Defining the caudal hypothalamic arcuate nucleus with a focus on anorexic excitatory neurons

Abstract

Key points

• Excitatory glutamate neurons are sparse in the rostral hypothalamic arcuate nucleus (ARC), the subregion that has received the most attention in the past. In striking contrast, excitatory neurons are far more common (by a factor of 10) in the caudal ARC, an area which has received relatively little attention.

• These glutamate cells may play a negative role in energy balance and food intake. They can show an increase in phosphorylated Stat‐3 in the presence of leptin, are electrically excited by the anorectic neuromodulator cholecystokinin, and inhibited by orexigenic neuromodulators neuropeptide Y, met‐enkephalin, dynorphin and the catecholamine dopamine.

• The neurons project local axonal connections that excite other ARC neurons including proopiomelanocortin neurons that can play an important role in obesity.

• These data are consistent with models suggesting that the ARC glutamatergic neurons may play both a rapid and a slower role in acting as anorectic neurons in CNS control of food intake and energy homeostasis.

Abstract

Here we interrogate a unique class of excitatory neurons in the hypothalamic arcuate nucleus (ARC) that utilizes glutamate as a fast neurotransmitter using mice expressing GFP under control of the vesicular glutamate transporter 2 (vGluT2) promoter. These neurons show a unique distribution, synaptic characterization, cellular physiology and response to neuropeptides involved in energy homeostasis. Although apparently not previously appreciated, the caudal ARC showed a far greater density of vGluT2 cells than the rostral ARC, as seen in transgenic vGluT2‐GFP mice and mRNA analysis. After food deprivation, leptin induced an increase in phosphorylated Stat‐3 in vGluT2‐positive neurons, indicating a response to hormonal cues of energy state. Based on whole‐cell recording electrophysiology in brain slices, vGluT2 neurons were spontaneously active with a spike frequency around 2 Hz. vGluT2 cells were responsive to a number of neuropeptides related to energy homeostasis; they were excited by the anorectic peptide cholecystokinin, but inhibited by orexigenic neuropeptide Y, dynorphin and met‐enkephalin, consistent with an anorexic role in energy homeostasis. Dopamine, associated with the hedonic aspect of enhancing food intake, inhibited vGluT2 neurons. Optogenetic excitation of vGluT2 cells evoked EPSCs in neighbouring neurons, indicating local synaptic excitation of other ARC neurons. Microdrop excitation of ARC glutamate cells in brain slices rapidly increased excitatory synaptic activity in anorexigenic proopiomelanocortin neurons. Together these data support the perspective that vGluT2 cells may be more prevalent in the ARC than previously appreciated, and play predominantly an anorectic role in energy metabolism.

Key points

• Excitatory glutamate neurons are sparse in the rostral hypothalamic arcuate nucleus (ARC), the subregion that has received the most attention in the past. In striking contrast, excitatory neurons are far more common (by a factor of 10) in the caudal ARC, an area which has received relatively little attention.

• These glutamate cells may play a negative role in energy balance and food intake. They can show an increase in phosphorylated Stat‐3 in the presence of leptin, are electrically excited by the anorectic neuromodulator cholecystokinin, and inhibited by orexigenic neuromodulators neuropeptide Y, met‐enkephalin, dynorphin and the catecholamine dopamine.

• The neurons project local axonal connections that excite other ARC neurons including proopiomelanocortin neurons that can play an important role in obesity.

• These data are consistent with models suggesting that the ARC glutamatergic neurons may play both a rapid and a slower role in acting as anorectic neurons in CNS control of food intake and energy homeostasis.

Introduction

The hypothalamic arcuate nucleus (ARC) plays a key role in two important aspects of homeostatic regulation. The ARC regulates endocrine secretions by releasing trophic factors into the portal vascular supply of the median eminence; these trophic factors are then carried by the blood to the pituitary to regulate hormone secretions. The ARC also plays a central role in regulating food intake and energy homeostasis (Woods et al. 1998; Meister, 2007; Andermann & Lowell, 2017). Within the ARC, there has been a strong focus on two key neuron types that participate in energy homeostasis, the anorexic proopiomelanocortin (POMC) cells (Huszar et al. 1997; Yaswen et al. 1999; Chen et al. 2000, Chen et al. 2004) and neurons that release agouti‐related peptide (AgRP) and neuropeptide Y (NPY), which play an orexigenic role (Clark et al. 1984; Woods et al. 1998; Andermann & Lowell, 2017).

For many years, most of the focus on signalling from neurons of the ARC has concentrated on the neuromodulators, particularly neuropeptides; fast neurotransmission within the ARC has received considerably less attention. More recent experiments provide strong evidence that genetically reducing or eliminating fast transmitters from specific peptidergic cells can exert a profound dysfunctional effect on the actions of the cells (Tong et al. 2008; Wu et al. 2009; Vong et al. 2011; Wu & Palmiter, 2011; Atasoy et al. 2012; Krashes et al. 2013), underlining the potential importance of fast transmitters. Most of the functional studies in the ARC to this point have focused on the fast inhibitory amino acid transmitter GABA (Acuna‐Goycolea et al. 2005); many ARC neurons synthesize GABA along with multiple peptide neuromodulators, and both the fast inhibitory transmitter and the slower neuropeptides appear necessary for the normal regulation of energy homeostasis (Wu & Palmiter, 2011; van den Pol, 2012; Krashes et al. 2013; Liu et al. 2013).

Here we examine in depth a unique class of ARC neurons that utilize glutamate as an excitatory neurotransmitter and have received relatively little attention. Excitatory neurons in the brain can be identified by the expression of vesicular glutamate transporters (vGluT) which include vGlutT1, ‐2 and ‐3; the primary vesicular glutamate transporter in the hypothalamus is vGluT2 (Fremeau et al. 2001). Glutamate is the primary fast neurotransmitter in the medial hypothalamus and also in the ARC, as determined by brain slice and culture recording and immunocytochemistry (van den Pol et al. 1990; van den Pol, 1991; van den Pol & Trombley, 1993; Belousov & van den Pol, 1997; Kiss et al. 2006). ARC neurons show robust expression of AMPA, kainate and NMDA ionotropic glutamate receptors (van den Pol et al. 1994a), as well as metabotropic glutamate receptors (van den Pol, 1994, van den Pol et al. 1994c; Romano et al. 1995), indicating multiple permutations of electrical responses to glutamate within the ARC.

Although there is no strict segregation of neuron phenotypes, most different types of neurons within the ARC show some anatomical preferences in distribution, with POMC cells in the lateral part of the ARC, AgRP neurons more medial (Elias et al. 1998; Saper & Lowell, 2014), tyrosine hydroxylase (TH)/dopamine neurons more dorsomedial, and TH cells that may contain growth hormone releasing hormone, but not dopamine, more ventrolateral (Zhang & van den Pol, 2016). Given the heterogeneous distribution of other cell types, we examined the anatomical distribution of vGluT2 cells within the ARC and found a far greater density in the caudal than in the rostral parts of the nucleus – a distribution imbalance not previously appreciated.

