Projections from ventral hippocampus to nucleus accumbens’ cholinergic neurons are altered in depression

The nucleus accumbens (NAc) is a large nucleus at the ventral part of the striatum where signals from a wide variety of sensory and motor inputs converge. The NAc functions as a funnel through which information must pass and where it is finely modulated (Zahm and Brog, 1992; O’Donnell et al., 1997). The NAc circuitry is centrally involved in the assessment of reward. It selects and reinforces the behaviors that lead to a positive outcome and abandons or modifies those that result in non-productive activities, and thus has a major role in the generation of motivated behaviors (Zahm, 2000; Meredith et al., 2008). The disrupted functioning of this network has been linked to the incapacity of experiencing pleasure (anhedonia), as seen, for example, in depression (Price and Drevets, 2010; Russo and Nestler, 2013). Thus, mapping complex NAc neuronal networks with other brain regions would be fundamental for understanding the behavioral changes in mood disorders and other neurological diseases. NAc connectivity was examined using the morphological and electrophysiological approaches, as well as non-selective anterograde and retrograde tracers (Thompson and Swanson, 2010; Britt et al., 2012; Bagot et al., 2015; Christoffel et al., 2015; LeGates et al., 2018). These classical studies showed that the main brain areas that project to the NAc are the prefrontal cortex (PFC), ventral hippocampal subiculum (vHIP/Sub), basolateral amygdala (BLA), ventral tegmental area (VTA), and thalamic nuclei (Smith et al., 2004; Sesack and Grace, 2010). These areas reciprocally interact through excitatory glutamatergic inputs of the cortico-striato-pallidal-thalamo-cortical loop and by the dopaminergic projections from the VTA and substantia nigra (SN) in the mesolimbic dopaminergic pathway. Although the connections of the NAc have been well characterized, the direct cell–cell interaction with other brain areas is still not well understood. Studies showed that many diseases, major depressive disorder among them, have a strong cell- and region-specific component (Warner-Schmidt et al., 2012), making it necessary to decipher the cell type–specific circuitry between the key brain regions. The gross anatomy of inputs to numerous brain areas has been studied using traditional tracers (Gerfen and Sawchenko, 1984; Bolam et al., 2000; Thompson and Swanson, 2010), but these techniques cannot distinguish connectivity to specific cell types. This has become achievable with the advent of techniques that allow cell-specific connectivity mapping (Wall et al., 2010), enabling the study of the processes in only one cell type without dissecting it out from its natural surroundings, preserving all the connections and processes that exist in vivo. Monosynaptic circuit tracing with the CRE-dependent modified rabies virus, developed by the Callaway laboratory (Wall et al., 2010) and modified by the Uchida laboratory (Watabe-Uchida et al., 2012), circumvent this limitation by using mouse lines that express CRE in a cell-specific manner. With the introduction of this cell-based approach for mapping connectivity inputs to different cell types in the dorsal striatum, including dopaminergic receptor 1 and 2 medium spiny neurons (MSNs) (Wall et al., 2013), and cholinergic interneurons (ChIs) (Guo et al., 2015; Klug et al., 2018) were elucidated. However, the cell-specific inputs to the ventral striatum have not been studied.

Striatal ChIs are giant aspiny interneurons that represent only about 0.3% of all striatal neurons (Rymar et al., 2004). Despite their under-representation, the tonically active ChIs have dominant modulating control over inputs and outputs of the main population of neurons in NAc, the medium spiny neurons (Russo and Nestler, 2013; Virk et al., 2016). Indeed, optogenetical inhibition of ChIs in the NAc increased medium spiny neurons activity and blocked the rewarding effects of cocaine (Witten et al., 2010). Chemogenetic and optogenetic manipulation of NAc ChIs interneuron activity opposed the motivating influence of appetitive cues, further emphasizing their importance for cue-motivated behaviors (Collins et al., 2019). These cells have been also shown to have a critical role in regulating depressive-like behaviors in mice (Alexander et al., 2010; Warner-Schmidt et al., 2012; Cheng et al., 2019). One of the molecular mechanisms of depressive-like behaviors is the alterations in expression levels of the calcium-binding protein S100a10 (also called p11), shown by our laboratory and others (Svenningsson et al., 2006; Zhang et al., 2011; Hanada et al., 2018; Cheng et al., 2019). Mice with a complete lack of p11 exhibit anhedonia and despair, but this depressive-like behavior can be reiterated with the region-specific knock down in NAc (Alexander et al., 2010), and specifically in the NAc ChIs (Warner-Schmidt et al., 2012; Hanada et al., 2018). Notably, this behavior appears to be governed specifically by the NAc and not caudate putamen/dorsal striatum (CPu) ChIs, since selective p11 knock down in CPu ChIs did not reproduce the depressive phenotype.

For all these reasons, it is important to examine the connectivity of these neurons to gain a better understanding of cell-specific circuitries regulating normal and pathologic behavioral outcomes. Here, we utilized the CRE-dependent monosynaptic modified rabies virus method and mapped all the brain regions and cells with direct inputs to NAc ChIs. We corroborated known areas with direct inputs to NAC but also found areas that project only to NAc but not the CPu ChIs. To trace the direct inputs to ChIs in the context of depression, we used ChAT-CRE mice with the depressive phenotype (ChAT-p11 cKO). We found that in ChAT-p11 cKO mice a significantly lower number of pyramidal cells from the ventral hippocampus (vHIP) projects to the NAc ChIs. Together, these data demonstrate the inputs to ChI neurons from a large number of discrete areas scattered throughout the brain and provide a possible mechanism of how accumbal ChI circuitry regulates depression and other NAc-related behaviors.

