Abstract
The nucleus accumbens (NAc) is critical to goal-directed behaviors as the main input structure for limbic pathways to the basal ganglia. The NAc shell is composed of inhibitory projection neurons that receive robust glutamatergic innervation from both the hippocampus and amygdala. In view of primate-specific changes in the neural composition of the NAc, it is still unclear how its circuits are organized in primates. We used a system-to-synapse approach to characterize amygdalar and hippocampal pathway distribution, innervation patterns, and synaptic characteristics in the NAc shell of rhesus monkeys (Macaca mulatta) of both sexes. Key findings showed that both the amygdalar and hippocampal pathways disproportionately innervated NAc shell interneurons relative to their population sizes, assessed via confocal systems’ analysis and at the synaptic level with electron microscopy. The synaptic features associated with the two pathways were distinct. The amygdalar projection was denser, with larger boutons that more often contained mitochondria than the hippocampal projection. The hippocampal pathway had larger postsynaptic densities and more frequently formed perforated synapses, which are features associated with high synaptic efficacy. In addition, hippocampal boutons more frequently formed multiple synapses, often with one projection neuron and one interneuron. These interactions with the NAc shell suggest distinct mechanisms for the processing of affective signaling from the amygdala and contextual information from the hippocampus.
Significance Statement
The nucleus accumbens (NAc) is a key structure for motivated behavior; it receives dense pathways from the amygdala, associated with emotional significance, and the hippocampus, associated with context. In the NAc shell, both pathways disproportionately innervated interneurons, relative to their population size. The amygdalar boutons were somewhat larger and enriched with mitochondria, associated with sustained activity. Hippocampal terminations formed larger synapses and were more often multisynaptic, suggesting high synaptic efficacy. These patterns diverge from previously described rodent circuit patterns. The findings suggest that internal emotional state and environmental cues related to context differentially affect circuits underlying goal-directed behavior in the NAc shell.
Introduction
The nucleus accumbens (NAc) is situated in the ventromedial striatum. As the main input structure for limbic afferents to the basal ganglia, several streams of information converge in the NAc (Groenewegen et al., 1982; Kelley and Domesick, 1982; Sadikot et al., 1992; Haber et al., 1995; Rudkin and Sadikot, 1999; Chiba et al., 2001; French and Totterdell, 2002, 2003; Grillner et al., 2013; Choi et al., 2017; Xia et al., 2019), which support a key role in motivated behavior. Among the sources of input to the NAc, the basal amygdala conveys signals associated with emotional processing needed to evaluate potential outcomes and aspects of learning (Jackson and Moghaddam, 2001; Ambroggi et al., 2011; Millan and McNally, 2011; Correia et al., 2016; Taswell et al., 2023). The hippocampus, which innervates the medial aspect of the NAc shell, provides contextual information needed to assess the relevance of stimuli during different events (Groenewegen et al., 1987; Haber and McFarland, 1999; Friedman et al., 2002; French et al., 2005; Salgado and Kaplitt, 2015).
Extensive work in mice and rats reveals the significance for integrating signals on salience and context for goal-directed behavior (Howland et al., 2002; Belujon and Grace, 2008; Talmi et al., 2009; Bagot et al., 2015; Calipari et al., 2016). Functional studies in humans and nonhuman primates have suggested the contribution of NAc circuits in pathological states including depression (Epstein et al., 2006; Bewernick et al., 2010, 2012; Young et al., 2016; Ironside et al., 2020), anxiety (Jensen et al., 2003; Levita et al., 2009, 2012), and addiction (Zou et al., 2015; Wang et al., 2019; Tolomeo and Yu, 2022). However, interaction of pathways with NAc neural subtypes vulnerable to disruption in pathological states is still unclear.
Like the dorsal striatum, the NAc has two major classes of neurons: the inhibitory medium spiny neurons (MSNs) that project downstream (Haber et al., 1990; Heimer et al., 1991, 2000; Usuda et al., 1998; Dallvechia-Adams et al., 2001; He et al., 2021) and a diverse group of local inhibitory interneurons (Kawaguchi et al., 1995; Wu and Parent, 2000; Ibáñez-Sandoval et al., 2011; Del Rey et al., 2022; reviewed in Castro and Bruchas, 2019). But there are differences between the rodent and primate NAc. A key difference is an increased population of interneurons in primates, with a specific expansion of the calretinin-positive (CR+) subclass by comparison with rats and mice (Wu and Parent, 2000; Petryszyn et al., 2014). Calretinin interneurons show an expansion in the primate cortex as well (Condé et al., 1994; Gabbott et al., 1997; Zaitsev et al., 2005), and their preponderance in the striatum suggests their parallel expansion in primate evolution. Further, the high presence of CR neurons in the primate striatum suggests a change in the local microcircuitry and afferent organization.
The interaction of pathways from the amygdala and hippocampus with projection neurons and interneurons in the NAc shell is not well understood in primates and raise further questions: What is the relative composition of the NAc shell of MSN projection neurons and local interneurons? To what extent does the amygdala and hippocampus innervate MSNs, as opposed to local interneurons? To address these issues, we performed a multilevel study of the distribution, innervation patterns, and ultrastructural features of amygdalar and hippocampal pathways to the NAc shell of rhesus monkeys. We conducted analyses in the context of estimated neuronal populations in the NAc shell and focused on MSNs and interneurons that express CR or parvalbumin (PV). Our findings revealed that both pathways innervated the CR+ and PV+ interneurons at a significantly higher rate than their respective population sizes, which differs by comparison with rodent studies (Sesack and Pickel, 1990; Johnson et al., 1994). Further, the pre- and postsynaptic characteristics of the amygdalar and hippocampal pathways were distinct in the NAc shell suggesting specializations in the transmission of signals on affective significance and context.
Materials and Methods
We studied the neural subtypes within the NAc shell and their interactions with amygdalar and hippocampal pathways in rhesus monkeys (Macaca mulatta). Figure 1 shows a schematic of the experimental design.
Experimental design. A, Medial surface of the rhesus monkey brain shows sites overlying tracer injection into amygdala and hippocampus with axons traveling to NAc. B, Cross section through the striatum shows CB labeling (left) and AChE staining (right), used to identify boundaries (dashed lines) of the NAc shell. C, Pathway distribution: schematic of stereology method used to estimate the density of labeled boutons. NAc shell is shaded in dark gray with stereological grid overlaid for systematic random sampling of sites framed in red/green boxes. An example of a sampling site is shown, outlined by a red/green border. Two boutons on a tracer-labeled axon from the amygdala within the sampling site are indicated by black arrowheads. D, Appositional analysis to identify pathway targets: confocal image of amygdalar axons (red) and CR neuropil (green) in the NAc shell. An apposition between an amygdalar bouton and CR structure is shown in the white box. Quantifying appositions between labeled boutons and structures labeled for CR or PV interneurons or DARPP-32 projection neurons. E, Ultrastructural analysis: inset shows an electron microscopy image of a DAB-labeled bouton from the amygdala forming a synapse (arrow) with a dendritic shaft in the NAc shell. AChE, acetylcholinesterase; AMY, amygdala; C, caudal; Cau, caudate; CB, calbindin; CC, corpus callosum; Co, core; CR, calretinin; D, dorsal; DAB, diaminobenzidine; DARPP-32, dopamine- and cAMP-regulated phosphoprotein, 32 kDa; HIPP, hippocampus; L, lateral; M, medial; MSN, medium spiny projection neuron; NAc, nucleus accumbens; Put, putamen; PV, parvalbumin; R, rostral; Thal, thalamus; V, ventral.
