our data so far highlight the VP→LH and VP→VTA projections as being potentially distinct neural populations. We therefore directly compared the differentially enriched genes between these two projections. We found 156 genes enriched within both VP→LH and VP→VTA neurons (Fig. 8A). GO term analysis illustrated that the 156 genes belonged to processes related to neuron development, synaptic transmission, GABA signaling pathways, chromatin assembly, and ion transport (Fig. 8B). This is consistent with the previous work identifying VP GABAergic projections to the LH and VTA (Faget et al., 2018; Prasad et al., 2020). Among the genes that are differentially expressed in these two projections, several are associated with synaptic transmission and cellular physiology, including genes encoding GABA receptors, ion channels, and transporters. A predicted transcription analysis on the genes that are differentially expressed in both VP→LH and VP→VTA highlights RFX3 as a top predictor regulator of several shared genes. This presents a possible target for shared genes within VP→VTA and VP→LH neurons.
Distinct sets of genes related to ion transport are enriched within VTA-projecting and LH-projecting VP neurons. A, Upregulated genes within VP→LH and VP→VTA projection neurons were compared in a Venn diagram. GO analysis identified common synaptic genes (B) in VP→VTA and VP→LH projection neurons (LFCs from each projection type displayed in a heat map) and RFX3 as a predicted upstream regulator of a subset of genes related to ion transport. GO term analysis reveals distinct sets of ion transport-related genes and their listed potential upstream regulators within the VP→LH (C) and VP→VTA (D) projection neurons. neur. devel., neural development; ion transmem. transp. act., ion transmembrane transporter activity; ext.-gly, gated ion ch. act, extracellular-glycine-gated ion channel activity; monov. inorg. cat. transp., monovalent inorganic cation transporter; reg. of synaptic trans., regulation of synaptic transmission; cell. metab. processes, cellular metabolic processes.
To identify genes and biological processes distinct to each neuronal type, GO term analysis was performed on the unshared genes within VP→LH (298 genes) and VP→VTA (333 genes) neurons. Interestingly, both sets of DEGs were associated with processes related to synaptic transmission and ion transport, which may lead to differences in the excitability between these projections (Fig. 8C,D). For example, both neuron types have enriched expression of potassium channels, despite distinct families expressed within each neuronal type. VP→LH neurons show increased expression of channels from the Kv3 family (KCNC1, KCNC2), while VP→VTA neurons show increased expression of other potassium channel subtypes (KCND2, KCNAB1). Since Kv3 channels are known to be important for high-frequency firing (Rudy and McBain, 2001; Lien and Jonas, 2003; Kaczmarek and Zhang, 2017), the increased expression of KCNC1,2 in VP→LH neurons may suggest higher excitability in these neurons.
Predicted transcription factor analysis was performed on synaptic genes enriched within VP→LH and VP→VTA neurons. We found that RNF138 was a top predicted regulator for a subset of VP→LH ion transport genes, and DLX6 was a top predicted regulator for a subset of ion transport genes within VP→VTA neurons. Altogether, these data demonstrate that VP→VTA and VP→LH neurons have sets of ion transport genes that are enriched within each projection neuron type and present novel molecular targets to study VP projection-specific neuronal function.
The VP→MDT projection receives less inhibitory input compared with the other VP projections
The projections of the VP may differ not only in their effect on cocaine preference (Fig. 1), input patterns (Fig. 4), or gene expression (Figs. 6–8) but also in the characteristics of the synaptic inputs they receive. To examine this, we used the whole-cell patch-clamp technique and recorded from each of the four VP projections the global spontaneous inputs they receive (Fig. 9A,B).
Projection-specific differences in the global synaptic input to VP neurons. A, Schematic representation of the experimental setup. In each mouse, a retrograde tracer (RetroBeads; red dots) was injected into one of the four targets of the VP examined here, and patch-clamp recordings were performed from retrogradely identified VP projections. B, Representative micrographs of RetroBeads injection sites. f, fornix; Hyp, hypothalamus; V3, 3rd ventricle. C–F, Groups consist of 24–28 cells from 8 to 10 mice per group. C, Cumulative probability of sPSC amplitudes in the four VP projections. There were no differences between the curves. D, Cumulative probability of the frequency of sPSCs in the four VP projections. The VP→MDT neurons showed significantly lower frequencies compared with the other projections. *p < 0.05 for MDT compared with each of the other projections. Inset: representative traces. E,F, Application of 100 μM picrotoxin (GABA-A receptor blocker) on the slice significantly reduced the frequency of spontaneous inputs in all projections (E), but the reduction was smaller for the VP→MDT projections (reduction to 47.2 ± 7% of baseline) compared with each of the other projections (F). Neurons were recorded from four to eight males and two to four females per group (on average 2.95 cells per male, 3.2 cells per female). *p < 0.05 compared with VP→MDT neurons.
