Lateral hypothalamic GAD65 neurons are spontaneously firing and distinct from orexin- and melanin-concentrating hormone neurons

Authors


M. Karnani: Department of Biological Sciences, Columbia University, 901 NWC Building, 550 West 120 Street, Box 4817, New York, NY 10027, USA. Email: mmk2185@columbia.edu

Key points

  • Lateral hypothalamus (LH) contains GABA neurons involved in controlling metabolism and sleep.

  • LH glutamic acid decarboxylase 65 (GAD65) GABA neurons are intrinsically depolarized, unlike classical GAD65 neurons of the cortex.

  • LH GAD65 GABA neurons are distinct from most studied LH neurons (orexin and melanin-concentrating hormone cells).

  • A subset of LH GAD65 neurons are glucose inhibited.

  • Our study adds new populations of glucose sensing neurons to the list of hypothalamic sugar sensors and introduces inhibitory circuit elements of the LH.

Abstract  GABAergic neurons are vital for brain function. Their neurochemical and electrical features have been classically characterized in the cortex, but in the lateral hypothalamic area (LHA), such knowledge is lacking, despite the emerging roles of LHA GABAergic cells in feeding and sleep. We used GAD65-GFP transgenic mice, developed for studies of cortical GABAergic cells, to determine fundamental properties of LHA GAD65 neurons, and compare them to ‘classical’ GABAergic cell types of the cortex, and to previously described classes of LHA cells. Whole-cell patch-clamp recordings in acute brain slices revealed that, unlike cortical GABAergic interneurons, LHA GAD65 neurons were intrinsically depolarized and fired action potentials spontaneously. Similar to cortical GABAergic cells, LHA GAD65 cells fell into four major subtypes based on evoked firing: fast spiking, late spiking, low threshold spiking and regular spiking. Three-dimensional reconstructions of biocytin-filled neurons, performed after the patch-clamp analysis, did not reveal striking morphological differences between these electrophysiological subtypes. Peptide transmitters expressed in known classes of LHA projection neurons, namely melanin-concentrating hormone (MCH) and hypocretin/orexin (hcrt/orx), were not detected in LHA GAD65 cells. Approximately 40% of LHA GAD65 cells were directly inhibited by physiological increases in extracellular glucose concentration. Glucose inhibition was most prevalent in the fast spiking subpopulation, although some glucose-responsive neurons were found in each electrophysiological subpopulation. These results suggest that LHA GAD65 neurons are electrically different from ‘classical’ GABAergic neurons of the cortex, are neurochemically distinct from LHA hcrt/orx and MCH cells, but partly resemble hcrt/orx cells in their glucose responses.

Abbreviations 
2DG

2-deoxy-d-glucose

ACSF

artificial cerebrospinal fluid

AHP

afterhyperpolarization

AP

action potential

AP5

(2R)-amino-5-phosphonovaleric acid

BSA

bovine serum albumin

eGFP

enhanced green fluorescent protein

FS

fast spiking

GAD65

glutamic acid decarboxylase 65

hcrt/orx

hypocretin/orexin

ISI

interspike interval

LHA

lateral hypothalamic area

LS

late spiking

LTS

low threshold spiking

MCH

melanin-concentrating hormone

PiTX

picrotoxin

RMP

resting membrane potential

RS

regular spiking

TTX

tetrodotoxin

Introduction

Animal survival depends on neural sensing of body energy levels and consequent alteration of behavioural drivers such as sleep and appetite. The lateral hypothalamic area (LHA) was historically identified as a centre regulating hunger and wakefulness (Moruzzi & Magoun, 1949; Delgado & Anand, 1953) which contains neurons directly sensitive to glucose changes (Anand et al. 1964). The LHA contains several cell types expressing different transmitters, including important projection neurons expressing peptide transmitters hypocretin/ orexin (hcrt/orx) and melanin-concentrating hormone (MCH), which are controlled in distinct ways by physiological signals such as glucose (Karnani & Burdakov, 2011), and in turn differentially control physiological variables such as arousal and feeding (Sakurai, 2007; Guyon et al. 2009). The electrical properties and glucose sensitivity of LHA neuropeptidergic cells have been studied in detail (van den Pol et al. 2004; Marston et al. 2011; Schone et al. 2011).

The LHA also contains GABAergic neurons (Rosin et al. 2003; van den Pol et al. 2004; Acuna-Goycolea et al. 2005), including those expressing the GABA-synthesizing enzyme glutamic acid decarboxylase 65 (GAD65; Shin et al. 2007). GABAergic inhibitory neurons are considered the most basic building block of neuronal circuits (Isaacson & Scanziani, 2011), but these cells have not received specific attention in the LHA, despite recent evidence implicating LHA GABA cells in the regulation of sleep and metabolism.

A large proportion of GABAergic LHA neurons are sleep-active in vivo (Hassani et al. 2010). Microinjection of the GABA-A receptor antagonist bicuculline to the perifornical area of LHA decreases sleep during the lights-on period and induces c-fos expression in many cells, most prominently in the wakefulness-promoting hcrt/orx neurons (Alam et al. 2005; Yi et al. 2009), which receive synaptic contacts from local GABAergic cells (Louis et al. 2010). LHA cells containing leptin receptor b are GABAergic (Leinninger et al. 2009) and project locally as well as to more distant areas such as the ventral tegmental area (Leinninger et al. 2009, 2011; Louis et al. 2010). In relation to energy balance, anatomical data suggest that LHA GABA neurons are targets of key indicators of energy balance such as leptin (Leinninger et al. 2009), and can control activity of hcrt/orx cells according to energy balance (Louis et al. 2010; Leinninger et al. 2011). Other evidence suggests that GABAergic cannabinoid receptor-expressing neurons might synapse preferentially on MCH rather than hcrt/orx cells (Huang et al. 2007). These data point to the existence of specialized energy-sensing subtypes of local GABAergic interneurons in the LHA. However, their electrical, morphological and neurochemical properties, as well as their responses to glucose, have not been studied in detail.

GABAergic neurons have been studied most extensively in the cortex, where they are extremely diverse (Markram et al. 2004; Ascoli et al. 2008; Klausberger & Somogyi, 2008). Many cell types are readily identifiable by their distinctive electrophysiology (Ascoli et al. 2008; Young & Sun, 2009) and, by virtue of these electrophysiological specializations, serve particular roles in cortical processing (Freund & Katona, 2007; Pouille et al. 2009). Similarly, a subset of thalamic GABAergic neurons express the low threshold transient calcium current, IT, and the hyperpolarization activated cation current, Ih, that endow them with the necessary membrane potential oscillations to orchestrate the cortical sleep spindles (Fuentealba & Steriade, 2005). To the best of our knowledge, an electrophysiological characterization of possible subtypes of hypothalamic GABAergic neurons has not been performed to date.

Here, we studied the electrophysiology, anatomical distribution, immunohistochemistry, morphology and glucose responses of LHA GAD65 neurons to assess their subtype classes and contribution to LHA glucose sensing. To target GABAergic neurons, we used the GAD65–green fluorescent protein (GFP) mouse line, which has been shown to reliably label GABAergic cells in the cortex (Lopez-Bendito et al. 2004) and hypothalamus (Bali et al. 2005). Although the precise expression levels of the GAD isoforms 65 and 67 differ according to brain region and subcellular compartment, both are expressed by most GABAergic neurons in the CNS (Erlander et al. 1991; Soghomonian & Martin, 1998) and therefore the GAD65 and 67 promoters are equally useful to drive GFP expression to mark GABAergic neurons. To study the firing phenotypes of GAD65 neurons in the LHA, we recorded their voltage responses to hyperpolarizing and depolarizing current steps. To explain differences between firing phenotypes we also analysed the voltage-gated conductances that are likely to underlie the differences. Finally, we show that ∼40% of LHA GAD65 neurons are directly inhibited by glucose through more than one ionic mechanism.