Methods

Ethical approval

All work with mice described here was approved by the Institutional Animal Care and Use Committee (approval reference 10117). For anaesthesia, ketamine (100 mg/kg) + xylazine (10 mg/kg) or pentobarbital (60 mg/kg) were used. All survival surgery included application of analgesics (meloxicam, 3 mg/kg) at the time of surgery and for the next 2–3 days to reduce pain or discomfort. Mice were killed with an overdose of ketamine (400 mg/kg) + xylazine (40 mg/kg) or pentobarbital (150 mg/kg).

vGluT2‐GFP transgenic mice

We used transgenic mice that express enhanced green fluorescent protein (GFP) in glutamatergic neurons of the hypothalamus under control of the vesicular glutamate transporter 2 (vGluT2). We have described this mouse previously and corroborated excitatory neuron selectivity with single cell RT‐PCR (Huang et al. 2006; Fu & van den Pol, 2008; Dumalska et al. 2008) in studies on the paraventricular nucleus of the thalamus (Huang et al. 2006), the hypothalamic ventromedial nucleus (Fu and van den Pol, 2008) and excitatory cells of the preoptic/septal area (Dumalska et al. 2008). Assessed by single cell RT‐PCR, 30 of 30 GFP‐expressing neurons of the hypothalamic ventromedial nucleus expressed vGluT2 mRNA in transgenic mice with GFP expression under control of a vGluT2 promoter (Fu & van den Pol, 2008), and all 18 GFP‐negative cells from the striatum showed no vGluT2 expression, consistent with the high proportion of GABA cells there. We also used a mouse that expresses Cre recombinase in vGluT2‐positive glutamatergic neurons of the ARC (Vong et al. 2011) obtained from Jackson Laboratory (Bar Harbor, ME, USA; Stock No. 016963). Mice were maintained on a 12/12 h light/dark cycle. Water and food were supplied ad libitum according to institutional regulations.

In some experiments we also used mice in which neurons synthesizing proopiomelanocortin (POMC) expressed GFP (Cowley et al. 2001), kindly provided by Dr M. Low.

vGluT2 RNA expression in the arcuate nucleus

Quantitative real‐time PCR (RT‐qPCR) was used to assess the level of vGlut2 mRNA in microdissected sections of the ARC isolated from vGlut2 mouse brain slices, prepared as described in the electrophysiological methods section below. Briefly, freshly cut slices immersed in ACSF were visualized under visible and fluorescent illumination using an Olympus SZX12 dissecting microscope equipped with a GFP filter set. The ARC was dissected from each slice. Microdissected tissues were stored at 4°C overnight in RNAlater (Qiagen, Carlsbad, CA, USA) and total RNA was extracted the following day using the RNeasy Micro kit (Qiagen). Total RNA was reverse transcribed with the Superscript III Reverse Transcriptase Kit (Invitrogen, Waltham, MA, USA) using the following gene‐specific oligos: vGluT2 5ʹ‐CTT ATA GGT GTA CGC GTC TTG‐3ʹ; EGFP 5ʹ‐GCT CAG GTA GTG GTT GTC‐3ʹ; beta‐actin 5ʹ‐CAA CGT CAC ACT TCA TGA TG‐3ʹ. Quantitative PCR (qPCR) was performed in 96‐well dishes on an iCycler iQ Multicolor Real‐Time Detection System (Bio‐Rad, Hercules, CA, USA) using TaqMan gene expression assays (Applied Biosystems, Waltham, MA, USA). The vGluT2 assay (Mm00499876_m1) was already commercially available from ABI, however the assay for detecting EGFP was custom synthesized and consisted of a forward (5ʹ‐GAG CGC ACC ATC TTC TTC AAG‐3ʹ) and reverse (5ʹ‐TGT CGC CCT CGA ACT TCA C‐3ʹ) primer pair and a FAM Dye labelled oligo probe (5ʹ‐ACG ACG GCA ACT ACA‐3ʹ). Results were normalized to beta‐actin expression using the Endogenous Control Assay (ABI No. 4352933E), mouse ACTB FAM Dye/MGB probe, non‐primer limited, amplicon size of 115 bp. All RT‐qPCRs were performed using the TaqMan Gene Expression Master Mix (ABI) according to the manufacturer's instructions. Corroboration from in situ hybridization studies of vGluT2, AgRP and POMC was obtained from the Allen Brain Institute Mouse Brain Atlas and used with permission. To determine whether vGluT2‐GFP cells also expressed alpha‐melanocyte stimulating hormone (MSH), a peptide found in POMC neurons, we utilized immunocytochemistry with a primary sheep antibody against alpha‐MSH (Millipore, Burlington, MA, USA) and a secondary anti‐sheep antibody bound to a red fluorophore.

Distribution of vGluT2 cells

Slides were mounted on a microprojector (Ken‐A‐Vision Inc., Raytown, MO, USA) and the hypothalamic region and the borders of the arcuate nucleus were traced, along with a rectilinear grid used as a scale reference. These tracings were then digitally scanned and the area of the ARC computed using ImageJ software (NIH). Slides were then examined under GFP fluorescent illumination with an Olympus IX70 microscope and the number and positions of the fluorescent cells were recorded onto copies of the previously drawn tracings of the ARC. The density of fluorescent cells in sections of the ARC were computed by dividing the number of fluorescent cells found by the area (in μm²) of the ARC. Slides were then compared to the anatomical plates from Paxinos & Franklin (2001) and a closest match for the ARC and immediately surrounding regions was found for each. Percentage values were used to indicate the position of each section along a rostral‐caudal axis running through the ARC with 0% representing the most rostral portion of the ARC (bregma –1.06 mm) and 100% representing the most caudal (bregma –2.80 mm). On photographic images, contrast and brightness were corrected with Adobe Photoshop using the whole frame of a photograph.

Brain slice preparation for electrophysiology

Mice 17–60 days old were anaesthetized with sodium pentobarbital or ketamine + xylazine and the brain was quickly removed from the skull and a block of tissue containing the hypothalamus was dissected out and placed in oxygenated (95% O₂, 5% CO₂) ice‐cold high‐sucrose solution that contained (in mm): sucrose, 220; KCl, 2.5; MgCl₂, 6; CaCl₂, 1; NaH₂PO₄, 1.25; NaHCO₃, 26; glucose, 10. Coronal hypothalamic slices (200–300 μm thick) containing GFP positive neurons in the ARC were then obtained using a vibratome; slices around 300 μm thick were used for experiments examining synaptic activity in microdrop and optogenetic experiments, and thinner slices around 200 μm were used to study responses to neuromodulators. Slices were transferred to a maintenance chamber kept at 37°C for 20 min in normal oxygenated (95% O₂, 5% CO₂) artificial cerebrospinal fluid (ACSF) that contained (in mm): NaCl, 124; KCl, 3; MgCl₂, 2; CaCl₂, 2; NaH₂PO₄, 1.23; NaHCO₃, 26; glucose, 10. Slices were then equilibrated for at least 1 h at room temperature in the same solution as above, and were then transferred to the recording chamber mounted on a BX51WI Olympus upright microscope equipped with both fluorescent and differential interference capabilities (DIC).

Electrophysiology

We used cell‐attached as well as whole‐cell voltage‐ and current‐clamp recordings from identified, GFP‐expressing, ARC vGlut2‐containing neurons. Green‐fluorescent cells were identified by excitation with blue light, and were then approached and patched under differential interference contrast microscopy. For cell‐attached and patch‐clamp recordings we used 3–5 MΩ glass pipettes (Sutter Instruments, Inc., Novato, CA, USA) pulled with a horizontal pipette puller (Sutter Instruments, Inc.).