Comparing the characteristics of the spontaneous postsynaptic currents (sPSCs; these include both inhibitory and excitatory currents) between the projections revealed that while they do not differ in the amplitude of the sPSCs (Fig. 9C; KS test, p > 0.16 for all pairwise comparisons between projections), the VP→MDT projection shows a significantly lower frequency of sPSCs compared with the other projections (Fig. 9D; KS test; d = 0.13, p = 0.0001 compared with VP→LHb; d = 0.16, p < 0.0001 compared with VP→LH; d = 0.14, p < 0.0001 compared with VP→VTA). This difference may be due to less GABAergic input because when we washed the slices with picrotoxin (100 μM), a GABA-A receptor antagonist, the reduction in the frequency of sPSCs was significantly smaller in VP→MDT neurons (52.8 ± 33% reduction) [Fig. 9E,F; one-way ANOVA main projection effect, F(3,79) = 4.06, p = 0.01; post hoc Tukey's multiple comparison—compared with VP→LHB (21.1 ± 3% of baseline), p = 0.001, compared with VP→LH (25.7 ± 3% of baseline), p = 0.02, compared with VP→VTA (26.9 ± 5% of baseline), p = 0.03]. Thus, our data indicate that the VP→MDT projection receives less overall inhibitory input than the other projections.
Synaptic input differences between projections that receive NAc D1-MSN or D2-MSN input
We and others have shown that VP neurons receive two parallel NAc GABAergic inputs—from medium spiny neurons (MSNs) expressing the D1 (D1-MSNs) or the D2 (D2-MSNs) dopamine receptor (Kupchik et al., 2015; Creed et al., 2016; Liu et al., 2022)— and that these two inputs convey different behavioral effects (Heinsbroek et al., 2017; Pardo-Garcia et al., 2019; Liu et al., 2022). We thus wanted to examine whether the differences in the synaptic input between VP projections we observed above also exist within the subpopulations of VP neurons receiving input from NAc D1-MSNs (D1-responders) or D2-MSNs (D2-responders).
We identified specific VP projections using retrograde tracing and expressed ChR2 Cre dependently in D1-MSNs (in D1-Cre mice) or D2-MSNs (in A2A-Cre mice) to identify optogenetically in brain slices D1-responders (Fig. 10A–J) or D2-responders (Fig. 10K–T). The proportions of neurons responding to D1-MSNs or D2-MSNs were largely similar between projections (Fig. 10B,L; average 73% responding for D1-MSNs, 69% responding for D2-MSNs), although almost all (91%) VP→MDT projections received D1-MSN input and a relatively high proportion (82%) of VP→LHb projections received D2-MSN input. We also did not find differences, in both D1- (Fig. 10C–F) and D2-responders (Fig. 10M–P), in the amplitude of optogenetically evoked inhibitory postsynaptic currents (oIPSCs) [one-way ANOVA, no main projection effect, F(3,36) = 0.48, p = 0.7 for D1-responders; F(3,39) = 0.96, p = 0.42 for D2-responders] or in the paired-pulse ratio (PPR) [one-way ANOVA, no main projection effect, F(3,36) = 0.32, p = 0.81 for D1-responders; F(3,39) = 1.83, p = 0.16 for D2-responders] or the coefficient of variation (CV) [one-way ANOVA, no main projection effect, F(3,36) = 0.45, p = 0.72 for D1-responders; F(3,80) = 0.54, p = 0.66 for D2-responders] of the oIPSCs [PPR and CV are two measures reflecting the probability of release; Berninger et al., 1999; Schinder et al., 2000].