Methods

Animals

GAD65-GFP mice expressed the GFP gene fused to the first or third exon of the GAD65 gene (Erdélyi et al. 2002). The GAD65-GFP animals were used for electrophysiology between postnatal days P21 and P60. We did not observe age-dependent variation in cell properties in this range. The mice were maintained on a standard 12 h light–dark cycle (lights on at 0700 h) and had free access to food and water. All animal procedures were performed in accordance with the Animals (Scientific Procedures) Act 1986 UK, following guidelines in Drummond (2009), and approved by local animal welfare committees of the University of Cambridge. Transgenic hcrt/orx-–enhanced GFP (eGFP) mice were used to identify and study hcrt/orx neurons. These mice express eGFP under the control of the prepro-hcrt/orx promoter, resulting in highly specific targeting of eGFP only to hcrt/orx neurons, as extensively characterized previously (Yamanaka et al. 2003; Burdakov et al. 2006; Williams et al. 2007, 2008).

Solutions

Artificial cerebrospinal fluid (ACSF) was always continuously gassed with 95% O2 and 5% CO2 and contained (in mm): 125 NaCl, 2.5 KCl, 2 MgCl2, 2 CaCl2, 1.2 NaH2PO4, 21 NaHCO3 and 1 d-(+)-glucose. The intracellular (pipette) solution contained (in mm): 120 potassium gluconate, 10 KCl, 0.1 EGTA, 10 Hepes, 5 K2ATP, 1 NaCl, 2 MgCl2, 0.2% biocytin; pH set to 7.3 with KOH. Liquid junction potential was estimated to be 10.1 mV and was subtracted from the voltage clamp measurements. The following drugs were added to the extracellular solution where indicated: 50 μm (2R)-amino-5-phosphonovaleric acid (AP5), 10 μm 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), 50 μm picrotoxin (PiTX), 1 μm strychnine and 1 μm tetrodotoxin (TTX). All drugs were obtained from Sigma (St Louis, MO, USA) or Tocris (Bristol, UK). All drugs were dissolved in water except PiTX, which was dissolved in ethanol (<0.1%, v/v, in the final solution). Chemicals were applied extracellularly by bath superfusion.

Preparation of acute brain slices

Mice were killed by cervical dislocation during the light phase, unless stated otherwise, and rapidly decapitated. Brains were quickly removed and placed into ice-cold ACSF. A block of brain tissue was glued to the stage of a Campden Vibroslice or Leica VT1200s for slicing while immersed in ice-cold ACSF. Coronal slices 250 μm thick containing the LHA were cut. After a 1 h recovery period at 35°C in ACSF, slices were kept at 22°C and used for recordings within ∼8 h.

Recording

Living GFP-containing neurons were visualized in brain slices using an Olympus BX50WI upright microscope equipped with oblique illumination optics, a mercury lamp and filters for visualizing GFP-containing cells. Somatic recordings were carried out at 37°C using a HEKA EPC 10 or EPC 9 patch-clamp amplifier controlled by Pulse or Patchmaster software (HEKA Elektronik, Lambrecht/Pfalz, Germany) after slices were allowed to equilibriate in the recording chamber for 10–15 min. Patch pipettes were made from borosilicate glass capillaries, and their tip resistances ranged from 3 to 5 MΩ. Slices were placed in a submerged-type chamber (volume ∼2 ml, solution flow rate 2.5 ml min−1) and anchored with a nylon string grid stretched over platinum wire. Only cells with access resistances below 20 MΩ were accepted for analysis. Signals were low-pass filtered at 3 kHz and digitized at 7 kHz.

In experiments where change in membrane potential was quantified, all cells were initially held at the same potential to facilitate comparison (approximately −40 mV, by applying a fixed holding current throughout the experiment). Breaks in some current clamp traces correspond to moments when the recording was paused to take voltage clamp measurements. In some current clamp experiments (e.g. Fig. 2B and C) the cells were periodically injected with constant hyperpolarizing current pulses to monitor membrane resistance.

Current–voltage (I–V) relationships were measured by sequentially stepping the pipette potential from −30 to −130 mV at 10 mV intervals (as shown in Fig. 7B). Net I–V relationships were obtained by subtracting the I–V curve during drug application from the one before.

Protocols for characterization of LHA GAD65 neurons

All cells that yielded stable recordings and had action potentials that overshot 0 mV were considered healthy and analysed.

Membrane time constant (ms): measured from the fitting of a monoexponential curve to the falling phase of the voltage response to a small amount of hyperpolarizing current (resulting in an ∼10 mV shift from resting potential).

Input resistance (MΩ): the slope of a line fitted to the I–V curve between −100 and −40 mV.

Resting membrane potential (RMP, mV): voltage at the I = 0 point on an I–V curve measured soon after membrane rupture.

Membrane capacitance (CM, pF): determined in voltage clamp using a 10 mV, 10 ms positive square test pulse and neutralizing the outward current transient due to charging of the membrane capacitance by the C-slow capacitance compensation circuit of the HEKA EPC10/9, yielding estimates for access resistance and CM.

The following parameters were analysed in current clamp mode from recordings using 1 s hyperpolarizing/depolarizing current injection steps.

Maximum firing frequency (Hz): the reciprocal of the average of first four interspike intervals (ISIs) as measured following the highest injected current before spike inactivation was observed.

Maximum steady-state frequency (Hz): the reciprocal of the average of the last four ISIs as measured following the highest injected current before spike inactivation was observed.

Adaptation ratio (dimensionless): the ratio of the maximum steady-state frequency to the maximum frequency.

H-current (Ih): analysed by quantifying Ih-mediated depolarization (ΔV(Ih), mV) from the most hyperpolarized point in the beginning of the hyperpolarizing step to the mean value during the last 100 ms of the step (as shown in Fig. 4C).

The following action potential parameters were analysed during a sustained current injection resulting in the lowest membrane potential where the cell still spiked spontaneously.

AP peak amplitude (mV): the difference between the threshold value and the peak of the action potential.

Afterhyperpolarization (AHP, mV): measured as the difference between the spike threshold value and the most negative potential following the spike during the smallest injected current to produce spiking.

Threshold (mV): the membrane potential as measured at the point where the interpolated rate of rise (dV/dt) of the spike was equal to 5 V s−1, during the smallest injected current to produce spiking.

Maximum rise slope (mV ms−1): as measured during the steepest part of the spike rising phase.

Decay time 90–10% (ms): measured from 90 to 10% of peak-to-AHP value.

Half-width (ms): measured halfway between threshold and the peak amplitude.

Input–output gain (tuning curve) plots were created in current clamp mode by injecting a ramp of positive current at a rate of 20 pA s−1, starting from the current required to keep the cell at −60 mV, until the cell stopped firing. Average frequencies of spikes during 10 pA bins were calculated and plotted.

Gain (Hz pA−1): quantified as the slope of a linear fit to the tuning curve from 50 to 150 pA injected current.

A-like current amplitude (pA): analysed in voltage clamp as the transient outward current during a 1 s voltage step from −90 to 0 mV (as shown in Fig. 4A).

Low-threshold Ca2+ current (IT): determined as present if a transient slow depolarization (V(IT)) was observed during current clamp subthreshold depolarizing steps from a holding potential of −80 mV (as shown in Fig. 4B).

Classification of GAD65 cells

Cells were initially divided into fast spiking (FS) and non-FS cells based on subjectively defined criteria of maximum firing frequency >200 Hz and spike half-width <0.6 ms, which are consistent with classification of cortical GABAergic cells (Young & Sun, 2009; Wozny & Williams, 2011; Avermann et al. 2012). The remaining non-FS cells were divided into distinct categories based on their responses to 1 s long negative current steps that hyperpolarized them to more negative potentials than −80 mV. As a vast majority of the LHA GAD65 neurons were spontaneously spiking, those cells that were not spiking at rest were first injected positive current (5–20 pA) to make them fire at a steady rate. Spike counts were calculated during a 500 ms window before the step and a 250 ms window after (Y and X, respectively). These were used to generate the spike ratio (= X/(Y/2)). We defined the following subjective criteria for categorizing non-FS cells: cells with spike ratio <0.5 were considered late spiking (LS), those with spike ratio between 0.5 and 1.5 were considered regular spiking (RS) and those with spike ratio >1.5 were considered low-threshold spiking (LTS) cells.