Cell‐attached recordings with a loose‐seal low‐resistance configuration (Perkins, 2006) were obtained by gently applying negative pressure to an ACSF‐filled electrode placed adjacent to the GFP‐expressing neuron. Whole‐cell configuration was achieved by approaching a target cell under positive pressure, and then releasing this pressure upon contact between the tip of the electrode and the cell membrane. In some cases, negative pressure through the electrode was necessary to obtain a gigaseal. Brief negative pressure was used to break through the membrane and obtain whole‐cell configuration (Acuna‐Goycolea & van den Pol, 2005; 2009).

For whole‐cell recordings, pipettes were filled with an internal solution containing (in mm): 130 KMeSO₄ or K‐gluconate, 1 MgCl₂, 10 HEPES, 1.1 EGTA, 2 Mg‐ATP, 0.5 Na₂‐GTP, and 10 sodium phosphocreatine, pH 7.3 with KOH. EPSCs and IPSCs were isolated by switching the membrane potential between –70 and 0 mV, respectively, and by application of selective receptor antagonists for ionotropic GABA (bicuculline: BIC; 30 μM) or glutamate receptors AP5 (50 μM) and CNQX (10 μM)(all from Tocris Bioscience, Minneapolis, MN, USA).

All electrophysiological recordings were performed with a single channel EPC10 amplifier (HEKA Elektronik, Lambrecht/Pfalz, Germany) that was controlled by Pulse and Patchmaster 2.2 software (HEKA Elektronik). Capacitive transients were automatically compensated. Data was acquired at 25 kHz and the liquid junction potential was calculated and corrected offline for all current‐clamp recordings.

Neuropetides cholecystokinin, neuropeptide Y, met‐enkephalin and dynorphin were obtained from Phoenix Pharmaceuticals (Burlingame, CA, USA), and dopamine from Sigma (Burlington, MA, USA). Drugs were either bath‐applied or delivered locally via a 250 μm tip diameter flow pipe.

Local stimulation with an excitatory microdrop

To selectively stimulate neurons in the ARC, we used a glutamate microdrop from a patch pipette with a small 5 μm diameter tip containing glutamate (10 mm) (Belousov & van den Pol, 1997). The microdrop was controlled by pressure (5 s, 69 kPa) through a Picospritzer (Parker). We first tested this system with a solution of fluorescent dye in the pipette to visualize the amount and location of liquid ejected and the efficiency of the washout from the chamber; the stimulation drop spread to a radius of about 15 μm from the application site. The drop of dye was restricted to the ARC, and no dye was found in nearby brain regions. During the experiments, the stimulation pipette tip was first placed close to the recorded neuron to directly activate the recorded neuron, and subsequently moved away from the recorded neuron within the ARC to selectively stimulate other ARC neurons.

Leptin response

To determine if leptin activated Stat‐3 in vGluT2 neurons, food was withheld from mice (n = 5) for 24 h; water was available. Mice then received an intraperitoneal injection of 200 μg mouse leptin (Santa Cruz, Dallas, TX, USA) (n = 5 leptin‐treated mice and 5 control mice). One hour later, mice were given an overdose of Nembutal, and perfused transcardially with 4% paraformaldehyde. After cryoprotection in sucrose, 20 μm thick sections were cut on a cryostat, rinsed, treated with 2% normal horse blocking serum, and then incubated with a 1:200 dilution of rabbit anti‐phosphorylated Stat‐3 antiserum (rabbit anti‐Phospho‐Stat3 (Tyr705) cat no. 9145; Cell Signaling Technology, Danvers, MA, USA) overnight at 4°C. After washing, sections were treated with a secondary antibody of 1:200 Alexa 594 donkey anti‐rabbit antibody for an hour, and then mounted on glass slides. Representative sections were studied with fluorescence microscopy, and all GFP‐positive vGluT2 cells (green) were examined to determine how many were also pStat‐positive (red).

Optogenetic activation of arcuate glutamate neurons in vitro

vGlut2‐Cre mice (5–6 weeks old) were anaesthetized with xylazine and ketamine and placed into a stereotaxic apparatus. (AAV)dj‐CAG‐DIO‐ChIEF‐tdTomato (50 nl, from Stanford University Viral Vector Core) was then microinjected over several minutes into the ARC (coordinates, caudal to bregma: −1.7 mm, down: −5.8 mm, lateral: ± 0.1 mm) through a glass pipette with a bevelled tip (20–40 μm diameter) connected to a micromanipulator. The pipette was slowly withdrawn 10 min after injection. Mice were returned to their home cage for at least 3 weeks of recovery before slice electrophysiology. Laser pulses (10 ms, 1–20 Hz) were used to stimulate ARC ChIEF‐expressing glutamate neurons during brain slice recording.

Statistical analyses

PulseFit 8.54 and Patchmaster 2.2 (both from HEKA Electronik), MiniAnalysis 6.03 (Synaptosoft, Decatur, GA, USA), Axograph 4.7 (Axon Instruments, Foster City, CA, USA), and Igor Pro 4.07 (WaveMetrics, Lake Oswego, OR, USA) software were used for electrophysiological data analysis. Spontaneous synaptic currents were detected using Axograph and MiniAnalysis (Zhang & van den Pol, 2016, 2017 for further details). Data are expressed as means ± SEM. For statistical analysis we used one‐way ANOVA or Student's t test, or paired t tests. Medcalc software was also used. Means for RT‐qPCR are shown along with SEM. Asterisks in figures indicate statistical significance: P < 0.05. Sample size for different experiments is identified in the text, figures and figure legends.

Results

vGluT2‐GFP protein and mRNA in arcuate neurons

The ARC extends along the ventral sides of the third ventricle, from just caudal to the suprachiasmatic nuclei back to the area adjacent to the mammillary recess of the third ventricle (Hof et al. 2000). Many types of cells, including those synthesizing GABA, NPY, TH, or POMC are distributed throughout the ARC with greater densities in one region or another. In contrast, the distribution of GFP‐positive glutamate cells in the ARC was quite different. There was a strong preponderance of positive cells in the caudal part of the nucleus (Fig. 1 B and C). Only a few rare positive cells were found in the rostral ARC (Fig. 1 A), consistent with multiple descriptions that focus on this more rostral area (Hrabovszky et al. 2005, 2007).

Figure 1.

vGluT2‐GFP expression in the arcuate nucleus of transgenic mouse brain

A, in rostral ARC, only a few neurons show GFP expression compared with the robust expression in the adjacent ventromedial hypothalamic nucleus (VMH). B, in contrast, in the more caudal ARC, robust expression of vGluT2‐GFP is found. A, B scale bar, 40 μm. C, higher magnification of vGluT2 neurons in caudal ARC. Scale bar, 14 μm. D, bar graph of RT‐qPCR results from microdissected regions of the arcuate nucleus from vGluT2 transgenic mice shows the relative levels of vGluT2 (filled bars) and GFP (open bars) mRNA. Four regions of the ARC were sampled, with the numbers indicating the position of the samples along a rostral‐caudal axis running through the ARC, with the most rostral as 1 and the most caudal as 100, similar to Fig. 2. The bars represent means; the first and third set of bars, n = 4; the second and fourth set n = 5 and n = 3, respectively. The error bars are SEM.

