Non-technical summary Orexin/hypocretin neurons are widely projecting, ‘multi-tasking’ brain cells that promote alertness, reward seeking and feeding. They are vital for stable consciousness in higher mammals. Loss of orexin/hypocretin cells produces narcolepsy. It was originally assumed that orexin/hypocretin neurons are one uniform population of cells, but recent studies hinted that they may be split into subsystems. To explore this, we performed unbiased statistical analysis of electrical properties of orexin/hypocretin cells in combination with 3-D analysis of their shape. Our results pointed to an existence of two subgroups of orexin/hypocretin neurons, that have unique ‘electrical fingerprints’ and distinct ways of receiving information from other neurons.
Abstract Hypothalamic hypocretin/orexin (Hcrt/Orx) neurons recently emerged as critical regulators of sleep–wake cycles, reward seeking and body energy balance. However, at the level of cellular and network properties, it remains unclear whether Hcrt/Orx neurons are one homogeneous population, or whether there are several distinct types of Hcrt/Orx cells. Here, we collated diverse structural and functional information about individual Hcrt/Orx neurons in mouse brain slices, by combining patch-clamp analysis of spike firing, membrane currents and synaptic inputs with confocal imaging of cell shape and subsequent 3-dimensional Sholl analysis of dendritic architecture. Statistical cluster analysis of intrinsic firing properties revealed that Hcrt/Orx neurons fall into two distinct types. These two cell types also differ in the complexity of their dendritic arbour, the strength of AMPA and GABAA receptor-mediated synaptic drive that they receive, and the density of low-threshold, 4-aminopyridine-sensitive, transient K+ current. Our results provide quantitative evidence that, at the cellular level, the mouse Hcrt/Orx system is composed of two classes of neurons with different firing properties, morphologies and synaptic input organization.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The emerging multiplicity of physiological functions of the Hcrt/Orx system led to a recent hypothesis that ‘Hcrt/Orx neurons’ are not one homogeneous neural population, but may comprise several groups of cells with distinct properties (Harris & Aston-Jones, 2006). At the cellular level, properties such as dendritic structure, synaptic input organization and ion channel expression are known to shape the physiological output of neural systems (Llinas, 1988; Mainen & Sejnowski, 1996; Chance et al. 2002). However, little is known about heterogeneities in these fundamental features of the Hcrt/Orx system. For example, the morphology of Hcrt/Orx neurons has not been quantitatively investigated. In turn, although some data exist on the electrical and synaptic properties of Hcrt/Orx neurons (Li et al. 2002; Yamanaka et al. 2003b; Horvath & Gao, 2005), their variation between different types of Hcrt/Orx neurons has not been examined.
In this study, we aimed to gather new information about the basic properties of the Hcrt/Orx cells and use it to explore whether, at the cellular level, these cells are a homogeneous or a heterogeneous population of neurons. To achieve this, we examined structural and functional properties of mouse Hcrt/Orx neurons by combining patch-clamp electrophysiology in living brain slices with fluorescence imaging, 3-dimensional reconstructions of cell morphology, and statistical cluster analysis of firing properties.
Identification of living Hcrt/Orx neurons in situ
All animal procedures were performed in accordance with the UK's Animals (Scientific Procedures) Act 1986, and with the guidelines in Drummond (2009). To identify Hcrt/Orx neurons in brain slices, we used mice expressing eGFP under the control of the prepro-orexin promoter, which results in highly specific labelling of Hcrt/Orx neurons with eGFP (this mouse line has been characterized and validated in previous studies: Yamanaka et al. 2003; Burdakov et al. 2006). Mice were maintained on a 12 h light–dark cycle (lights on at 08.00 h) and had free access to food and water. Coronal or sagittal slices (250 μm thick) containing the lateral hypothalamus were prepared from 14- to 28-day-old mice as previously described (Williams et al. 2008). Briefly, mice were killed by cervical dislocation during the light phase and rapidly decapitated. Brains were quickly removed and immersed in ice-cold artificial cerebrospinal fluid (ACSF). A block of brain tissue was glued to the stage of a Campden Vibroslice brain slicer using cyanoacrylate glue and sliced while immersed in ice-cold ACSF. After a 1 h recovery at 35°C in ACSF, slices were used for recordings.