Analysis

Electrophysiological data were analysed and statistical testing performed with Pulse, Patchmaster (both HEKA Elektronik), Matlab (Mathworks, Natick, MA, USA), Microsoft Excel (Microsoft, Redmond, WA, USA) and Origin (Microcal, Northampton, MA, USA) software. Averaged data are presented as mean ± SEM. Statistical significance was evaluated using one-way ANOVA followed by Tukey's honestly significant test in order to determine intergroup significance in the case of comparisons between multiple subtypes, unless indicated otherwise.

Immunohistochemistry in thin sections

Mice (>8-week-old male and female GAD65-GFP) were anaesthetized with ketamine (100 mg kg−1, i.p.) and xylazine (20 mg kg−1, i.p.), and transcardially perfused with 0.9% NaCl followed by phosphate-buffered 10% formalin. Brains were removed, postfixed for 4 h, and then submerged overnight in 30% sucrose in PBS. Coronal sections 25 μm thick were cut on a freezing sliding microtome (collected in 1:5 equal series). Sections were stored in antifreeze solution (30% sucrose and 30% ethylene glycol in PBS) at −20°C until tissue was processed for detection of GFP and hcrt/orx or MCH immunoreactivity.

Fluorescence immunohistochemistry was performed as follows: sections were washed in PBS, blocked and permeabilized in PBS with 0.3% Triton X-100 and 1% bovine serum albumin (BSA) for 1 h, incubated with primary antibodies overnight, washed, incubated with secondary antibodies for 3.5 h, washed and mounted. Antibodies were applied in PBS with 0.3% Triton X-100 and 0.2% BSA. Primary antibodies used were goat anti-GFP (1:1000; Abcam, Cambridge, UK) and rabbit anti-MCH or rabbit anti-hcrt/orx-A (both 1:1000; Phoenix Pharmaceuticals Europe, Karlsruhe, Germany). Secondary antibodies were Alexa 568 conjugated donkey anti-rabbit and Alexa 488 conjugated donkey anti-goat (both 1:1000; Molecular Probes, Eugene, OR, USA).

Analysis was performed with an Olympus BX61WI laser scanning confocal microscope (Olympus UK, Southend-on-Sea, UK). Bregma levels were assigned using information from Paxinos & Franklin (2008) and digital images were captured using Fluoview 2.1b (Olympus UK). Images were merged (where appropriate) using ImageJ (National Institutes of Health, Bethesda, MD, USA). Coexpression was analysed manually.

Immunohistochemistry in thick sections

For obtaining reconstructions of neuronal morphology and post-recording immunostaining, we fixed slices in 4% paraformaldehyde in PBS for 1 h at room temperature after recording. Slices were then stored at 4°C in PBS with 0.02% sodium azide for 1–6 months before further processing for detection of biocytin and hcrt/orx-A as follows. Cells were permeabilized and background was blocked in PBS with 0.3% Triton X-100 and 1% BSA for 1 h. For detection of biocytin and hcrt/orx-A, tissue was incubated in rabbit anti-hcrt/orx-A (1:250; Phoenix Pharmaceuticals Europe) overnight, then washed with PBS and incubated in Cy5 conjugated goat anti-rabbit (1:500; Molecular Probes) and streptavidin conjugated to Alexa 555 (1:500; Molecular Probes) for 3.5 h before final washing and mounting. For detection of biocytin, tissue was incubated for 3.5 h in streptavidin conjugated to Alexa 555. Antibodies and streptavidin were applied in PBS with 0.3% Triton X-100 and 0.2% BSA.

Morphological reconstruction

Images were taken on an Olympus BX61WI confocal microscope (Olympus FluoView v 2.1b software) using a 25× water immersion objective (NA 1.05, Olympus). Alexa 555 was excited with a diode-pumped solid-state laser at 559 nm, and fluorescence emission collected at 570–670 nm using a spectral detector (Olympus). Three-dimensional stacks of confocal images were xy-scanned at 0.248 μm per pixel resolution at 1 μm z-intervals. Kalman filtering (two images per frame) was applied during acquisition to improve image clarity.

When needed, several stacks from different parts of the neuron were integrated using the Volume Integration and Alignment System (Rodriguez et al. 2003). ImageJ was used to manipulate image stacks at various stages. Neuromantic (Myatt & Nasuto, 2007) was used to semi-manually reconstruct three-dimensional neuronal morphology and L-measure (Scorcioni et al. 2008) was used to quantify morphological parameters. Shrinkage was not corrected.

Results

LHA GAD65-GFP neurons are neurochemically distinct from hcrt/orx and MCH cells

The LHA contains two non-overlapping populations of neurochemically identified neurons: hcrt/orx and MCH neurons (Elias et al. 1998). To see whether GAD65 neurons in the LHA coexpress either of these neuropeptides, we stained thin sections from perfusion fixed GAD65-GFP mouse brains (n = 3 animals for each neuropeptide) for hcrt/orx-A and GFP or MCH and GFP. As GFP was not coexpressed with either peptide (537–786 MCH or hcrt/orx-A neurons were counted from each brain), GAD65 neurons are a completely distinct population from either hcrt/orx or MCH neurons (Fig. 1).

Figure 1.

LHA GAD65-GFP neurons are distinct from hcrt/orx and MCH neurons 
A, schematic diagram of location (boxed area) of B and D (approximately Bregma −1.6 mm). B, representative confocal micrograph of a GAD65-GFP LHA section stained for GFP (green) and MCH (red). Scale bar, 100 μm. C, magnified image of boxed area in B. Scale bar, 20 μm. D, representative confocal micrograph of a GAD65-GFP LHA section stained for GFP (green) and hcrt/orx A (red). Scale bar, 100 μm. E, magnified image of boxed area in D. Scale bar, 20 μm. DMH, dorsomedial hypothalamic nucleus; f, fornix; VMH, ventromedial hypothalamic nucleus; LHA, lateral hypothalamus; ARC, arcuate nucleus; 3V, third ventricle.

LHA GAD65 neurons are intrinsically depolarized and spiking

To probe the electrophysiological diversity of LHA GAD65 neurons, we randomly sampled 139 GAD65-GFP neurons from the LHA in acute slices. Most (124 of 139 neurons) GAD65 neurons were spontaneously active in slice under whole cell current clamp without any holding current, firing at a mean frequency of 9.1 ± 0.6 Hz (range 0.4–29.2 Hz, n = 124, Fig. 2A). We confirmed that this is not an artefact caused by our experimental conditions, by recording from some GAD65-GFP neurons in layers 1 and 2/3 of parietal cortex as controls. Consistent with literature on cortical interneurons (Ascoli et al. 2008), these were not firing at rest and had a mean resting membrane potential of −68.3 ± 1.6 mV (n = 12), but they were considered healthy as when depolarized they fired action potentials that overshot 0 mV and recordings were stable. In the cell-attached mode, wherein the cytosol of the recorded cell is left unperturbed, LHA GAD65 neurons had a mean firing rate of 6.7 ± 2.6 Hz (n = 8 of 8 cells, P > 0.8 compared to whole cell conditions by Student's t test, Fig. 2A), indicating that our whole cell intracellular solution did not affect spontaneous activity.

Figure 2.

LHA GAD65 neurons are intrinsically active and depolarized 
A, above, merged oblique illumination contrast and fluorescence micrograph of spontaneously spiking LHA GAD65-GFP neuron being recorded. Middle, whole cell current clamp recording from the neuron at rest. Below, cell attached recording from another LHA GAD65-GFP neuron. B, tonic activity of LHA GAD65 neurons continues in the presence of synaptic blockade. Above, representative current clamp trace (n = 5). Below, firing rate histogram from above trace. C, GAD65 neurons are hyperpolarized in the presence of action potential blockade by TTX. Left, representative current clamp trace (n = 4). Right, quantified RMPs based on I–V data before and after TTX. Data points with error bars are means ± SEM; other points are from individual cells. *P < 0.05.

