Neurochemical Heterogeneity Among Lateral Hypothalamic Hypocretin/Orexin and Melanin-Concentrating Hormone Neurons Identified Through Single-Cell Gene Expression Analysis

Abstract The lateral hypothalamic area (LHA) lies at the intersection of multiple neural and humoral systems and orchestrates fundamental aspects of behavior. Two neuronal cell types found in the LHA are defined by their expression of hypocretin/orexin (Hcrt/Ox) and melanin-concentrating hormone (MCH) and are both important regulators of arousal, feeding, and metabolism. Conflicting evidence suggests that these cell populations have a more complex signaling repertoire than previously appreciated, particularly in regard to their coexpression of other neuropeptides and the machinery for the synthesis and release of GABA and glutamate. Here, we undertook a single-cell expression profiling approach to decipher the neurochemical phenotype, and heterogeneity therein, of Hcrt/Ox and MCH neurons. In transgenic mouse lines, we used single-cell quantitative polymerase chain reaction (qPCR) to quantify the expression of 48 key genes, which include neuropeptides, fast neurotransmitter components, and other key markers, which revealed unexpected neurochemical diversity. We found that single MCH and Hcrt/Ox neurons express transcripts for multiple neuropeptides and markers of both excitatory and inhibitory fast neurotransmission. Virtually all MCH and approximately half of the Hcrt/Ox neurons sampled express both the machinery for glutamate release and GABA synthesis in the absence of a vesicular GABA release pathway. Furthermore, we found that this profile is characteristic of a subpopulation of LHA glutamatergic neurons but contrasts with a broad population of LHA GABAergic neurons. Identifying the neurochemical diversity of Hcrt/Ox and MCH neurons will further our understanding of how these populations modulate postsynaptic excitability through multiple signaling mechanisms and coordinate diverse behavioral outputs.


Introduction
The lateral hypothalamic area (LHA) is a powerful modulator of behavioral state through its ability to integrate diverse central and peripheral signals and orchestrate adaptive responses to challenges in the environment. Such functions include the organization of sleep-wake states, feeding, energy balance, and motivated behavior (Berthoud and Münzberg, 2011;Hurley and Johnson, 2014;Brown et al., 2015;Bonnavion et al., 2016;Stuber and Wise, 2016), the dysregulation of which are cardinal symptoms of both neuropsychiatric illnesses and metabolic disorders (Wulff et al., 2010;Spiegelhalder et al., 2013;. The LHA thus plays a vital role in coordinating physiologic and behavioral homeostasis to ensure survival (Sternson, 2013).
Underlying the diverse functions of the LHA is a heterogeneous population of neurons, of which only a few have been definitively identified and characterized. Hypocretin/ orexin (Hcrt/Ox) and melanin-concentrating hormone (MCH) neurons make up two neuropeptidergic neuronal populations which, despite comprising a relatively small proportion of cells within the LHA, have a profound impact on brain function. Both populations regulate sleep-wake states, but have also been implicated in feeding, metabolism, stress and reward-related behaviors (Burt et al., 2011;Barson et al., 2013;Jones and Hassani, 2013;Brown et al., 2015;Konadhode et al., 2015;Bonnavion et al., 2016;Stuber and Wise, 2016) Hcrt/Ox neurons project widely, serve as key integrators of a diverse set of inputs from both the CNS and periphery, and exert homeostatic control over arousal, feeding, reward and stress. Furthermore, Hcrt/Ox signaling has been implicated in the stabilization of sleep-wake states and its dysfunction has been directly linked to narcolepsy in animal models and humans (Alexandre et al., 2013;Giardino and de Lecea, 2014;Mahler et al., 2014;. Importantly, the electrical activity of Hcrt/Ox neurons is directly correlated with behavioral state, firing at high frequencies during high arousal states and at low frequencies or not at all during low arousal states and sleep (Lee, 2005;Mileykovskiy et al., 2005;Takahashi et al., 2008;Hassani et al., 2009Hassani et al., , 2016González et al., 2016). MCH neurons, in turn, are also unique to the LHA and are implicated in regulating sleepwake states, as well as feeding and metabolism (Pissios et al., 2006;Monti et al., 2013). Optogenetic experiments show that, whereas activation of Hcrt/Ox neurons promotes wakefulness and behavioral arousal, stimulation of MCH neurons promotes sleep (Adamantidis et al., 2007;Konadhode et al., 2013;Tsunematsu et al., 2014;Vetrivelan et al., 2016).
Despite the crucial role of these neurons in maintaining behavioral homeostasis, outstanding questions remain concerning the neurochemical diversity of Hcrt/Ox and MCH neurons, particularly in regard to their coexpression of other neuropeptides and necessary components for the synthesis and release of the fast amino acid neurotransmitters GABA and glutamate. Anatomic data reveal that both populations may be neurochemically heterogeneous in terms of their neuropeptide coexpression and markers for one or more fast neurotransmitters, including machinery for GABA synthesis (GAD65/67), vesicular GABA release (VGAT) and vesicular glutamate release (VGLUT1, VGLUT2, or VGLUT3;Brown et al., 2015;Bonnavion et al., 2016;Stuber and Wise, 2016). A systematic assessment of the neurochemical phenotype of Hcrt/Ox and MCH neurons is lacking owing to the challenge of profiling the expression of many genes in cleanly isolated neurons, which are scattered throughout the LHA and intermingled with other cell populations. Here, we optimized a robust procedure for isolating single, transgenically-labeled Hcrt/Ox and MCH neurons and employed a single-cell quantitative polymerase chain reaction (qPCR) profiling approach to discern their expression of a panel of 48 key genes that encode neuropeptides, fast neurotransmitter components and other key markers. The goal of our study is, through a single-cell gene expression profiling approach, to more fully resolve the neurochemical profile of Hcrt/Ox and MCH neurons, which would shed light on the full extent of their signaling repertoire at postsynaptic targets and inform their role in driving complex, innate behavior.

Ethics statement
All experiments were performed in accordance with the guidelines described in the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of the University of Connecticut.

Animals
Transgenic mice used to identify Hcrt/Ox neurons expressed enhanced green fluorescent protein (EGFP), under the control of the human prepro-orexin promoter Yamanaka et al., 2003), referred to here as Ox-EGFP. To identify MCH neurons, we used then photographed and the borders of the tissue punch were subsequently mapped onto mouse brain atlas images (Paxinos and Franklin, 2012). Microdissected tissue punches were kept in sucrose dissociation solution (with trypsin inhibitor/BSA) on ice until trituration. Immediately before dissociation, tissue punches were incubated for ϳ10 min in a 37°C water bath, then triturated with a series of small bore fire-polished glass Pasteur pipettes in a volume of 400 -600 l trypsin inhibitor/BSA sucrose dissociation solution. Single-cell suspensions were passed through 60 m nylon mesh filters to remove any cell aggregates, and kept on ice until fluorescence-activated cell sorting (FACS). to obtain sufficient number of cells for FACS, punches from multiple animals were pooled together for each FACS collection. Ox-EGFP and Pmch-Cre;EYFP mice were FACS sorted in two collections each (two to three mice per collection) while Vgat-Cre;EYFP and Vglut2-Cre;EYFP were sorted in single collections (two mice per collection).