We injected tracers into the basal amygdala and anterior hippocampus to quantitatively study their pathways as they innervate neuron subtypes of the NAc shell (Fig. 1A). We used series of sections stained for acetylcholine esterase (AChE) and calbindin (CB; examples shown in Fig. 1B), to map the boundaries of the NAc shell (Meredith et al., 1996; Heimer et al., 1997; Haber and McFarland, 1999). We then tested if there is overlap in expression of dopamine and cAMP-regulated neuronal phosphoprotein, 32 kDa (DARPP-32), CR, PV, and CB in neurons in the NAc shell. At the systems’ level, we studied the distribution of axons from the amygdala or hippocampus across the NAc shell and the density of boutons on these axons with bright-field microscopy (Fig. 1C). We then used confocal microscopy to quantify appositions between fluorescently labeled pathways from the amygdala and hippocampus and neuron subtypes in the NAc shell (Fig. 1D). We then studied innervation patterns and synaptic features in the NAc shell through ultrastructural analysis with electron microscopy (EM, Fig. 1E).
Animals, surgery, and tracer injections
The studies were conducted on rhesus monkeys (Macaca mulatta; Table 1) of both sexes. Eight cases were used to study the amygdala and hippocampal pathways to the NAc. Previously labeled sections from seven cases were used for neuron population estimates. Protocols were approved by the Institutional Animal Care and Use Committee at Boston University School of Medicine and in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. All procedures were designed to minimize animal suffering and the number of animals. Multiple tracers were injected into each hemisphere for cases used in this study and unrelated studies, to ensure minimal use of precious primates.
Cases, sex, age, injection site, tracer, hemisphere, and analyses
Animals were sedated for magnetic resonance imaging (MRI) and placement in a stereotax (model #1430, David Kopf Instruments) with 10–15 mg/kg intramuscular ketamine hydrochloride, followed by a loading dose of 2.5–5 mg/kg intravenous propofol, and then a continuous infusion (0.25–0.4 mg/kg/min). The MRI scans were aligned to the stereotaxic space for each animal to calculate the coordinates for future injection of neural tracers in the amygdala and hippocampus.
After 1 week, the animals were prepared for tracer injections. Each animal was sedated (as above) with ketamine and propofol and then placed under general anesthesia with isoflurane. The animals were monitored for respiratory rate, oxygen saturation, heart rate, and temperature. Under sterile surgical conditions, each animal was placed in the same stereotax apparatus and positions to match the MRI scan used to calculate injection placement. Above each injection site, an area of cortex was exposed. Microsyringes (Hamilton, 5 or 10 µl) were loaded with tracer (Invitrogen). To avoid tracer leakage during insertion, we loaded syringes with a small bubble of air after aspiration of the tracer. Microsyringes were mounted on the stereotax by a microdrive attachment and positioned at the injection site coordinates. We injected a mixture of 3 kDa (optimal for retrograde labeling of cell bodies, not studied here) and 10 kDa (optimal for anterograde labeling of axon terminals) tracer variants to allow for bidirectional tracing. We injected 4–6 µl of each tracer at each injection site. After injections, the syringes were left for 5–10 min at the injection site to allow for diffusion of the tracer into the tissue and minimize backflow of the tracer during syringe retraction. Postoperatively, animals were monitored and given antibiotics and analgesics as needed.
For optimal tracer transport, animals recovered for 18–20 d after tracer injections. Then the animals were sedated, given a lethal dose of anesthetic (sodium pentobarbital, to effect), and transcardially perfused with either 4% paraformaldehyde, 0.2% glutaraldehyde in 0.1 M PB, pH 7.4, or 4% paraformaldehyde in 0.1 M cacodylate buffer, pH 7.4, or 4% paraformaldehyde in 0.1 M PB, pH 7.4. The brains were removed and photographed and then cryoprotected via a series of increasingly hypertonic solutions of 10–20% sucrose in 0.05% sodium azide in 0.1 M PB, pH 7.4. The sections were then frozen in −80°C isopentane and sectioned on a freezing microtome (AO Scientific Instruments/Reichert Technologies) at 40 or 50 µm. We collected sections systematically into matched series and placed them in antifreeze (30% ethylene glycol, 30% glycerol, 0.05% sodium azide in 0.01 M PBS, pH 7.4) for long-term storage.
We also used previously prepared slides from one rhesus monkey (Case SF) that had been treated with the irreversible AChE blocker diisopropylfluorophosphate (DFP) prior to sacrifice (Poirier et al., 1977; Mesulam et al., 1983). We used this case to map cholinergic neurons, as described previously (Mesulam et al., 1983).
Labeling procedures for bright-field, confocal, and electron microscopy
AChE staining was previously described (Wang and Barbas, 2018). To label neurons, we used primary antibodies raised in mouse, rabbit, or goat to PV, CR, CB, or DARPP-32. All antibodies used are listed in Table 2. The phosphoprotein DARPP-32 is a faithful marker of MSNs in the striatum (Ouimet et al., 1984, 1998; Walaas and Greengard, 1984; Ouimet and Greengard, 1990; Anderson and Reiner, 1991; Anderson and Turner, 1991; Tamura et al., 2004; Bateup et al., 2008). Previous studies have shown that CR and PV label nonoverlapping interneurons in the striatum (Wu and Parent, 2000; Petryszyn et al., 2018; Del Rey et al., 2022), while CB labels some projection neurons and some interneurons (Kawaguchi et al., 1995; Deng et al., 2010).
Antibodies, RRID (when available), use, and dilution
The immunolabeling protocol is as follows: Prior to immunohistochemical procedures, we rinsed residue cryoprotectant from tissue sections in phosphate buffer (0.1 M PB), pH 7.4. For tissue processing for bright-field and confocal microscopy, we performed antigen retrieval in sodium citrate solution, pH 8.5 (10 mM, Sigma-Aldrich) using a 70–80°C water bath. We rinsed tissue sections in 0.1 M PB, followed by incubation in glycine (0.5 mM, 4°C, 1 h; Sigma-Aldrich) to bind free aldehydes. For tissue labeling for bright-field microscopy, sections were incubated in 0.3% hydrogen peroxide in 0.1 M PB for 30 min to quench endogenous peroxidases. For tissue processed for diaminobenzidine (DAB), we used Avidin/Biotin blocking solution (catalog #SP-2001, Vector Laboratories; RRID: AB 2336231) to block the biotin conjugate on biotinylated dextran amine (BDA; step is omitted when tissue is being processed to visualize the BDA tracer). After multiple washes in 0.1 M PB, sections were incubated for 1 h at 4°C in a preblock solution containing 5% bovine serum albumin (BSA, Sigma-Aldrich), 0.1% BSA-c (Aurion), 0.1% cold-water fish gelatin (only for EM, Aurion; to stabilize the ultrastructure), 5% normal serum (donkey or goat matched to the secondary antibody host animal, Sigma-Aldrich), and 0.5% Triton X-100 (only for bright-field and confocal, Sigma-Aldrich) or 0.025% reduced Triton X-100 (only for EM, Roche Applied Science). Tissue sections were incubated in the primary antibody for 3–4 d. We omitted the primary antibodies for control sections. Sections were rinsed in 0.1 M PB and incubated in secondary antibodies for 3 h to 1 d. During both incubation periods, the sections were microwaved at the start of incubation and once each day of the incubation or at the midpoint if <1 d to enhance antibody penetration (2 min at 150 W, 4°C, PELCO BioWave with SteadyTemp thermocooler, Ted Pella).
For bright-field microscopy, after the secondary antibody incubation, sections were rinsed in 0.1 M PB, incubated in avidin-biotin complex (ABC, 1:100 in 0.1 M PB; AB-HRP, Catalog #PK-6100, Vector Laboratories, RRID: AB 2336827), and processed with DAB for 1–3 min (catalog #SK-4100, Vector Laboratories, RRID: AB 2336382). The sections were rinsed in dH2O and mounted on glass slides coated with gelatin. After drying, some sections were processed for Nissl staining (Thionin stain) as described previously (García-Cabezas et al., 2016). Sections were coverslipped with Entellan (Sigma-Aldrich). In control sections incubated without primary antibodies, we found no evidence of nonspecific immunolabeling.