The asymmetric distribution of GFP‐positive cells was further corroborated by RT‐PCR to examine vGluT2 mRNA expression in microdissected ARC. Similar to expression of GFP in vGluT2 transgenic mice, vGluT2 mRNA was also more robust in the caudal ARC. A bar graph of RT‐PCR results from microdissected regions of the ARC from vGluT2 transgenic mice shows the relative levels of vGluT2 (filled bars) and GFP (open bars) mRNA (Fig. 1 D). Four regions of the arcuate nucleus were sampled, with the numbers indicating the position of the samples along a rostral‐caudal axis running through the ARC, as in Fig. 1 A–C. That the caudal regions are considered part of the ARC is supported by several lines of evidence. First, multiple mouse and rat brain atlases show the ARC extending caudally along the mammillary recess of the third ventricle (Hof et al. 2000; Paxinos & Franklin, 2001; Franklin & Paxinos, 2008; Mouse Brain Library: http://www.mbl.org/atlas170/atlas170_frame.html); the ARC appears as a continuous cell‐dense structure from rostral to caudal. Transplants of the anterior pituitary are only viable in hypophysiotropic regions of the brain, particularly in the ARC extending back to the mammillary recess of the third ventricle (Szentagothai et al. 1972).

The number of GFP‐positive cells in the caudal half of the ARC is > 10‐fold greater than in the rostral half (Fig. 2). To visualize the distribution of vGluT2 throughout the ARC, the cells were identified and depicted on a series of coronal sections of the ARC (Fig. 2), again indicating a substantial anatomical heterogeneity in the ARC relating to the distribution of these neurons.

Figure 2.

Schematic drawing depicting the density of GFP fluorescent neurons (red circles) in fixed coronal sections of the ARC from vGluT2 transgenic mice

Anatomical plates taken from Paxinos & Franklin (2001). Percentage values (in black, lower left of each section) indicate the position along a rostral‐caudal axis running through the ARC with 0% equal to the most rostral portion of the arcuate (bregma –1.06 mm) and 100% equal to the most caudal (bregma –2.80 mm). The number (n) of GFP neurons found in the ARC in each section of a representative brain is displayed in red (lower centre of plate) and the calculated density (n/A = number of vGluT2 neurons per 10,000 μm² of arcuate area A) is displayed in blue (middle of each section). ARC, arcuate nucleus; 3 V, third ventricle; VMH, ventromedial hypothalamic nucleus; DMH, dorsomedial hypothalamic nucleus; F, fornix.

Consistent with the GFP expression, in situ hybridization for vGluT2 mRNA showed only modest expression in the rostral half of the ARC, and much stronger expression in the caudal ARC (Fig. 3 A and B) (Allen Institute Mouse Brain Atlas: http://mouse.brain-map.org). In contrast, AgRP expression was strong in the rostral half of the ARC, but still robust in the caudal ARC (Fig. 3 C and D). POMC expression was very strong in the rostral ARC and reduced but still found in the caudal ARC (Fig. 3 E and F). The presence of both AgRP and POMC cells in what we are including as part of the ARC is consistent with the inclusion of the robust population of vGluT2 neurons described here. Although here we focus on the excitatory neurons in the caudal ARC, there are also a substantial number of GABA neurons in the same general area (Acuna‐Goycolea et al. 2005).

Figure 3.

vGluT2 comparison with AgRP and POMC in situ hybridization

A, there is little vGluT2 mRNA expression in rostral ARC but strong expression in ventromedial nucleus (VMH). Scale bar, 200 μm. B, many more vGluT2 positive cells in caudal ARC. C and D, AgRP mRNA expression is robust in rostral and caudal ARC. Scale bar, 210 μm. E and F, POMC mRNA expression is strong in rostral ARC, but less in caudal ARC. Rostral ARC is shown on the left (A, C, E), more caudal on the right (B, D, F). Inverted images, all from Allen Institute Mouse Brain Atlas with permission; Lein et al. 2007.

Leptin response

Leptin is a key signal from fat tissue that is carried to the brain where it signals the peripheral nutritional state to neurons bearing leptin receptors. To determine if vGluT2 neurons responded to leptin, vGluT2‐GFP mice were food‐restricted overnight, and then given leptin. Mice receiving leptin were then compared with controls. Leptin induces the phosphorylation of Stat‐3, and the pStat‐3 then moves into the nucleus where this translocation can be detected with immunostaining (Hubschle et al. 2001; Munzberg et al. 2003, Munzberg & Myers, 2005; Frontini et al. 2008). We found that leptin induced an increase in pStat‐3 in the ARC, seen as a strong increase in the number of cells with red nuclei, as immunostained with Alexa 594. A number of cells with pStat‐3 immuno‐positivity did not express GFP, as expected from previous studies showing other cell types with pStat‐3 responses to leptin (Anderson et al. 2003; Ellacott et al. 2006). Some of the vGluT2 neurons did show pStat‐3 immunostaining. To quantify changes in the pStat‐3 response to leptin in vGluT2 neurons, the number of green GFP‐expressing vGluT2 neurons that also showed red immunofluorescence in the nucleus was determined in multiple sections from each of 5 leptin‐treated and 5 control mice that did not receive leptin (Fig. 4). The number of vGluT2 cells that were positive for pStat‐3 increased with leptin exposure by 3‐fold (paired t test; P < 0.001), from a baseline of 4.7 ± 0.7 % (n = 17) to 14.7 ± 0.8% (n = 20) after leptin stimulation (Fig. 4) based on the means of the slides quantified; these results are derived from the detection of 77 pStat‐3‐positive neurons from a total of 1478 vGluT2 cells from controls, and 310 pStat‐3‐positive cells out of 1882 vGluT2 neurons from leptin‐treated mice. These data suggest that some vGluT2 neurons respond to leptin with an increase in phosphorylated Stat‐3, indicating a sensitivity to leptin. Positive cells were counted independently by two individuals and a similar 3‐fold increase in pStat‐3‐positive vGluT2 cells was corroborated.

Figure 4.

Leptin increases pStat‐3 in arcuate vGluT2 neurons

As studied with fluorescent microscopy, a subpopulation of green vGluT2 neurons (A; arrows) showed pStat‐3 expression (red; B) after leptin treatment, merged in C. D, each data triangle shows the percentage of pStat‐3‐positive (red) vGluT2 neurons (green) in an ARC histological section. The red line shows the mean number of pStat‐3‐positive cells for the control and leptin‐treated groups, respectively, based on the means of the different sections. ^*** P < 0.001 for the leptin‐treated group compared with the control group using a Student's two‐tailed t test. Scale bar, 20 μm.

Physiological characteristics of vGluT2 neurons

Electrical properties of vGluT2‐expressing arcuate cells

Using electrophysiological recordings from GFP‐expressing cells, we characterized the physiological properties of ARC glutamatergic neurons in brain slices. Almost all of the recordings were done in the caudal part of the nucleus where most of the vGluT2 cells exist. In cell‐attached recording, the majority of the cells (42 of 46) showed spontaneous activity, with a mean spike frequency of 2.1 ± 0.2 Hz (Fig. 5 A and B). In whole‐cell current clamp (Fig. 5 C), the mean resting membrane potential of GFP‐vGluT2 neurons was −56.9 ± 2.2 mV (n = 20, Fig. 5 D). To examine the active properties of vGluT2 neurons, pulses of positive current (30–50 pA for 700 ms) were injected in whole‐cell current‐clamp mode. Relatively strong somatic depolarization triggered a quick burst of action potentials showing some spike frequency adaptation (Fig. 5 E), with the interspike interval increasing by 60 ± 15% between the first and second halves of a series of spikes (n = 10).