Adult ChAT-CRE (GM60) mice obtained from GENSAT (https://www.gensat.org; gift from Dr. Nathaniel Heintz, The Rockefeller University). We utilized cell-type–specific p11 knockout mouse line (ChAT-p11 cKO) as a model of depression by breeding the p11 floxed mice to ChAT-CRE GM60 mice, as previously described (Svenningsson et al., 2006; Warner-Schmidt et al., 2012). In addition, this mouse line was also bred to GENSAT ChAT-TRAP line (Doyle et al., 2008). Additionally, the Cux2-CRE line was used (kindly donated by Ulrich Muller, Johns Hopkins School of Medicine, Baltimore, MD, USA). Camk2a-CRE line was obtained from the Jackson laboratory (B6.Cg-Tg(Camk2a-cre)T29-1Stl/J). The mouse lines were maintained in a C57BL/6 background. Mice were kept at the 12-h light/dark cycle and water and food ad libitum at The Rockefeller University animal facility. Mice of both sexes were selected for experiments when they were 8- to 16-wk old, and the selection for each experiment and mouse line was randomized. All procedures were approved by the institutional IACUC and adhere to the ARRIVE guidelines.

Adenoviral vectors AAV2.CMV.PI.Cre.rBG and AAV2.CMV.HI.GFP-Cre.SV40 adenoviral vectors (Watabe-Uchida et al., 2012) were purchased from the University of North Carolina at Chapel Hill Vector Core. Rabies virus vectors EnvA G-deleted Rabies-GFP or EnvA G-deleted Rabies-mCherry (Wall et al., 2010) were purchased from the GT3 Core Facility at Salk Institute.

We performed three types of injection: helper virus was injected into NAc, followed by the injection of the rabies virus at a different angle to avoid infection and replication of the virus in the dorsal striatum. Stereotaxic surgeries were performed under general ketamine–xylazine anesthesia. Mice were placed in the stereotaxic apparatus and injection was guided by the AngleTwo program. Coordinates for the NAc injection were +1.10 mm mediolateral, +1.80 mm anteroposterior, and −4.60 mm dorsoventral from bregma, with the head tilt of 3–5°. Coordinates for CPu injections were +1.80 mm lateral, 0.70 anterior, and −3.50 mm dorsoventral from the bregma. Each mouse was injected with 0.3–1 μl of the viral suspension using the 10-μl Hamilton syringe and 33-g Hamilton needle over 10 min and an infusion pump (World Precision Instruments). Two consecutive injections were performed on each mouse. The helper virus mixture, consisting of an equal volume of AAV1.EF1-Flex-TVA-Cherry.ape (titer 4 × 10¹² vg/ml) and AAV1 CA-Flex-RG (titer 3 × 10¹² vg/ml) adenoviral vectors (Watabe-Uchida et al., 2012) (University of North Carolina at Chapel Hill Vector Core), was injected at the volume of 0.8 μl, followed by the 0.3 μl of the EnvA G-deleted Rabies-GFP (titer 1.54 × 10⁹ tu/ml) or EnvA G-deleted Rabies-mCherry (titer 1.58 × 10⁹ tu/ml) (Wall et al., 2010) (GT3 Core Facility at Salk Institute) 4–5 wk after the first injection. The titer of the viruses was 0.9–1 cfu ×10⁶. For the injections into NAc, the second injections were performed at the 25° angle to avoid infecting the cells in the CPu with the rabies viral vector. Mice were euthanized and analyzed 1 wk after the rabies virus injection. Since this technique involves stereotaxic injection into the ventral part of the striatum, there is a possibility of unintended infection of the ChIs in the dorsal striatum along the injection route. Thus, two sets of injections were done: one into NAc and one into the dorsal striatum right above the NAc injection with the smaller (0.3 μl) and larger (0.8 μl) volume of the rabies virus. The analysis with the larger volume was used for the scanning of the whole brain for the representative images because it enables easier visualization of the projection sites.

1 wk after the injection of the EnvA G-deleted Rabies viral vector, animals were deeply anesthetized with Nembutal and intracardially perfused with 4% paraformaldehyde solution and cryopreserved through a series of sucrose dilutions. Brains were cryopreserved in a series of sucrose solutions in 0.01 M PBS and embedded in the embedding media (Tissue-Tek). A series of 30-µm–thick coronal sections throughout the whole brain were sectioned on the cryostat. Immunocytochemistry was performed as described before (Milosevic et al., 2017). In short, cryostat sections were incubated with the solution of 5% normal goat serum in PBS for 1 h, followed by overnight incubation with the primary antibodies and appropriate Alexa Fluor (Invitrogen) secondary antibodies. Primary antibodies used were anti-GFP (1/500, raised in chicken, cat#ab13970; Abcam) and mCherry (1/500, raised in rabbit, cat#167453; Abcam). Sections were coverslipped with a ProLong mounting solution.