For sections processed for confocal microscopy, we incubated the tissue in secondary antibodies conjugated with fluorophores for 3 h. We rinsed the sections in deionized water (dH2O) and mounted them on glass slides to dry overnight. The sections were coverslipped with ProLong Gold Antifade (catalog #36930, Invitrogen), and the coverslip edges were sealed using fast-drying clear nail polish (Sally Hansen), which reduces the infiltration of air bubbles into the coverslipping medium.
We analyzed two cases per pathway for electron microscopy (AMY: cases BM and BL; HIPP: cases BQ and BT). We performed double immunolabeling for the injected tracer and a marker of each neuron subtype (DARPP-32 for MSNs; PV or CR for interneurons). The tracer was processed with DAB, which was electron dense and filled axons with a dark, uniform precipitate visualized with EM. For neuron markers, a primary antibody was paired with a gold-conjugated secondary antibody, which was visualized as a globular dark precipitate. After primary and secondary antibody incubations, sections were rinsed, postfixed with a reduced 3% glutaraldehyde, 1% paraformaldehyde in 0.1 M PB solution, and microwaved for 2 min at 150 W, 4°C. Sections were then incubated in glycine. To enhance the gold secondary antibody with silver, we incubated sections in the enhancement conditioning solution (ECS, 1:10, 10 min; Enhancement Conditioning Solution 10_500.055, catalog #25830, Electron Microscopy Sciences), silver enhancement solution (90 min; R-GENT SE-EM kit, catalog #500.033, catalog #25520, Electron Microscopy Sciences), second ECS, and then 0.1 M PB rinses. Tissue was then processed for DAB to visualize tracer, as above. After rinsing, we used 0.3% hydrogen peroxide to quench any remaining peroxidases. For Case BT, one section was triple-labeled for Fluoro-Emerald tracer injected into the hippocampus (visualized with DAB), CR (visualized with gold), and PV (visualized with tetramethylbenzidine, TBM), sections were rinsed in 0.1 m PB, then underwent another 0.3% H2O2 to quench remaining peroxidases, and AB-blocking again to block remaining HRP-binding sites. We then incubated the tissue with the third secondary antibody, which had a biotin conjugate to visualize TMB. After incubation, sections underwent rinses and then TMB staining. Tissue sections processed for TMB staining were preincubated in a solution of 0.005% TMB, 0.004% NH4Cl, 5% ammonium paratungstate in 0.1 M PB at pH 6 (pH adjusted with HCl acid). Tissue was then incubated in a solution of 0.005% TMB, 0.004% NH4Cl, 5% ammonium paratungstate, and 0.005% H2O2 in 0.1 M PB, at pH 6 (mixed into solution in the order listed). The reaction was quenched in 0.1 M PB at pH 6 at 4°C (5 min). To stabilize the reaction product TMB, the tissue was incubated in a DAB cobalt chloride solution (0.05% DAB, 0.02% cobalt chloride, 0.004% NH4Cl, 0.005% H2O2; in 0.1 M PB, pH 6) for at least 5 min, as previously described (Medalla et al., 2007; Joyce et al., 2023). Tissue was rinsed in 1.0 M PB with pH 6 at 4°C (1 min) and then rinsed in 0.1 M PB at pH 7.4 at 4°C (3 × 10 min).
For all tissue processed for EM, we then performed microwave postfixation (6% glutaraldehyde, 2% paraformaldehyde in 0.1 M PB, 150 W, 15°C) until the sample temperatures reached 30–35°C. The sections were left in the fixative for 30 min to reach room temperature. The sections were rinsed with 0.1 M PB and underwent EM processing, as described previously (Joyce et al., 2020; Wang et al., 2021). For this protocol, tissue rinses, which occur between each change of solution, were done in dH2O. First, the labeled tissue was postfixed in 2% osmium tetroxide (Electron Microscopy Sciences) with 1.5% potassium ferrocyanide in 0.1 M PB under vacuum for 36 min. Osmium tetroxide acts as a fixative for several compound types including proteins, lipids, carbohydrates, and nucleic acids (Cano-Astorga et al., 2024). We also used additional osmium tetroxide as a contrasting agent. The tissue was microwaved at 100 W at 4°C for 6 min at the start of the fixation. The sections were rinsed in dH2O and then incubated for 30 min in 1% thriocarbohydrazide in dH2O (Sigma Millipore). The second osmification included incubation in 2% osmium tetroxide in dH2O under vacuum for 36 min, microwaved at 100 W at 4°C for 6 min at the start of the incubation. The sections were incubated overnight in 1% uranyl acetate (Electron Microscopy Sciences) in dH2O at 4°C. Then the sections were incubated in lead aspartate solution for 30 min at 60°C. The lead aspartate solution consisted of 0.066 g lead nitrite (Electron Microscopy Sciences), in 10 ml of 0.4% ʟ-aspartic acid in dH2O, and was titrated to a pH 5.5 using a 20% potassium hydroxide solution (Sigma Millipore). The sections were dehydrated by incubation in a series of solutions with increasing ethanol levels (50%, 75%, 85%, 95%, 100%; 3 × 5 min/solution). After dehydration, the sections were placed in propylene oxide for 2 × 10 min (Electron Microscopy Sciences), then a mixture of equal parts propylene oxide and LX112 resin (LX112 Embedding Kits, Ladd Research Industries) for 1 h, and then a mixture of two parts LX112 resin and one part propylene oxide at 25°C overnight. The last infiltration was with 100% LX112 for 4 h under vacuum, and then the tissue was flat embedded in fresh LX112 resin between sheets of Aclar (Ted Pella). The embedded tissue was placed in the 60°C oven for at least 48 h to solidify and then placed in binders at room temperature for long-term storage.
Neural subtype identification and stereology
We identified four neuron subtypes through cytological features seen with Nissl staining and immunolabeling for DARPP-32, CR, and PV. On Nissl-stained sections, several cytological features (cytoplasm surrounding the nucleus, presence of nucleoli or heterochromatin) were used to identify neurons versus other cell types, such as glia or endothelial cells, as described in the algorithm developed by Garcia-Cabezas and colleagues (García-Cabezas et al., 2016). The large cholinergic interneurons can also be identified on Nissl-stained sections based on size. There is a small population of large CR+ neurons with a diameter >20 µm, reported to coexpress the cholinergic marker choline acetyltransferase in primates (Cicchetti et al., 1998), but few were identified within the sampled areas with bright-field and confocal imaging in this study. Large CR+ interneurons that were identified in sampled regions were not counted toward the population of small- to medium-sized, putative GABAergic CR+ interneurons.
We performed unbiased stereology using the commercial system StereoInvestigator with optical fractionator, coupled to a light microscope (Stereo Investigator, RRID:SCR_002526) to estimate the Nissl-stained neuron population, DARPP-32+ MSN population, and PV+ and CR+ interneuron populations. Using this software system, we employed systematic random sampling at regular intervals to extrapolate the number of markers and volume of the region of interest (Fig. 1C). We performed stereology and counting on series of sections spaced 1 mm apart through the NAc shell at 400×. We estimated population sizes using the number weighted section thickness, measured at each counting site. The coefficient of error (Gundersen error, m = 1) was below 10% for all estimated populations, as recommended (Gundersen, 1986). We performed exhaustive counting of the small population of large cholinergic interneurons, stained for Nissl (3 cases: AV, AU, AN) or AChE (1 case: SF). For each case and for each neuron subtype, we calculated the ratio between the estimated density in the medial NAc shell and the lateral NAc shell and the average medial-lateral ratio.
Bright-field microscopy: pathway mapping and bouton diameter measurement
First, we exhaustively mapped the distribution of tracer-labeled axons originating from either the basolateral amygdala (case BM) or the anterior hippocampus (case BT) at 1,000× magnification with oil immersion, through coronal tissue sections spaced 1 mm apart.