Projection-specific differences in the synaptic input to VP neurons contacted by NAc D1-MSNs or D2-MSNs. A–J, Experiments in D1-Cre mice, examining VP projection neurons receiving input from NAc D1-MSNs. Groups consist of 8–12 cells from three to four mice per group. A, Schematic representation of the experimental setup. A retrograde tracer (RetroBeads) was injected into one of the four targets of the VP examined here, and ChR2 was expressed Cre dependently in NAc D1-MSNs of D1-Cre mice. Recordings were performed from identified VP projections that showed optogenetically evoked synaptic input from NAc D1-MSNs. Inset: representative micrograph of ChR2 expression in the NAc. B, The proportions of VP projections receiving synaptic input from NAc D1-MSNs. C–E, The amplitude (C), paired-pulse ratio (D), and coefficient of variation (E) of the evoked oIPSC did not differ between projections. F, Application of 100 μM picrotoxin (GABA-A receptor blocker) on the slice completely blocked the oIPSCs in all VP projections, indicating these are GABA-only inputs. Inset: representative traces of oIPSCs before (baseline) and after application of picrotoxin. G, Cumulative probability of sPSC amplitudes in the four VP projections receiving D1-MSN input. There were no differences between the curves. H, Cumulative probability of the frequency of sPSCs in the four VP projections. The VP→MDT and VP→LHb neurons showed significantly lower frequencies compared with VP→LH and VP→VTA projections. *p < 0.01 for VP→MDT or VP→LHb compared with either VP→LH or VP→VTA. I,J, Application of 100 μM picrotoxin on the slice significantly reduced the frequency of spontaneous inputs in all projections (I), but the reduction was smaller for the VP→MDT projections (reduction to 55.3 ± 12% of baseline) compared with all other projections (J). *p < 0.05; #p < 0.05 (one-way ANOVA). K–T, Experiments in A2A-Cre mice, examining VP projection neurons receiving input from NAc D2-MSNs. Groups consist of 9–13 cells from three to four mice per group. K, Schematic representation of the experimental setup. RetroBeads were injected into one of the four targets of the VP examined here, and ChR2 was expressed Cre dependently in NAc D2-MSNs of A2A-Cre mice. Recordings were performed from identified VP projections that showed optogenetically evoked synaptic input from NAc D2-MSNs. L, The proportion of VP projections receiving synaptic input from NAc D2-MSNs. M–O, The amplitude (M), paired-pulse ratio (N), and coefficient of variation (O) of the evoked oIPSC did not differ between projections. P, Application of 100 μM picrotoxin on the slice completely blocked the oIPSCs in all VP projections, indicating these are GABA-only inputs. Q, Cumulative probability of sPSC amplitudes in the four VP projections receiving D2-MSN input. The amplitude of sPSCs recorded in VP→LHb neurons was larger than that recorded in the other projections. R, Cumulative probability of the frequency of sPSCs in the four VP projections receiving D2-MSN input. The VP→VTA and VP→LHb neurons showed significantly higher frequencies compared with VP→LH and VP→MDT projections. *p < 0.01 for VP→VTA or VP→LHb compared with either VP→LH or VP→MDT. S,T, Application of 100 μM picrotoxin on the slice significantly reduced the frequency of spontaneous inputs in all projections (S), but the reduction was strongest for the VP→LHb projections (reduction to 14.5 ± 3% of baseline) (T). *p < 0.05; #p < 0.05 (one-way ANOVA). Neurons were recorded from one to four males and one to two females per group (on average 3.2 cells per male, 3.3 cells per female).
The lower GABA spontaneous input to VP→MDT projections seen in the general VP population was observed also in the subpopulation of D1-responders. This is evidenced by the lower frequency of sPSCs in VP→MDT D1-responders compared with VP→LH (KS test; d = 0.17, p = 0.0001) or VP→VTA (KS test; d = 0.16, p = 0.0004; Fig. 10H) together with a weak suppressing effect (to 55.3 ± 11% of baseline) of picrotoxin [Fig. 10J; one-way ANOVA main projection effect, F(3,30) = 4.25, p = 0.01; post hoc Tukey's multiple comparison—compared with VP→LHB (27.9 ± 6% of baseline), p = 0.07, compared with VP→LH (26.3 ± 6% of baseline), p = 0.08, compared with VP→VTA (17.4 ± 5% of baseline), p = 0.02]. The VP→LHb D1-responders, as the VP→MDT D1-responders, showed a low frequency of sPSCs compared with VP→LH (KS test; d = 0.15, p = 0.0005) or VP→VTA (KS test; d = 0.13, p = 0.005; Fig. 10H). However, VP→LHb D2-responders showed high sPSC frequency (Fig. 10R; KS tests; cumulative distribution compared with VP→MDT, KS test; d = 0.53, p < 0.0001; compared with VP→LH, d = 0.46, p < 0.0001) together with large sPSC amplitude (Fig. 10Q; KS tests; cumulative distribution compared with VP→MDT, KS test; d = 0.37, p = 0.04; compared with VP→LH, d = 0.33, p = 0.1; compared with VP→VTA, d = 0.41, p = 0.02). Since our chloride-rich internal solution favors large GABA chloride-mediated currents but not glutamate sodium-mediated currents, these data suggest that VP→LHb D2-responders receive stronger GABA input than the other D2-responding projections. This is also corroborated by the finding that picrotoxin showed the strongest decrease in sPSC frequency (to 14.5 ± 3% of baseline) in VP→LHb D2-responders [Fig. 10T; one-way ANOVA, main projection effect, F(3,36) = 0.3.55, p = 0.02; post hoc Tukey's multiple comparison—compared with VP→MDT (43.3 ± 10% of baseline), p = 0.02, compared with VP→LH (27.1 ± 6% of baseline), p = 0.6, compared with VP→VTA (32.6 ± 7% of baseline)], p = 0.2).