Data acquisition Orexin-eGFP neurons were visualized in brain slices using an Olympus BX50WI upright microscope equipped with oblique illumination optics, a xenon lamp and filters for visualizing eGFP-containing cells. Somatic recordings were carried out at 37°C using an EPC-10 amplifier controlled by Patchmaster software (HEKA Elektronik, Lambrecht/Pfalz, Germany). Patch pipettes were made from borosilicate glass, and their tip resistances were 4–6 MΩ. Slices were placed in a submerged-type chamber (volume ∼2 ml, solution flow rate 3 ml min−1) and anchored with a nylon string grid stretched over platinum wire. Only cells with access resistances of <20 MΩ were used for analysis. Signals were low-pass filtered at 3 kHz and digitized at 10 kHz.
Chemicals and data analysis ACSF was 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. For standard whole-cell recordings, pipettes were filled with intracellular solution containing (in mm): 130 KCl, 10 Hepes, 0.1 EGTA, 2 MgCl2, 5 K2ATP, 2 NaCl. Inhibitory mPSCs were isolated using 50 μm (2R)-amino-5-phosphonovaleric acid (AP5), 10 μm 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), 10 μm dizocilpine maleate (MK801), and 1 μm tetrodotoxin (TTX) at a holding potential of –60 mV, and verified as GABAA receptor mediated by blockade with 3 μm gabazine. Glutamatergic mPSCs were recorded in the presence of 50 μm picrotoxin (PiTX) and 1 μm TTX at a holding potential of –60 mV, and verified as AMPA receptor-mediated glutamatergic currents by blockade with 10 μm CNQX. Ten millimolar 4-aminopyridine (4-AP) was used to block A-type K+ currents. All drugs were obtained from Sigma-Aldrich or Tocris Bioscience (UK). Chemicals were applied extracellularly by bath superfusion. All drugs were dissolved in water except PiTX, which was dissolved in ethanol (0.1% final concentration).
Averaged data are presented as means ± SEM. Statistical significance was evaluated using Student's t test unless stated otherwise. Cluster analysis is described separately below. Statistical analysis of mPSCs was performed using the Mini Analysis program (Synaptosoft, Dacatur, GA, USA) and Matlab (The Mathworks, Natick, MA, USA). Amplitude thresholds of 10 pA for mEPSCs and 20 pA for mIPSCs, as well as area thresholds of 20 pA ms for mEPSCs and 50 pA ms for mIPSCs, were used to automatically detect events with Mini Analysis. ‘Rise time’ and ‘decay τ’ of mPSCs (Figs 3B and 4B) are, respectively, the time to rise from 20 to 80% of peak amplitude, and the monoexponential time constant of decay. For mEPSCs, 500 events per cell were analysed, and for mIPSCs, 86–110 events per cell were analysed.
Normalized A-type charge density (pC pF-1) was calculated as described in Liss et al. (2001), by obtaining the area under the curve (nA × ms = pC) of the inactivating component by integration, and then dividing this value by the whole-cell capacitance (pF). Input resistances were calculated from slopes of whole-cell current–voltage relationships obtained by performing voltage-clamp ramps from −60 to −100 mV at a rate of 0.1 mV ms−1.
To measure the voltage dependence of A-current activation (Fig. 5B), we made use of the fact that the A-current is activated by depolarizing steps from –90 mV, but not from –40 mV (from graph in Fig. 5C). Thus the A-current was isolated using a two-step voltage protocol that involved subtracting currents activated by depolarization from –40 mV from those activated by depolarization from –90 mV. To determine the voltage dependence of A-current inactivation, we measured peak A-currents evoked by steps to 0 mV from different holding potentials. A-type conductance (GA) was calculated from peak current (IA), as follows: GA= peak IA/(Vtest–Vrev), where Vrev=–100 mV (EK). To obtain the voltage dependence of A-current inactivation (Fig. 5C), we measured peak A-currents evoked by steps to 0 mV from different prepulse potentials (shown on the x-axis of Fig. 5D, prepulse duration was 500 ms). In the graphs showing the voltage dependence of activation and inactivation (Fig. 5B and C, respectively), data were fitted using the sigmoidal function y= 1/(1+exp[(x+k1)/k2]), where y is the normalized conductance (Fig. 5B) or current (Fig. 5C), and x is the test potential (Fig. 5B) or prepulse potential (Fig. 5C). For activation, the best fit (Fig. 5B) was obtained with k1= 20 and k2=–8. For inactivation, the best fit (Fig. 5C) was obtained with k1= 80 and k2= 8.