We used unbiased stereological methods at 1,000× with oil immersion to estimate the density of labeled boutons on axons originating in the basal amygdala (three cases: BM, BT, BL) and anterior hippocampus (three cases: BT, BU, BQ). Boutons were identified as varicosities on tracer-labeled axons (“Beads on a String,” Fig. 1C). We used stereology (Stereo Investigator, RRID:SCR_002526) in a series of sections spaced 1 mm apart through the NAc shell with parameters: disector 25 × 25 or 50 × 50 µm with a grid size 250 × 250 µm, disector height of 10–15 µm, guard zone of 2 µm. To determine the estimated density, we divided the estimated bouton population by the estimated tissue volume for each case. The coefficient of error (Gundersen error, m = 1) was below 10% for all estimated populations, as recommended (Gundersen, 1986). We divided the NAc shell approximately in half into medial and lateral regions based on the parcellation study by Xia et al. (2019).
We measured the major diameter of boutons using light microscopy for the same three cases per pathway that were used for bouton density estimation. We took a series of images in 0.5 µm increments through each NAc shell section at 1,000× with oil (six stacks per case). We imported image series as z-stacks with a step size of 0.5 µm to the software Reconstruct (SynapseWeb; RRID:SCR_002716) to measure the major diameter of each labeled bouton within the section (Fiala, 2005).
Confocal microscopy: neural subtypes and appositional analysis
We double labeled two cases (BN and BU) for a combination of the markers DARPP-32, CR, CB, or PV to test that they represent nonoverlapping populations in the primate striatum. We imaged the sections using a laser-scanning confocal microscope (Axio Observer Z1, LSM 880, Zeiss) at 200× (ZEN Digital Imaging for Light Microscopy, Zen 2.1 package; RRID:SCR_013672). Imaging for fluorophores was done with lasers from Zeiss: Blue fluorophores with a Diode 405-30 nm laser, green fluorophores with a 488 nm argon ion laser, red fluorophores with a 568 nm DPSS 561-10 laser, and far-red fluorophores with a 633 nm helium neon laser. We deconvolved stacks of imaged sections with the program Huygens Professional (17.10, Scientific Volume Imaging) to minimize the effect of point spread function.
For appositional analysis, we double- or triple-labeled tissue sections (from cases BM and BL for the amygdala pathway, cases BQ and BU for the hippocampal pathway) for the injected tracer and a marker of interest (DARPP-32, CR, or PV). We imaged the labeled tissue on the same laser-scanning confocal microscope as described above, but at a magnification of 630× with oil immersion. We acquired the image stacks with a step size of 0.31–0.33 µm. We adjusted the laser power, gain, pinhole, and offset only before starting an imaging session. As above, we deconvolved image stacks on Huygens Professional (17.10, Scientific Volume Imaging) and saved as Tiff stacks for further analysis on ImageJ. Appositions were identified as the close juxtaposition of a tracer-labeled bouton and marker-labeled structure. We confirmed an apposition by rotating the image 90° in two directions on Huygens Professional. To calculate the apposition rate between the pathways and the three different neuron subtypes, we divided the number of boutons with appositions by the total number of boutons in their respective image stacks. We identified boutons by the same features as seen with bright-field microscopy, described above.
Electron microscopy: neural subtypes and synaptic analysis
We analyzed two cases per pathway (amygdala, cases BM and BL; hippocampus, cases BQ and BT). To prepare samples for serial sectioning, we cut small cubes of tissue from the resin-embedded section, mounted the tissue on premade LX112 resin blocks with fresh resin, and placed the blocks in the 60°C oven for at least 48 h. After solidifying, we stored the blocks at room temperature. We cut the blocks to expose the surface of the tissue in the resin with an ultramicrotome (Ultracut UCT, Leica Microsystems), then cut the tissue into 50–60 nm sections, and sequentially mounted them on pioloform-coated copper slot grids to form an uninterrupted series of sections. We imaged tissue mounted on grids at 80 kV on the transmission electron microscope (100CX, JEOL) at a magnification of 20,000–33,000×. We systematically identified DAB-labeled boutons and imaged through multiple sections for 3D reconstruction and analysis using a digital camera (DigitalMicrograph, Gatan). We converted the acquired image stacks from 16 to 8 bit images on ImageJ and then aligned manually and analyzed the series in the software Reconstruct.
For block face imaging, we cut the tissue embedded in resin into small cubes and placed them on aluminum pins with a conductive epoxy glue (catalog #CW2400, Chemtronics), which solidified in the 60°C oven over 48 h. On the ultramicrotome (Ultracut UCT, Leica Microsystems), we exposed the tissue surface and then covered the tissue and surrounding epoxy with a conductive silver paint (catalog #16035, Ted Pella) to reduce charging artifacts during imaging. We imaged the pins using the 3View 2XP System (Gatan) coupled to a 1.5 kV scanning electron microscope (GeminiSEM 300, Zeiss). First, the surface of the tissue was imaged using a backscatter detector to create a map and then we manually placed 2–15 regions of interest (ROIs). ROIs measured 8,000 by 8,000 pixels and were imaged at 2.5 nm resolution with 2.0 µs per pixel imaging time. After ROIs were imaged on the surface of the tissue, a 100 nm section of tissue was cut from the surface by the microtome automatically. In this way, 50–200 sequential sections were imaged for each ROI. We imported ROI stacks to Gatan Microscopy Suite (GMS) to automatically align them using the GMS algorithm (GMS 3.0, Gatan). We converted the stack 16 to 8 bit images on ImageJ before importing the stacks to Reconstruct for analysis.
Using Reconstruct, we manually outlined the labeled boutons to measure the major diameter, volume, and postsynaptic density (PSD) surface area. When necessary, we identified excitatory and inhibitory synapses using classical features. Excitatory synapses are asymmetric, have round presynaptic vesicles, and have wide synaptic clefts; inhibitory synapses are symmetric, have pleomorphic presynaptic vesicles, and have narrow synaptic clefts (Peters et al., 1991). We reconstructed a select number of outlined structures in 3D. In addition to labeling for NAc markers, postsynaptic structures were presumed to be MSNs if they had spiny dendrites and interneurons if they had aspiny or sparsely spiny dendrites, which serve as reliable morphological features in the striatum (Dimova et al., 1980; Kawaguchi, 1993; Kawaguchi et al., 1995; Petryszyn et al., 2018). Identification by spine formation was further corroborated in the 3D image series obtained and analyzed, where nearly no CR+ or PV+ dendrites had spines, while DARPP-32+ MSNs were often densely spiny, as expected.
Data analysis and statistics
We used Python for data analysis, statistics, and graphing. We used two-tailed t tests to compare the proportion of boutons in medial versus lateral NAc shell. We compared bouton diameters, volumes, PSD areas, and morphologic features between amygdala and hippocampal boutons using a two-tailed t test. We used one-way or two-way ANOVA with post hoc pairwise Bonferroni for analysis of synaptic targets within and across pathways. We used linear regression to view the relationship between the PSD surface area and bouton volume or diameter. Means were reported with SEM. We used ImageJ (RRID:SCR_003070) or Adobe Photoshop (RRID:SCR_014199) to adjust contrast, brightness, and saturation of images. We used Adobe Illustrator CC (RRID:SCR_010279) to prepare figures.
Results
Neuron populations in the NAc shell
The NAc is divided into the core and the shell based on neurochemical, connectional, and functional properties (Meredith et al., 1996; Liu et al., 2021; reviewed in Záborszky et al., 1985; Zahm, 1999; Floresco, 2015; Haber, 2016). In this study we focused exclusively on the NAc shell, which encompasses the main area of convergence of hippocampal and amygdalar projections in the striatum (Groenewegen et al., 1987; Brog et al., 1993; Friedman et al., 2002). A schematic of the experimental design is shown in Figure 1. Cross sections in Figure 1B show decreased CB immunoreactivity in the medial NAc corresponding to the shell and intense AChE staining at its medial sector, which we used to determine the borders of the NAc shell, as described (Martin et al., 1991; Heimer et al., 1997; Haber and McFarland, 1999).