VP→LH neurons are more excitable and VP→VTA neurons are less excitable than the average VP neuron
We next tested various properties that contribute to the excitability of the VP projection neurons (Fig. 11A). The resting membrane potential (Fig. 11B) did not differ between VP projections [one-way ANOVA, F(3,103) = 0.49, p = 0.69] and ranged between −45.8 and −47.2 mV on average. In contrast, the firing rate at rest (Fig. 11C,D) did differ between projections [one-way ANOVA, F(3,103) = 3.47, p = 0.02]. In particular, the VP→LH neurons had the highest firing rate (10.9 ± 8.1 Hz), which was significantly higher than the firing rate of VP→MDT [6.1 ± 4.8 Hz, post hoc Holm–Sidak's multiple comparison test, t(109) = 2.9, p = 0.03] and VP→VTA neurons [6.5 ± 5.2 Hz; post hoc Holm–Sidak's multiple comparisons test, t(109) = 2.6, p = 0.049].
VP→LH neurons are more excitable and VP→VTA neurons are less excitable than the average VP neuron. A, Experimental setup. VP projections were labeled by injecting a retrograde tracer (RetroBeads) in the target region and recorded from using whole-cell patch clamp electrophysiology. B,C, Groups consist of 22–31 cells from eight to nine mice per group. B, Resting membrane potential was not different between the four VP projections we examined. C, Action potential firing frequency at rest differed between the different VP projections. Post hoc analyses show that VP→LH neurons fire in higher frequencies than VP→MDT and VP→VTA neurons. *p < 0.05; #p < 0.05 (one-way ANOVA). D, Representative traces. E,F, Groups consist of 24–33 cells from seven to nine mice per group. When applying a series of increasing depolarizing steps (inset, 20, 40, 60, and 80 pA), we found a main projection effect both in the number of action potentials per step (E) and in the minimal latency to fire the first action potential (F). Note that VP→LH neurons had among the highest number of action potentials and shortest latency to fire, both of which indicate high excitability. #p < 0.05 (ANCOVA) (G) Representative traces. H, An “excitability index” was calculated by first calculating for each cell a z-score in each of the parameters recorded in panels B,C,E,F and then calculating the average z-score for each cell. There was a main projection effect for the z-scores, and post hoc analyses revealed that the VP→LH and the VP→VTA projections were significantly different in their excitability. In addition, the excitability index of VP→LH neurons was significantly higher than zero, while that of VP→VTA neurons was significantly lower than zero. Thus, VP→LH neurons are more excitable and VP→VTA neurons are less excitable than the average VP neuron. Groups consist of 24–37 cells from eight to nine mice per group. Neurons were recorded from three to seven males and two to five females per group (on average 3.05 cells per male, 2.9 cells per female). *p < 0.05 compared with zero; #p < 0.05 (one-way ANOVA).
When we applied a series of depolarizing current steps to the VP projections (Fig. 11E–G), we found that the projections differed both in the number of action potentials they generated per step [analysis of covariance (ANCOVA), F(3,260) = 1.06, p = 0.37 for slope comparison; F(3,260) = 6.1, p = 0.0005 for elevation comparison] and in the latency to the first action potential [ANCOVA, F(3,260) = 4.65, p = 0.003 for slope comparison]. Note that the VP→LH projections, which had the highest firing rates at rest (Fig. 11C), also showed strong firing and quick response when depolarized.