The time constants of A-current activation and inactivation (τm and τh, respectively, Fig. 5D) were measured by fitting the time course of A-current (measured as 4-AP-sensitive component in Fig. 4A) with Hodgkin–Huxley-based equations (Pulsefit software), where A-current (IA) varies with time t according to IA(t) =m(t)3h(t), where m(t) = 1 − exp(−t/τm) and h(t) =k+ exp(−t/τh). The accuracy of this method was confirmed with simple monoexponential fits of the A-current decay, which generated very similar values of τh (data not shown). The fits of the voltage dependence of τm and τh in Fig. 5D are described by the following functions: τm= 15.5 – 14.2/[1 + exp((V+ 41.5)/–4.8)] and τh= 230 – 180/[1 + exp((V+ 35.9)/–20.2)].
Control experiments To ensure that the two cell types that we describe in Fig. 1 are not an artefact related to the whole-cell recording mode, presynaptic influences, or a developmental stage, we repeated the experiment in Fig. 1 under two additional experimental conditions. Condition 1: as in previous section, but using nystatin-perforated recordings, where pipettes were filled with (in mm): 128 potassium gluconate, 10 KCl, 2 MgCl2, 2 NaCl, 10 Hepes, 300μg ml−1 nystatin in Pluronic F-127 and DMSO (0.05% final concentration), pH 7.3 with KOH; performed in the presence of synaptic blockers: 50 μm AP5, 10 μm CNQX, 50 μm PiTX, 3 μm strychnine. Condition 2: as in previous section (whole-cell recordings), but using adult (14-week-old) mice. The two types of Hcrt/Orx cells were clearly observed under both of these additional experimental conditions (n > 10 cells/type/condition).
We determined the number of distinct clusters that best separate the data using Bayesian model selection. This involved calculating the model evidence (the probability of the observed data under the proposed model) for each considered number of clusters. We chose to consider a maximum of six clusters. Following Corduneanu & Bishop (2001), we used the variational Bayes algorithm, implemented in the Infer.NET software framework (Infer.NET 2.4, Microsoft Research Cambridge, 2010; http://research.microsoft.com/infernet), to find the approximate model evidence for each number of clusters. Our model was a mixture of Gaussian distributions with diagonal covariance structure, where the posterior probabilities over the parameters for each cluster are learnt as part of the variational Bayes approximation. Since variational approximations are known to be sensitive to initialization, 10 repeats were run for each number of clusters with different random initializations, and the maximum model evidence value recorded. This is valid because the approximate model evidence calculated by variational Bayes is in fact a lower bound on the true model evidence, so the maximum value of the 10 repeats is the most accurate. We considered clustering based on two firing-related variables described in Results and shown in Fig. 1A. The black line in Fig. 1C shows the decision boundary dividing the two clusters. Points to the left of the decision boundary are assigned to the left cluster, and vice versa. Points on the line are assigned equally to both clusters. The boundary line is given by the equation:
where y is the spike ratio, x is the time to 1st spike, l1=–5 × 10−5, l2= 3.1, m1=–12.6, m2=–0.92, and r= 2.12. The coefficients l, m and r are derived from the cluster analysis and are related to the cluster means, variances and weights estimated by the variational Bayes algorithm. The equation was found by equating the probability density functions of the two clusters and solving for y. The contours in Fig. 1C show the probability densities of each cluster. Figure 1B shows the posterior probabilities of different numbers of clusters, given the two-variable representation of the samples. The posterior probabilities are calculated using Bayes's rule, assuming each number of clusters is equally likely a priori. The two-cluster assumption clearly produced the highest posterior probability (0.741, Fig. 1B), showing that this assumption best explains the observed data.
Staining Hcrt/Orx neurons were filled with biocytin by adding it to the pipette solution (0.5% biocytin, Tocris) and keeping the cells in the whole-cell mode for 20 min. After recovery for at least 20 min, the tissue was fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) overnight. Three repeated washing steps in PBS were followed by 1h blockade of unspecific binding and permeabilization using 1% bovine serum albumin (BSA) and 0.3% TritonX-100 in PBS. Streptavidin linked to Alexa 555 (Invitrogen) was diluted 1:500 in blocking solution. Slices were incubated overnight at 4°C on an orbital shaker in 300 μl of the streptavidin solution per slice. After washing 3 times in PBS the slices were mounted in Vectashield (Vector Laboratories, Inc., Burlingame, CA, USA). PBS contained (in mm): 2.7 KCl, 1.5 KH2PO4, 137 NaCl, 8 Na2HPO4, pH 7.4.
Image acquisition Images were taken using an Olympus BX61WI confocal miroscope (Olympus FluoView v 2.1b software) in a dynamic range of 16 bit using a 25× water immersion objective (NA 1.05, Olympus). Alexa 555 was excited with a diode-pumped solid-state (DPSS) laser at 559 nm, and fluorescence emission collected at 570–670 nm using a spectral detector (Olympus). With the objective we used, the full dendritic field was visible within a single field of view; the section was xy-scanned at 0.3 μm pixel size then z-scanned at 1 μm intervals as deep as the deepest dendrite. Kalman filtering (×2 per frame) was applied during acquisition to improve image clarity.
3-D reconstructions and analysis Initial image processing was performed with ImageJ (NIH ImageJ v 1.41a). Image stacks were created in Olympus FluoView image acquisition software at 16 bit, and then converted using ImageJ to 8-bit in order to reduce image file-size and processing time. Although images were taken at the Nyquist limit, we found that the structures to be traced were not very densely compacted or highly complicated, and a lower resolution was sufficient for reconstruction. The 3-D images of soma and dendrites were then traced in semiautomated mode with Imaris (Bitplane, Zurich, Switzerland; v. 7.0, including FilamentTracer; for detailed description see http://www.bitplane.com/go/products/filamenttracer). Briefly, after manual selection of the soma and dendrite endpoints, we used the automatic tracing mode of Imaris to trace the dendrites and quantify their properties. The reconstructed length of each dendrite was verified visually against the raw image to ensure that dendrites are correctly assigned. Dendrite straightness (Table 1) was defined as the ratio between dendrite length and radial distance between two branch points. Soma sphericity (Table 1) was defined as the ratio of soma surface area to the surface area of a sphere with same volume as the soma. Axons, which were readily identified by their characteristic thin and constant diameter (e.g. see Fig. 2B), were excluded from the reconstructions and analysis. The ‘concentric sphere’ method of Sholl (1953) was used to analyse the branching patterns of dendritic trees. Briefly, concentric spheres of a constant interval of 1 μm, with their centre lying at the soma, were used to count the number of dendritic intersections through each sphere (Fig. 2C). A modified Sholl analysis was performed using straightened dendritic trees, thereby plotting the number of dendritic intersections versus the dendrite arc length from the soma (Fig. 2D).
Table 1. Properties of the two types of orexin neurons
Total dendrite length
Dendrite segment length
No of dendrite branches
Soma surface area
Spontaneous firing rate
AP half width
mEPSC rise time
mEPSC decay τ
mEPSC interevent interval
mIPSC rise time
mIPSC decay τ
mIPSC interevent interval
Cells from sagittal slices
Cells from coronal cells
Soma depth in slice
No of dendritic amputations
Control experiments To ensure that the morphological differences we describe are not an artefact resulting from the choice of slicing plane, we analysed the shape of Hcrt/Orx cells recorded and filled in both sagittal and coronal brain slices (see Table 1); the sagittal and coronal data were pooled in plots in Fig. 2C and D. We also measured the number of dendritic amputations by slice surface and the depths of soma in the slice; neither of these parameters was different between cell types (see Table 1), suggesting that the differences in dendritic architectures in the two cell types are not artefacts of variation in these parameters.
Statistical evidence and definitions for two populations of Hcrt/Orx cells
To probe for presence of functional subgroups in the Hcrt/Orx system, we recorded firing properties using a simple, widely used protocol involving an injection of hyperpolarizing current (Fig. 1A, Minami et al. 1986; Miki et al. 2001; Burdakov & Ashcroft, 2002), and then applied statistical cluster analysis to the data. To create the same biophysical conditions for all cells, we used zero holding current and then applied current injections that hyperpolarized cells to –80 mV (Fig. 1A). We quantified two easily measurable parameters, ‘spike ratio’ (number of spikes in 500 ms after injection divided by that in 500 ms before injection), and ‘time to first spike’ (the time from 50 ms after the end of injection to occurrence of first spike). Plotting ‘spike ratio’ against ‘time to first spike’ revealed two distinct cell clusters (Fig. 1B and C, n= 34 cells), as confirmed by posterior probability analysis of cluster number showing the highest probability peak at 2 (Fig. 1B).
We called the two clusters ‘H-cells’ and ‘D-cells’, because, on visual inspection, the two cell types were clearly different in that they exhibited hyperpolarizing and depolarizing post-inhibitory rebounds, respectively (Fig. 1D). We did not find any clear differences in anatomical distribution of the two cell types in mouse brain slices; both cell types were intermixed in all regions of the Hcrt/Orx field and were often found near each other in the same slice. The resting firing rates were also not different in the two cell types (see Table 1 for all parameters that we quantified in addition to those shown in figures).
Based on our new statistical analysis, we can objectively and quantitatively define the difference between H-cells and D-cells, as those located to the right and to the left, respectively, of the cluster separation boundary (see Fig. 1C, equation of the boundary line is given in Methods). In the subsequent Result sections, we used the definitions of H and D Hcrt/Orx cells established here to examine the functional and morphological properties of Hcrt/Orx neurons in greater detail.
Little is known about the morphology of Hcrt/Orx neurons. To explore the morphological properties of Hcrt/Orx cell subtypes, we filled 18 H-cells and 23 D-cells with biocytin and examined their shapes using confocal imaging of streptavidin-linked Alexa-555, followed by 3-dimensional (3-D) reconstructions (Fig. 2A–D, see Methods). Several morphological parameters were chosen for quantification (Table 1). The dendrite diameter, dendrite straightness, cell volume and total dendritic length were not significantly different between D-cells and H-cells (Table 1). However, the number of dendrite branch points, soma volume, and soma area were significantly larger in H-cells, which accordingly also showed a greater membrane capacitance, while soma sphericity was greater in D-cells (see Table 1 for a list of all parameters that were quantified).
We also quantified the dendritic complexity of Hcrt/Orx cells by looking at the intersections between dendrites and the surface of an imaginary sphere whose centre is the cell body (Sholl analysis; Sholl, 1953). Plotting the number of intersections against the distance from the soma revealed that H-cells have significantly more dendrites near the soma than D-cells (Fig. 2C). We also plotted the number of dendritic branches against arc length from the soma, which, unlike the traditional Sholl analysis, takes into account the dendrite straightness (Fig. 2D). As expected from the similar dendrite straightness in the two cell types (Table 1), both plots produced similar results, showing that H-cells have significantly more dendritic branches in the proximal region (∼150 μm from soma) than D-cells (Fig. 2C and D).
AMPA and GABAA receptor-mediated synaptic inputs
To examine the differences in the activity of presynaptic terminals that control Hcrt/Orx cells, we looked at the amplitude and frequency of miniature postsynaptic currents (mPSCs). GABA and glutamate-mediated mPSCs were isolated using standard pharmacological tools and recorded in voltage-clamp at a holding potential of –60 mV (Figs 3 and 4, see Methods). We also quantified the input resistance, which was similar in the two cell types (Table 1), suggesting that any differences in mPSC amplitude were not due to differences in membrane length constant (see Discussion).
For glutamate mPSCs, we observed significantly larger amplitudes and frequencies in H-type than D-type cells (Fig. 3C and D, P < 0.001 in K-S test). The time constants of rise and decay of glutamate mPSCs were similar in the two cells types (Fig. 3B, P > 0.1 in unpaired t test). Of note, the mPSC decay times were below 10 ms, confirming that in both cell types, the glutamate mPSCs that we analysed were mediated by AMPA rather than NMDA receptors (Dingledine et al. 1999), as expected from the negative holding potential (−60 mV) and the concentration of extracellular Mg2+ used in our experiments. The AMPA receptor identity of glutamatergic mPSCs was also directly confirmed by complete blockade with 10 μm CNQX (n= 6).
For GABA mPSCs, which were confirmed as mediated by GABAA receptors by complete blockade with 3 μm gabazine (n= 7), we observed similar frequencies (Fig. 4D, P > 0.05 in K-S test) but significantly different amplitudes (H-type > D-type, Fig. 4C, P < 0.001 in K-S test). The rise and decay time constants of GABA mPSCs were similar in the two cells types (Fig. 4B, P > 0.1 by unpaired t test). These data suggest that there are functional differences in both GABA- and glutamate-mediated synaptic inputs in the two types of Hcrt/Orx cells (see Discussion).
We also examined the frequencies of spontaneous PSCs, which were recorded at –60 mV with high-chloride pipette solutions (and thus contained both glutamate and GABA components), without TTX or synaptic blockers. Spontaneous PSC frequency was significantly smaller in D-cells compared to H-cells (D-cells: 15.2 ± 1.7 Hz, n= 15; H-cells: 22.3 ± 3.4 Hz, n= 10; P < 0.05). However, no significant correlation between spontaneous PSC frequency and the total dendrite length or number of dendritic branchpoints was found (P > 0.05 for both, Pearson correlation test).
Presence of low-threshold A-type K+ current
Previous work showed that at least some Hcrt/Orx cells express low-threshold A-type K+ channels (Burdakov et al. 2004). To explore possible differences in the expression of functional A-type channels among Hcrt/Orx neurons, we looked at relative magnitude of whole-cell A-currents. We isolated the A-current pharmacologically by blockade with 10 mm 4-aminopyridine (4-AP). This drug eliminated the large transient current evoked by a step from –90 to –40 mV in H-type cells (Fig. 5A, n= 7), but did not have a significant effect on currents elicited by this change in membrane potential in D-type cells (Fig. 5A, n= 7). Quantification of 4-AP-sensitive charge densities revealed that, at the test potential of –40 mV, H-type, but not D-type, cells had significant 4-AP-sensitive currents (Fig. 5A, P < 0.001 for H-cells, P > 0.1 for D-cells, n= 7 for both). These data imply that H-type cells, but not D-type cells, express a low-activation-threshold, 4-AP-sensitive A-current.
To characterize the properties of the A-current displayed by H-type cells, we measured the voltage dependence of the extent and the speed of A-current activation and inactivation in standard ways (see Methods, and also Liss et al. 2001, Burdakov & Ashcroft, 2002). The A-current activated between –60 and –50 mV, with half-maximal activation observed at –21 ± 3 mV (n= 5, Fig. 5B). Half-maximal inactivation of the A-current occurred at −80 ± 2 mV, and the current was completely inactivated at potentials positive to −40 mV (Fig. 5C, n= 5). The rates of both activation (τm) and inactivation (τh) of the A-current accelerated with depolarization (Fig. 5D, n= 7), but this effect was more pronounced for activation than for inactivation (Fig. 5D), as reported for other neurons (Liss et al. 2001).
Last, we examined the effect of A-current block on rebound firing and action potential half-width in H- and D-cells. 4-AP (10 mm) significantly reduced the post-rebound ‘time to first spike’ (defined in Fig. 1A) in H- but not D-cells (Fig. 6A; H-cells, control 420 ± 50 ms, 4-AP 90 ± 25 ms, n= 5, P < 0.01 by paired t test; D-cells, control 50 ± 4 ms, 4-AP 54 ± 9 ms, n= 6, P > 0.5 by paired t test). 4-AP (10 mm) also significantly increased the half-width of spontaneously-generated spikes in H- but not D-cells (Fig. 6B, H-cells, 20 ± 3.5% increase, P < 0.005 by paired t test; D-cells, 4.9 ± 4.5% increase, P > 0.1 by paired t test).
Our study provides unbiased statistical evidence for, and quantitative definition of, two distinct types of Hcrt/Orx neurons in the mouse brain. This conclusion is based on formal cluster analysis of firing properties, as well as differences in dendritic structure, synaptic input organization, and ionic currents.
Although hypothalamic neurons have been previously classified in the literature based on differences in their firing responses to inhibitory currents (Minami et al. 1986; Miki et al. 2001; Burdakov & Ashcroft, 2002; Williams et al. 2008), to the best of our knowledge, the subjective definitions of these differences have not been systematically validated with formal statistical analysis. Our study replaces subjective criteria with unambiguous quantitative standards. This is advantageous because it allows different groups to compare data from the same cell types. Another advantage of our proposed classification is that the simple measurements required can be obtained either in current-clamp (if classification is based on firing, Fig. 1), or in voltage-clamp (if classification is based on A-current density, Fig. 5). In principle, the statistical cluster analysis we used can be applied to any electrophysiological measurements. However, the measurements shown in Fig. 1A are advantageous because they minimize the effects of variations in spontaneous firing rates and input resistances, by measuring firing ratios within each cell and hyperpolarizing all cells to the same potential. We thus hope that this method can be used widely to compare different types of neurons.
Morphologically, we found that H-cells have a more complex dendritic architecture than D-cells. However, in both cell types, the dendritic trees had relatively few branch points. This simple dendritic structure appears to be broadly similar to other types of hypothalamic projection neurons (Armstrong, 1995; Stern, 2001), and to midbrain dopamine neurons (Grace & Onn, 1989). In contrast, dendritic trees of Hcrt/Orx cells are strikingly different from those of, for example, cortical and cerebellar projection neurons, which have much more complex dendrites (Roth & Hausser, 2001; Wang et al. 2010).
Electrophysiologically, H-cells had larger amplitudes and frequencies of AMPA receptor-mediated mPSCs than D-cells. H-cells also had larger amplitudes of GABAA receptor-mediated mPSCs, whereas the GABAA mPSC frequency was similar in the two cell types. Overall, H-cells thus receive greater amounts of both synaptic excitation and inhibition than D-cells. The amplitude of mPSCs is usually interpreted as a measure of postsynaptic receptor density, but it also depends on dendritic filtering properties, which may cause distal synaptic inputs to appear smaller in amplitude. However, in our dataset, mPSCs from H- and D-cells had similar rise times (Figs 3, 4), and the membrane resistances of the two cell types were also similar (Table 1), arguing against major differences in dendritic filtering in the two cell types. Rather, our data suggest that, on average, H-cells have more GABAA and AMPA receptors per synapse than D-cells. In turn, the greater frequency of glutamate mPSCs in H-cells suggests that glutamate synapses on H-cells either have a higher release probability per synapse, and/or are more numerous, than glutamate synapses on D-cells.
In terms of intrinsic ionic conductances, H-cells, but not D-cells, express high levels of 4-AP-sensitive, low-activation threshold A-current. The A-current causes, in H-cells, a hyperpolarized post-inhibitory rebound (Fig. 6A) and a narrowing of spike half-width (Fig. 6B). The latter effect of 4-AP sensitive A-current was previously reported in other neurons (e.g. Reyes et al. 1994; Christie & Westbrook, 2003) and may have important functional implications, since spike width regulates the efficiency of transmitter release (Wang et al. 2009). Methodologically, poor space-clamp in neurons may make A-current appear artefactually smaller. However, this problem would be worse in larger cells. In contrast, we see larger A-current in the larger H-cells, thus making it unlikely that space-clamp problems explain the differences in A-current size in H- and D-cells. The similarity in kinetics of AMPA and GABAA currents also argues for similar efficiency of voltage-clamp in the two cell types.
Although our study identified several fundamental heterogeneities in the Hcrt/Orx system, at present we can only speculate about their physiological implications. For example, since H-cells have more complex dendrites and stronger synapses than D-cells, they could be a preferential source of neural input to the Hcrt/Orx system. H-cells may also process these inputs differently from D-cells, since they express greater densities of low-threshold A-type current, which is known to be involved in dendritic information processing (Kim et al. 2007). Of note, H- and D-cells are also likely to be differentially involved in physiology and pathology of the Hcrt/Orx system. In particular, our previous work also suggests that glucose has a longer-lasting inhibitory effect on H-like than on D-like Hcrt/Orx cells, although the physiological significance of this remains to be determined (Williams et al. 2008; Karnani & Burdakov 2011). Furthermore, our recent work suggests that the two cell types are differentially affected in a mouse model of Huntington's disease (Williams et al. 2011). To fully understand the relative functional roles of the two cell types, ultimately it will be also important to examine whether they differentially project to the numerous CNS regions innervated by Hcrt/Orx cells (Peyron et al. 1998; van den Pol, 1999).
In conclusion, our study highlights several previously undefined features of Hcrt/Orx neurons, providing a new functional and anatomical framework for further understanding the physiological and pathological impact of these brain cells.
C.S. performed most of the experiments and data analysis, and contributed to experimental and analytical design. A.V. performed experiments shown in Fig. 5A. D.K. performed the cluster analysis. M.M.K. contributed data shown in Figs 3 and 4. D.B. conceived and designed the study and wrote the paper. All authors approved the final version for publication. The work was carried out at the University of Cambridge, UK.
This work is funded by a European Research Council (FP7) grant to D.B. We thank Dr Antonio Gonzalez for critical reading of the manuscript and many constructive suggestions.