Figure 5.

Physiological properties of arcuate glutamate cells

A and B, extracellular (cell‐attached) recording from a representative, spontaneously active, vGlut2 neuron (A), and summary histogram showing the spontaneous firing rate of 46 cells (B). C and D, whole‐cell current‐clamp recording from a representative glutamatergic neuron showing spontaneous spikes (C) and histogram of the resting membrane potential of 20 cells recorded in whole‐cell current‐clamp configuration (D). E, somatic current injection in ARC vGluT2 neuron. Arrow highlights the relatively long membrane time constant of these cells.

Injections of negative current through the recording electrode (−60 pA for 700 ms) revealed relatively long membrane time constants (247 ± 63 ms, range: 93–306 ms; Fig. 5 E, arrow). In contrast, nearby POMC cells activated with similar current pulses showed a relatively faster membrane time constant (127 ± 31 ms, not shown. See also Acuna‐Goycolea & van den Pol, 2009). Thus, the average intrinsic properties of vGlut2 cells studied here appear to be different than those of the neighbouring POMC neurons.

Synaptic input

Excitatory and inhibitory synaptic input to vGluT2 cells may play a critical role in regulating their activity. We therefore studied the properties of synaptic inputs to these neurons in vitro. Our pipette solution with low Cl⁻ concentration allowed us to sequentially monitor both excitatory and inhibitory transmission in a given neuron. At 0 mV holding potential, the driving force for ionotropic glutamate receptor‐mediated current is expected to be minimal, whereas the driving force for GABA‐gated Cl⁻ channels should be relatively robust given that the reversal potential for this ion is normally near −70 mV. We therefore held vGlut2‐expressing neurons at 0 mV to examine GABAergic currents (IPSCs) (Fig. 6 A, left). These currents were picrotoxin‐sensitive, indicating that they were mediated by GABA_A type receptor (Fig. 6 A, middle). The rise‐time of the IPSCs was 1.4 ± 0.5 ms (10–90%) (Fig. 6 A, right). AMPA currents were examined with picrotoxin in the bath to block GABA responses and cells were held at −70 mV, a potential where the contribution of the NMDA receptor should be minimal due to the voltage‐dependent Mg²⁺ blockade (Fig. 6 B). NMDA currents were pharmacologically isolated after application of CNQX and picrotoxin to the bath to block AMPA‐ and GABA‐mediated synaptic currents, respectively, while holding the recorded cells at −30 mV (Fig. 6 C) to decrease the Mg²⁺ blockade. These NMDA currents had a much slower rise time than the AMPA currents (10–90% rise time: 1.5 ± 0.4 ms for AMPA, 8.6 ± 2.1 ms for NMDA), and more prolonged decay kinetics (Fig. 6 B and C).

Figure 6.

ARC glutamatergic cells receive a relatively balanced excitatory/inhibitory input under resting conditions

A–C, GABA_A‐ (A), AMPA‐ (B), and NMDA‐mediated currents (C) in vGlut2 cells from the ARC. Means of 20 consecutive events are shown on the right. Note the difference in rise‐time and decay kinetics between them. D, frequency of IPSCs/EPSCs in typical vGlut2 (black) and POMC (red) neurons. E–G, vGluT2‐GFP and alpha‐MSH immunostaining show little overlap. E, caudal ARC vGluT2‐GFP‐expressing neurons, green. Green arrowheads indicate GFP expression. F, immuno‐staining for POMC neuron antigen alpha‐MSH, red. Red arrowheads show alpha‐MSH immunoreactive neurons. G, dual red and green imaging shows no overlap between the cells examined here.

We determined the relative frequency of inhibitory and excitatory synaptic currents in individual cells by sequentially switching the holding potential between the reversal potential for GABA (near −70 mV, to measure EPSCs) and glutamate currents (0 mV, to measure IPSCs). Such analysis revealed that vGlut2‐expressing cells received a well‐balanced excitation/inhibition ratio slightly in favour of EPSCs, as shown for a typical cell in Fig. 6 D (black). The mean ratio of EPSCs to IPSCs for vGluT2 cells was 1.4 ± 0.3. To determine whether this was a specific property of glutamatergic cells in the ARC, or a general property of ARC cells, we repeated the analysis in neighbouring POMC cells (Fig. 6 D; red). Individual POMC cells (n = 5) received more inhibitory synaptic activity than excitatory activity under resting conditions, with an EPSC/IPSC ratio of 0.6 ± 0.1. This result suggests that in contrast to vGlut2 cells, POMC neurons in the arcuate are tonically inhibited, as previously described (Cowley et al. 2001). These results also support the view that, as a population of neurons, vGlut2 cells represent a distinct cell type within the ARC; their intrinsic as well as network properties appear to be different from those of the average ARC POMC cell.

In parallel experiments, we used an antibody against alpha‐MSH to identify POMC cells in sections prepared from vGluT2‐GFP mice (Fig. 6 E–G). Almost none (<1%) of the GFP‐positive green vGluT2 cells (n = 321) expressed alpha‐MSH, and similarly, alpha‐MSH immunoreactive red cells (n = 355) showed little (<1%) GFP expression. Together with the physiological comparison above, these results are consistent with the view that few if any of the vGluT2 cells studied here also synthesize alpha‐MSH and thus are unlikely to include POMC neurons.

Microdrop and optogenetic activation of arcuate glutamate neurons excites POMC and other ARC neurons

To test the hypothesis that ARC glutamate neurons excite POMC neurons, we studied the excitatory postsynaptic current (EPSC) induced by local stimulation of ARC glutamate neurons. All experiments were done in the presence of the GABA_A receptor antagonist BIC (30 μM) to block IPSCs and to eliminate the possibility of indirect effects mediated by inhibitory neurons. When the excitatory glutamate microdrop, which generated an area of stimulation with a radius of about 15 μm (Belousov & van den Pol, 1997), was placed directly on or adjacent to the recorded POMC neuron, a direct depolarization was observed (Fig. 7 Aa), as expected; sometimes the direct depolarization included an increase in EPSC frequency (Fig. 7 Ab). An advantage of the microdrop stimulation is that it should only activate neuron cell bodies and dendrites, but not axons of passage from other brain regions (Christian & Dudek, 1988). The EPSC was blocked by glutamate receptor antagonists AP5 (50 μm) and CNQX (10 μm) (Fig. 7 D). When the microdrop application pipette tip was moved away from the recorded neuron, but still within the ARC, local stimulation significantly increased EPSC frequency from 5.1 ± 0.9 Hz to 13.6 ± 1.4 Hz (Fig. 7 B and E, n = 7; P < 0.0001, one‐way ANOVA) and amplitude from 19.3 ± 2.3 pA to 31.3 ± 3.1 pA (Fig. 7 B, n = 7; P = 0.002, one‐way ANOVA), with no detectable direct activation of the recorded POMC cell (Fig. 7 B). To further confirm that this enhancement of EPSCs was due to presynaptic glutamate neuron activation, we tested the effect of local stimulation on EPSCs in the same neurons in the presence of TTX (1 μm). Bath application of TTX blocked the EPSC frequency (Fig. 7 C and E, n = 5) and amplitude (Fig. 7 C, n = 5) increase in response to the microdrop; TTX did not block the effect of the microdrop when applied directly to the cell body (not shown), supporting the view that the effect was based on increased action potentials from excitatory neurons. These results thus indicate that activation of ARC neurons increases glutamate release onto POMC neurons.