To evaluate the overall excitability of the projections, we generated an “excitability index.” In each of the measured parameters, we pooled all cells together and applied z-scores to all cells. Thus, each neuron had four z-scores (one from each experiment), and the excitability index of a neuron is the average of its four z-scores. Comparison of the excitability indexes between VP projections (Fig. 11H) demonstrated that it differed between projections [one-way ANOVA, F(3,146) = 3.21, p = 0.03] and that the VP→LH and VP→VTA projections had significantly different excitability indexes, 0.25 and −0.21, respectively [Tukey's multiple comparison test, q(146) = 4.3, p = 0.02]. Moreover, when comparing the excitability indexes to zero, which reflects the average excitability of a VP neuron, the average excitability index of the VP→LH neurons was significantly higher than zero, while that of VP→VTA neurons was significantly lower than zero [one-sample t tests compared with zero, t(36) = 2.18, p = 0.035 for VP→LH; t(31) = 2.15, p = 0.039 for VP→VTA]. Thus, our data suggest that VP→LH neurons may be a subtype of VP neurons that is more excitable than the average VP neuron, while VP→VTA neurons may be a different subtype of VP neurons that is less excitable than the average VP neuron. This, together with the rest of our current findings, supports the hypothesis that VP→LH and VP→VTA neurons may be largely different subpopulations.
Excitability differences between VP projections in D1-responders and D2-responders
Using the same tools to identify D1- and D2-responding VP projections as in Figure 10 (Fig. 12A,F), we examined whether the difference in excitability between the VP projections is preserved also in D1-responders (Fig. 12A–E,K) or D2-responders (Fig. 12F–J,K). The membrane potential did not differ between projections in either D1- or D2-responders [Fig. 12B,G; one-way ANOVA—for D1-responders, F(3,33) = 0.38, p = 0.77; for D2-responders, F(3,26) = 0.43, p = 0.73], as we also saw in the general population. The firing frequency, which was significantly higher in VP→LH neurons at the general population, did not differ between projections in D1- or D2-responders [one-way ANOVA—for D1-responders, F(3,33) = 1.52, p = 0.23; for D2-responders, F(3,26) = 0.94, p = 0.44], although in both subpopulations, the VP→LH projections showed the highest average firing rate (12.0 Hz for D1-responders and 11.2 Hz for D2-responders).
Excitability of VP projections does not differ between D1- and D2-responders. A–E, VP projections receiving NAc D1-MSN input. A, Experimental setup. Excitability was recorded from VP projections (identified by retrograde labeling) that showed postsynaptic currents to optogenetic activation of NAc D1-MSNs in D1-Cre mice. B,C, Groups consist of 5–17 cells from three to four mice per group. Resting membrane potential (B) and resting firing frequency (C) were not different between the four VP projections we examined among D1-responders. D,E, Groups consist of 10–19 cells from three to four mice per group. When applying a series of increasing depolarizing steps (20, 40, 60, and 80 pA), we found a main projection effect both in the number of action potentials per step (D) and in the minimal latency to fire the first action potential (E). #p < 0.05 (ANCOVA). F–J, VP projections receiving NAc D2-MSN input. F, Experimental setup. Excitability was recorded from VP projections (identified by retrograde labeling) that showed postsynaptic currents to optogenetic activation of NAc D2-MSNs in A2A-Cre mice. G,H, Groups consist of 7–8 cells from three to four mice per group. Resting membrane potential (G) and resting firing frequency (H) were not different between the four VP projections we examined among D2-responders. I,J, Groups consist of 7–8 cells from three to four mice per group. When applying a series of increasing depolarizing steps (20, 40, 60, and 80 pA), we found a main projection effect in the number of action potentials per step (I) but not in the minimal latency to fire the first action potential (J). #p < 0.05 (ANCOVA). K, An “excitability index” was calculated as explained in Figure 11 for D1-responders and D2-responders separately. In a two-way ANOVA test (projection × input), there was a main projection effect (#p < 0.05) but no main input effect. Note that VP→LH projections showed high excitability indexes in both D1- and D2-responders, while VP→VTA projections showed low excitability only in D2-responders. Post hoc analyses revealed that the difference in excitability between VP→LH and the VP→VTA projections that was seen in the general VP population was preserved only in D2-responders. *p < 0.05. Groups consist of 8–19 cells from three to four mice per group. Neurons were recorded from one to three males and one to two females per group (on average 2.8 cells per male, 2.6 cells per female).