Imaging was performed in two ways. For representative images of the brain areas, tissue sections were first labeled with the anti-GFP or anti-mCherry antibodies, depending on the type of RV injected. Then, images were selected based on the quality of expression and imaged at the confocal microscope (LSM710 and LSM900; Zeiss). Images were taken with 10×, 20×, and two oil objectives, 40× and 63×, with the fluorescence-free immersion oil Immersol 518 (Zeiss) at room temperature. Images shown throughout the paper were processed in ImageJ (for the scale bars) and Photoshop. In Photoshop, regions with labeled cells were enlarged to enable a better visual representation of projecting neurons. Images were modified only with the brightness and contrast of Photoshop’s image modification tool. Schematic diagrams were created with https://Biorender.com.

For quantification, three biological and technical replicate samples were used for control and ChAT-p11 cKO. Imaging for quantification was performed on a series of sections, immunolabeled with anti-GFP or anti-mCherry antibodies, and counterstained with DAPI to better delineate histological layers and anatomical structures. Sections that were used for quantification represented every other 30-µm–thick section throughout the whole brain. Imaging was performed with the Mirax Scan scanning system (Zeiss). The resulting low magnification images represented 5 × 5 tiled images of a whole brain section. These images were loaded into the PanoramicView software and areas with GFP-labeled cells were outlined in their respective brain regions. The expression sites delineations were made in accordance with The Mouse Brain atlas (Franklin and Paxinos, 2008) and Scalable Brain Atlas (http://scalablebrainatlas.incf.org/) as reference. MetaMorph Image Analysis software was used for the automatic quantification of GFP- or mCherry-labeled cell bodies in each region, and ImageJ/Fiji (Schindelin et al., 2012) was used for the semiautomatic conformation of the cell body numbers. Differential enrichment of cell bodies between different regions and respective graph bar charts was completed using the Prism software. For this analysis, only areas that appear in at least two samples were analyzed. Areas at the injection sites, as well as septal, basal forebrain, extended amygdala, and cortical–amygdala transitory areas that contain a large number of ChI neurons were not included in the analysis. We used Paxinos atlas to outline the brain regions, and the abbreviations for these regions are shown in Table 1.

Mice were anesthetized with a ketamine/xylazine cocktail and underwent stereotaxic surgery to inject serotype 5 adeno-associated viruses (AAV) encoding CaMK2a-ChR2 (H134R) (AAV5/CaMKII-hChR2(H134R)-eYFP-WPRE, titer 6 × 10¹² vg/ml; UNC Viral Vector Core), and control virus rAAV5/CaMKII-EFYP (titer 4.3 × 10¹² vg/ml; UNC Viral Vector Core). The virus was injected bilaterally into the vHIP (from bregma: anterior/posterior: −3.8, lateral: +3.0, dorsal/ventral: 4.5 from top of skull) at a rate of 0.1 ml per minute for 10 min. Mice recovered for 5–6 wk before being subjected to electrophysiological experiments. For electrophysiological recordings, field light stimulation of ChR2-expressing vHIP terminals in NAc neurons was done through a 40× objective using a SPECTRA X LED light engine (Lumencor).

Mice between 8 and 12 wk of age were euthanized with CO₂. Coronal slices (300 μm thickness) were cut using a Vibratome 1000 Plus (Leica Microsystems) at 2°C in a NMDG-containing cutting solution (in mM): 105 NMDG (N-Methyl-D-glucamine), 105 HCl, 2.5 KCl, 1.2 NaH2PO₄, 26 NaHCO₃, 25 glucose, 10 MgSO₄, 0.5 CaCl₂, 5 L-ascorbic acid, 3 sodium pyruvate, 2 thiourea (pH was around 7.4, with osmolarity of 295–305 mOsm). After cutting, slices recovered for 15 min in the same cutting solution at 35°C and for 1 h at room temperature (RT) in aCSF recording solution containing (in mM): 125 NaCl, 25 NaHCO₃, 2.5 KCl, 1.25 NaH2PO₄, 2 CaCl₂, 1 MgCl₂, and 25 glucose (bubbled with 95% O₂ and 5% CO₂) (see below). Whole-cell patch-clamp recordings were performed with a Multiclamp 700B/Digidata1550A system (Molecular Devices), an upright Olympus BX51WI microscope equipped with the appropriate filters (Olympus) and a SPECTRA X LED light engine (Lumencor). The slice was placed in a recording chamber (RC-27L; Warner Instruments) and constantly perfused with oxygenated aCSF at 24°C (TC-324B; Warner Instruments) at a rate of 1.5–2.0 ml/min. The intracellular solution contained (in mM): 126 K-gluconate, 4 NaCl, 1 MgSO₄, 0.02 CaCl₂, 0.1 BAPTA, 15 glucose, 5 HEPES, 3 ATP, 0.1 GTP (pH 7.3). We used a whole-cell current-clamp for electrophysiology recordings in layer II/III PFC neurons. For measuring the membrane potential, 30 s of recording were binned into 0.5 ms bins and fitted with a Gaussian. For measuring the action potential firing, small currents were injected into the cells to bring the membrane potential to −70 mV. Consecutive 1-s current steps of 50 pA starting from −100 pA were injected to induce depolarization. The action potential threshold was measured from the first action potential to avoid any confounding effects of adaptation. For electrophysiology recording in NAc, we used whole-cell current-clamp to record the tonic activity of ChI neurons.

Consecutively, we used whole-cell voltage-clamp configuration to record evoked glutamatergic responses in ChI or medium spiny neurons. For this, we used field light stimulation of ChR2-expressing vHIP terminals in NAc in the presence of 30 µM bicuculine (to block GABAergic transmission). Consecutively, we used whole-cell voltage-clamp configuration to record evoked synaptic responses in ChIs or medium spiny neurons. For this, we used field light stimulation of ChR2-expressing vHIP terminals in NAc in the presence of 30 µM bicuculine (to block GABAergic transmission). The photostimulation-induced currents were blocked completely by the subsequent addition of glutamate receptor antagonists AMPA (DNQX 20 μm) and NMDA (APV 50 μm) receptor antagonists.