Figure 7.

Local stimulation reveals excitatory synaptic inputs from arcuate glutamate neurons to POMC neurons

A, trace examples showing glutamate microdrop application (Glu) close to the recorded POMC neurons activated POMC cells postsynaptically only (Aa), or both presynaptically and postsynaptically (Ab). Recording was performed in voltage‐clamp mode in the presence of GABA_A receptor blocker Bic (30 μM). B, representative traces show local microdrop stimulation within the ARC increased synaptic glutamate release onto ARC POMC neurons in the absence of any direct effect on the recorded cell. C, representative traces showing no effect of local stimulation on synaptic glutamate release in the presence of Bic plus TTX (1 μM). D, representative trace showing no synaptic current in the presence of TTX, Bic, AP5 and CNQX. E, bar graph summary of data showing EPSC frequency under the following conditions: control (Ctrl, n = 7), microdrop stimulation (Stim., n = 7), TTX (n = 5) and TTX plus microdrop stimulation (TTX + Stim., n = 5). ^*** P < 0.001. F and G, ARC neurons with ChIEF expression were recorded under whole‐cell voltage‐clamp configuration during optogenetic stimulation. F, blue light (10 ms pulse, 1–20 Hz) stimulation of ChIEF‐positive glutamate neurons evoked excitatory synaptic activity in other ARC neurons. The laser‐evoked EPSCs were completely abolished by the application of ionotropic glutamate receptor antagonists AP5 and CNQX. The cells were stimulated for 10 s at each indicated frequency. EPSCs are shown and analysed during the 10 s photostimulation period, or in the absence of photostimulation in control slices. G, example of optostimulation evoking an EPSC (blue line) which was blocked by glutamate receptor antagonists AP5 + CNQX (red line).

To demonstrate further that arcuate glutamate neurons exert excitatory actions on neighbouring neurons, we used a channelrhodopsin variant, ChIEF. A floxed AAV‐ChIEF‐tdTomato was employed to selectively generate ChIEF expression in ARC glutamate neurons of vGluT2‐Cre mice. One month after AAV injection, we recorded the activity of ChIEF‐negative ARC neurons surrounded by ChIEF‐positive axon terminals in ARC brain slices. Blue light at 1–20 Hz evoked excitatory synaptic currents on neighbouring neurons, with a positive correlation between frequency of optostimulation and EPSC frequency (Fig. 7 F and G). The evoked synaptic currents were abolished with AP5 (50 μM) and CNQX (10 μM) (Fig. 7 F and G), suggesting that optogenetic activation of ARC glutamate neurons generates synaptic glutamate release onto neighbouring neurons.

Neuromodulator responses

To investigate further the potential for ARC vGluT2 neurons to participate in energy homoeostasis, we studied the actions of several neuropeptides that are involved in the modulation of food intake and energy homeostasis. We first tested an anorectic neuropeptide, cholecystokinin (CCK), and then tested several orexigenic peptides including neuropeptide Y (NPY), met‐enkephalin, alpha‐MSH and ghrelin, and finally we tested the catecholamine dopamine.

Cholecystokinin (CCK)

Previous studies have suggested that cholecystokinin (CCK) plays a role in modulating energy homeostasis, primarily by anorectic actions (Kissileff et al. 1981; de Graaf et al. 2004; Morley, 1982). We therefore tested the actions of this neuropeptide on the activity of vGluT2 cells. We first measured vGluT2 neuron responses in cell‐attached mode. As shown in Fig. 8 A, as well as in the summary plot in Fig. 8 B, CCK‐8S (1 μM) increased the rate of vGluT2 neuron firing from 1.6 ± 0.4 to 5.3 ± 0.7 Hz (P < 0.01). The firing rate of these cells returned to 3.4 ± 0.7 Hz upon peptide washout. Mean data for all neurons (n = 13) is shown in Fig. 8 B. Traces from a representative experiment are presented in the upper part of Fig. 8 A. CCK also depolarized vGluT2 cells in the presence of 1 μM TTX as shown in the bottom part of Fig. 8 A, indicating a direct effect; the mean depolarization was 5.2 ± 0.8 mV (n = 5).

Figure 8.

Cholecystokinin excites vGluT2 neurons

A and B, CCK‐8S increased spontaneous firing in the absence of TTX and depolarized the membrane potential of vGluT2 cells in 1 μM TTX (A, bottom). C and D, CCK‐8S also increased the synaptic release of glutamate. ^* P < 0.05; ^** P < 0.01.

In addition, we measured the actions of CCK on both excitatory (Fig. 8 C and D) and inhibitory synaptic inputs to vGluT2 neurons. CCK increased the frequency of spontaneous EPSCs from 1.7 ± 0.3 to 2.9 ± 0.5 Hz (n = 8), and reversibly increased the frequency of IPSCs from 1.3 ± 0.2 to 2.3 ± 0.2 Hz (n = 10) (not shown). These results suggest that, in addition to raising the spontaneous firing rate of vGluT2 cells, CCK enhanced the release of both glutamate and GABA onto these neurons under our experimental conditions.

Neuropeptide Y (NPY)

NPY is synthesized and released by AgRP cells of the ARC and increases food intake and body weight (Clark et al. 1984; Woods et al. 1998). ARC NPY neurons also use GABA as a fast inhibitory transmitter (Wu & Palmiter, 2011; Krashes et al. 2013; Liu et al. 2013). The actions of NPY on the firing of vGluT2‐expressing cells were examined with both cell‐attached and whole‐cell current‐clamp recordings. In 9 vGluT2 neurons recorded in cell‐attached mode, NPY (1 μM) decreased the firing rate from 1.5 ± 0.4 to 0.3 ± 0.1 Hz (Fig. 9 A and B) (P < 0.05). In whole‐cell mode, NPY (1 μM) hyperpolarized ARC vGluT2 cells by 9.4 ± 4.1 mV (n = 5) (P < 0.05), and after peptide washout the membrane potential partially returned to control levels (Fig. 9 C).

Figure 9.

Neuropeptide Y depresses glutamate cells from the arcuate nucleus

A, an inhibitory action of NPY on spontaneous firing recorded in cell‐attached mode in a representative cell, followed by washout recovery. B, summarized reduction in firing rate induced by NPY in vGluT2 cells (n = 9) from the ARC. ^* P < 0.05, Control vs. NPY. C, actions of NPY on the membrane potential of the same cell subtype, recorded in whole‐cell current‐clamp mode (n = 5). D, NPY hyperpolarized vGluT2 cells in TTX (1 μM), suggesting a direct effect. E, membrane potential hyperpolarization evoked by NPY (n = 6) and recovery after peptide washout in the presence of TTX. B, C and E, ^* P < 0.05 comparing pre‐NPY baseline value to the value during NPY application.

To determine if the NPY inhibition was due to a direct action of NPY on vGluT2 cells, the peptide was applied in the presence of TTX (1 μM) to block spike‐generated transmitter release. In the presence of TTX, NPY (1 μM) hyperpolarized vGluT2 cells by 3.2 ± 1.5 mV (n = 6) (Fig. 9 D and E) (P < 0.05), and the membrane potential showed partial recovery after peptide washout, suggesting that NPY exerted a direct inhibitory effect on the cells.