Applying a series of depolarizing current steps to the VP projections to test how strongly (APs/step) and quickly (average latency to the first action potential) the projections respond, we found that the difference between projections in APs/step that was seen in the general population existed also in D1-responders [Fig. 12D; ANCOVA, F(3,59) = 5.35, p = 0.003 for elevation comparison] and D2-responders [Fig. 12I; ANCOVA, F(3,87) = 3.26, p = 0.03 for elevation comparison]. In both, VP→LH and VP→VTA projections showed the highest rate of firing, as in the general population. The difference in the latency to the first action potential seen in the general VP population was significant only in D1-responders [Fig. 12E; ANCOVA, F(3,197) = 2.85, p = 0.04 for slope comparison] but not D2-responders (Fig. 12J). Nevertheless, note that in both D1- and D2-responders, the VP→LH projections were among the quickest to respond, while the VP→VTA projections were the slowest.
We lastly calculated the excitability index for the D1-responders and D2-responders (separately) and compared it between projections in both subpopulations (Fig. 12K). A two-way ANOVA test showed a main projection effect [F(3,86) = 3.29; p = 0.02] but no main input effect [F(1,86) = 2.82; p = 0.1] or input × projection interaction [F(3,86) = 0.98; p = 0.4]. As in the general population, the VP→LH projections showed positive average excitability values in both D1- and D2-responders, although this did not reach significance as in the general population [one-sample t tests compared with zero, t(12) = 1.63, p = 0.13 for D1-responders; t(8) = 1.3, p = 0.23 for D2-responders]. The VP→VTA projections, which were hypoexcitable in the general VP population, showed this hypoexcitability only in the D2-responders [one-sample t tests compared with zero, t(9) = 0.6, p = 0.56 for D1-responders; t(8) = 2.12, p = 0.06 for D2-responders]. The significant difference in excitability between VP→LH and VP→VTA projections that was found in the general population was also seen in D2-responders [Tukey's multiple comparison test, q(86) = 1.3, p = 0.03] but not D1-responders.
Discussion
The VP is a structure largely composed of GABAergic neurons and has numerous downstream projection targets. It has been assumed that the cellular composition of the different projections is largely similar, despite the VP's projection-specific roles in behavior. In this manuscript, we provide for the first time a thorough multimodal examination of the four major outputs of the VP and compare directly their genetic profile, input sources, physiological characteristics, and roles in cocaine motivation after withdrawal. We find that the different VP projections show distinct characteristics in all properties examined. Each projection has distinct sets of enriched genes (Figs. 6–8), receives different patterns of inputs from various regions of the brain (Fig. 4), exhibits different levels of excitability and inhibitory input (Figs. 9–12), and has differential roles in regulating cocaine CPP behavior after abstinence (Fig. 1). In particular, the VP projections to the LH and VTA stand out as being largely distinct cell populations. They show a minimal level of projection overlap (Fig. 2), have differences in input sources (Fig. 4), express different sets of enriched genes (Fig. 8), have different excitability levels (Fig. 9), and show opposite effects on cocaine CPP (Fig. 1).
VP heterogeneity
Our data suggest that the four projection neurons of the VP examined here differ from each other in multiple aspects, potentially suggesting some of these projections contain distinct subtypes of VP neurons. This supports previous studies attributing different behavioral roles to distinct VP outputs (Leung and Balleine, 2015; Tooley et al., 2018; Prasad et al., 2020; Pribiag et al., 2021; Vachez et al., 2021; Engeln et al., 2022). Examining expression of gene markers from VP projection neurons (Fig. 7) revealed that VP→MDT and VP→LH projection neurons have enriched GABAergic gene marker expression, whereas VP→VTA neurons show high expression for the pENK gene, which is expressed in a subset of GABAergic VP neurons (Heinsbroek et al., 2020; Soares-Cunha and Heinsbroek, 2023). VP→LH projection neurons were the only ones with high expression of Slc17a6, encoding vGluT2. Consistent with this, glutamatergic VP neurons were observed to project to the LH to regulate wakefulness (Luo et al., 2023). However, our GO term analysis indicates that both VP→VTA and VP→LH neurons show enrichment for genes related to glutamatergic release and previous studies implicate glutamatergic VP neurons projecting to the VTA or to the LHb in behavior regulation (Faget et al., 2018, 2023; Tooley et al., 2018). Thus, while the VP→LH projection may have a higher proportion of glutamatergic neurons, these neurons exist and have strong behavioral effects also in other VP projections.