We first performed a quantitative analysis of the NAc shell neurons. MSNs constitute the largest group of neurons in the NAc shell and are reliably labeled with DARPP-32, an intracellular signaling phosphoprotein (Anderson and Reiner, 1991; Ouimet et al., 1992; Greengard et al., 1999). We also focused on three interneuron types: two subtypes of small- to medium-sized GABAergic interneurons, which are CR+ or PV+, and the large cholinergic interneurons. The MSNs and three interneuron types represent nonoverlapping populations (Wu and Parent, 2000; Petryszyn et al., 2018; Del Rey et al., 2022). Figure 2 shows representative images of all four neuron subtypes studied here. We also performed staining for CB, another commonly labeled calcium-binding protein, and found that ∼63% of the CB+ population coexpressed DARPP-32 (n = 165 CB+ neurons), while only a small population showed colocalization with CR (n = 84 CB+ neurons) or PV labeling (n = 71 CB+ neurons), consistent with previous studies (Kawaguchi et al., 1995; Deng et al., 2010).
Images of NAc shell neuron subtypes. A, Bright-field photomicrograph of Nissl-stained NAc neurons. Arrowheads point to two large, putative cholinergic interneurons. B, AChE-stain shows large, cholinergic interneurons (arrowheads), counterstained with neutral red. C, Bright-field photomicrograph of DARPP-32+ neurons (arrowhead points to one of many). D, Bright-field photomicrograph of three CR+ interneurons (brown, arrowheads), counterstained with Nissl. E, Bright-field photomicrograph of two PV+ immunolabeled interneurons (brown, arrowheads) with Nissl counterstain. F, Confocal immunofluorescence shows DARPP-32+ neurons (magenta), CB+ neurons (green, double arrowhead), and DARPP-32+/CB+ MSNs (arrowheads). G, DARPP-32 MSNs (magenta, arrowhead) and CR+ interneurons (green, double arrowhead). H, CR+ neurons (magenta, arrowhead) and PV+ neurons (green, double arrowhead). I, DARPP-32+ MSNs (magenta, arrowhead) and PV+ interneuron (green, double arrowhead). J, CR+ interneurons (magenta) and CB+ neurons (green). Magnified sections show CR+ interneurons (arrowheads) and CB+ neurons (double arrowheads) closely located to each other. K, PV+ interneurons (magenta) and CB+ neurons (green). Magnified sections show CB+ neuron (top, double arrowhead) and PV+ interneuron (bottom, arrowhead). Scale bars, 50 µm. AChE, choline acetyltransferase; CB, calbindin; CR, calretinin; DARPP-32, dopamine- and cAMP-regulated phosphoprotein, 32 kDa; PV, parvalbumin.
The average density of the total neuron population, identified by Nissl-stained morphological features, was 4.2 × 104 neurons/mm3. The most prominent neuron subtype was the DARPP-32+ MSNs, which accounted for ∼86% of all neurons in the NAc shell. Approximately 8.3% of neurons were CR+ interneurons and 2.8% were PV+. Large cholinergic interneurons accounted for 1.2% of all neurons, estimated with exhaustive sampling. Cholinergic interneurons were identified by morphology visualized with Nissl staining (cases AN, AU, AV) and AChE-stained sections (case SF), which yielded similar results (Fig. 3), leaving ∼2% of unidentified neurons, which may be other interneuron subtypes. These findings are consistent with previous primate studies (Del Rey et al., 2022). Table 3 shows neuron populations by case. The neuron populations did not show significant differences in neuron subtype density between the medial and lateral aspects of the NAc shell (Fig. 3D).
Neuron populations in the NAc shell. A, Stereological estimates of the density of all neurons and density of MSNs (DARPP-32+) and three interneuron subtypes (CR+, PV+, Cholinergic) in the NAc shell. Cholinergic population averages from three cases estimated by morphology seen with Nissl-staining (cases AV, AU, and AN) and with AChE staining (case SF). B, Expanded graph of the interneuron population density. C, Pie chart shows the proportions of each neuron subtype out of the total neuron count. D, Ratios of the densities in the medial shell to lateral shell for the total neuron population and the four subtypes of interest (DARPP-32+ MSNs and CR+, PV+, or cholinergic interneurons). Dashed line at a ratio of 1:1, which would signify no difference in density between the medial and lateral aspects of the shell. ACh, cholinergic; CR, calretinin; DARPP-32, dopamine- and cAMP-regulated phosphoprotein, 32 kDa; LAT, lateral NAc shell; MED, medial NAc shell; PV, parvalbumin.
Density of all neurons, DARPP-32+ MSNs, and CR+, PV+, and cholinergic interneurons
Pathways from the amygdala and hippocampus to the NAc shell
We studied terminations originating from the basolateral, basomedial, and cortical nuclei of the amygdala (four cases: BM, BT, BN, BL). The hippocampal pathway originated from anterior portions of the CA1/CA2/CA3, dentate gyrus, subiculum, and prosubiculum (five cases: BT, BU, BQ, BS, BW). We analyzed pathways to the NAc shell ipsilateral to the injection sites. The cases used have been described previously for unrelated experiments (Wang and Barbas, 2018; Timbie et al., 2020; Wang et al., 2021; Joyce et al., 2022, 2023).
We exhaustively traced axons from the basal amygdala (case BM; Fig. 4A) and anterior hippocampus (case BT; Fig. 4B) to produce a map of their termination patterns. The levels of the coronal sections mapped are shown on the outlined rhesus macaque brain (medial side) for each case. Qualitatively, the amygdalar pathway showed a uniformly dense innervation pattern, while the hippocampus showed preferential innervation in the medial aspect of the NAc shell.
Pathway maps and bouton density. A, Exhaustive tracings of labeled axons from the basolateral amygdala (blue) to the medial NAc. Top inset shows the outline of the medial aspect of the macaque brain with vertical lines at sites of the coronal sections used for tracings. Darker gray signifies a more anterior section. Bottom inset shows the injection site in the amygdala (blue). B, Exhaustive tracing of hippocampal axons (red) across the NAc shell. Top inset shows sites of coronal sections. Bottom inset shows hippocampal injection site (red). C, Amygdalar axons with varicosities (black arrowheads) in NAc shell visualized with DAB and counterstained with Nissl. D, Hippocampal axons with varicosities (black arrowhead) in NAc shell visualized with DAB. E, Density of labeled boutons from the amygdalar and hippocampal pathways. F, Ratio of the estimated densities of amygdalar and hippocampal boutons in the medial to lateral NAc shell for each case. Dashed line at 1:1 signifies equal distribution of boutons; cases with higher bouton densities at the medial shell are above 1.0. Key for cases and injection sites at top right. AMY, amygdala; BL, basolateral; C, caudal; Caud, caudate nucleus; Core, NAc Core; D, dorsal; HIPP, hippocampus; L, lateral; LAT, lateral NAc shell; M, medial; MED, medial NAc shell; Put, putamen nucleus; R, rostral; V, ventral.
Stereological analysis of termination density supported the trends seen qualitatively. The density of tracer-labeled boutons on amygdalar axons was generally higher (3.2 × 106 boutons/mm3; cases BM, BT, BL), with the average density being two-fold that of the hippocampal pathway (1.3 × 106 boutons/mm3; cases BT, BS, BW). All three amygdala cases showed dense and relatively uniform innervation across the NAc shell. Pathways from the basolateral amygdala (BLA) showed the highest density of terminations. The cortical (Co)/basomedial (BM) projection showed the lowest density among the amygdala cases but was still denser than two of the three hippocampal cases (Fig. 4E, Table 4). Several factors contribute to the density of labeled boutons, including the specific site and size of the injection of tracer. Nevertheless, this finding is consistent with previous data and suggests that the BLA provides the strongest amygdalar input to the ventral striatum (Fudge et al., 2002).