Fluorescence-activated cell sorting (FACS)
EGFPϩ or EYFPϩ neurons were sorted on a BD FACSAria II Cell Sorter (UConn Flow Cytometry Facility, Storrs CT, RRID: SCR_012341) equipped with a sapphire 488 nm excitation laser (Coherent). Cells were selected based on scatter and for singlets, then gated on the presence of EGFP/EYFP fluorescence (PE-A vs GFP-A) and sorted in single-cell precision mode with a 10 K threshold into a sterile 96-well plate containing 2.5 l lysis buffer [0.5% NP-40, 1 U/l RNAsinϩ (Promega), 0.25ϫ pooled TaqMan assays], snap frozen on dry ice, and stored at Ϫ80°C until processing for single-cell qPCR. To control for free-RNA contamination in FACS-sorted droplets, calibrite allophycocyanin (APC) beads (Beckton Dickenson) were added to all LHA single-cell suspensions and single beads were sorted based on red/orange fluorescence via FACS using the same gating parameters as single-cells. Beads were snap frozen and underwent qPCR alongside single cells. In initial experiments, single EGFPϩ cells were FACS-sorted using the same gating parameters into flat-bottomed wells of a Terasaki plate (Yaron et al., 2014;Tasic et al., 2016) containing 5 l of 10 mg/ml trypsin inhibitor/BSA in dissociation sucrose and visualized under a fluorescence microscope.

qPCR data analysis
Raw cycle threshold (Ct) values were obtained from the Fluidigm Biomark software and inverted (35-Ct) to generate a log 2 -based scale for gene expression analysis. To eliminate cells with low/no cDNA yield, we filtered our dataset to include only cells within the 95% confidence interval (CI) for Gapdh expression after removing cells absent for the transcript. Hierarchical clustering was performed using Ward's method with complete linkage (Ward, 1963). For principle component analysis (PCA), gene expression was z score normalized and processed using the princomp function in R. To examine potential subclusters and/or batch effects, we used both multiple hypothesis testing analysis using custom routines and the fisher.test function in R as well as PCA analysis using the princomp function in R. To quantitatively compare gene expression between Hcrt/Ox and MCH neurons, we performed multiple hypothesis testing on the 48 genes using Fisher's exact test (Agresti, 1992) to report adjusted p values, with the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) to control the false discovery rate (FDR) at 5%. All statistical analyses were performed using R (The R Project for Statistical Computing; www.rproject.org, RRID: SCR_001905).

Statistical power analysis
We performed power analysis to assess whether the numbers of neurons used in this study are adequate to achieve sufficient statistical power in detecting differential gene expression. To this end, we used a simulation in which the sample sizes are fixed at the same values of the real data (Hcrt/Ox: 69; MCH: 89), and the true difference between the two probabilities of expression is set to various levels (0%, 15%, 25%, and 35%). With each simulation, presence/absence data are randomly generated, for which the Fisher's exact test (Agresti, 1992) was performed at 5% significance level. The simulations were repeated 1000 times under each setting of true probabilities and effect size, and the proportion of times that the test is rejected is then an estimate of the corresponding power. Power analysis via simulation was performed using custom routines in R.

Fluorescence in situ hybridization (FISH)
To prepare tissue sections for FISH, male juvenile (postnatal days P21-P24) wild type C57BL/6 mice were anesthetized with isoflurane, decapitated, and brains were dissected out into ice-cold sucrose. Brains were rapidly frozen on dry ice, embedded in OCT compound and cryosectioned at a thickness of 14 m onto SuperFrost Plus microscope slides. Sections were fixed with 4% paraformaldehyde (PFA) at 4°C for 15 min, and then dehydrated in 50%, 70%, and 100% ethanol. RNAscope 2.5 Assay (Advanced Cell Diagnostics, ACD, RRID: SCR_012481) was used for all FISH experiments according to manufacturer's protocols (Wang et al., 2012). All RNAscope FISH probes were designed and validated by ACD.

Imaging and image quantification of FISH data
Confocal images of FISH experiments were obtained using a Leica TSC Sp8 and confocal image files (lif) containing image stacks were loaded into ImageJ (version 2.0.0, NIH, RRID: SCR_003070) and processed to analyze percentage colocalization between mRNA transcripts for various neuropeptide or neurotransmitter components and Hcrt or Pmch. Cells were marked and categorized using the ImageJ cell counter plugin. Cells showing expression in the orange (TRITC) channel for one of the following mRNA: Cartpt, Nucb2, Pdyn, Penk, Gad1, Slc32a1, or Slc17a6 were counted and marked. Expression was denoted as binary yes/no dependent on the fulfillment of a defined criteria; the presence of at least five punctate fluorescent dots accompanying a nucleus labeled by 4',6-diamidino-2-phenylindole (DAPI; Vector Laboratories, RRID: AB_2336788). Subsequently, cells showing expression in the green (FITC) channel (for either Hcrt or Pmch) were similarly marked. Cells that were double marked through this process were identified as cells exhibiting colocalization. This process was repeated for multiple lif files for each FISH experiment, and the total number of colocalized cells versus the total number of Hcrt-or Pmch-expressing cells was used to create pie charts displaying percentage colocalization, where n values describe the total number of Hcrt-or Pmchexpressing cells analyzed.

Brain slice preparation for patch-clamp recording and manual harvest
Coronal brain slices containing the LHA were prepared from male juvenile (P21-P23) Ox-EGFP and Pmch-Cre; EYFP mice. Mice were deeply anesthetized with ketamine/xylazine before being transcardially perfused with ice-cold high-sucrose cutting solution before slicing using a vibrating microtome as described previously. Hypothalamic slices containing the LHA (225 m thick) were then placed in a recovery chamber containing artificial CSF (ACSF) consisting of 125 mM NaCl, 2.5 mM KCl, 25 mM NaHCO 3 , 1.25 mM Na 2 PO 4 , 25 mM glucose, 2 mM CaCl 2 , and 1 mM MgCl 2 , saturated with 95% O 2 /5% CO 2 , and allowed to recover for 45 min at 34 -35°C and at least 30 min at RT before recording. Cells were visualized with an upright Olympus BX51 microscope (Olympus America) with infrared differential interference contrast and identified by EGFP expression with epifluorescence illumination. The slice in the recording chamber was superfused with bubbled ACSF at physiological temperature (34 -35°C), at which all recordings were performed. All recordings were acquired using a Multiclamp 700B amplifier (Molecular Devices). Whole-cell patch-clamp recordings from EGFPϩ Hcrt/Ox neurons were conducted using thinwalled borosilicate glass pipettes (World Precision Instruments) with R pipette values of 2-6 M⍀. Pipettes were backfilled with 1 l RNase-free internal solution containing 130 mM K-gluconate, 10 mM HEPES, 0.1 mM EGTA, 10 mM NaCl, and 2 mM MgCl 2 with 20 g/ml glycogen and 1 U/l RNase inhibitor made with RNase-free H 2 O. All recordings were conducted using a Multiclamp 700B amplifier (Molecular Devices), digitized at 10 kHz and filtered at 2 kHz using a Digidata 1440 digitizer (Molecular Devices) and were acquired using pClamp10 software (Molecular Devices, RRID:SCR_011323). Data were analyzed using ClampFit (Molecular Devices) and custom procedures in IgorPro (Wavemetrics, RRID: SCR_00325). Only cells with an R access below 15 M⍀ were accepted. In current-clamp mode, neurons were subjected to hyperpolarizing and depolarizing step pulses (Ϯ40 pA). RNA was collected at the end of recording (typically 3-5 min after initial break-in) by applying light suction until the cell cytoplasm had been aspirated (Citri et al., 2011;Cadwell et al., 2015;Fuzik et al., 2015). The contents of the pipette were expelled using positive pressure into an RNase free PCR tube containing lysis buffer (described above), snap frozen on dry ice and stored at Ϫ80°C until further processing for single-cell qPCR.

Pmch-Cre;EYFP transgenic mice
To identify labeled Hcrt/Ox and MCH neurons for FACS and for recording in brain slices, we used Ox-EGFP and Pmch-Cre;EYFP mice. In both cases, we quantitatively evaluated the fidelity with which each transgenic mouse line expresses EGFP or EYFP in Hcrt/Ox and MCH neurons, respectively, through IHC. To assess specificity of Ox-EGFP mice, we quantified the number of EGFPϩ cells which were Hcrt/Ox-immunoreactive (IR) and found the specificity to be ϳ98% (n ϭ 3 mice). To assess penetrance, we quantified the number of Hcrt/Ox-IR cells that were EGFPϩ and found the penetrance to be ϳ49% (Fig.  1C). We performed a similar analysis using Pmch-Cre; EYFP mice ( Fig. 1B) and found the specificity of this transgenic line to be ϳ77% and the penetrance to be ϳ81% (n ϭ 3 mice; Fig. 1C). Thus, we found the Ox-EGFP transgenic mouse to label Hcrt/Ox neurons with high specificity but relatively low penetrance, broadly consis- tent with the description of the specificity and penetrance of previously described Ox-nLacZ transgenic mice that share the same promoter . In turn, Pmch-Cre;EYFP mice were effective in selectively labeling Pmch-expressing neurons with reasonably high specificity and penetrance. Although we found penetrance to be in line with the original description of this mouse line (Kong et al., 2010), specificity was less than described. We used both transgenic lines in our subsequent singlecell gene expression profiling and electrophysiological analysis.

Precise and reproducible microdissection and FACS of Hcrt/Ox and MCH neurons from brain slices
To isolate and capture Hcrt/Ox and MCH neurons for single-cell expression profiling, we optimized a method for the precise and reproducible microdissection of the LHA from coronal brain slices from juvenile mice (P21-P23) and subsequent isolation of single neurons for qPCR via FACS ( Fig. 2A). In our microdissection, we used a 1 mm diameter circular tissue punch to dissect the LHA from 225-m-thick coronal brain slices containing the LHA from Ox-EGFP and Pmch-Cre;EYFP mice. Following enzyme treatment of whole brain slices, the LHA was visually identified under a dissecting microscope using anatomic landmarks. In all cases, bilateral tissue punches from each slice were subsequently imaged and transposed onto a schematic representation of each of the three representative atlas images to record the accuracy and reproducibility of each microdissection (Fig. 2B,C). To ensure accuracy of the microdissection in capturing the highest concentration of labeled Hcrt/Ox (Fig. 2B) and MCH ( Fig. 2C) neurons, representative tissue punches were mounted and imaged. Finally, the tissue punches from each preparation were triturated to yield suspensions of healthy LHA neurons with intact fluorescence, shown in representative images (Fig. 2B,C).
Following gentle trituration and a filtration step to remove aggregates, single-cell suspensions were loaded onto a BD FACSAria II Cell Sorter. To remove debris and minimize the chance of collecting more than a single EGFPϩ neuron, a 10K threshold was applied to maximally filter debris in the media. Cells were selected based on scatter, and for singlets, gated on the presence of EGFP/ EYFP fluorescence (PE-A vs GFP-A) and sorted into 96-well PCR plates in single-cell purity mode. A representative FACS run, sorting Ox-EGFP neurons under the described conditions, is shown (Fig. 2D). To further test the reliability of our FACS configuration to sort only a single EGFPϩ neuron in each well, we performed control experiments in which we visualized sorted single cells in 72-well Terasaki plates. As shown in a representative example (Fig. 2E), the flat, optically clear bottoms of the wells of Terasaki plates allowed us to visualize the entire field of each well. In initial experiments using Ox-EGFP neurons, we observed only single EGFPϩ neurons in each well and failed to detect doublets or any visible debris. Using this method, we were able to successfully sort intact, single EGFPϩ or EYFPϩ neurons into each well with minimal debris. Further controls for potential contamination were conducted in our subsequent qPCR analysis.

Panel of 48 key genes used in qPCR analysis of Hcrt/Ox and MCH neurons
To characterize the neurochemical phenotypes of Hcrt/Ox and MCH neurons, we subjected FACS-purified, single Ox-EGFP and Pmch-Cre;EYFP neurons to qPCR analysis of the expression of 48 key genes. For this purpose, we used TaqMan Gene Expression Assays on 48.48 dynamic arrays using a microfluidic Biomark HD system (Fluidigm). We developed a custom, curated gene panel (Table 1) comprising a wide variety of genes associated with neuronal function and neurochemical phenotype, selected for their potential ability to serve as positive and negative controls for distinct LHA cell populations based on previous molecular and anatomic characterization. These include housekeeping genes (Gapdh and Hprt), neuronal markers (Rbfox3, Map2, and Tubb3) and glial markers (Gfap and S100b) to ensure the analysis of intact, single neurons with minimal glial contamination. We included key neuropeptides known to be expressed in the LHA (Hcrt, Pmch, Pdyn, Gal, Nts, Cartpt, etc.) in addition to others which have not been clearly described in the LHA. Markers of fast amino acid neurotransmission were represented including genes necessary for GABA synthesis (Gad1 and Gad2) and vesicular GABA transport (Slc32a1) as well as glutamate vesicular transport (Slc17a7, Slc17a6, and Slc17a8) and metabolism (Gls). Genes associated with cholinergic and monoaminergic synthesis and release were also included (Chat, Th, Hdc, Tph, Slc18a2, etc.) to largely serve as negative controls given the relative dearth of such cell populations in the LHA (Abrahamson and Moore, 2001;Lein et al., 2007). Calcium-binding proteins have shown cell type-specific expression in many other brain regions (Hof et al., 1999), and a Pvalb-expressing population of glutamatergic neurons in the LHA has been described (Mészár et al., 2012) and were thus also included (Pvalb, Calb2, and Calb1).
Several key transcription factors are known to be relatively specific for LHA cell types, such as Nkx2.1 for MCH neurons (Croizier et al., 2011) and Lhx9 for Hcrt/Ox neurons Dalal et al., 2013;Sokolowski et al., 2015), and thus serve as important positive controls. To validate findings from previous work (Reti et al., 2002;Crocker et al., 2005;Honda et al., 2009;Dalal et al., 2013;Sokolowski et al., 2015), we included a number of genes known to be robustly expressed in Hcrt/Ox neurons (Nptx2, Igbp3, Celf6, and Nnat), as well as three key receptor genes (Lepr, Ghsr, and Mc4r), implicated in LHA function. Two of these (Lepr and Mc4r) have been shown to be expressed in overlapping populations of LHA neurons, separate from Hcrt/Ox and MCH neurons (Leinninger et al., 2009;Cui et al., 2012). To ensure only intact neurons with robust gene expression were used for our analysis, all cells collected by FACS were prescreened for the expression of Gapdh and outliers (CI ϭ 95%) were removed, resulting in a total of 89 Pmch-Cre;EYFP and 70 Ox-EGFP neurons that went on to further qPCR analysis using the full 48-gene panel.

Single-cell qPCR for transcripts of 48 key genes in Hcrt/Ox and MCH neurons
Following single-cell qPCR, with an output of inverted Ct values plotted on a log 2 -based scale, we first asked what proportion of neurons sorted from Ox-EGFP mice expressed Hcrt, and what proportion of Pmch-Cre;EYFP neurons expressed Pmch. We found that virtually every labeled and sorted Ox-EGFP neuron (98.6%; 69/70) that passed our Gapdh prescreen, robustly expressed Hcrt (the 1 Hcrt-cell was removed from the dataset), and every labeled and sorted Pmch-Cre;EYFP neuron (100%; 89/89) expressed Pmch, although we observed variability in Pmch expression levels. Nevertheless, these data indicate that our dissociation and FACS-sorting procedure, using Ox-EGFP and Pmch-Cre;EYFP transgenic lines, is capable of isolating cells that appear to faithfully and selectively express transcripts for their defining neuropeptides.
We then asked whether the gene expression profiles of Hcrt/Ox and MCH neurons are distinct or overlapping based on our 48-gene panel. To do this, we performed unsupervised hierarchical clustering using Ward's method with complete linkage (Fig. 3A). We found that Hcrt/Ox Gfap Glial marker Mm01253033_m1 S100 calcium binding protein B (S100␤) S100b Glial marker Mm00485897_m1 To formally determine which genes differentiate the Hcrt/Ox and MCH neuronal populations based on our 48-gene panel, we performed principal component analysis (PCA) and examined the loadings of genes on the PC axes. In our PCA of the single-cell qPCR data, Hcrt/Ox and MCH neurons separate clearly along the first principal component (PC1) into distinct clusters on the basis of several key markers (Fig. 3B). The variance captured in PC1 accounts for 26% of the total variance in the dataset. This is in contrast to PC2, which largely captures the within-group variation in Hcrt/Ox and MCH cells, which only captures 8.5% of the total variance. Ellipses represent 95% coverage for each population. Not surprisingly, Pmch and Hcrt are the most discriminatory markers that differentiate each cluster from one another and show mutually exclusive expression patterns, further confirming  Table 2) and the Benjamini-Hochberg procedure to control false discovery rate (FDR) at 5%. In all panels, red denotes Hcrt/Ox neurons and blue denotes MCH neurons. that these two populations are nonoverlapping. Hcrt/Ox neurons are further distinguished by the expression of key discriminatory genes (Igfbp3, Pdyn, Nptx2, Lhx9, and Nnat), confirming previous molecular and anatomic characterizations of Hcrt/Ox neurons (Reti et al., 2002;Crocker et al., 2005;Honda et al., 2009;Dalal et al., 2013;Sokolowski et al., 2015). However, many of these same markers were also expressed in MCH neurons to varying extents. In particular, we found that Celf6, a gene that encodes an RNA-binding protein implicated in autism-related behavior  and found to be widely expressed in neuromodulatory cell types throughout the mouse brain , is expressed in all Hcrt/Ox and MCH neurons. In our PCA analysis, we found that MCH neurons are defined by the expression of other discriminatory genes (Nkx2.1, Gad1, and Pnoc). The highly specific expression of Nkx2.1 in MCH neurons is consistent with previous molecular and anatomic work (Croizier et al., 2011). We note that using PCA analysis, we found that neither Hcrt/Ox nor MCH populations form clear subclusters attributable to the expression of specific discriminatory genes, implying a lack of discrete subpopulations, at least based on our limited 48-gene panel.
We next addressed two potential confounds in the interpretation of our data. One is whether we have sufficient statistical power to quantitatively evaluate differential gene expression between Hcrt/Ox and MCH neurons. Another is whether there are systematic batch effects within each population. To address the first point, we performed power analysis via simulation (see Materials and Methods) to determine if the numbers of neurons we collected for each population are adequate to achieve sufficient statistical power. We found that with the sample sizes of each population (69 and 89) used in this study, Fisher's exact test (Agresti, 1992) has adequate power (Ͼ90%) to detect differential gene expression between the two types of neurons for even moderate effect sizes.
To address the issue of batch effects, we used two complementary approaches. First, we used multiple hypothesis testing to examine whether there is a systematic difference between the two batches within each condition. Specifically, for each population, and for each of the 48 genes, we used Fisher's exact test (Agresti, 1992) to assess whether the probability of gene expression is the same in the two batches. The Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) was used to control the FDR at 5%. For Hcrt/Ox neurons, only one gene (Rbfox3) shows significant difference among the two batches when the FDR is controlled at 5%. For MCH neurons, three genes (Gal, Trh, and Nnat) show significant difference among the two batches when the FDR is controlled at 5%. Therefore, we found minimal systematic difference between two batches of data for each population. We also performed PCA analysis of batch-effects within each population and found that cells from different collections share nearly identical gene expression patterns and these differences do not contribute to the differences between the Hcrt/Ox and MCH neuron populations.
Finally, another potential confound in our single-cell analysis is the possibility that the droplet of media that accompanies FACS-sorted single cells is a source of contaminating, ambient mRNA from ruptured cells in the cell suspension. To address this possibility, we added fluorescent APC beads to the single-cell suspensions from each population and sorted single beads using parameters described above. Single sorted beads, eight from each condition, were subjected to qPCR alongside our cell samples. Beads sorted from Pmch-Cre;EYFP preparations yielded undetectable levels of any of the 48 genes in our panel. Among beads sorted from Ox-EGFP preparations, a single bead contained detectable levels of Hcrt and Nptx2 while another bead contained detectable levels of Pmch. These data demonstrate that media contamination is unlikely to contribute to the patterns of gene expression we observed in our single-cell qPCR analysis.

Differential gene expression between Hcrt/Ox and MCH neurons
Having demonstrated that Hcrt/Ox and MCH neurons are transcriptionally distinct on the basis of our 48-gene panel, we averaged the single-cell expression values from each population to more clearly describe differential gene expression (Fig. 3D). Aside from housekeeping genes (Gapdh and Hprt), Hcrt/Ox and MCH neurons show robust expression of neuronal markers (Tubb3, NeuN, and Map2) and minimal expression of glial markers (Gfap and S100b), further demonstrating that our capture of single, labeled neurons was largely free of glial contamination. Other broad categories of genes (neuropeptides, receptors, neurotransmitter components etc.) exhibited striking patterns of both common and differential gene expression. To complement average expression level, we quantified proportional difference in gene expression using Fisher's exact test (Agresti, 1992) to examine whether the probability of gene expression is different among the two populations, while the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) was used to control the FDR at 5%. Table 2 shows all 48 genes in descending order of proportional difference, with significance level indicated as adjusted p value. Differential gene expression will be addressed in more detail in the subsequent discussion of the coexpression of neuropeptides (Figs. 5, 6) and fast neurotransmitter components (Figs. 7, 8). Taken together, our single-cell qPCR analysis thus indicates that our procedure for isolating single, intact, FACSsorted Hcrt/Ox and MCH neurons from two transgenicallylabeled mouse lines is selective and robust, with minimal contamination by glial cells and ambient transcripts in the media.

Comparison of single-cell qPCR analysis of Hcrt/Ox neurons with bulk TRAP/microarray data
As further validation of the pattern of gene expression that we observed in our single-cell qPCR analysis of Hcrt/Ox neurons, we compared our dataset with a previously described bulk transcriptomic analysis of Hcrt/Ox neurons (Dalal et al., 2013). In this work, Dalal and coauthors generated a novel transgenic mouse line (Hcrt:: eGFP-RpL10a), isolated ribosome-bound transcripts from homogenized diencephalon through translating ribosome affinity purification (TRAP) methodology, followed by systematic translational profiling through microarrays. Using this published data, we identified 41 genes that were common to both their dataset and our gene panel and directly compared their gene expression values with our averaged single-cell qPCR data (Fig. 4). We found a strong positive linear correlation between these two datasets, with a Spearman's correlation coefficient ϭ 0.71 (p ϭ 1.2 ϫ 10 Ϫ8 ). Among the most strongly correlated genes were Hcrt, Pdyn, Nptx2, Igfbp3, Lhx9, and Nnat, all known to be enriched in Hcrt/Ox neurons from Dalal et al. (2013) and from previous work (Reti et al., 2002;Crocker et al., 2005;Honda et al., 2009;Sokolowski et al., 2015). This robust correlation of known discriminatory markers, observed using two independent methodologies and transgenic reagents, is additional supportive evidence that our single-cell qPCR approach is capable of capturing pure populations of rare and intermingled neuropeptidergic neurons within the LHA.
As described above, we found that every labeled and sorted Ox-EGFP neuron expressed Hcrt and every labeled and every sorted Pmch-Cre;EYFP neuron expressed Pmch to varying degrees ( Fig. 5; Table 2). However, we also found that a small number of Hcrt/Ox neurons also expressed Pmch (8.7%), and an even smaller number of MCH neurons also expressed Hcrt (5.6%; Fig. 5; Table 2). This potential crosstalk is mitigated by our observation of highly selective expression of two key discriminatory transcription factors, Lhx9 in Hcrt/Ox neurons (100% in Hcrt/Ox neurons; 3.4% in MCH neurons) and Nkx2.1 in MCH neurons (60.1% in MCH neurons; 0% in Hcrt/Ox neurons; Fig. 3D; Table 2).
Neuropeptides/receptors exhibit varying degrees of specificity and abundance within the MCH and Hcrt/ Ox populations, summarized in a bubble plot ( Fig. 5E; Table 2). For example, Pdyn is unsurprisingly both highly abundant and highly specific to Hcrt/Ox neurons, consistent with previous work (Chou et al., 2001;Muschamp et al., 2014) and the only neuropeptide transcript to exhibit this profile. Other neuropeptides (Sst and Pnoc), although much less abundant, also exhibit highly specific differential expression (Sst in a small subpopulation of Hcrt/Ox neurons and Pnoc in a subpopulation of MCH neurons). Our observation that Pnoc, encoding nociceptin/orphanin FQ (N/OFQ), exhibits moderate to low expression in MCH but is undetectable in Hcrt/Ox neurons, contrasts with a previous report showing that N/OFQ-IR is found in most Hcrt/Ox but not MCH neurons in the rat (Maolood and Meister, 2010), but is consistent with a study showing that N/OFQ-IR and Hcrt/Ox-IR neurons are intermingled but separate (Gerashchenko et al., 2011). Other neuropeptides (Cartpt and Penk) are abundant and show only moderate specificity, with Cartpt being moderately specific to MCH neurons, with clear Cartptϩ and Cartpt-subpopulations (especially prominent in Fig. 5A), while Penk is moderately specific to Hcrt/Ox neurons. Detailed analyses in rodents demonstrated extensive and selective expression of Cartpt in a subpopulation of MCH neurons, with little to no expression in Hcrt/Ox neurons (Broberger, 1999;Vrang et al., 1999;Brischoux et al., 2001;Elias et al., 2001;Cvetkovic et al., 2004). Penk expression has been reported in the rat LHA (Abrahamson and Moore, 2001;Horjales-Araujo et al., 2014), although colocalization with Hcrt/Ox and MCH neurons has not been previously shown.
In contrast, other neuropeptide transcripts were either abundant or absent from both populations. Nucb2 shows significantly higher expression in Hcrt/Ox neurons (Table 2) but is highly abundant in both populations, in contrast to previous work showing strong specificity for MCH neurons (Foo et al., 2008;Fort et al., 2008). Gal was also relatively abundant but seemingly nonspecific, contrasting with previous work (Laque et al., 2013). Similarly, other neuropeptide transcripts (Tac1, Crh, Nts, Trh) also show little specificity but are much less abundant, while Tac2 is undetectable in either population. Our failure to detect appreciable Nts transcript in Hcrt/Ox neurons contrasts with a previous report showing that a majority of Hcrt/Ox neurons express Nts mRNA and NTS-IR (Furutani et al., 2013). Receptor expression in both populations is relatively low; Mc4r is expressed in small proportion of both populations with no specificity while both Ghsr and Lepr are neither abundant nor specific. The low expression of Mc4r, Lepr, and Nts in either population is consistent with MCH and Hcrt/Ox neurons being entirely separate from overlapping populations of largely leptin-sensitive, neuropeptidergic LHA neurons (Leinninger et al., 2009Cui et al., 2012;Laque et al., 2013). Taken together, these data suggest that both Hcrt/Ox and MCH neurons exhibit a surprisingly high degree of coexpression with other neuropeptide transcripts, with only a few showing both high abundance and specificity for one population over another.

Hcrt/Ox and MCH neurons exhibit markers of both GABAergic and glutamatergic fast neurotransmitter phenotypes
While MCH and Hcrt/Ox neurons are widely thought to exhibit GABAergic and glutamatergic phenotypes, respectively, recent evidence suggests greater ambiguity in their fast neurotransmitter phenotype than previously appreciated (Bonnavion et al., 2016). To address this conflicting evidence using our single-cell qPCR dataset, we compared the expression patterns of genes necessary for fast amino acid neurotransmitter synthesis and vesicular transport among MCH (Fig. 7A) and Hcrt/Ox neurons (Fig.  7B) and summarized their differential expression in the form of bubble plots (Fig. 7E).
Consistent with the widely held view that MCH neurons are GABAergic in nature, we found that virtually all MCH neurons express the GABA synthetic enzyme Gad1 (97.8%) and a smaller subpopulation expresses Gad2 (21.3%; Fig. 7; Table 2). However, none of the MCH neurons exhibited detectable expression of the vesicular GABA transporter Slc32a1 (0%) This suggests that while these cells may be capable of GABA synthesis, they lack the machinery for its vesicular packaging and release through canonical pathways. In contrast, the vesicular glutamate transporter 2 (Slc17a6) was expressed in virtually every MCH neuron (97.8%) with a very small subpopulation of neurons expressing the vesicular glutamate transporter 3, Slc17a8 (9.0%; Fig. 7A), suggesting a functional glutamatergic phenotype. Similarly, we found that every Hcrt/Ox neuron expressed Slc17a6 (100%) approximately half of which also express Slc17a8 (47.8%; Fig. 7; Table 2), consistent with evidence showing that these cells are glutamatergic. Surprisingly, over half of the Hcrt/Ox neurons we sampled expressed Gad1 (56.5%) and a smaller subpopulation expressed Gad2 (15.9%). As with MCH neurons, we found that Slc32a1 expression was virtually undetectable in Hcrt/Ox neurons (1.5%). Finally, Slc17a7 expression was undetectable in either population. Given the abundant expression of Gad1 in both populations, and failure to detect Slc32a1, we asked whether Hcrt/Ox and MCH neurons expressed the vesicular monoamine transporters Slc18a1 and Slc18a2, the latter of which has been shown to be capable of the transport and vesicular release of GABA through a noncanonical pathway (Tritsch et al., 2012). We found, however, that only a very small fraction of MCH (2.5%) and Hcrt/Ox (1.5%) neurons express Slc18a2 transcript at detectable levels.
Having observed the surprising result that these two populations of neuropeptidergic neurons in the LHA uniformly and abundantly express the machinery for both glutamate release (Slc17a6) and GABA synthesis (Gad1), we asked if this pattern of coexpression is a common feature of glutamatergic neurons in the LHA. To this end, we performed an identical single-cell qPCR analysis of Slc17a6-expressing LHA neurons from Vglut2-Cre;EYFP mice. As these mutant mice label cells with both transient developmental and stable expression of Slc17a6, we isolated 117 single labeled neurons from microdissected slices that passed Gapdh prescreening and confined our analysis to the 86 neurons expressing the Slc17a6 transcript (73.5%). Of these cells, none expressed Slc32a1 (0%; 0/86; Fig. 7C). However, exactly half of these Slc17a6-expressing neurons expressed Gad1 (50%; 43/86) along with another small, partially overlapping population expressing Gad2 (14.0%; 12/86). Although the sample size is limited, these data suggest that approximately half of Slc17a6-expressing, putative glutamatergic neurons in the LHA are potentially capable of GABA synthesis.
Finally, considering that in our analysis of Hcrt/Ox, MCH and glutamatergic neurons, we did not detect expression of Slc32a1, we asked whether this was the result of a true lack of expression or the failure of our probe to detect Slc32a1 expression. Furthermore, we wished to confirm the neurotransmitter phenotype of canonical GABAergic neurons in the LHA (i.e., those capable of both the synthesis and vesicular release of GABA). For this purpose, we performed a single-cell qPCR analysis of Slc32a1-expressing neurons from the LHA of Vgat-Cre; EYFP mice. Similar to the Vglut2-expressing cells collected above, we isolated 212 single neurons and focused our analysis on the 118 cells expressing the Slc32a1 transcript (55.7%). We found that virtually none of these cells expressed vesicular glutamate transporters (Slc17a6: 0%; 0/118, Slc17a7: 0%; 0/118, Slc17a8: 0.8%; 1/118). However, all Slc32a1-expressing neurons robustly and uniformly expressed the triad of canonical markers of GABAergic phenotype (Slc32a1: 100%; 118/118, Gad1: 100%; 118/118, Gad2: 100%; 118/118). These data confirm that we can detect Slc32a1 in GABAergic neurons and that LHA GABAergic neurons uniformly and abun- dantly express markers of canonical GABA synthesis and release (Slc32a1, Gad1, and Gad2).

Comparing electrophysiological diversity with neurochemical diversity in Hcrt/Ox neurons
Finally, we asked whether the neurochemical heterogeneity that we identified among Hcrt/Ox and MCH neurons correlates with known functional diversity. To approach this question, we turned to the well-characterized intrinsic membrane properties of Hcrt/Ox neurons.
Previous work in brain slices from Ox-EGFP mice has shown that, on the basis of key intrinsic membrane properties, morphologic characteristics and measures of synaptic input, Hcrt/Ox neurons may be parsed into two functional subpopulations (Schöne et al., 2011). In particular, Schöne and coauthors found that the signature of one population ("D-type") is a depolarizing rebound in firing following a hyperpolarizing step in current-clamp mode. Another population ("H-type"), in turn, exhibited a characteristic long latency to firing following the same hyperpolarizing step as a result of prominent expression of A-type potassium current Schöne et al., 2011). Neither population was shown to segregate on the basis of their projection targets (González et al., 2012). We asked whether we could (1) recapitulate this broad categorization; (2) harvest cytoplasm from recorded cells for qPCR analysis using the same 48-gene panel and correlate gene expression with FACS-sorted cells; and (3) determine if D-type and H-type signatures segregated between neurochemically distinct Hcrt/Ox neurons at the single-cell level. For example, can Gad1ϩ and Gad1-Hcrt/Ox neurons be identified on the basis of key electrical signatures?
We obtained slices from Ox-EGFP mice and performed patch-clamp recordings from 16 EGFPϩ neurons. From current-clamp recordings, we found that 62.5% (10/16) Hcrt/Ox neurons exhibited a depolarizing postinhibitory rebound following a 1 s hyperpolarizing step, consistent with the D-type signature. In turn, 37.5% (6/16) Hcrt/Ox neurons exhibited a latency (Ͼ100 ms) to firing following the hyperpolarizing step, consistent with the H-type signature (Fig. 9A,B). We also noted the location of each recorded cell and mapped them onto an atlas image of the LHA (Fig. 9C). We then harvested cytoplasm from each of these neurons through the recording pipette and performed single-cell qPCR analysis using our 48-gene panel, in an otherwise identical manner to FACS-sorted cells. We found that 93.7% (15/16) of these Hcrt/Ox neurons expressed Hcrt in addition to other key discriminatory genes known to be enriched in Hcrt/Ox neurons including Lhx9, Nptx2, Pdyn, and Nnat. In performing unsupervised hierarchical clustering of the gene expres- sion data, we failed to detect a clear correlation between neurochemically diverse Hcrt/Ox neurons and electrophysiological signatures (Fig. 9D). Despite this, when we compared gene expression patterns between recorded cells and FACS-sorted cells, we found a strong, positive linear correlation with a Spearman's correlation coefficient ϭ 0.89 (p ϭ 2.2 ϫ 10 Ϫ16 ; Fig. 9E). Taken together, these data suggest that although distinct electrical signatures of Hcrt/Ox neurons may not readily track with the known markers of neurochemical phenotype included in our gene panel, the gene expression profile of manually harvested single neurons, that have undergone whole-cell recordings, closely match FACSsorted single cells.

Discussion
Underlying functionally discrete neuronal cell types in the brain are suites of differentially expressed genes that determine and maintain cell type identity (Nelson et al., 2006;Fishell and Heintz, 2013). An important category of these discriminatory genes are those that specify neurochemical phenotype, i.e. the expression of neuropeptides and markers of cholinergic, monoaminergic, purinergic and fast amino acid neurotransmission. In the hypothalamus, neuronal populations are often defined by their expression of neuropeptides, but an ever growing body of molecular, anatomic and functional evidence suggests that many hypothalamic neuronal cell types are equipped with a broader palette of neurochemical signaling mechanisms that may include multiple neuropeptides and fast neurotransmitters (Schöne and Burdakov, 2012;van den Pol, 2012;Arrigoni and Saper, 2014;Bonnavion et al., 2016). In the LHA, mounting evidence suggests that Hcrt/Ox and MCH neurons exhibit a complex but poorly understood repertoire of neuropeptide and fast neurotransmitter signaling mechanisms (Bonnavion et al., 2016). We undertook a quantitative single-cell gene expression profiling approach to better understand the neurochemical phenotype of Hcrt/Ox and MCH neurons. We optimized a robust procedure for microdissecting the LHA from transgenically-labeled mouse lines, cleanly isolating single Hcrt/Ox and MCH neurons via FACS and quantifying the expression of a panel of 48 genes in each cell to define neurochemical phenotype. We found that, in addition to expressing a number of neuropeptide transcripts, all MCH and a subpopulation of Hcrt/Ox neurons express markers for both inhibitory and excitatory fast neurotransmission. Furthermore, we found this profile is characteristic of a subpopulation of LHA glutamatergic neurons, but contrasts with a broad population of LHA GABAergic neurons. Finally, we found that electrical signatures that define Hcrt/Ox subpopulations do not appear to track with neurochemical phenotype.

Comparison with previous approaches, technical considerations, and limitations
Previous studies of single-cell expression profiling in rodents were performed in rats; one probing the expression of five genes in manually harvested cells labeled with antibodies (Volgin et al., 2004) and another examining twelve genes from cells in fixed tissue, isolated using laser capture microdissection (LCM;Harthoorn et al., 2005). Two recent reports, however, describe single-cell microarray analysis  and RNA sequencing (Yelin-Bekerman et al., 2015) of zebrafish Hcrt neurons, which exhibit conserved function and expression of the key transcription factor gene Lhx9. Nevertheless, our approach represents an improvement over previous singlecell methods in the rodent LHA in terms of scale, throughput and precision. Furthermore, LCM has been shown to be significantly more prone to contamination, from glial cells in particular, as compared to FACS-sorted cells (Okaty et al., 2011). We found that robust expression of neuronal markers, with little to no expression of glial markers in these samples, was evidence for minimal contamination by glial cells. Furthermore, our failure to detect significant signal from single, fluorescent, FACS-sorted beads that were intermixed with our cell suspensions was evidence for minimal contamination by ambient transcripts in the media. With respect to the specificity of single-cell capture, we found that the vast majority of Hcrt/Ox and MCH neurons exhibited highly specific expression of their respective neuropeptides consistent with our demonstration that key transcription factor genes Lhx9 and Nkx2.1 exhibited selective and mutually exclusive expression.
However, the interpretation of our single-cell gene expression analysis must be considered in light of several important caveats. One is the specificity and penetrance of the mutant mouse lines that we used. Despite the presence of EYFP potentially reflecting both transient developmental and stable Pmch promoter-driven cre expression, Pmch-Cre;EYFP mice are very effective in selectively labeling Pmch-expressing neurons and capturing a large percentage of the MCH-IR population. Although Ox-EGFP mice show high specificity, the penetrance of their EGFP expression is comparatively low and we captured only about half of the Hcrt/Ox-IR population. Whether this represents a random cross-section of the Hcrt/Ox population or a specific, molecularly-defined subpopulation remains to be determined. Nevertheless, in each case, we independently assessed Hcrt/Ox or MCH expression through qPCR as an internal control to account for any lack of specificity of the transgenic line. Furthermore, the neurochemical profile of Hcrt/Ox neu-rons, that we determined through single-cell analysis, closely matched bulk RNA-TRAP data derived from another transgenic reporter with higher penetrance (Dalal et al., 2013;Fig. 4).
Another set of technical considerations in the interpretation of this data include: (1) the age and gender of the mice; (2) circadian, metabolic and behavioral state; (3) sample size; and (4) detection of mRNA as a reflection of translated protein. With regard to age, we chose to collect neurons from juvenile mice (P21-P23) to approximate baseline gene expression patterns in adult mice yet optimize the isolation of healthy and intact neurons following FACS sorting, which we found to be compromised in older animals. Although we chose to limit our study to males, future work comparing male and female mice will be required to probe possible sexual dimorphism in gene expression. In another effort to limit variability and obtain an approximation of baseline gene expression, we collected cells from mice at the same time of day in the absence of metabolic or behavioral challenges, which likely alters gene expression in LHA neurons (Leibowitz, 2007;Barson et al., 2013). Our qPCR and FISH data may be considered a snapshot of gene expression in juvenile male mice under these baseline conditions. With regard to sample size, we found that the number of single cells that we collected, along with minimal between-batch variability, had sufficient statistical power to detect even moderate levels of differential gene expression. However, further work, ideally employing whole transcriptome analysis of a large number of single neurons would be required to allow a higher resolution interrogation of neurochemical heterogeneity among these neurons. A final caveat in interpreting our single-cell qPCR data, as with any gene expression data, is that quantifying mRNA content with high sensitivity may or may not be a faithful, linear representation of translated protein (Citri et al., 2011). Also, given the stochastic nature of gene expression in single cells (Raj and van Oudenaarden, 2008), there may be an element of noise in the gene expression patterns that we observe, the extent to which, and biological significance of, remains to be determined.

Neuropeptide/receptor coexpression in Hcrt/Ox and MCH neurons
Single MCH and Hcrt/Ox neurons express a variety of neuropeptide transcripts including some with clear differential expression. Hcrt/Ox neurons exhibit robust and selective expression of Pdyn, as shown previously (Chou et al., 2001;Muschamp et al., 2014), and moderately specific expression of Penk. MCH neurons, in turn, showed moderately specific expression of Cartpt, with apparent Pmchϩ populations that were either Cartptϩ or Cartpt-, consistent with previous work (Broberger, 1999;Vrang et al., 1999;Brischoux et al., 2001;Elias et al., 2001;Cvetkovic et al., 2004), although we found more expression in Hcrt/Ox neurons than expected. Other transcripts (Nucb2 and Gal) exhibit surprisingly abundant expression among the two populations. Previous studies showed strong colocalization between NUCB2-IR and MCH-IR but not with Hcrt/Ox-IR (Foo et al., 2008;Fort et al., 2008).
While GAL-IR has been observed in Hcrt/Ox neurons (Håkansson et al., 1999), subsequent work showed no coexpression in Hcrt/Ox or MCH neurons (Laque et al., 2013). Despite the heterogeneity in neuropeptide transcript expression, clear subpopulations did not emerge from our unsupervised cluster analysis. Further work is required to determine if neuropeptide coexpression, or a constellation of other key markers, may be the primary driver(s) of population diversity among these and other LHA neurons.
Nevertheless, MCH and Hcrt/Ox neurons likely have the capacity to release multiple neuropeptides, evidenced by neuropeptide colocalization in single terminals (Guan et al., 2002;Balcita-Pedicino and Sesack, 2007) or even in the same dense-core vesicle (Muschamp et al., 2014). Furthermore, the release of multiple neuropeptides may significantly impact the ability of single neurons to tune their postsynaptic targets (van den Pol, 2012). For example, Hcrt/Ox and dynorphin cotransmission has been proposed to differentially regulate postsynaptic excitability among target neurons (Li and van den Pol, 2006;Arrigoni et al., 2010;Muschamp et al., 2014;Ferrari et al., 2015). Although Hcrt/Ox release from Hcrt/Ox fibers has been demonstrated using an optogenetic approach , evoked release of multiple neuropeptides, and determining their role in circuit function, remains a significant technical challenge (van den Pol, 2012; Arrigoni and Saper, 2014).

Fast neurotransmitter coexpression in MCH and Hcrt/Ox neurons
MCH neurons have long been thought to exhibit a GABAergic phenotype, evidenced by their expression of GABAergic markers such as GAD1 mRNA (Elias et al., 2001;Harthoorn et al., 2005;Sapin et al., 2010). Immunohistochemical evidence has shown that a small proportion of MCHϩ varicosities found in the locus coeruleus (LC) contain VGAT and another small proportion oppose gephryin puncta on THϩ cell bodies (del Cid-Pellitero and Jones, 2012), suggesting the capacity of at least a small proportion of MCH neurons for synaptic GABA release. Consistent with these findings, optogenetic experiments showed that selective stimulation of MCH fibers evoked GABA release onto hypothalamic histaminergic (HA) neurons in brain slices . However, further work showed that MCH neurons may not fit the profile of canonical GABAergic neurons. First, MCH-IR was shown not to overlap with intermingled VGAT-expressing LHA neurons in Vgat-Cre;EYFP mice (Jennings et al., 2015). Furthermore, recent work shows that the vast majority of MCH neurons express mRNA for VGLUT2 while virtually none express VGAT, and that optogenetic activation of MCH fibers in the lateral septum induces monosynaptic release of glutamate, not GABA (Chee et al., 2015). The most parsimonious interpretation of our single-cell results is that MCH neurons are functionally glutamatergic (Slc17a6ϩ) and uniformly capable of GABA synthesis (Gad1ϩ), but that GABA release (Slc32a1-) is either mostly nonsynaptic or occurs through a noncanonical pathway as discussed further in the next section.
Convincing anatomic evidence suggests that Hcrt/Ox neurons are glutamatergic (Horvath et al., 1999;Guan et al., 2002;Rosin et al., 2003;Torrealba et al., 2003;Balcita-Pedicino and Sesack, 2007) including immunohistochemical evidence that a small proportion of Hcrt/Oxϩ varicosities in the LC contain VGLUT2 and another small percentage oppose postsynaptic density protein-95 (PSD-95) puncta (Henny et al., 2010). Furthermore, optogenetic stimulation of Hcrt/Ox fibers induced glutamate release onto HA neurons (Schöne et al., 2012). As with MCH neurons, other data suggest ambiguity regarding the fast neurotransmitter phenotype of Hcrt/Ox neurons. Ultrastructural data in rats shows that while a majority of Hcrt/Ox terminals have asymmetric synapses, symmetric synapses are also present (Guan et al., 2002;Torrealba et al., 2003;Balcita-Pedicino and Sesack, 2007). Furthermore, recent work demonstrated that optogenetic activation of Hcrt/Ox neurons suppresses firing in nearby MCH neurons through light-evoked inhibitory synaptic events, which in some cases were time-locked to the stimulus with relatively short latencies, abolished by gabazine, but unaffected by glutamate receptor blockade. Although this may be explained by a disynaptic mechanism, GABA-IR was found in ϳ10 -20% of Hcrt/Ox-IR neurons . Our single-cell qPCR data are broadly consistent with the notion that Hcrt/Ox neurons are functionally glutamatergic (Slc17a6ϩ), and that a subpopulation (ϳ50%) are potentially capable of GABA synthesis (Gad1ϩ), but not release through a canonical pathway (Slc32a1-).

Implications for neurochemical flexibility and plasticity among MCH and Hcrt/Ox neurons
We found that virtually all MCH neurons and approximately half of Hcrt/Ox neurons exhibit mixed fast neurotransmitter profiles (Slc17a6ϩ, Slc32a1-, Gad1ϩ). We also showed that this profile is characteristic of the broader population of LHA VGLUT2 neurons, half of which are Gad1ϩ, suggesting that Gad1 expression may not be a reliable marker for GABAergic phenotypes in this region. Formally, one explanation is that both MCH and Hcrt/Ox neuron populations have a functionally glutamatergic phenotype (Slc17a6ϩ, Slc32a1-). Alternatively, given the prominent expression of Gad1 among these neurons, it is tempting to speculate that MCH neurons and a subpopulation of Hcrt/Ox neurons may be capable of glutamate/ GABA cotransmission through noncanonical GABA release pathways described in other systems (Seal and Edwards, 2006;Hnasko and Edwards, 2012;Münster-Wandowski et al., 2013;Barker et al., 2016;Tritsch et al., 2016;Granger et al., 2017). Anatomic and functional evidence for glutamate/GABA corelease from single neurons has been described in the developing auditory brainstem and among hippocampal mossy fibers during development and in the adult (Seal and Edwards, 2006;Hnasko and Edwards, 2012;Münster-Wandowski et al., 2013). Recent reports describe functional, monosynaptic glutamate/ GABA cotransmission in basal ganglia outputs to the lateral habenula (LHb), with VGLUT2-IR and GAD-IR colocalized in terminals, as well as evidence that VGAT is necessary for cotransmission (Shabel et al., 2014). In addition, evidence for glutamate/GABA cotransmission was shown in two recent studies of ventral tegmental area (VTA) projections. In one, VGLUT2, VGAT and GAD appear to colocalize in single mesohabenular axons to facilitate glutamate/GABA corelease (Root et al., 2014). In the other, VGLUT2ϩ VTA neurons exhibit target-dependent monosynaptic cotransmission, with glutamate/GABA corelease in the LHb, but exclusive glutamate release in the nucleus accumbens (Yoo et al., 2016). In these latter cases, glutamate/GABA cotransmission appears to depend on colocalization of VGLUT2 and VGAT. However, among MCH and Hcrt/Ox neurons, Slc32a1 mRNA was undetectable in our qPCR analysis and minimal in our FISH analysis. If MCH and a subset of Hcrt/Ox neurons are indeed capable of synaptic GABA corelease, it may occur through an alternative pathway. Noncanonical vesicular GABA release through VMAT2 (Tritsch et al., 2012) is an unlikely candidate in this case as Slc18a2 is virtually undetectable among Hcrt/Ox and MCH neurons. Another alternative pathway is through reversal of the plasma membrane GABA transporter GAT-1 (Wu et al., 2007), which remains to be investigated. Finally, highly informative recent single-cell RNA sequencing studies of neuronal populations in the medial hypothalamus (Romanov et al., 2017;Campbell et al., 2017) and entopenduncular nucleus of the basal ganglia  have revealed compelling evidence for mixed fast neurotransmitter gene expression profiles, suggesting that such forms of cotransmission may be more common than previously thought.
Regarding the Gad1ϩ, Slc32a1-profile of MCH and Hcrt/Ox neurons, an informative comparison may be made with proopiomelanocortin (POMC) neurons in the arcuate nucleus. Detailed anatomic, electrophysiological and optogenetic analysis of the neurochemical phenotype of POMC neurons by Hentges and coworkers has revealed that this neuronal population may be parsed into two functional subpopulations: GABAergic (Gad1ϩ and/or Gad2ϩ), and a much smaller population of glutamatergic POMC neurons (Slc17a6ϩ), half of which also express Gad2 (Hentges et al., 2004(Hentges et al., , 2009Jarvie and Hentges, 2012). Interestingly, POMC GABAergic neurons are capable of fast synaptic release of GABA (Hentges et al., 2009;Dicken et al., 2012) yet Slc32a1 is virtually undetectable (Jarvie and Hentges, 2012), but may be expressed in small subpopulations (Jeong et al., 2016). Nevertheless, this suggests that GABA may largely be released from these cells through a noncanonical pathway. These same phenotypic markers were shown to change over postnatal development (Dennison et al., 2016) and can shift following metabolic perturbation (Dicken et al., 2015;Jarvie et al., 2017). Given evidence of alterations in neuropeptide gene expression in the LHA following metabolic and behavioral challenges (Leibowitz, 2007;Barson et al., 2013) and neurochemical phenotype switching in other systems (Demarque and Spitzer, 2012), it is plausible that MCH and Hcrt/Ox neurons also have the capacity for dynamic expression of key genes that specify neurochemical profile.
The possibility that MCH and/or Hcrt/Ox neurons exhibit neuropeptidergic and glutamate/GABA cotransmission, target-dependent cotransmission or the capacity for plasticity in neurotransmitter phenotype following metabolic or behavioral challenges all remain intriguing questions. Furthermore, the mechanisms of cotransmission and neurochemical heterogeneity may be the basis for functional classification of neuronal cell types in the LHA, as well as an important means of understanding their role in regulating postsynaptic excitability. Overall, our singlecell qPCR analysis of MCH and Hcrt/Ox neurons suggests the capacity for neurochemical flexibility: the ability to draw on a broad palette of signaling mechanisms with which to rapidly respond to fluctuating intrinsic and extrinsic physiologic challenges.