Fig. S1 contains images of the different brain regions that contained GFP- or mCherry-labeled cells, representing neurons projecting to the NAc and CPu ChIs.

We sought to determine direct inputs onto the ChI neurons of the NAc using the monosynaptic circuit tracing method with the CRE-dependent modified rabies virus technique (Wall et al., 2010; Watabe-Uchida et al., 2012). This method utilizes the spread of rabies virus (RV) through the synapses onto the projecting neurons. However, this spread is limited to the direct, primary input onto the infected cells due to several modifications in the rabies virus genome. We used the previously characterized ChAT-CRE mouse line GM60 (https://www.gensat.org). The experimental design was adapted to allow the optimal infection of the sparce cell population. ChI neurons were labeled with stereotaxic injection of a mixture of two viruses, one carrying the G-protein and another carrying the TVA receptor. This was followed by injection of a rabies virus into the same NAc region 5 wk later (Fig. 1, A and B). We confirmed that helper viruses, carrying the mCherry tag, infected ChI neurons by immunocytochemical labeling with anti-mCherry and anti-ChAT antibodies (Fig. 1 C). Next, we confirmed that viral injections infected only NAc ChI neurons using the mouse line ChAT-CRE crossed to ChAT-TRAP (https://www.gensat.org) in which all ChI neurons are tagged with the GFP (Doyle et al., 2008). We immunolabeled a series of coronal sections with an anti-mCherry antibody and detected no mCherry⁺ GFP⁺ cells in the neighboring brain regions with ChI neurons (Fig. 1 D). Furthermore, because the injection into NAc infects the dorsal striatum–caudate putamen ChI cells along the needle track, a set of control injections into the CPu was performed to determine the overlap with the CPu ChIs projections.

To generate a comprehensive list of all projection sites and to quantify the number of neurons that project to NAc ChIs, we analyzed the whole brain, as described in detail in the Materials and methods section. The cells retrogradely labeled with RV-GFP virus were detected in distinctive subregions of the cortex, hippocampus, thalamus, hypothalamus, amygdala, and brainstem (Fig. 2 and Fig. S1 A). The representative images of the NAc ChAT projection sites are shown in Fig. 2, A–G, while a complete list of all regions and proportions of labeled cells in each area is shown in Fig. 2 H. In addition, to gain insight into the similarities and differences between the NAc and CPu projections, we also analyzed the brains injected into the CPu (Fig. S1 B). Notably, the regional distribution of cells labeled with the rabies virus after the CPu and NAc injections had some very distinct areas with labeled cells (Fig. S1, A and B). This suggested that NAc ChIs have a distinct set of regions that project exclusively to these cells and not the CPu ChIs.

A large proportion of labeled cells in the brains that have been injected into NAc was found in the discrete areas of the PFC (Fig. 2 H). These included medial (Fig. 2 A1) and lateral orbital cortex (Fig. 1 A1), agranular insular cortex, prelimbic cortex (Fig. 2 A2), and anterior cingulate cortex (Fig. 2 A3). The majority of the projections came from the ipsilateral side (Fig. S1 A), as opposed to CPu where a substantial projection from the contralateral side was reported (Schmidt et al., 2012).

In the orbital areas of the PFC, in the NAc-injected brains, large number of labeled cells were found in the medial and ventral orbital (MO-VO) region of the PFC, while they were absent in the CPu-injected brains (Fig. 3, A1 and A2). Prefrontal cortical areas contained a large number of labeled neurons in layer 2/3 (Fig. 2, A1 and A2; and Fig. 3 A1). This is the opposite of what was found after CPu injection, where labeled cells were concentrated in the deep cortical layers (Fig. 3 A2 and [Guo et al., 2015; Klug et al., 2018]). To confirm that layer 2/3 cells in MO-VO were indeed glutamatergic pyramidal cells that project to NAc ChIs, two sets of experiments were performed. First, we adapted the injection approach to test the connectivity. For this purpose, we used another mouse line with a CRE expression under the regulation of the Cux2, a specific marker for layer 2/3 pyramidal neurons (Franco et al., 2012). Helper virus, tagged with mCherry, was injected into the upper layers of the MO-VO cortex of the Cux2-CRE mouse line (gift from Ulrich Muller, Johns Hopkins School of Medicine), followed by the injection of the RV-GFP into the NAc 5 wk later, as shown in the schematic diagram in Fig. 3 B. This injection approach enabled layer 2/3 projecting neurons to be infected with the helper virus via axonal ends in the NAc. The consecutive RV-GFP injection and recombination enabled projecting neurons labeled with the helper virus to express the GFP. The resulting double-labeled cells (mCherry/GFP) in the medial orbital cortex layer 2/3 are shown in Fig. 3 C. Second, a set of electrophysiological recordings were done to confirm our initial data (Fig. 3, D–H). We used a whole-cell patch clamp to record from layer 2/3 neurons infected with RV-GFP and the helper virus, respectively. The firing frequency, as well as the membrane potential and action potential threshold of the recorded neurons, were characteristic of pyramidal neurons (Fig. 3 D) and were not changed between neurons infected with the different viruses (Fig. 3, E–H). These experiments confirmed that layer 2/3 in the MO-VO are indeed pyramidal projecting neurons and that their basic physiological properties are not altered by the virus injection protocol. Interestingly, Wfs1-expressing cells with the same layer and regional specificity project to the motor cortex and CPu and receive projections from the lateral amygdala, posterior thalamic group, and several cortical areas (Shrestha et al., 2015), suggesting that multiple subtypes of the layer 2/3 neurons from the medial orbital cortex project outside of the cortex.