Estimated densities of labeled terminations originating from the amygdala or hippocampus
The average density of hippocampal boutons was significantly higher for all three cases in the medial shell (1.5 × 106 boutons/mm3) than in the lateral shell (1.1 × 106 boutons/mm3; Fig. 4E,F; Table 4). There was no significant difference in the density of amygdalar boutons in the NAc shell for medial (3.1 × 106 boutons/mm3) and lateral (3.3 × 106 boutons/mm3) sites.
Amygdala and hippocampal pathways interact with NAc shell neurons
We then studied how pathways from the amygdala and hippocampus interact with MSNs and interneurons within the NAc shell. We first quantified the appositions to infer synaptic connectivity on a broad scale (review Stepanyants and Chklovskii, 2005). Appositions were identified as two closely juxtaposed structures, one of which was a tracer-labeled termination and the other was positive for one of the striatal neuron markers (DARPP-32, CR, or PV). Figure 5, A and B, depicts appositions formed by hippocampal and amygdalar boutons in the NAc shell. Approximately half of identified tracer-labeled boutons for both the amygdalar (47.8%) and hippocampal (48.9%) pathways apposed DARPP-32+ MSNs, the projection neurons (Fig. 5C). Further, 33.4% of amygdalar boutons apposed CR+ structures and 17.8% apposed PV+ structures (cases BM, BL, and BN). Notably, for the projection originating from the Co/BM of the amygdala (case BL), only 11.3% of boutons apposed PV+ structures, but in the cases with injection sites in the BLA (case BM) and BLA/BM (case BN), about a quarter of boutons apposed PV+ structures (24.9%). For the hippocampal pathway, 20.8% of boutons apposed PV+ structures (cases BQ and BU), while 35.2% apposed CR+ structures (Fig. 5C). Overall, both pathways apposed the interneurons disproportionately to their respective populations in the NAc shell (Fig. 5D).
Amygdalar and hippocampal boutons innervate interneurons and MSNs in the NAc shell. A, Image of a hippocampal axon (green) forming two boutons that appose one PV+ structure (magenta). B, Amygdalar bouton (magenta) forming an apposition with a CR+ structure (green). Scale bars, 10 µm. White arrowheads indicate apposition location. C, Rates of appositions between labeled boutons on axons originating from the hippocampus (red, 2 cases) or amygdala (blue, 3 cases) and labeled target structures (DARPP-32+ MSNs, CR or PV+ interneurons) with average ± SEM. D, Ratio of apposition frequency to population sizes of MSNs, CR+ interneurons, and PV+ interneurons for hippocampal (red) and amygdalar (blue) pathways. E, Frequency of amygdalar boutons (n = 179; pooled data from all analyzed tissue series) forming synapses with dendritic shafts and spines shown at left (white/gray). The three columns to the right (white/blue shades) show innervation frequencies from double-labeled tissue for tracer/DARPP-32, or tracer/CR or tracer/PV. F, In a similar organization as E: Frequency of hippocampal boutons (n = 156; pooled data from all tissue series) forming synapses with dendritic shafts and spines (white/gray). The three columns to the right show frequency of postsynaptic target identities from double-labeled tissue (tracer/DARPP-32 or tracer/CR or tracer/PV). G, Pie chart shows estimated innervation pattern for the amygdalar pathway in the NAc shell in tissue labeled for DARPP-32 (to label MSNs) or CR or PV to label nonoverlapping interneuron types. Proportions are based on data from double-labeled tissue in each column, as indicated. H, Pie chart shows estimated innervation pattern for the hippocampal pathway in the NAc shell. Proportion of innervated DARPP-32+, CR+, or PV+ structures are based on data from double-labeled tissue series (as in F).
Interactions at the synaptic level
We first tested if it was possible to accurately categorize postsynaptic spines as belonging to MSNs and postsynaptic aspiny dendritic shafts as belonging to interneurons. We found one innervated CR+ spine and no PV+ innervated spines out of the 109 postsynaptic spines identified in tissue samples labeled for CR or PV across pathways. Only a few innervated dendritic shafts were spiny or positive for the MSN marker DARPP-32. This finding corroborated the assumption that postsynaptic spines belong to MSNs and labeled dendritic shafts suggest interneuron identity for the two pathways. Further, approximately half of the innervated spines were DARPP-32+ with the majority of those identified at the parent dendrite. Innervated spines that could not be followed to the parent dendrite in the series were recorded as DARPP-32 negative but likely belonged to MSNs. Only one targeted dendritic shaft showed DARPP-32 labeling (<5% of targets in tissue labeled with DARPP-32). Figure 5, E and F, shows by pathway the frequency of synapses on postsynaptic dendritic spines and shafts across all samples and separately for samples stained for DARPP-32, CR, or PV.
About half of amygdalar boutons formed synapses with spines belonging to putative MSNs (45.8 ± 3.3%), with the remaining synapses formed with putative interneuron dendritic shafts (54.2%) based on pooled data from series labeled with either DARPP-32, CR, or PV. In series labeled for DARPP-32, the amygdala pathway formed 24.4% (±7.1) of synapses with DARPP-32+ spines (Table 5). No innervated dendritic shaft was positive for DARPP-32 or showed spiny morphology in these sections. The amygdala formed synapses with CR+ structures slightly more frequently than with PV+ structures, but it was not statistically significant (CR: 19.7 ± 5.6%; PV: 16.9 ± 5.5%; Fig. 5E). Given that CR+ and PV+ interneurons are nonoverlapping populations, this leaves ∼17.6% of innervated dendritic shafts, presumed to belong to interneurons, that were not labeled for CR or PV (Fig. 5G), which may represent an interneuron subtype not explored here.
Amygdalar and hippocampal innervation patterns at the synaptic level
Across two cases (BQ, BT), about half of hippocampal synapses were on spines (54.2 ± 3.7%) from putative MSNs (45.1% on dendritic shafts from putative interneurons) based on pooled data from series labeled with DARPP-32, CR, or PV, and one series labeled for both CR and PV (Table 5). In sections labeled for DARPP-32, nearly half of the innervated spines were positive for DARPP-32 and one dendrite was DARPP-32+ (Fig. 5F). The hippocampal pathway formed synapses with PV+ dendrites (20.4 ± 4.5%) at a slightly higher rate than CR+ dendrites (16.9 ± 5.6%; Fig. 5H), though there was variability between case BQ (PV:26.2%, CR:22.2%) and BT (PV: 10.5%, CR:13.2%). Labeling in case BT was overall lighter than BQ, and there was no significant difference between the proportion of synapses with dendritic spines between the two cases.
Additionally, the hippocampal boutons more often formed multiple synapses. In one case (BQ), we observed nine instances of hippocampal boutons forming synapses with a PV-negative spine and a PV+ dendritic shaft. Out of all labeled series, 19.7% of hippocampal boutons formed more than one synapse, in contrast to only 7.4% of amygdalar boutons. Two hippocampal boutons formed three synapses (case BT), one of which formed synapses with two spines and one PV+ dendrite (Fig. 6A) and the other formed synapses with one spine and two unlabeled dendrites. In CR-labeled tissue, we have not yet identified a multisynaptic hippocampal bouton that innervated a CR+ target. In tissue labeled for PV, three amygdalar boutons formed synapses with one MSN and one unlabeled presumed interneuron. However, we did not observe any multisynaptic amygdalar boutons that targeted a PV+ interneuron.
Hippocampal boutons form multiple synapses. A, Photomicrograph of a hippocampal bouton (orange) and a reconstruction of that bouton forming synapses with one spine (blue arrowhead), one unlabeled dendritic shaft (green arrowhead) and one PV+ dendritic shaft (yellow arrowhead) in NAc shell. Reconstruction shows a normal (top, left) and expanded view (top, right) of the detached complex with PSD (purple). B, Hippocampal bouton forming two synapses, one on a PV+ dendritic shaft (gray, black arrow) and one with an unlabeled dendritic spine (blue, black arrowhead).