Opioid peptides dynorphin and met‐enkephalin inhibit vGluT2 cells

Dynorphin (Dyn) synthesized by hypothalamic neurons plays an orexigenic role in food intake (Morley et al. 1982; Morley & Levine, 1983; Inui et al. 1991; Silva et al. 2002). We therefore examined the effects of the opioid receptor agonist Dyn. In cells tested with cell‐attached recording, Dyn (5 μM) reduced the action potential firing from 3.6 ± 0.4 to 0.8 ± 0.3 Hz (n = 6) (Fig. 10 A and B). Next, we tested the effect of Dyn on membrane potential in current clamp in the presence of TTX (1 μM) to block normal synaptic responses. In TTX, Dyn hyperpolarized the membrane potential by 5.8 ± 0.6 mV, a significant effect (n = 5; P < 0.001), suggesting a direct effect of Dyn.

Figure 10.

Dynorphin and met‐Enkephalin inhibit vGlut2 cell firing

A, top: the inhibitory effect of Dyn on the firing (detected in cell‐attached configuration) of a typical vGluT2 cell is shown. A, bottom: Dyn applied in the presence of TTX (1 μM) hyperpolarized the membrane potential in current‐clamp mode, suggesting a direct action. B, bar graphs show mean spike frequency before, during and after Dyn application, recorded in cell‐attached mode. C, recordings from a representative neuron showing the inhibitory effects of Dyn on EPSCs. D, Dyn reduces spontaneous EPSCs; bar graphs show EPSC frequency in control, dynorphin and washout conditions. E, mEnk reduced spike frequency of a representative vGluT2 cell. F, summary bar graph showing the inhibitory actions of mEnk on the mean firing rate of 6 vGluT2‐expressing cells. B, D and F, ^* P < 0.05.

The effect of Dyn on the frequency of EPSCs was examined (Fig. 10 C and D). Dyn reduced the frequency of excitatory currents from 2.3 ± 0.3 to 1.4 ± 0.4 Hz (n = 5, P < 0.01), and showed recovery after washout. Dyn reduced the frequency of inhibitory currents onto vGluT2 cells from 5.1 ± 1.6 to 0.9 ± 0.1 Hz (P = 0.02; n = 6, data not shown) and recovered after neuropeptide washout. Thus, Dyn evoked a direct inhibitory action on vGluT2 cells, and also reduced both EPSC and IPSC frequency.

Met‐enkephalin (mENK) is synthesized and released by cells of the ARC and can increase food intake (Baile & McLaughlin, 1987). mENK (1 μM) reduced spontaneous firing recorded in cell‐attached mode (Fig. 10 E and F). The spontaneous firing frequency was reduced from 2.5 ± 0.6 Hz to 1.0 ± 0.2 Hz, a significant reduction (P = 0.02, n = 6, paired t test), and recovered to 2.4 ± 0.5 Hz after peptide washout (Fig. 10 E and F). In the presence of TTX (0.5 μM) in current‐clamp recording, the membrane potential was not substantially altered by mENK (−0.5 ± 0.7 mV; n = 7, P = 0.53) (data not shown). These data suggest that whereas other neuromodulators act on both the vGluT2 cell and presynaptic partners, mENK may modulate the activity of vGluT2 cells indirectly by acting on synaptic partners presynaptic to vGluT2 neurons.

No effect of MC3‐4 agonist/antagonist or ghrelin

POMC neurons release alpha‐MSH which activates melanocortin MC3/4 receptors that play key roles in reducing food intake (Chen et al. 2000, Chen et al. 2004; Cone, 2005). We studied the actions of the MC3/4 receptor agonist MTII (1 μM), as well as the effect of the MC3/4 receptor antagonist SHU9119 (1 μM), on the spontaneous firing of ARC vGluT2‐containing cells in cell‐attached mode. The cells showed no change in firing rate in control, MTII and washout conditions, firing at 2.6 ± 0.9, 2.8 ± 1.2, and 2.7 ± 0.9 Hz, respectively (n = 6). Thus, the firing rate of these cells did not significantly change when MC3/4 receptors were activated. Similarly, the firing rate of vGluT2 cells was also not significantly changed when the MC3/4 antagonist SHU9119 was added to the bath. In control conditions, glutamatergic cells in the ARC fired at 2.6 ± 0.6 Hz, and after addition of SHU9119 these hypothalamic cells did not alter their firing rate, 2.5 ± 0.9 Hz (n = 7). Thus, neither the agonist nor the antagonist of MC3/4 receptors altered vGluT2 cell activity.

We also tested ghrelin, a peptide released by the stomach that stimulates hunger and food intake (Kojima et al. 1999). Ghrelin (250 nM) had no detectable effect on the membrane potential in vGluT‐2 cells (in the presence of ghrelin +0.5 ± 3.0 mV; n = 5).

Dopamine inhibition

Dopamine has been associated with the hedonic aspect of food intake, and dopamine in other brain systems has generally been considered orexigenic (Wise, 2006; Palmiter, 2007). Depending on the dopamine receptor expressed, dopamine can be inhibitory or excitatory. When dopamine (30 μM) was applied to vGluT2 neurons it evoked a substantial inhibitory action, reducing spike frequency considerably, varying from complete block of spikes to a reduction by half (see Fig. 11 A and B) (P < 0.0001, one‐way ANOVA post hoc test). Membrane potential was significantly hyperpolarized by the application of dopamine (Fig. 11 C). After dopamine washout, action potential frequency and membrane potential recovered to baseline states.

Figure 11.

Dopamine inhibits vGluT2 cells

A, a representative trace shows dopamine hyperpolarizes and inhibits vGluT2 neurons in posterior ARC. B, the mean firing rates of vGluT2 neurons (n = 9) in control buffer, in the presence of dopamine (30 μM), and after dopamine washout. C, the mean resting membrane potentials of vGluT2 neurons (n = 9) in baseline conditions, in the presence of dopamine, and after washout. ^*** P < 0.001.

Discussion

Glutamate neurons in the arcuate nucleus

The glutamatergic neurons of the hypothalamic ARC appear to form a unique population within the nucleus on the basis of localization, intrinsic membrane properties, synaptic input and responses to neuromodulators that play a role in energy homeostasis.