To better understand the circuitry at the root of depressive behavior, we investigated projections between control and depressive-like mouse model ChAT-p11 cKO mice. Quantitative analysis was performed on four brain regions: MO-VO, amygdala, vHIP, and dorsal raphe. These areas were selected because they have been shown to have an important role in depressive-like behaviors (Price and Drevets, 2010; Stuber et al., 2011; Chaudhury et al., 2015; Hultman et al., 2018) and because MO-VO, amygdala, and vHIP project specifically to the NAc ChIs. Representative images of projections from MO-VO, vHIP, and dorsal raphe used for this quantification are shown in Fig. 4 A. Quantification of projections has shown a significantly larger number of projections from vHIP in the ChAT-CRE control (CTR) mice than in ChAT-p11 cKO (Fig. 4 B). A significant difference was not reached for MO-VO, amygdala, and dorsal raphe, although there is a trend toward a lower number of projections from the amygdala and higher number of projections from MO-VO in ChAT-p11 cKO.

Virtually, all labeled cells in the hippocampus were confined to the ventral part, namely subiculum and CA1 (Fig. 2 F; Fig. 4 A; and Fig. S1, A4 and A5). No direct input to the CPu ChIs was detected. This corroborated and expanded on previous findings that the ventral subiculum has a strong projection to NAc (Britt et al., 2012; Bagot et al., 2015; LeGates et al., 2018). To verify if the connections from vHIP to the NAc ChI neurons are functionally altered by the p11 deletion, we expressed channelrhodopsin (ChR) in vHIP pyramidal neurons (Fig. 4, C and D) of the Ca²⁺/calmodulin-dependent protein kinase (Camk2a)-CRE mouse that specifically label projection neurons because Camk2a promoter has been traditionally used as an excitatory neuron-specific promoter (Basu et al., 2008; Tye et al., 2011). First, we confirmed that optically evoked currents are blocked completely by glutamate receptor antagonists (CNQX/AP5), and we amended the Materials and methods section to show that. We have shown before that deletion of p11 from ChI neurons results in a decrease in their firing frequency (Cheng et al., 2019), so we used this to identify the ChI neurons by their tonic firing in acute slices containing NAc from CTR and p11 KO mice (Fig. 4 E). Next, we recorded photostimulation-evoked glutamatergic currents of the vHIP terminals in NAc ChI neurons from CTR and p11 KO mice. In the CA1 area of the hippocampus, it is highly expressed exclusively in pyramidal neurons (Wang et al., 2013), and the direct glutamatergic projection between the pyramidal neurons in vHIP and medium spiny neurons in NAc has been previously described (Bagot et al., 2015; Muir et al., 2020). We applied a paired-pulse protocol with two light pulses (1–2 ms) separated by 100-ms intervals to verify first if the connections between vHIP pyramidal neurons and NAc neurons are functional and second if the connections show normal short-term synaptic plasticity. While in the CTR ChI neurons, we were able to elicit glutamatergic postsynaptic responses in 70% of the recorded neurons, we could not find any response in p11 KO ChI neurons in multiple mice (Fig. 4, F–H). To validate our recording protocol, we took advantage of the well-established connection between medium spiny neurons (MSN) and HIP (Bagot et al., 2015; LeGates et al., 2018) by recording photostimulation-evoked glutamatergic currents in neighboring MSN neurons from the same slices. In each NAc slice in which we recorded photostimulation-evoked currents in ChI neurons, we recorded multiple neighboring MSN neurons as a control for the vHIP-NAc projection. Some of the patched MSN neurons (16.7%) did not respond to photostimulation; however, we had the same percentage of responsive MSN neurons in each slice from which we presented ChIs data in both genotypes. Thus, we are positive that the percentage of non-responsive ChI neurons is not random between genotypes and it is not due to the absence of glutamatergic input from vHIP. Using this paired-pulse protocol, we evoked glutamatergic responses in 83.3% of the tested MSN neurons in both CTR and p11 KO mice, and both the amplitude of the response and the paired-pulse ration were unchanged between the genotypes (Fig. 4, I–K). Taken together, these data show for the first time the existence of functional glutamatergic synaptic connections between vHIP pyramidal neurons and ChIs in NAc and further emphasize the alterations of these projections induced by the deletion of p11 in these neurons.

Large number of areas contained cells labeled by retrogradely transferred rabies virus after injections into the NAc and CPu (Fig. S1, A and B).