We also analyzed several structural features of axonal terminals and postsynaptic sites that correlate with synaptic function, including bouton size, mitochondrial content, and PSD shape. The amygdalar pathway tended to have larger boutons with a wider distribution of bouton diameters, measured from series of bright-field images and EM images (Fig. 7A,B). Measurement of images from EM showed that the average bouton diameter for the amygdala pathway was 1.11 ± 0.04 µm (range, 0.39–2.37 µm; n = 155) and average volume was 0.38 ± 0.03 µm3 (n = 84). The average diameter for the hippocampal pathway was slightly smaller at 0.98 ± 0.04 µm (0.48–1.71 µm; n = 126); the average volume was 0.29 ± 0.03 µm3 (n = 74). There were slight differences between these absolute values obtained with EM imaging at the synaptic level compared with those obtained via bright-field at the system level, but trends were consistent across levels of resolution and cases (Fig. 7).
Presynaptic features of pathways from the amygdala and hippocampus to NAc shell. A, Major diameter frequency distributions of boutons on amygdalar (dashed blue, n = 2, 109 boutons) and hippocampal (red, n = 2,136) axons, measured from bright-field photomicrographs. Example of amygdalar axon with outlined boutons (red) in the NAc shell (top, right). B, Major diameter frequency distributions of boutons forming synapses in the NAc shell measured from 2-D EM images (amygdalar boutons, dashed blue, n = 155; hippocampal boutons, red, n = 126). C, Relationship of PSD surface area (amygdalar, blue, n = 84; hippocampal, red, n = 74) and bouton volumes measured from 3D serial EM images. Solid points represent perforated synapses and open points are round synapses. D, Relationship of PSD surface area (amygdalar, blue, n = 152; hippocampal, red, n = 102) and EM-measured bouton diameters. Solid points represent perforated synapses, and open points represent round synapses. E, EM-measured amygdalar and hippocampal bouton average diameters (solid squares) and average diameters of boutons with and without mitochondria (open squares), measured in EM. F, EM photomicrograph shows a labeled hippocampal bouton forming a synapse with a spine in the NAc shell. The bouton contains a large mitochondrion. *p < 0.05 (independent two-sample t test). AMY, amygdala; EM, electron microscopy; HIPP, hippocampus; NAc, nucleus accumbens; PSD, postsynaptic density.
Bouton volume, measured with 3-D reconstruction from serial EM imaging, and major diameter, measured on 2-D EM photomicrographs, were positively correlated with PSD surface area and showed that hippocampal boutons tended to have larger PSD surface areas compared with similarly sized amygdalar boutons (Fig. 7C,D). Larger PSDs are associated with higher AMPA receptor density, are more often perforated, and suggest higher synaptic efficacy (Geinisman et al., 1993). The average PSD surface area was significantly larger in the hippocampal pathway compared with the amygdalar pathways (HIPP: 0.15 ± 0.01 µm2, n = 102; AMY: 0.12 ± 0.01 µm2, n = 152; t test p = 0.002). In turn, a larger percentage of synapses formed by the hippocampus were perforated (22.7%) than the amygdala (8.9%). Figure 8 shows several examples of structures formed by amygdalar and hippocampal terminations, including round and perforated synapses. There were no statistically significant differences in the number of perforated synapses on interneurons versus MSNs for either pathway. In our dataset, the hippocampal pathway formed more perforated synapses with spines (∼57%) on putative MSNs, than with dendritic shafts, which included putative interneurons that were PV+ or CR+. The amygdalar pathway rarely formed perforated synapses and, when present, they were more often on dendritic shafts (∼58%) than on dendritic spines, and some were CR+.
Structure of hippocampal and amygdalar terminations. A, EM photomicrograph of a labeled hippocampal bouton (orange-shaded) forming a synapse (white arrows) with a spine (light blue-shaded). B, 3-D reconstruction of hippocampal boutons (orange) from one axon forming two synapses (purple, arrows) on the same dendrite (gray) and two synapses (red, double arrows) on two different dendritic spines (blue). C, Reconstruction of a hippocampal bouton and axon (orange) winding around and forming a nonperforated synapse (purple, arrow) with a dendritic shaft. D, Amygdalar bouton (blue) forming a synapse (orange, arrow) on a dendritic shaft (gray). E, Amygdalar bouton (blue) forming a perforated synapse (orange) on a spine (purple, left). F, Amygdalar bouton (blue) forming a nonperforated synapse (purple, arrow) with a dendritic shaft (gray). Scale bar, 1 µm. Scale cubes, 1 µm3.
The majority of amygdala boutons contained mitochondria (74.2%, n = 163) of which half formed synapses with putative interneurons. Only half of hippocampal boutons contained mitochondria (52.4%, n = 126) of which only a third formed synapses with putative interneurons. The postsynaptic targets of hippocampal boutons that contained mitochondria were mainly on spines (66%). Across the pathways from the hippocampus and amygdala, boutons that contained mitochondria were significantly larger than those without mitochondria (Fig. 7E,F). There was no significant difference between the sizes of amygdalar and hippocampal boutons with mitochondria or between those without mitochondria (Fig. 7E).
Discussion
The amygdala and hippocampus densely innervated the NAc shell, with comparatively denser and broader amygdalar terminations, consistent with other primate studies (Russchen et al., 1985; Fudge et al., 2002, 2004). Both pathways disproportionately innervated interneurons relative to their respective populations, suggesting robust influence on the internal NAc shell computations. In turn, only half of synapses were with the dominant MSN projection neurons (Fig. 9A). The amygdalar pathway had large boutons with mitochondria, associated with high activity and stability (Pierce and Lewin, 1994; Lees et al., 2019). The hippocampal pathway formed large synapses, associated with high probability of multivesicular release and postsynaptic effects (O'Donnell and Grace, 1995; Desmond and Weinberg, 1998; Ganeshina et al., 2004; Nava et al., 2014; Medalla and Luebke, 2015). These amygdalar and hippocampal circuit patterns differ from rodents and suggest their specialization in the NAc shell, as elaborated below.
Summary. A, Pie chart of population proportions of CR+ and PV+ interneurons (innermost circle), amygdalar synapses with CR+ and PV+ structures identified on EM images (middle circle), and hippocampal synapses with CR+ and PV+ structures identified on EM images (outer circle). B, Amygdalar and hippocampal innervation patterns across striatum based on this study and literature, with downstream targets of the shell and core. AMY, amygdala; CR, calretinin; HIPP, hippocampus; PV, parvalbumin; SNpr, substantia nigra pars reticulata; STN, subthalamic nucleus; VP, ventral pallidum; VTA, ventral tegmental area.
Innervation of the internal NAc shell circuitry
Disproportionate innervation of CR+ and PV+ interneurons by the amygdala and hippocampus was unexpected, in view of their smooth dendrites and likely smaller surface area available for synapses than the MSN dendrites, which are enriched with spines (Chang and Kitai, 1985; Prensa et al., 1998; Rymar et al., 2004; Gertler et al., 2008; Krstonošić et al., 2023). This pattern also differs from the substantially lower level of innervation by amygdala to hippocampus (Wang and Barbas, 2018) or hippocampus to amygdala (Joyce et al., 2023). This pattern is also distinct from the predominant innervation of MSNs in the rat striatum (Sesack and Pickel, 1990; Johnson et al., 1994).
PV+ interneurons accounted for fewer than 3% of all NAc shell neurons, but ∼20% of terminations were onto them. Fast-spiking PV+ interneurons innervate perisomatic elements of hundreds of MSNs (Koós and Tepper, 1999), exerting strong inhibition (Parthasarathy and Graybiel, 1997; Planert et al., 2010) that organizes MSN population activity (Stern et al., 1998; Humphries et al., 2010; Howe et al., 2011; Scudder et al., 2018; Pisansky et al., 2019; Duhne et al., 2021); review Schall et al., 2021). These circuit patterns likely help select highly active, coincident excitatory input, as in mice (Trouche et al., 2019), increasing the dynamic network range (Pouille et al., 2009) to support cognitive tasks (Banaie Boroujeni et al., 2020).