To see if the intrinsic firing of GAD65 neurons is driven by synaptic activity we first blocked fast synaptic receptors with a cocktail of synaptic blockers (CNQX, AP5, PiTX, strychnine, Fig. 2B). This did not change firing frequency (control 5.3 ± 1.3 Hz and synaptic blockade 5.4 ± 1.4 Hz, n = 5, P > 0.8 by paired Student's t test), indicating that ionotropic glutamate, GABAA or glycine receptors do not control tonic firing of GAD65 cells. To further probe if these cells are tonically depolarized in an action potential dependent manner, we blocked action potentials with TTX (Fig. 2C). This produced a 6.2 ± 0.8 mV hyperpolarization (RMP in control −49.4 ± 2.4 mV and in TTX −55.6 ± 2.2 mV, n = 4, P < 0.01 by paired Student's t test). Taken together, these data suggest that ongoing activity in the slice slightly depolarizes these neurons via other mechanisms than fast synaptic transmission mediated by GABA, glutamate or glycine. We note that the action potential-dependent depolarization was so small that it is unlikely to fully account for the depolarized and tonically firing state of these neurons.

As some LHA GABAergic neurons are known to be active only during sleep (Hassani et al. 2010), we performed an experiment where tissue was prepared at night, 3 h after lights off, during the waking period of mice. The neurons were firing at a mean frequency of 12.3 ± 1.6 Hz (n = 7) in tissue prepared during night suggesting that these cells do not contain an intrinsic mechanism that silences them during the night.

Evoked firing of LHA GAD65 cells: comparison with other LHA cells and cortical GABA neurons

Cortical GABAergic interneurons are divided into five main categories according to their evoked firing patterns: fast spiking, late spiking, low-threshold spiking (also called burst spiking), irregular spiking and regular spiking neurons (Kawaguchi, 1995; Markram et al. 2004; Ascoli et al. 2008; Young & Sun, 2009; Wozny & Williams, 2011). To be consistent with this classification scheme, we adapted protocols routinely used in subjective categorization of GABAergic interneurons in cortex (Ascoli et al. 2008; Young & Sun, 2009) and, because LHA GAD65 cells were found to be spontaneously active unlike cortical GABAergic interneurons (Fig. 2), we also modified protocols previously used to characterize spontaneously spiking neurons in LHA (Schone et al. 2011) in the following analysis of LHA GAD65 neurons. To begin, we looked at action potential (AP) waveforms by quantifying, for each neuron, the average AP half-width when the cell was spiking at near-threshold membrane potential (achieved by sustained current injection), and maximum firing frequency during maximal depolarization (see Methods for definition of parameters and precise protocols used for measurement). These electrical characteristics formed continuous distributions without any clear clustering (Fig. 3A and B). However, to describe the main changes along these continua we imposed on the data the standard electrophysiological subtype classification scheme used on inhibitory interneurons in other brain areas (Kawaguchi, 1995; Markram et al. 2004; Ascoli et al. 2008; Young & Sun, 2009; Wozny & Williams, 2011). We thus grouped GAD65 neurons that fired narrow APs (half-width < 0.6 ms) at high maximum rates (>200 Hz), which correspond to the cortical fast spiking GABAergic interneurons (Young & Sun, 2009; Wozny & Williams, 2011; Avermann et al. 2012), as fast spiking (FS, Fig. 3A and B). FS cells comprised approximately 14% (20/139) of all LHA GAD65 cells and had mean maximum firing frequency of 257.5 ± 11.6 Hz (n = 20, compared to 108.7 ± 4.3 Hz for the other GAD65 neurons, n = 119). The mean AP half-width of FS neurons was 0.41 ± 0.02 ms (n = 20, compared to 0.65 ± 0.02 ms for the other GAD65 neurons, n = 119).

Figure 3.

Electrophysiological classification of LHA GAD65 neurons 
A, fast spiking LHA GAD65 neurons. Maximal firing rates of a typical LHA GAD65 neuron (left) and a fast spiking LHA GAD65 neuron (right) when a maximally depolarizing current step is applied (shown in red; see Methods for details). B, left, representative action potential waveforms of FS (cyan) and non-FS neurons (black). Both traces are mean of 10 action potentials aligned at peak. Right, summary of action potential half-width plotted against maximum firing frequency for 139 LHA GAD65 cells (FS cells in cyan) as well as hcrt/orx D and H type cells for comparison (orxD and orxH, respectively). The representative data shown on left and in A are circled in red. C, classification of non-FS cells by hyperpolarizing current steps. Representative responses of RS (left), LS (middle) and LTS (right) to the current step. D, summary of spike ratio calculated from the X (250 ms) and Y (500 ms) time windows shown in C (see Methods) as spike amounts during X/(Y/2). Open circles denote individual cells, closed circles mean ± SEM. E, fractions of each cell type (n = 139 total). F, representative action potentials of FS, LTS, RS and LS cells. Each trace is the mean of 10 action potentials from one cell aligned at peak. G, same scatter plot as right panel in B, with all GAD65 cell types shown, without hcrt/orx cells. H, representative cell showing protocol for gain analysis. Left, cells were injected positive going ramps of current at 20 pA s−1 (red trace) until they stopped firing. Right, firing rates were analysed from these traces in 10 pA bins yielding a tuning curve. Gain was quantified as the slope of a linear equation fitted to the regime 50–150 pA (red dotted line). I, left, mean tuning curves for each LHA GAD65 subtype (n = 10 cells for each) and, for comparison, hcrt/orx H and D type cells (n = 6 and 12 cells, respectively). Right, quantification of gain (red dotted line in H) for each cell type (n = 10 for each) and hcrt/orx H and D type cells (n = 6 and 12 cells, respectively). Open circles denote individual cells, closed circles mean ± SEM *P < 0.05, *P < 0.01.

To compare these parameters to those of other cells in the local circuit, we also compared them to hcrt/orx neurons. All hcrt/orx neurons were tonically active in slice, firing at a mean frequency of 10.4 ± 0.9 Hz (range 5.3–18.0 Hz, n = 18) and when depolarized maximally (see Methods) their mean maximum firing frequency was 88.9 ± 6.7 Hz (range 29.2–152.7 Hz, n = 29; by subtype (Williams et al. 2008; Schone et al. 2011): H-type, 88.5 ± 12.2 Hz, n = 10; D-type, 89.4 ± 9.9 Hz, n = 19). AP half-widths were 0.84 ± 0.04 ms (range 0.54–1.28 ms, n = 29; H-type, 0.83 ± 0.09 ms, n = 10; D-type, 0.85 ± 0.05 ms, n = 19; Fig. 3B). These data indicate that hcrt/orx cells do not have the ability to fire narrow action potentials at high frequencies like FS GAD65 neurons.

To further characterize the remaining non-FS GAD65 cells, we recorded firing responses after hyperpolarizing current steps lasting 1 s. This allows us to see how release from inhibition would affect activity. There were three kinds of firing patterns after the hyperpolarizing step: increased, decreased or unchanged firing compared to before the step (Fig. 3CE). Approximately 24% (33/139) of the population increased firing frequency by more than 50%, with a mean increase of 206.4 ± 21.6% (n = 33). These cells were termed low-threshold spiking (LTS). About 11% (15/139) of all neurons did not commence firing immediately after the step, but rather showed a slow ramp depolarization back to baseline resting membrane potential, resulting in a firing rate of less than 50% of that before the step (on average 11.3 ± 4.4%, n = 15). These were termed late-spiking (LS) neurons. The remaining 51% (71/139) did not change firing by more than 50% in either direction after the hyperpolarizing step (average change 7.5 ± 3.5%, n = 71) and were termed regular spiking (RS) neurons. In similar classification schemes used on cortical interneurons, a further cell type called irregular spiking cell is often found (Parra et al. 1998; Galarreta et al. 2004; Young and Sun, 2009) which fires APs at stochastically varying intervals when depolarized by sustained current injection. However, we did not find this subtype among LHA GAD65 neurons which were always spiking at a steady rate (as exemplified in Fig. 2) or not at all (15/139 neurons), in which case they spiked at a steady rate when depolarized to threshold. To be more specific, the coefficient of variation of inter-spike-intervals in traces containing at least 10 spikes was 0.08 ± 0.01 (range, 0.04–0.16; n = 45), whereas cortical irregular spiking GAD65 neurons, identified using the same mouse line we used (Galarreta et al. 2004), were reported to have a corresponding value of 0.58 ± 0.05 (n = 21; data from Galarreta et al. 2004).