We surprisingly detected enrichment for dopamine biosynthesis in our VP→LHb projection neurons. This high expression of tyrosine hydroxylase (TH) and dopamine transporter relative to bulk VP neurons may indicate dopaminergic cell types present in the VP. However, follow-up studies examining the presence of dopaminergic markers at the level of protein will be required, as other brain regions, including the LHb, are known to express TH mRNA, but not at the level of protein (Quina et al., 2009).
We did not detect any differential expression of cholinergic markers (ChAT, AchE) in our projection neurons. About 10% of the neurons in the VP are cholinergic (Kupchik and Prasad, 2021), but it is possible that the projections examined here do not express cholinergic markers as these were so far shown for other VP projections, such as to the basolateral amygdala (Root et al., 2015; Ji et al., 2023). Nevertheless, given that our comparisons are relative to bulk VP samples, the projection neurons we examined may project cholinergic markers at relatively similar levels.
Our GO term analysis suggests that all projection types highly express GABA-related receptors, which would be consistent with our finding that 70% of the neurons in the VP indeed receive synaptic inhibitory input from either NAc D1-MSNs or D2-MSNs (Fig. 10), similar to estimations in previous studies (Kupchik et al., 2015; Creed et al., 2016). Nevertheless, our viral tracing experiments show that more striatal neurons converge on VP→VTA neurons than on VP→LH or VP→MDT neurons (Fig. 4), and our RNAseq experiments indicate that VP→VTA and VP→LHb neurons may receive more dopamine input than others as they are enriched for dopaminergic receptors (Fig. 7). The VP→VTA projections also have particularly high expression of glutamate receptor activity, implying stronger glutamatergic input into these neurons.
Altogether, this is the first set of data to examine cell-type differences of VP projection neurons using unbiased transcriptome-wide analysis. By combining this analysis with physiological and anatomical data, we have confirmed known cell types within different VP projections, revealed new cell types potentially present in these projections, and described differences and similarities between the projections.
Input–output interaction in the VP
The identity of VP neurons may be determined not only by their projection target but also by the input they receive from the NAc. Our data in Figures 10 and 12 indicate that the interprojection differences in the synaptic input and membrane properties may differ when examining VP neurons receiving D1-MSN input or D2-MSN input. For example, the VP→LH projection shows higher excitability than the average VP neuron in both D1- and D2-responders, but the low excitability of VP→VTA projections is driven mainly by D2-responders (Figs. 11, 12). Also, the synaptic input to VP→LHb projections tested in the general VP population (Fig. 9) is not different than the other projections, but this is changed in D1- and D2-responders. VP→LHb projections of D1-responders receive significantly less synaptic input than other projections, while in D2-responders they receive more input than others (Fig. 10H,R). The high frequency input to D2-responding VP→LHb neurons is mostly GABAergic (Fig. 10T), suggesting that this projection is under stronger tonic inhibition compared with other D2-responding projections.
The difference between VP→LH and VP→VTA projections
Our results highlight the VP→LH and the VP→VTA projections as being potentially distinct from each other. Chemogenetic inhibition of these populations resulted in different behavioral effects (Fig. 1); they show a low level of overlap (Fig. 2), different input patterns (Fig. 4), and possible differential sensitivity to and release of different neurotransmitters (Fig. 7), and a large number of upregulated genes within these populations are nonoverlapping (Fig. 8). We also detected differences in excitability, with VP→LH neurons being more excitable than VP→VTA neurons (Fig. 10), possibly due to specific upregulation of genes encoding for the Kv3 potassium channel family, known to allow high firing frequencies (Rudy and McBain, 2001; Lien and Jonas, 2003; Kaczmarek and Zhang, 2017).