Labeled cells were distributed in distinct areas of the amygdala. In NAc-injected brains, central amygdaloid nucleus (CeN) prominently displayed a large number of GFP⁺ cells (Fig. 2 B and Fig. S1 A). When brains injected into the NAc and CPu were examined, labeled cells were distributed in distinct areas. In NAc-injected brains, CeN contained a vast majority of labeled cells (Fig. 2 B). We observed a very small number of GFP⁺ cells in the BLA, suggesting that these inputs synapse on other types of accumbal neurons. It is well-established that the amygdala, especially the basolateral amygdaloid nucleus project to the striatum (Stuber et al., 2011). More recently, cell-based tracing methods showed that both central and BLA project specifically to dopaminergic receptor 1 medium spiny neurons in the dorsal striatum (Wall et al., 2013). Functional studies investigating the role of amygdalar subdivisions linked BLA and CeN to fear conditioning (Haubensak et al., 2010; Tye et al., 2011) and memory consolidation (Paré, 2003), while lesions of the CeN reduced stress response and anxiety and fear response to chronic unpredictable stress (Ventura-Silva et al., 2013). In the brains injected into the CPu, labeled neurons were predominantly in the anterior cortical, central, and medial amygdaloid nuclei. Interestingly, we did not observe cells in basolateral or central amygdaloid nuclei, suggesting distinct inputs to ChI neurons from the dorsal and ventral striatum. The data presented here demonstrate strong connectivity of CeN to the NAc ChIs, and connectivity of BLA to other accumbal cells would suggest the intricate modulation of fear conditioning, anxiety, and stress responses in the striatum.

The majority of labeled cells in the thalamus of the brains injected into the NAc were concentrated in parafascicular, mediodorsal, and centromedial thalamic nuclei (Fig. 2 C; and Fig. S1, A2 and A4). It is notable that inputs to the CPu ChIs came from a different set of nuclei (Fig. S1, B2, B4, and B5; and [Guo et al., 2015; Klug et al., 2018]), with few exceptions such as parafascicular nucleus, which projects to both dorsal and ventral striatal ChIs. Others have shown that both dopaminergic receptor 1 and 2 medium spiny neurons receive afferents predominantly from parafascicular and mediodorsal nuclei, as well as a smaller number of inputs from ventromedial, anterodorsal, and anteroventral nuclei (Wall et al., 2013). This would imply that the thalamo–striatal connectivity targets both projection neurons and interneurons, bringing an additional level of complexity in control over striatal activity.

The hypothalamus of mice injected in the NAc contained a very small number of labeled cells, scattered mostly in the lateral hypothalamic area (Fig. 2 H and Fig. S1 A3). The lateral hypothalamic area has been shown to connect with basomedial and anterior cortical amygdaloid nuclei (Niu et al., 2012) and has a role in food intake regulation (Stanley et al., 2010; Pérez et al., 2011). This would suggest that inputs to ChI neurons in NAc have a role in modulating the behavior related to feeding and/or food intake.

Strong labeling of neurons in SN and VTA after NAc injection were observed (Fig. S1 A6). The striatal–VTA circuitry was implicated in the regulation of stimulus-dependent learning (Brown et al., 2012), as well as reward behavior (Chaudhury et al., 2013). The projections from SN and VTA to the neurons of the dorsal striatum cells are well documented (Wall et al., 2013; Guo et al., 2015; Klug et al., 2018). Here, we have shown for the first time that the part of this circuitry involves the SN/VTA-NAc ChI connectivity.

Dorsal raphe contained labeled cells in the brains from NAc and CPu injection. This affirmed previous findings about relatively weak serotoninergic inputs into the striatum (Waselus et al., 2006), and specifically ChI cells (Cachope et al., 2012; Guo et al., 2015; Klug et al., 2018). Interestingly, and not previously reported, we found virally labeled cells to be confined only to the ipsilateral side of the brain (Fig. 2 E).

Medulla, and specifically locus coeruleus, the main noradrenergic output in the brain contained few labeled cells. In addition, we detected very few neurons in other areas of the pons and medulla, most notably the area of the gigantocellular reticular nucleus. However, these neurons were sparse and thus difficult to be assigned to the appropriate nuclei. This is probably due to the technical limitations of the technique used in this study but may also represent the actual in vivo representation of connectivity within the brain.

Mapping complex networks between the neurons in different regions of the central nervous system is essential to understanding the behavior and its changes in neurological diseases. Here, we have generated a detailed, exhaustive brain-wide map of direct projections to the cholinergic neurons in the NAc (Fig. 5 A). Importantly, this mapping revealed several areas that project only to the ventral but not dorsal striatal ChIs, such as layer 2/3 from the medial and ventral subdivisions of the orbital PFC, CeN, and subiculum of the vHIP. Most importantly, the circuitry between NAc ChIs and vHIP is significantly altered in the mouse model of depression, ChAT-p11 cKO (Fig. 5 B), with implications for cognizing the specific morphological and anatomical changes in psychiatric diseases involving these brain networks.