After the initial delineation of subtypes of LHA GAD65 neurons we studied their properties further. The AP waveforms (when cells were injected current to elicit minimal AP firing) of LS cells were broader (Fig. 3F and G, half-width 0.80 ± 0.08 ms, n = 15, P < 0.01 compared to RS, 0.64 ± 0.03 ms, n = 71; LTS, 0.60 ± 0.03 ms, n = 33; or FS, 0.41 ± 0.02 ms, n = 20) than the others and LS cells also had significantly lower maximum firing rates (67.2 ± 7.7 Hz, n = 15, P < 0.01 compared to RS, 105.2 ± 5.2 Hz, n = 71; LTS, 135.7 ± 7.8 Hz, n = 33; or FS, 257.5 ± 11.6 Hz, n = 20). FS cells showed varied responses to the hyperpolarizing current injection used to classify the other subgroups. Their mean change in firing after the step was 160.5 ± 48.7% (n = 20, range −71.4 to +900.0%), indicating that according to our classification scheme of non-FS cells the FS cells would contain subtypes. Cortical FS neurons have a further distinct property of very little spike rate adaptation (Young & Sun, 2009), but the LHA GAD65 FS neurons had prominent adaptation with mean adaptation ratio 0.39 ± 0.03 (n = 20, range 0.17–0.58). This was the case also for other subtypes (LTS, 0.39 ± 0.03, n = 33, range 0.12–0.82; RS, 0.50 ± 0.02, n = 71, range 0.16–1.04) apart from LS, which adapted the least (0.79 ± 0.07, n = 15, range 0.42–1.52).

To get more comprehensive information about evoked firing characteristics, we measured input–output gain of GAD65 neurons, which has been used as a further defining characteristic of cortical FS cells (Young & Sun, 2009). We measured gain by first setting the cell at −60 mV and then injecting depolarizing ramps (20 pA s−1) of current, which resulted in steadily increasing firing frequency (Fig. 3H). To quantify gain, we examined the change in firing rate between 50 and 150 pA of injected current by fitting a linear equation to the tuning curve (Fig. 3H). We chose this range because it is a mostly linear part of the tuning curve and it corresponds to a physiologically meaningful dynamic range of firing since at these injected current values (1) all cells had begun firing, (2) many cells showed increases from their resting firing frequency of roughly 9 Hz (Table 1) and (3) in vivo recordings of LHA GABAergic cells across the sleep–wake cycle show changes in firing from around 5 to 20 Hz (Hassani et al. 2010). In this regime, FS cells had significantly higher gain (0.31 ± 0.03 Hz pA−1, n = 10) than RS (0.16 ± 0.02 Hz pA−1, n = 10, P < 0.01), LTS (0.16 ± 0.02 Hz pA−1, n = 10, P < 0.01) or LS (0.19 ± 0.04 Hz pA−1, n = 10, P < 0.05) cells. To compare gain to that of other nearby cells, we also measured it from the two subtypes of hcrt/orx neurons (Fig. 3I). Hcrt/orx H-type cells (n = 6) had significantly lower gain than hcrt/orx D-type or GAD65 FS cells (H-type, 0.13 ± 0.01 Hz pA−1, n = 6, P < 0.05 compared to D-type, 0.26 ± 0.14 Hz pA−1, n = 12). Other significant differences in gain were not detected (P > 0.5), but as is clear from Fig. 3I, GAD65 FS and LS cells as well as hcrt/orx H-type cells were able to sustain firing at high currents that caused hcrt/orx D-type, GAD65 LTS and RS cells to stop firing due to depolarization block. These data indicate that LHA GAD65 FS cell firing can be modulated with smaller input than non-FS cells, similarly to cortical FS cells (Young & Sun, 2009).

Table 1.  Summary of electrophysiological parameters for LH GAD65-GFP neurons
 FSLSLTSRS 
  1. Values are means ± SEM (n). Statistical testing shown as > = P < 0.05 and >> = P < 0.01. *LTS cells without V(IT). Abbreviations: AP, action potential; RMP, resting membrane potential.

Spike ratio2.6 ± 0.5  (20)0.1 ± 0.0  (15)3.1 ± 0.2  (33)1.1 ± 0.0  (71)LTS, FS>>RS>>LS
Maximum firing frequency (Hz)257.5 ± 11.6  (20)67.2 ± 7.7  (15)135.7 ± 7.8  (33)105.2 ± 5.2  (71)FS>>LTS>RS>>LS
Steady-state firing frequency (Hz)103.6 ± 9.2  (20)49.4 ± 3.9  (15)50.6 ± 4.4  (32)50.7 ± 3.1  (71)FS>>LTS, RS, LS
Adaptation ratio0.39 ± 0.03  (20)0.79 ± 0.07  (15)0.39 ± 0.03  (33)0.50 ± 0.02  (71)LS>>RS>FS, LTS
AP threshold (mV)−40.4 ± 0.7  (20)−36.8 ± 0.9  (15)−38.8 ± 0.5  (33)−38.1 ± 0.6  (71)LS>FS
AP half-width (ms)0.41 ± 0.02  (20)0.80 ± 0.08  (15)0.60 ± 0.03  (33)0.64 ± 0.03  (71)LS>>RS, LTS>>FS
AP afterhyperpolarization (mV)−25.1 ± 0.8  (20)−26.6 ± 1.4  (15)−23.4 ± 0.8  (33)−25.7 ± 0.6  (71)LS>LTS
AP peak amplitude (mV)73.8 ± 2.3  (20)59.4 ± 3.7  (15)63.0 ± 1.8  (33)59.2 ± 1.8  (71)FS>>LS, LTS, RS
AP max. rise slope (mV ms−1)334.2 ± 18.1  (20)195.1 ± 19.8  (15)226.9 ± 16.9  (33)220.4 ± 11.3  (71)FS>>LS, LTS, RS
AP 10–90% decay time (ms)0.23 ± 0.02  (20)0.69 ± 0.09  (15)0.44 ± 0.03  (33)0.47 ± 0.03  (71)LS>>LTS, RS>>FS
Firing frequency at rest (Hz)8.9 ± 1.8  (18)7.9 ± 1.6  (13)8.1 ± 1.1  (30)9.5 ± 0.8  (63) 
Input resistance (MΩ)463 ± 89  (15)628 ± 94  (13)416 ± 39  (29)646 ± 54  (58)RS>LTS
RMP (mV)−53.7 ± 3.0  (15)−47.6 ± 1.7  (10)−51.7 ± 1.9  (30)−48.3 ± 1.7  (54) 
Membrane time constant (ms)28.5 ± 1.6  (12)24.2 ± 3.6  (14)25.0 ± 2.6  (18)25.0 ± 4.2  (15) 
Membrane capacitance (pF)16.9 ± 1.3  (17)17.8 ± 1.5  (14)16.3 ± 0.8  (34)15.5 ± 0.6  (68) 
A-like current amplitude (pA)160 ± 45  (14)1575 ± 296  (12)217 ± 40  (29)472 ± 49  (61)LS>>RS>FS; LS>>LTS
I h-mediated depolarization (mV)6.6 ± 0.9  (16)5.6 ± 1.1  (13)9.7 ± 1.3  (15)*3.0 ± 0.9  (21)LTS*>RS, LS
Number of cells with V(IT)4  (16)0  (13)10  (26)9  (65) 

Subtypes of LHA GAD65 cells have different voltage-gated conductances

It has been shown that H-type hcrt/orx cells have a higher density of a low-threshold, transient outward current (termed here A-like current) than D-type cells (Schone et al. 2011). To see whether the similar phenotype of LS cells is linked to the same current, we analysed A-like current amplitudes in LHA GAD65 neurons in voltage clamp by stepping from −90 to 0 mV. This elicited transient outward currents of varying magnitude (Fig. 4A). LS neurons had significantly higher A-like currents than all other subtypes (1575 ± 296 pA, n = 12, P < 0.01 for all comparisons), and RS cells had significantly higher currents (472 ± 49 pA, n = 61, P < 0.05) than FS cells (160 ± 45 pA, n = 14). LTS cells had A-like current of 217 ± 40 pA (n = 29). These data suggest that the LS firing phenotype is linked to their high A-like current.

Figure 4.