The differences we report on VP→LH and VP→VTA neurons challenge several previous notions of the VP. First, an anatomical study suggests that ∼90% of VP projections, including those to the VTA, pass through the LH (Tripathi et al., 2013). In fact, the LH is positioned between the VP and the VTA (Paxinos and Franklin, 2001). Our data suggest that the VP→VTA projections do not collateralize to the LH and are strictly passing through the structure. Our CPP work does demonstrate that the VP exerts projection-specific effects on behavior, as previously seen (Prasad and McNally, 2016; Faget et al., 2018; Tooley et al., 2018; Pribiag et al., 2021). However, there are conflicting reports on the effects of inhibiting VP projection neurons on drug seeking. These differences may be due to different cell-type-specific manipulations, as Prasad et al. (2020) had found that inhibition of VP→LH GABAergic neurons, but not VP→LH parvalbumin-expressing neurons, reduces renewal of operant-based alcohol seeking in rats. This may also be due to species-specific effects, as inhibition of VP→VTA neurons has previously been shown to enhance cocaine seeking in mice (Pribiag et al., 2021), but impair cocaine seeking in rats (Mahler et al., 2014). Alternatively, it is possible that different circuits and cell types are affected due to extinction training (Mahler et al., 2014; Prasad et al., 2020) and forced abstinence (Pribiag et al., 2021). Further investigation into these underlying differences is required to better understand the role of VP in drug seeking.
The mechanism allowing the VP→LH and the VP→VTA projections to have different behavioral roles in drug reward after abstinence is still not understood. One possibility is potential opposite effects on dopamine release from VTA neurons. VP→VTA neurons target both GABA and dopamine neurons (Watabe-Uchida et al., 2012; Hjelmstad et al., 2013; Levi et al., 2020). It is still not known how activation of the VP input to the VTA affects dopamine levels, but it is plausible that it enhances dopamine release as it increases the firing rate of putative dopamine neurons (Mahler et al., 2014). LH GABAergic neurons that project to the VTA increase firing of VTA dopamine neurons (Nieh et al., 2016). These neurons are also the main target of VP→LH projections (Jennings et al., 2013). Thus, activation of VP→LH neurons, which are mostly GABAergic (Kupchik and Prasad, 2021), may result in a net decrease in dopamine release. Therefore, VP→VTA and VP→LH are expected to have opposite effects on dopamine release from the VTA. This may underlie their opposite effect on cocaine CPP.
Altogether, our multimodal interrogation of the VP demonstrates that VP projection neurons have relatively distinct pathways, cellular properties, and molecular compositions. These baseline differences may reflect their projection-specific roles of cocaine-related behaviors. Our behavioral findings illustrate how VP projection neurons have differential roles in cocaine reward, which is consistent with the previous work in the field. This work emphasizes the importance of further investigating and characterizing the VP projections and cell types in addiction processes.
Methodological considerations
In drawing conclusions from this study, we must emphasize two methodological considerations that may limit the conclusions. The first limitation is the double-retrograde staining approach we used to identify overlap between VP projection neurons (Fig. 2). This method highly depends on the precise location of injection and the efficiency of viral expression. It is possible, for example, that many VP→VTA neurons also project to a specific subregion of the LH, but in our injection of the retrograde virus in the LH, we hit that subregion only partially, resulting in smaller overlap than in reality. We also show that when activating VP→LH neurons optogenetically, 40% of VTA neurons show direct synaptic input, thus calling for more significant overlap between VP→LH and VP→VTA neurons. In addition, we only tested pairs of VP projections and not all four projections in the same mouse. Since we found 15–20% overlap between most pairs of projections (Fig. 2), it is possible that most VP neurons project to more than one target.
The second limitation of our study regards to the behavioral experiments and conclusions. We demonstrate that inhibiting different VP projections has different effects on cocaine CPP after abstinence from cocaine (Fig. 1). While this suggests that different VP projections may have different behavioral roles, it is only a preliminary attempt at understanding the different behavioral roles of VP projections, and many future experiments are still required. For example, it is still not known whether the interprojection differences stem from the mere exposure to cocaine or from the abstinence process. We also do not show whether the behavioral differences are seen when enhancing the activity of the VP projections, for example, with Gq-DREADDs, or when testing for preference of natural rewards. Therefore, the behavioral conclusions that can be made from our experiments are partial, and future studies are required to thoroughly understand the different behavioral relevance of the various VP projections.
Footnotes
This study was supported by the U.S.–Israel Binational Science Foundation (Grant No. 2015239 and 2017252 to Y.M.K. and M.K.L.), Israel Science Foundation (Grant No. 1381/15 and 1117/21 to Y.M.K.), IMRIC Center for Addiction Research (ICARe), and National Institutes of Health (RO1DA047843 and U01DA051947 to M.K.L. and F32DA056191 to R.C.).
↵*N.B. and R.C. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Yonatan M. Kupchik at yonatan.kupchik{at}mail.huji.ac.il or Mary Kay Lobo at mklobo{at}som.umaryland.edu.