The NAc is involved in mediating an altered perception of social status (Berton et al., 2006), inability to experience pleasure and positive emotions, heighten negative emotional processing, psychomotor retardation (Epstein et al., 2006), and behaviors related to reward (Russo and Nestler, 2013), all distinctive traits of mood disorders and depression. The final output from the NAc and regulation of behavioral outcomes is determined by the ChIs activity. This is accomplished by integrating the large number of inputs from various regions and different neurotransmitter systems that allow fine coordination of ChIs tonic activity. For example, silencing the activity of NAc ChIs increases the activity of medium spiny neurons, the main output cells of the NAc (Witten et al., 2010). Based on these data, functional implications of the NAc ChIs circuit mapped in this study and its role in depressive behavior most likely depend on the projections unique to NAc, namely, MO-VO PFC areas, vHIP, and CeN. Circuits between these areas and the NAc are known for their role in a wide range of behaviors, including depressive behaviors. The complex regulation of the NAc output governed by the groups of neurons within the orbitofrontal cortex and their extensive circuitry with other brain regions has been described (Pinto and Sesack, 2000; Sesack and Grace, 2010; Hultman et al., 2018). For example, the PFC-NAC circuitry provides executive control by mediating task-switching behaviors and inhibiting motor responses (Sesack and Grace, 2010). An increase in the synchronized activity between vHIP and NAc predicted vulnerability to stress in naïve mice and was a marker for mice susceptible to social stress (Hultman et al., 2018). In this study, we reported several notable findings regarding the connectivity of this region with NAc. We found that layer 2/3 pyramidal cells project directly to the NAc ChIs. This is similar to the wolframin- and Ntf3-expressing layer 2/3 pyramidal neurons, which project to the dorsal striatum, and regulate stress-induced depressive-like behavior in mice (Shrestha et al., 2015). Furthermore, a large body of work has shown that layer 2/3 pyramidal neurons are mostly involved in intracortical circuits (Yoshimura et al., 2005; Shepherd, 2013). Our data and data from others (Thompson and Swanson, 2010; Shrestha et al., 2015) found projections from layer 2/3 to subcortical regions. This adds another layer to the cortical circuitry complexity, as the PFC is to our knowledge the only area where layer 2/3 pyramidal neurons project outside of the cortex. Also notable is the heterogeneity between layer 2/3 pyramidal cells since wolframin and Ntf3 expressing neurons do not project to the NAc, while we found neurons in the same region and layer specificity that synapse on the NAc ChIs. Furthermore, it was reported that direct inputs to CPu ChIs arise primarily from the motor areas, and much less from the PFC areas, such as lateral orbital, prelimbic, and cingulate (Guo et al., 2015; Klug et al., 2018). This may also imply that prefrontal cortical areas project more heavily to the NAc ChI neurons then hippocampal neurons or amygdala, as suggested before (Britt et al., 2012). Our data support this hypothesis and further show that the vHIP and the MO-VO cortices project directly and specifically to ventral, but not dorsal ChIs. These results point to the complex circuitry between the medial orbital cortex and striatum involved in depressive-like behaviors in mice.

However, it is also important to consider that the regulation of depressive phenotype in ChAT-p11 cKO depends mainly on the glutamatergic inputs from the vHIP, as they were shown here to be significantly reduced in these mice. The role of glutamatergic projections to the NAc in the regulation of addiction, anxiety, reward, and depressive behavior has been reported previously (Britt et al., 2012; Bagot et al., 2015; LeGates et al., 2018; Muir et al., 2020) although the specific mechanism reported differs between studies. Britt et al. suggested that ChIs and possibly other interneurons in NAc do not receive direct input from the vHIP but indirectly through connections with the medium spiny neurons (Britt et al., 2012). On the other hand, direct inputs from vHIP to NAc were found by others (LeGates et al., 2018), and confirmed in this study. LeGates et al. uncovered that the strength of the ventral hippocampal input to dopaminergic receptor 1 medium spiny neurons weakens in chronic stress, and this is linked to changes in reward response (LeGates et al., 2018). The role of this circuitry in depressive-like phenotype was further supported by another report showing that glutamatergic transmission from vHIP to NAc medium spiny neurons is increased in mice susceptible to stress (Bagot et al., 2015). Moreover, this effect is specific to vHIP-NAc circuitry since the stimulation of either mPFC or amygdala afferents to the NAc led to resilience. Our study shows for the first time cell-specific direct glutamatergic input from the vHIP to the NAc ChIs. Furthermore, this glutamatergic projection appears to have a decisive role in regulating the activity of the NAc ChIs in the ChAT-p11 cKO mice, as we found a significant reduction in the number of projections from this region, while projections from PFC, amygdala, and dorsal raphe are not changed significantly. It has been proposed that the behavioral output of the NAc depends on the amount of glutamate released from the inputs coming from the PFC and amygdala (Britt et al., 2012). It is possible that a decreased number of inputs from the vHIP to NAc ChIs disrupts the balance of signals received from other major glutamatergic inputs, including prefrontal cortical areas, CeN, and parafascicular thalamic nucleus. This would suggest that glutamatergic input disbalance is a key circuitry mechanism of depressive-like phenotype in ChAT-p11 cKO mice (Fig. 5). In addition, septum, basal forebrain, and habenula are rich in cholinergic neurons that do not express p11 in ChAT-p11 cKO mice. It is known that the basal forebrain (Campbell and Lobo, 2023; Morais-Silva et al., 2023), septum (Li et al., 2023), and habenula (Cui et al., 2018) play a key role in regulating depressive-like phenotype in mice. In this context, the absence of p11 in the ChIs in these regions and their inputs to NAc may also affect the physiology of the NAc ChIs and the behavioral phenotype related to depression. This cell-specific deletion of p11 and its mechanism of action in modulating the behavior, especially depressive- and anxiety-like phenotypes remains to be examined.