Voltage-gated conductances of LHA GAD65 neurons 
A, A-like current. Left side: top, example of A-like current from an LS cell. Middle, FS cell with minimal A-like current. Bottom, voltage clamp protocol used in the two traces above. Right, quantified A-like current amplitudes by cell type (n = 14, FS; 12, LS; 29, LTS; 61, RS). **P < 0.01, *P < 0.05. B, transient low-threshold inward current-mediated slow depolarization, V(IT). Left side: top, example of V(IT) (arrow) in an LTS cell. Middle, RS cell with no V(IT). Bottom, current clamp protocol used in the two traces above. Right, fraction of cells with V(IT) by cell type (total n = 16, FS; 13, LS; 26, LTS; 65, RS). C, Ih-mediated depolarization, ΔV(Ih). Left side: top, example of ΔV(Ih) from an LTS cell. Middle, RS cell with minimal ΔV(Ih). Bottom, current clamp protocol used in the two traces above. Right, quantified ΔV(Ih) by cell type (n = 16, FS; 13, LS; 15, LTS; 21, RS). **P < 0.01, *P < 0.05.

We found that in 19.2% (23/120) of GAD65 neurons depolarizing current steps from a holding potential of −80 mV elicited a slow transient depolarization (Fig. 4B). This was termed V(IT) and is probably due to a low-threshold Ca2+ current which is also seen in LTS subtypes of cortical GABAergic interneurons (Kawaguchi, 1995; Parra et al. 1998; Goldberg et al. 2004). V(IT) was most frequently expressed in LTS cells (38.5%, 10/26 cells), in 13.8% of RS cells (9/65), 25.0% (4/16) of FS cells and never in LS cells (0/13). The presence of V(IT) therefore correlates most strongly with LTS cells, but does not always lead to the LTS phenotype as some LTS cells can increase firing after hyperpolarization without V(IT).

To test if the h-current (Ih, a hyper-polarization-activated cation conductance found in many neurons; Pape, 1996) could account for the LTS phenotype of those LTS cells without V(IT), we quantified the spontaneous depolarization during a hyperpolarizing current step below −80 mV (ΔV(Ih), Fig. 4C). LTS cells had the largest ΔV(Ih) (9.1 ± 1.1 mV, n = 28), in particular the LTS cells that did not express V(IT) had mean ΔV(Ih) of 9.7 ± 1.3 mV (n = 15) which was significantly larger than ΔV(Ih) of LS (5.6 ± 1.1 mV, n = 13, P < 0.05) or RS (3.0 ± 0.9 mV, n = 21, P < 0.01) cells. FS cells had ΔV(Ih) of 6.6 ± 0.9 mV (n = 16). These data suggest that IT and Ih are important contributors to the LTS phenotype. Electrophysiological properties of LHA GAD65 subpopulations are summarized in Table 1.

We note that, even when using the same current clamp protocol that unmasked clear V(IT) in 19.2% of GAD65 neurons, we never saw V(IT) in hcrt/orx neurons (n = 29). This is consistent with V(IT) not being reported in other studies of mouse hcrt/orx neurons (Burdakov et al. 2005; Williams et al. 2008; Schone et al. 2011).

Locations and morphology of LHA GAD65 neurons

We filled recorded neurons with biocytin and reconstructed the neurons to see their morphology (Fig. 5A). LHA GAD65 cells had very few dendritic spines (analysed in the whole dendritic arbour in a Z-projection of a confocal Z-stack of the whole cell) or sometimes none at all (Fig. 5B and C). RS cells had on average 4.0 ± 1.8 spines per cell (range 0–17, n = 9 cells), FS cells 7.2 ± 2.7 (range 2–19, n = 6 cells), LS cells 3.9 ± 1.1 (range 1–11, n = 9 cells) and LTS cells 16.7 ± 8.3 (range 3–76, n = 9 cells). These differences in spine amounts were not statistically significant (P values for all comparisons > 0.1). Although spine counts obtained by biocytin filling may be underestimates, these results suggest that LHA GAD65 cells are mostly devoid of large spines visible with light microscopy. Between one and three neurons in each subtype (out of 6–9 cells per subtype) showed sparsely situated flattened regions in some dendrites (example in Fig. 5D). We analysed parameters describing neuronal morphology, but did not find striking differences between cell types other than a trend for FS cells to have slightly longer neurites than LS and RS cells (summarized in Table 2; parameters explained in detail in Methods). These results suggest that as with many GABAergic interneuron subtypes in the cortex (Markram et al. 2004; but see Young & Sun, 2009; Wozny & Williams, 2011), the LHA GAD65 neurons do not show obvious morphological specializations specific to electrophysiological subtype. However, we note that a complete description of neuronal morphology cannot be obtained in the coronal slice preparation alone due to truncation of processes.

Figure 5.

Morphology and locations of LHA GAD65 neurons 
A, pictures of four reconstructed neurons of each subtype in a slice drawn on the same coronal section (although they were not always found in this particular section) to illustrate their location in the coronal plane with respect to fornix (f) and third ventricle (3V). Cells are colour coded according to subtype: RS, black; LS, blue; LTS, red; FS, green. Scale bars = 500 μm. B, representative z-projections of confocal micrograph z-stacks through GAD65 cells to show general rarity of dendritic spines. Left, RS cells with very few spines. Right, LTS cell with highest number of spines. Scale bars = 100 μm. C, enlarged views of indicated dendrites showing scarcity of spines. Scale bars = 10 μm. D, enlarged view of a dendrite showing flattened regions. Scale bars = 10 μm. E, approximate locations of recorded GAD65 neuron somata in LHA. RS cells marked by plus signs; LTS, red open circles; LS, blue upright triangles; FS, green inverted triangles. LHA, lateral hypothalamic area; VMH, ventromedial hypothalamic nucleus; DMH, dorsomedial hypothalamic nucleus; f, fornix; 3V, third ventricle.

Table 2.  Summary of morphological parameters for LHA GAD65-GFP neurons
 RS (n = 9)FS (n = 6)LS (n = 9)LTS (n = 9) 
  1. Values are means ± SEM. Statistical testing shown as > = P < 0.05. Soma surface (μm2): surface area of the soma. N_stems: number of processes branching out from soma. N_bifs: number of bifurcations, i.e. branching points. N_branch: number of branches, i.e. segments that lie between two branching points or between one branching point and a termination point. N_tips: number of termination points. Width (μm): maximum distance between two points in the horizontal dimension. Height (μm): maximum distance between two points in the vertical dimension. Depth (μm): maximum depth between two points. Length (μm): sum of the lengths of all components of the structure. Surface (μm2): total surface area of the neuron. Volume (μm3): total volume of the neuron.

Soma surface (μm2)859 ± 118774 ± 1131409 ± 3291234 ± 408 
N_stems3.6 ± 0.44.7 ± 0.84.1 ± 0.35.1 ± 0.4 
N_bifs17.3 ± 7.758.7 ± 25.719.0 ± 9.020.6 ± 6.4 
N_branch40.8 ± 15.4125.7 ± 52.159.3 ± 25.857.3 ± 15.6 
N_tips20.9 ± 7.763.3 ± 26.123.1 ± 9.125.7 ± 6.4 
Width (μm)316 ± 63606 ± 107369 ± 68396 ± 33FS>LS, RS
Height (μm)314 ± 33368 ± 55330 ± 58397 ± 48 
Depth (μm)83 ± 783 ± 1484 ± 1076 ± 4 
Length (μm)1910 ± 3904940 ± 17401920 ± 4402430 ± 370FS>LS, RS
Surface (μm2)4540 ± 88012000 ± 49204720 ± 9506430 ± 1060FS>LS, RS
Volume (μm3)1440 ± 3503200 ± 15101620 ± 3502260 ± 450 

We noted the location of each GAD65 cell soma after recording and plotted them in Fig. 5E. The distribution of recorded GAD65 neurons was similar to that of hcrt/orx and MCH neurons, with most cells located in the LHA between Bregma levels −2.06 and −1.46 mm and a small fraction of cells in dorsomedial hypothalamic nucleus. There were no clear subtype-specific patterns of distribution.

Forty percent of LHA GAD65 neurons are inhibited by glucose

Since the lateral hypothalamus is a well-known locus of glucose-sensing neurons (Oomura et al. 1974), we tested the responses of GAD65 neurons to glucose by first switching from 1 to 5 mm glucose in the ACSF. This led to a hyperpolarization of 10.5 ± 1.2 mV (range 4.1–21.0 mV, n = 18, Fig. 6A) from resting membrane potential in about 40% (18/46) of the tested GAD65 neurons. About 60% (28/46) of tested neurons did not respond to glucose (mean hyperpolarization of 0.8 ± 0.3 mV, range −3.3 to 3.3 mV, n = 28, Fig. 6B). To show that the response in responsive neurons is not mediated by synaptic communication, we first blocked AP-dependent synaptic release with 0.5 μm TTX, which did not block the glucose response (12.3 ± 2.6 mV hyperpolarization, n = 4, P > 0.8 compared to control using Student's t test, Fig. 6A). To further show that AP-independent presynaptic release does not account for the response, we blocked all synaptic communication with a cocktail of synaptic blockers (CNQX, AP-5, PiTX, strychnine) and TTX, which did not block the glucose response (10.5 ± 1.6 mV hyperpolarization, n = 4, P > 0.6 compared to control by Student's t test, data not shown).

Figure 6.

Approximately 40% of LHA GAD65 neurons are directly inhibited by physiological concentrations of glucose 
A, typical response of a glucose-responsive LHA GAD65 neuron to 5 mm glucose in control conditions (n = 18) and with TTX in the bath (n = 4). B, typical response of a glucose non-responsive LHA GAD65 neuron to 5 mm glucose (n = 28). C, typical response of a glucose-responsive LHA GAD65 neuron to 2.5 mm glucose (n = 4). D, left, proportions of glucose-responsive (grey) and non-responsive (black) neurons in FS vs. non-FS LHA GAD65 neurons. Right, action potential half-width and maximum firing frequency plotted for glucose-responsive (grey) and non-responsive (black) cells. E, approximate locations of glucose-responsive (grey circles) and non-responsive (black squares) GAD65 neurons.

To probe the sensitivity of glucose sensing in the responsive neurons, we also used a smaller change in glucose of 1–2.5 mm (Fig. 6C). This led to an average hyperpolarization of 9.4 ± 2.6 mV in the same cells that hyperpolarized by 16.9 ± 2.9 mV in response to 5 mm glucose (n = 4, P < 0.05 by paired Student's t test), indicating that the response is robust and dose-dependent at physiologically relevant glucose concentrations.

Cell type preference of LHA GAD65 cell glucose inhibition

When we characterized the electrophysiological fingerprints (Fig. 3) of glucose-responsive and non-responsive GAD65 neurons (a more than 4 mV hyperpolarization in response to glucose was defined as responsive), we found the largest proportion (10/16) of glucose-responsive cells in the FS subtype. All the non-FS classes also had some glucose-responsive cells but had predominantly non-responsive cells: LTS 3/10, LS 4/11 and RS 3/10 cells responding to glucose (Fig. 6D). The mean maximum firing rates and AP half-widths of glucose-responsive versus non-responsive cells were 196.4 ± 19.2 Hz (n = 20) versus 152.6 ± 20.6 Hz (n = 28, P = 0.14 using Student's t test) and 0.56 ± 0.05 ms (n = 20) versus 0.65 ± 0.04 ms (n = 28, P = 0.14 using Student's t test), respectively (Fig. 6D). To compare both parameters in the same statistical test, we calculated the spike width/maximum firing frequency ratio, which was 0.011 ± 0.002 ms Hz−1 for glucose-responsive and 0.003 ± 0.001 ms Hz−1 for non-responsive cells (P < 0.05 by Student's t test). This suggests that, only when these properties are analysed together, GAD65 neurons that are inhibited by glucose tend toward FS-like characteristics, i.e. to be capable of firing narrow spikes at higher frequencies than non-responsive cells. Finally, we also noted the locations of glucose-responsive and non-responsive neurons (Fig. 6E), but did not find a consistent pattern in their spatial distribution.

Since the response of GAD65 cells to glucose is similar to that of hcrt/orx cells, we used post hoc immunostaining to see if the glucose-responsive GAD65 neurons also express hcrt/orx. None of the six cells tested was positive for hcrt/orx whereas hcrt/orx-positive cells were identified in close vicinity to LHA GAD65 neurons (Fig. 7A). This is consistent with the lack of coexpression of hcrt/orx and GFP in GAD65-GFP animals (Fig. 1) as well as the differences in electrophysiological parameters of all LHA GAD65 subtypes compared to hcrt/orx cells; for example, whole-cell capacitances of glucose-responsive GAD65 neurons (19.6 ± 1.9 pF) are lower than for hcrt/orx cells (26.7 ± 1.1 pF for D-type and 34.9 ± 2.1 pF for H-type cells, as reported in Schone et al. 2011).

Figure 7.

Mechanisms of glucose inhibition in LHA GAD65 neurons 
A, example of glucose-sensitive GAD65 neuron filled with biocytin and double stained for hcrt/orx and biocytin. All panels are confocal micrographs of the same field of view. Biocytin (left) and hcrt/orx (middle) are not colocalized (right). Figure is representative of n = 6 replicates. Scale bar = 10 μm. B, above, typical response of a glucose-sensitive LHA GAD65 neuron to 5 mm glucose in control conditions showing where I–V curves in C were taken (arrows). Below, example of I–V measurement protocol. C, left, I–V relationships before (black), during (red) and after (blue) glucose application. Right, net glucose-activated current in this cell (I–V before subtracted from I–V during glucose). D, left, collection of example net glucose activated I–V curves from different GAD65 cells colour coded according to slope. Ones with positive slope in black, negative slope in red and ones with mixture of both in blue. Right, I–V curves grouped according to slope as in left panel and averaged; error bars are SEM. E, mean of all glucose-induced net I–V curves. F, LHA GAD65 neuron glucose sensing is mimicked by the non-metabolizable 2DG. Typical response of a glucose-sensitive LHA GAD65 neuron to 5 mm 2DG (n = 9). G, mean net I–V curves of glucose (grey) and 2DG (black) activated currents from the same cells (n = 5 cells).

Ionic mechanisms of GAD65 cell glucose inhibition

To study the mechanism of this glucose response, we analysed the I–V relationship of the conductance change during the response (Fig. 7B and C). The resulting net I–V relationships of the glucose-induced current showed considerable variance across cells, suggesting that several different ionic species are involved in the responses of different GAD65 cells (Fig. 7D). To analyse this in more detail, we grouped the net glucose-activated I–V curves for averaging into those with positive, negative or mixed slope (Fig. 7D). The majority of cells (n = 8, Fig. 7D black traces) showed an I–V relationship suggesting an activated K+ conductance (mean reversal potential −102.0 ± 8.4 mV; EK = −107.6 mV with our solutions), whereas others showed an inactivated excitatory, most probably mixed cationic, conductance (n = 3, Fig. 7D red traces, mean reversal potential −29.2 ± 11.3 mV). Some cells (n = 6, Fig. 7D blue traces) showed either an I–V parallel to the x-axis or one resembling a parabola suggesting a near equal mixture of the two currents outlined above. Indeed when an average of net glucose-induced I–V curves is taken across all cells (Fig. 7E), the resulting curve is again similar to a parabola, consistent with the presence of two conductance changes with well-separated reversal potentials. We thus speculate that each cell has a different mixture of at least two different populations of ion channels that act to hyperpolarize the cell when glucose concentrations rise. We observed no consistent correlation between electrophysiological subtype and the shape of the I–V response to glucose.

To test if the GAD65 cell glucose response is independent of glucose metabolism like that of hcrt/orx neurons (González et al. 2008), we exposed them to the non-metabolizable glucose analogue 2-deoxy-d-glucose (2DG). 2DG at 5 mm hyperpolarized the glucose responsive cells by 9.5 ± 1.5 mV (n = 9, Fig. 7F) and 1 mm 2DG by 2.4 ± 0.9 mV (n = 6, data not shown). As a negative control it has been shown (González et al. 2008) that cortical neurons are not hyperpolarized by 5 mm 2DG. The I–V relationship of the 2DG-induced conductance change was similar to that induced by glucose in the same cells (n = 5, Fig. 7G), suggesting that the same mechanism is driving both responses. Thus, the data suggest that like hcrt/orx neurons, GAD65 neurons are also hyperpolarized by glucose via a metabolism-independent mechanism.

Discussion

LHA GAD65 neurons are intrinsically active and are distinct from hcrt/orx and MCH cells

We showed that LHA GAD65 neurons are distinct from the main LHA cell types containing MCH or hcrt/orx. Besides immunohistochemistry, this was also reflected in their high firing rates at rest and smaller size visually and in terms of membrane capacitance. Since GAD65 and 67 are both expressed by most GABAergic neurons in the CNS (Soghomonian & Martin, 1998), the lack of coexpression of MCH and GAD65 is remarkable in light of the evidence for coexpression of GAD67 and MCH (Sapin et al. 2010). The high spontaneous average firing rate of GAD65 neurons was found to not depend on synaptic inputs as it was unchanged by synaptic blockade. Firing was also not abolished by changing the time of day when slices were prepared.

Four electrophysiological subtypes of LHA GAD65 neurons

Although most of the electrophysiological properties we observed were continuously distributed without clear clustering across all LHA GAD65 neurons, they can be divided into four electrophysiological subtypes consistent with the literature on GABAergic interneurons: FS cells with high maximum firing rates, narrow APs and high gain; LS cells with low maximum firing rates, broad APs, slow recovery from hyperpolarization and large A-like currents; LTS cells with V(IT) or prominent ΔV(Ih) and accelerated spiking after hyperpolarization; and the large group of RS cells that did not belong to any of the other subclasses. Our preliminary analysis of neuronal morphology did not reveal striking differences in morphological properties of these subtypes. However, differences might be found in a more thorough analysis using, for example, different cutting angles instead of the coronal slices we studied here.

Implications of subtype differences to processing in LHA

Maximum AP frequency was much higher in FS GAD65 than hcrt/orx cells, whereas AP half-width showed the inverse correlation. The same difference was found between FS cells and the other GAD65 subpopulations by definition, but, in particular, LS cells had broader APs and a lower maximum firing rate than other GAD65 subpopulations. This suggests different processing roles for these types of neurons in the local circuit: faster APs and capability to fire more rapidly are associated with specialization to temporally precise feedforward inhibition in cortical circuits (Freund & Katona, 2007). In addition to maintaining temporal fidelity of integration windows (Gabernet et al. 2005), the cortical feedforward inhibitory FS cells have been reported to suppress saturation of pyramidal cell responses to strong inputs by delivering disynaptic inhibition concurrently with the strong input to pyramidal cells, thus increasing their dynamic range (Pouille et al. 2009). On the other hand, hcrt/orx and LS GAD65 cells with their slower APs and maximum firing rates might be expected to perform slower processing than FS cells (Freund & Katona, 2007). Furthermore, the marked differences in gains and excitatory synapse arrangement in hcrt/orx (Schone et al. 2011) and GAD65 neuron subtypes imply that these neurons play different computational roles in the local network. Potential differences in input–output gains of glucose sensing GAD65 and D and H hcrt/orx cells (Fig. 3I) could lead to graded sensing by these different cell types when the ambient glucose concentration surrounding them changes. Thus, it is conceivable that changes in CSF glucose levels are encoded as subtle, cell type-specific changes in firing in the LHA.

While hcrt/orx cells never showed the V(IT), many GAD65 cells did. It was observed most frequently in the LTS group, although other groups except LS also showed it. The lack of clear correlation might mean, for example, that cells undergo a change of phenotype from time to time and during these transitions expression levels of conductance-producing proteins change slowly. A combination of Ih and IT can lead to an intrinsic bursting phenotype, as in, for example, cholinergic basal forebrain neurons (Jones, 2004) and thalamic and inferior olivary neurons (Luthi & McCormick, 1998). The combination of V(IT) and ΔV(Ih) was here found also in many cells (e.g. Fig. 3C right panel). Therefore, although we did not find any spontaneously bursting neurons, it is possible that V(IT) and ΔV(Ih) expressing GAD65 neurons do this in vivo, if, for example, their RMPs are in a hyperpolarized range where thalamic neurons with IT and Ih burst-fire (McCormick & Bal, 1997). Our failure to observe this might be due to a methodological artefact, e.g. loss of synaptic connections necessary to drive the oscillatory behaviour or run-down of either conductance. Since there is evidence to suggest synaptic connectivity between GABAergic and hcrt/orx neurons (Louis et al. 2010; Leinninger et al. 2011) and ongoing inhibition of hcrt/orx neurons by the local GABAergic interneurons during sleep (Alam et al. 2005; Yi et al. 2009), it is possible that intrinsically bursting GABAergic inputs might phase the tonic activity of hcrt/orx neurons into complex patterns during wakefulness.

Glucose-inhibited LHA GAD65 neurons

A subset of ∼40% of LHA GAD65 neurons were shown to be directly inhibited by physiological concentrations of glucose (Fig. 6). The FS category had the largest proportion of glucose-sensitive cells, but this distinction was not clear cut with all four electrophysiological categories having some glucose-sensitive cells (Fig. 6D). Because of the predominance of glucose-sensitive cells in the FS cell group, these cells might serve a specialized role in processing metabolic information in the LHA just as hippocampal FS cells are thought to serve a special role in generating gamma oscillations (Freund & Katona, 2007; Mann & Paulsen, 2007), which might be important for information processing (Jensen et al. 2007). The ionic mechanisms behind the GAD65 cell glucose response seem to involve at least two populations of ion channels that might be expressed at varying densities in each cell. The I–V relationships of glucose-induced currents were heterogeneous across cells; some cells showed an activated hyperpolarizing conductance (Fig. 7D, black curves), similar to the K+ current mediating the glucose response in hcrt/orx cells (Burdakov et al. 2006), while some GAD65 cells showed inhibition of a depolarizing conductance (Fig. 7D, red curves). Some GAD65 neurons had glucose-activated I–V relationships suggesting a mixture of both above mentioned ionic mechanisms (Fig. 7D, blue curves). Since the non-metabolizable glucose analogue 2DG mimicked the action of glucose on glucose-sensitive GAD65 cells, this response is unlikely to be mediated by intracellular metabolism of glucose beyond its phosphorylation, as in hcrt/orx cells (Gonzalez et al. 2008). However, we confirmed that glucose-inhibited GAD65 neurons do not express hcrt/orx (Figs 1D and E and 7A), in spite of the properties shared between these LHA cell groups. Some LHA neuropeptide Y neurons are likewise hyperpolarized by glucose (Marston et al. 2011), but the glucose-sensitive GAD65 neurons are likely to be a distinct population due to the larger soma size and more dorsolaterally biased distribution of LHA neuropeptide Y neurons (Marston et al. 2011).

Recent data indicate that GABAergic and hcrt/orx neurons of the LHA are likely to be interconnected (Leinninger et al. 2009, 2011; Louis et al. 2010). Future work will show how these components interact to detect changes in glucose levels.

Translational perspective

Neurons producing the inhibitory neurotransmitter GABA are vital for brain function. They are a critical component of ‘higher’ brain centres, such as the cortex, where detailed knowledge of their properties has been experimentally obtained. In contrast, in vital and evolutionarily older brain areas such as the hypothalamus, such knowledge is lacking. We measured different functional properties of a subset GABA cells in the LH, identified by the marker GAD65. Surprisingly, these cells were found to be a different population from key known LH neurons, orexin and MCH cells. Furthermore, the hypothalamic GABA neurons were intrinsically active, different from ‘classical’ GABA neurons of the cortex. Glucose silenced a subset of LH GABA cells. Our results identify and characterize a new functional element of LH networks, which are essential for avoiding disorders of sleep and body weight. This lays groundwork for further understanding of this important brain area, and highlights similarities and differences between GABA cells in different brain regions.

Appendix

Author contributions

M.M.K. designed and performed experiments, analysed data and wrote the paper. D.B. obtained funding, designed and directed experiments and contributed to writing the paper. G.Z. and F.E. engineered the GAD65-GFP mice. All authors approved the final version.

Acknowledgements

We thank Professor Ole Paulsen for sharing GAD65-GFP mice, Professors Lise Jensen and Lars Fugger for the hcrt/orx-eGFP mice, the European Research Counsil for funding and the Osk. Huttunen Foundation for M.M.K.'s studentship.

Author's present address

M. M. Karnani: Department of Biological Sciences, Columbia University, New York, NY, USA.

Ancillary