The implications of both pathways innervating CR+ interneurons are unclear. Primate CR+ interneurons have increased in number and morphologic variation compared with rats (Figueredo-Cardenas et al., 1996; Hussain et al., 1996; Garas et al., 2018; Lecumberri et al., 2018). Sequencing studies have identified a primate-specific interneuron that partially overlaps with CR+ populations (Krienen et al., 2020; Schmitz et al., 2022; Garma et al., 2024). If the primate striatum has a similar evolutionary trajectory as the cortex (DeFelipe, 1997; del Río and DeFelipe, 1997; Meskenaite, 1997), the GABAergic CR+ interneurons may inhibit interneurons and thus disinhibit MSNs. Consequently, activation of CR+ interneurons by amygdalar or hippocampal pathways may fine-tune integration in MSN ensembles to promote behavioral flexibility.
Some innervated putative interneurons were not CR+ or PV+ and likely belong to other subtypes not explored here, including low-threshold spiking (LTS) or cholinergic interneurons. LTS interneurons make up 1–2% of striatal neurons; often express neuropeptide Y, somatostatin (SOM), or nitric oxide synthase (NOS); are tonically active; and innervate the distal dendrites of MSNs and cholinergic interneurons (Beal et al., 1986; Elghaba et al., 2016; Straub et al., 2016). In rats and mice, amygdalar and hippocampal pathways to NAc innervate LTS interneurons that are SOM+ (Scudder et al., 2018) or NOS+ (French et al., 2005), which may promote synaptic plasticity via nitric oxide release (Calabresi et al., 1999; Nishi et al., 2005; Sagi et al., 2014). Cholinergic interneurons may also contribute to unlabeled postsynaptic sites (Guo et al., 2015; Baimel et al., 2022), which can modulate striatal activity broadly (Prensa et al., 1998; Pakhotin and Bracci, 2007).
Functional implications
The density and mitochondrial content of the amygdalar pathway suggests capacity for sustained activity and regulation of calcium levels for vesicular release and synaptic remodeling (Tang and Zucker, 1997; Brodin et al., 1999; Scotti et al., 1999); review Thomson, 2000). The basolateral amygdala has distinct projection neurons with tonic or burst firing (Rainnie et al., 1993; Pare et al., 1995; Gonzalez Andino and Grave de Peralta Menendez, 2012). Amygdalar stimulation induces subthreshold MSN depolarization (O'Donnell and Grace, 1995; Britt et al., 2012). The amygdalar pathway may help signal the current internal state, as well as salient events through increased activity in NAc shell to shift postsynaptic MSNs toward a plastic state (Popescu et al., 2007; Xia et al., 2020; Yu et al., 2022) and enhance memory consolidation of cues and outcome values (Ambroggi et al., 2008). Low amygdala activity may engage interneurons more than MSNs, which have a hyperpolarized resting membrane potential, to increase inhibitory tone in the NAc shell (review Tepper et al., 2018).
The hippocampal terminations had perforated synapses and large PSD sites, suggesting a potent driver of NAc activity (Desmond and Weinberg, 1998). In mice, brief stimulation of hippocampal terminations elicits larger amplitude postsynaptic potentials of MSNs compared with amygdalar or medial prefrontal terminations (Britt et al., 2012), which may be conserved in monkeys (Deng et al., 2010). Several multisynaptic hippocampal boutons found here uniquely innervated a spine and a PV+ dendrite. Simultaneous MSN and PV+ interneuron excitation may support feedforward inhibition, providing a brief window for input summation at MSNs. Consequently, signals associated with context from hippocampus may gate or enhance other inputs in the NAc shell, as in rats (O'Donnell and Grace, 1995; French and Totterdell, 2002, 2003). Efficient activation of MSNs coupled to the constraint on MSN dendritic integration and output occurring with simultaneous activation of MSN and PV+ interneurons suggests that temporal specificity is a key feature of hippocampal signal transmission.
Implications for behavior and disease
The NAc core and shell have dissociable roles in motivated behavior (Corbit et al., 2001, 2016; Hall et al., 2001; Corbit and Balleine, 2005). The NAc core may have a role in approach behavior, while the shell is implicated in outcome valuation, suppressing irrelevant stimuli (Gal et al., 2005; Floresco et al., 2008; Jones et al., 2010), and basic motivational drives like feeding (Urstadt et al., 2013), where contextual signals from hippocampus are likely relevant (Fig. 9B). The broad amygdala termination in the NAc shell suggests that internal state representations may be integrated with different aspects of motivated behavior. Integration of amygdalar and hippocampal signals may occur through convergence onto individual MSNs and local interneurons (Harvey and Lacey, 1997; Floresco et al., 2001; Roozendaal et al., 2001; French and Totterdell, 2003; LaLumiere et al., 2005), potentially through the expanded interneuron innervation described here. These pathways also shape activity influenced by other dense glutamatergic (Xia et al., 2020; Yu et al., 2022) or neuromodulatory inputs (O'Donnell and Grace, 1995; Christoffel et al., 2021).
Disruption of NAc shell circuits is evident in disease states (Meredith, 1999; Cadoni et al., 2005; Zinsmaier et al., 2022) including addiction and relapse, which vary between pathways and environment (Bertran-Gonzalez et al., 2008; Calipari et al., 2016; LeGates et al., 2018; Wright et al., 2020). Cocaine use potentiates hippocampal-MSN synapses that express D1-receptors (Pascoli et al., 2014). This change may be particularly consequential since D1+ MSNs project to the ventral tegmental area (Lu et al., 1998) and may alter dopamine release in lateral striatum to influence cognitive processes (Haber et al., 2000; Ikeda et al., 2013; Wouterlood et al., 2018). Maturation of silent amygdala-MSN synapses occurs during cocaine withdrawal and is correlated with increased craving and risk of cue-associated relapse (Lee et al., 2013; Wright et al., 2020). Increased sensitivity to substance use-related cues through strengthened hippocampal inputs and increased density of active amygdalar synapses may overactivate circuits associated with addiction-related behavior.
Innervation of interneurons by the amygdala and hippocampus in the healthy primate NAc suggests that maladaptive modifications likely extend to these synapses. Modeling studies of interneuron innervation in other structures have functional implications and dysregulation in disease (John et al., 2016, 2018, 2024). In the striatum, PV+ interneuron disruption with increased MSN excitability is associated with disorders of motivation and compulsivity (Kalanithi et al., 2005; Gittis et al., 2011; Burguiere et al., 2013). Striatal CR+ interneurons are also altered in disease states in monkeys and rats (Petryszyn et al., 2016; Boracı et al., 2020), but it is unclear how they may influence symptomatology.
Discussion of function is based on studies relating anatomy, physiology, and behavior in rodents and primates. Glutamatergic input to interneurons is an integral component in NAc shell computations in primates, suggesting divergence from circuit mechanisms in mice and rats, where the amygdala and hippocampus influence NAc output through direct MSN innervation (Sesack and Pickel, 1990; Johnson et al., 1994). The distinct pathway features from the amygdala and hippocampus to the primate NAc shell suggest that internal emotional states and environmental cues differentially mediate circuits underlying goal-directed behaviors in the NAc shell.
Footnotes
We thank Mary-Kate Joyce (PhD), Jess Holz (MFA), and Tara McHugh (MA) for expert help with electron microscopy and technical assistance. This work was supported by National Institutes of Health, National Institute of Mental Health grants (R01MH136013; R01MH9500310563; and R01MH9500305870).
The authors declare no competing financial interests.
- Correspondence should be addressed to Helen Barbas at barbas{at}bu.edu.