A previous in situ hybridization study reported strong expression of vGluT2 in various hypothalamic nuclei, but noted “no vGluT2 expression was seen in the arcuate nucleus” (Ziegler et al. 2002). Similarly, other reports describe only a small number of vGluT2 neurons in the ARC (Hrabovszky et al. 2005, 2007). It is important to note that these studies focused on more rostral regions of the ARC, and similarly we found only a small number of vGluT2 cells there. This is further echoed by recent papers that show only small numbers of vGluT2 cells in the rostral ARC (Vong et al. 2011; Fenselau et al. 2017). In contrast to glutamate, a large number of GABA cells are located in the rostral ARC (Vong et al. 2011). What previously published reports may not have appreciated is the dramatic increase in the number of vGluT2 cells in the caudal part of the ARC with a > 10‐fold greater number of vGluT2 cells in the caudal ARC compared with the rostral ARC. We substantiated this with transgenic mice and independently with mRNA analysis. That many more glutamate neurons are found in the caudal ARC is also consistent with in situ hybridization of vGluT2 mRNA (Allen Brain Atlas). Together, these three independent sets of data corroborate the robust increase in glutamatergic neurons in the caudal ARC. That this more caudal area merits inclusion as an extension of the ARC is consistent with multiple mouse and rat anatomical brain atlases that define this region as part of the ARC (Hof et al. 2000; Paxinos & Franklin, 2001; Franklin & Paxinos, 2008; Mouse Brain Library: http://www.mbl.org/atlas170/atlas170_frame.html). The extension of the ARC more caudally than previous papers on ARC glutamate neurons have done is also consistent with the presence of other neurons that are associated selectively with the ARC, including POMC and AgRP/NPY neurons in the same caudal region (Allen Mouse Brain Atlas; Cravo et al. 2011). Tyrosine hydroxylase neurons, which synthesize dopamine regulate pituitary prolactin secretions (Ben‐Jonathan & Hnasko, 2001) or may contribute to energy homeostasis (Zhang & van den Pol, 2016), are also found in both rostral and caudal ARC (Allen Mouse Brain Atlas). Some ARC vGluT2 cells may also contain kisspeptin (Cheong et al. 2015), and the preponderance of ARC kisspeptin neurons is in the caudal ARC (Cravo et al. 2011; Cheong et al. 2015). Injections of dye into the axons of the median eminence label hypophysiotropic cells that project to the median eminence and labelled cells are found in both the rostral and caudal ARC (Wiegand & Price, 1980). The demonstration of a robust excitatory synaptic response in the hypothalamic paraventricular nucleus (PVN) evoked by activated ARC cells (Fenselau et al. 2017) is difficult to explain by the small number of vGluT2 cells in the rostral part of the ARC, whereas the robust incidence of vGluT2 cells in the caudal ARC helps support these results.

Anorectic role of arcuate glutamate neurons

That the ARC vGluT2 neurons are involved in energy homeostasis is supported by multiple converging lines of evidence: Leptin plays an important role signalling fat availability from peripheral sites. We show that leptin administration increases Stat‐3 phosphorylation in a number of vGluT2 neurons, thus indicating their leptin sensitivity. Leptin also activates POMC neurons (Cowley et al. 2001) as well as other cells in the ARC (Elias et al. 1999, 2000; Elmquist, 2001).

Our electrophysiological recording showed responses to neuromodulators that are associated with regulation of energy homeostasis. These include orexigenic neuromodulators NPY, metENK, dynorphin and dopamine, which all inhibited the ARC vGluT2 cells, and neuromodulators that reduce food intake such as CCK, which excited the vGluT2 cells. The responses to these feeding‐related neuromodulators include both direct actions as well as modulation of excitatory or inhibitory synaptic input to the ARC vGluT2 neurons. These electrophysiological results, inhibition with orexigenic neuromodulators and excitation with an anorectic neuromodulator, are consistent with an anorectic role for the ARC vGluT2 cell in energy regulation.

POMC neurons play a key anorexic role in both rodent models and in humans; severe human obesity can be caused by defects in POMC neurons or MC3/MC4 receptor signalling (Hager et al. 1998; Yeo et al. 2000; Krude et al. 2003; Cone, 2005; Mencarelli et al. 2012). The anorectic POMC cells generate a fairly slow inhibition of food intake that occurs over a number of hours (Aponte et al. 2011; Zhan et al. 2013; Koch et al. 2015). Interestingly, local microdrop activation of ARC glutamatergic neurons generated an increase in excitatory synaptic responses in the nearby POMC neurons, consistent with the view that one mechanism of how the ARC vGluT2 cells may regulate energy homeostasis is by excitation of the anorexic POMC neurons.

This is consistent with the view that ARC vGluT2 cells may play both a rapid role in reducing food intake as well as more long‐term inhibitory modulation of feeding through the POMC neurons (Aponte et al. 2011; Zhan et al. 2013; Koch et al. 2015). Fenselau et al. (2017) showed that ARC vGluT2 cell activation of MC4R‐expressing paraventricular nucleus neurons results in a rapid reduction in food intake.

Our data showing that ARC glutamatergic cells excite POMC neurons therefore suggests that the ARC vGluT2 cells can potentially inhibit food intake by three related mechanisms: First by excitation of anorectic PVN cells (Fenselau et al. 2017); second by excitation of anorectic ARC POMC cells (our study), and third by enhancing release of alpha‐MSH from POMC cells by glutamate excitation that would potentiate actions of ARC excitatory input to MC4R‐expressing PVN neurons.

Optogenetic stimulation of the ARC vGluT2 cells generated an increase in excitatory synaptic activity in other non‐vGluT2 ARC neurons. No inhibitory GABA responses were detected with optogenetic activation, supporting the view that the Cre recombinase‐dependent expression was selective to glutamate neurons. GABA has been reported as the primary fast transmitter in POMC cells (Hentges et al. 2004); a small subpopulation of POMC cells may express some characteristics of glutamate neurons (Dicken et al. 2012; Dennison et al. 2016). In immunocytochemical analysis, we found almost no expression of the POMC neuron‐associated alpha‐MSH in vGluT2‐GFP neurons in the transgenic mouse used here. In our experiments with optogenetics, we found light‐mediated release of glutamate but no induced release of GABA, consistent with the absence of glutamate release in the PVN with POMC neuron activation (Atasoy et al. 2014; Fenselau et al. 2017). POMC cells, used as control neurons, show a preponderance of GABA‐mediated inhibitory synaptic inputs, whereas vGluT2 cells show a slightly greater ratio of excitatory to inhibitory spontaneous synaptic currents.

Conclusion

Our results described here including leptin induction of Stat‐3 phosphorylation, a unique localization pattern in the caudal ARC, excitation by an anorexic neuromodulator or inhibition by orexigenic neuropeptides and dopamine, together with the excitatory input to anorexic POMC neurons, support the perspective (Fenselau et al. 2017) that these ARC vGluT2 neurons may play a role in attenuating caloric intake. Here we focus on energy homeostasis; given the role of the ARC in multiple homeostatic functions, ARC vGluT2 cells may also contribute to other ARC functions that were not addressed in the current work.

Additional information

Competing interests

The authors have no competing interests.

Author contributions

All authors contributed to data collection, analysis, conception and design, and interpretation. All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

Research support was provided by US National Institute of Health grants DK103176 and DK115933.

Authors’ present addresses

X. Zhang: Department of Psychology, Florida State University, Tallahassee, FL 32304, USA

C. Acuna: University of Heidelberg, Heidelberg, Germany.

Biographies

Anthony van den Pol is a professor in the Department of Neurosurgery at Yale University.

Claudio Acuna did some of his PhD thesis work at Yale in the Department of Neurosurgery, and his postdoctoral work at Harvard and Stanford Universities before starting his own lab in the Center for Neuroscience at the University of Heidelberg.

John Davis did graduate work at the University of Virginia and worked at Bayer Corp before coming to Yale.

Hao Huang received his MD from the First Military Medical School in China and did postdoctoral work in Yale Neurosurgery before his present position as a biostatistician in the Yale school of public health.

Xiaobing Zhang received his PhD from the University of Science and Technology of China in Hafei, then worked at Yale for several years before starting his own lab at Florida State University. All authors except Dr Huang continue to work on hypothalamic contributions to homeostatic regulation, including energy balance.