One fascinating outcome outside of the scope of this study is that p11 ablation in ChIs affects specifically the ChAT-vHIP circuitry but not others examined in this study. One possibility is that p11 affects the development of ChIs in a way that prevents the formation of inputs from the vHIP. Genetic ablation impact on ChAT circuitry development has been shown before, where the TrkA ablation impairs the basal forebrain-vHIP ChAT circuitry, specifically its laminar pattern and the number of projections (Sanchez-Ortiz et al., 2012). In rodents, ChIs development is tightly regulated by the NGF and BDNF signaling cascades trough their receptors, p75, TrkA, and TrkB, where the NGF/p75 determines the number of striatal ChIs (Ward and Hagg, 1999; Sanchez-Ortiz et al., 2012), while the BDNF/TrkB regulate growth and complexity of neurons, including loss of dendritic spines, and diminished nigral–striatal projections (Li et al., 2012). The mechanism by which p11 affects ChIs development may rely on a complex cascade involving plasmin, BDNF, NGF, and Anxa2, an important binding partner of p11. Results from the in vitro studies support this hypothesis. For example, it has been shown that NGF-induced neuritogenesis depends on the Anxa2-mediated plasmin generation (Jacovina et al., 2001), while p11 has been shown to be necessary for the BDNF-mediated effect on dendritic length and spine density (Park et al., 2016). However, the exact mechanism of p11-induced changes in NAc ChIs circuitry shown here remains to be elucidated.

The difficulty of labeling sparse populations and areas deep within the brain, such as NAc, has so far prevented the mapping of the interneuronal subtypes in NAc. The optimal infection of scant populations, such as ChIs, requires a larger number of viral particles and a higher volume for injection, which then brings a risk of off-target infections in adjacent areas with the same type of cell. For example, olfactory bulb granule cells as well as olfactory tubercle neurons were very often labeled due to the infection of olfactory tubercle neurons and piriform cortex. To circumvent this constraint, various volumes of rabies viral particles were injected. While labeled cells were detected in the same areas of the brain, a much smaller number of cells was observed with a smaller volume. Thus, smaller volumes ensured the proper regional quantification, but the optimal visualization of the projection sites in the whole brain tissue had to be performed with a higher volume. Furthermore, in this study, even the smaller volume of the virus injected prevented us to discern between the projections to the core and shell of the NAc as the virus infected both areas of the NAc. We also detected some areas, particularly in the regions further away from the injection site, such as locus coeruleus or medulla, with the number of projections in single digits. This implied that the number of viral particles defined by the titer of the virus and the volume injected is very important for its propagation and consequently a number of labeled cells and quality of labeling, as suggested previously (Wall et al., 2013). Therefore, while this study used the approach in experimental design and rigorous analysis of virally labeled projection neurons, previously designed by others (Haubensak et al., 2010; Guo et al., 2015; Klug et al., 2018), these circuitries will need further confirmation by other methods, such as electrophysiology. In addition, the mouse line that used BAC-generated ChAT-CRE has been shown to overexpress the vesicular acetylcholine transporter, but this primarily has a downstream effect due to the excess acetylcholine and less effect on ChIs themselves (Crittenden et al., 2014; Straub et al., 2014). Thus, we believe that our study utilizing this transgenic mouse line closely represents this circuitry in wild type mice.

In summary, given the key role of the ChIs, and especially NAc ChIs in mood disorders, we elucidated the connectivity of the accumbal ChIs on the level of the whole brain and examined this connectivity in the context of the mouse model of depression. Specific behavioral patterns and clinical symptoms of depression depend on the correlations between specific networks between multiple brain areas, their interactions, and cell-specific circuits between these regions. In this context, the functional significance of our data warrants further mechanistic studies that will elucidate the role of the inputs to NAc ChIs from various brain regions and how they govern reward behavior and dysfunction in mood disorders. Furthermore, this study may further our understanding of the functional changes in the patients suffering from neurodegenerative diseases, such as Alzheimer’s, Parkinson’s, and Huntington’s, which all have disrupted acetylcholine signaling in the striatum and often exhibit psychiatric symptoms.

Data underlying this study are available in the published article.

Jeanne M. Nerbonne served as editor.

We thank Dr. Nathaniel Heintz for a generous gift of the ChAT-CRE and ChAT-TRAP mice lines (project Gene Expression Nervous System Atlas (GENSAT) Project, National Institute of Neurological Disorders and Stroke contracts N01NS02331 & HHSN271200723701C to The Rockefeller University). We are grateful to Dr. Urlich Muller (John Hopkins School of Medicine, Baltimore, MD, USA) for the Cux2-CRE mouse line. We thank Dr. Katia Manova-Todorova, the head of MSKCC Molecular Imaging Core Facility, Dr. Jerry Chang, and Dr. Thomas Liebman for help with the imaging, Mahira Tiwana for technical help with the experiments, Drs. Inez Ibanez-Tallon, Luca Parolari, and Jennifer Warner-Schmidt for helpful suggestions during this study, and Dr. Marc Flajolet who read the manuscript and offered thoughtful suggestions. We extend special thanks to Dina Becaj, a participant in the Rockefeller University Summer Science Research Program, who helped with the cell quantification experiments.

This work was supported by the Fisher Center for Alzheimer’s Research Foundation (to P. Greengard), National Institute on Drug Abuse P30 DA035756 Pilot Project Award and National Institute on Aging 1RF1AG059770-01 (to A. Milosevic).

Author contributions: L. Medrihan: Conceptualization, Investigation, Validation, Visualization, Writing - original draft, Writing - review & editing, M.G. Knudsen: Data curation, Investigation, Writing - review & editing, T. Ferraro: Data curation, Validation, Writing - review & editing, P. Del Cioppo Vasques: Formal analysis, Investigation, Validation, Y. Romin: Formal analysis, Resources, S. Fujisawa: Data curation, Formal analysis, P. Greengard: Conceptualization, Funding acquisition, A. Milosevic: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing.