Single-Cell Gene Expression Analysis of Cholinergic Neurons in the Arcuate Nucleus of the Hypothalamus

The cholinoceptive system in the hypothalamus, in particular in the arcuate nucleus (ARC), plays a role in regulating food intake. Neurons in the ARC contain multiple neuropeptides, amines, and neurotransmitters. To study molecular and neurochemical heterogeneity of ARC neurons, we combine single-cell qRT-PCR and single-cell whole transcriptome amplification methods to analyze expression patterns of our hand-picked 60 genes in individual neurons in the ARC. Immunohistochemical and single-cell qRT-PCR analyses show choline acetyltransferase (ChAT)-expressing neurons in the ARC. Gene expression patterns are remarkably distinct in each individual cholinergic neuron. Two-thirds of cholinergic neurons express tyrosine hydroxylase (Th) mRNA. A large subset of these Th-positive cholinergic neurons is GABAergic as they express the GABA synthesizing enzyme glutamate decarboxylase and vesicular GABA transporter transcripts. Some cholinergic neurons also express the vesicular glutamate transporter transcript gene. POMC and POMC-processing enzyme transcripts are found in a subpopulation of cholinergic neurons. Despite this heterogeneity, gene expression patterns in individual cholinergic cells appear to be highly regulated in a cell-specific manner. In fact, membrane receptor transcripts are clustered with their respective intracellular signaling and downstream targets. This novel population of cholinergic neurons may be part of the neural circuitries that detect homeostatic need for food and control the drive to eat.


Introduction
The hypothalamus integrates circulating hormonal and nutritional signals to regulate energy homeostasis [1][2][3][4]. In particular, the arcuate nucleus of the hypothalamus (ARC) containing the melanocortin system plays a key role in the control of energy intake and expenditure [5][6][7][8][9][10]. In addition to the melanocortin system, there exists the cholinoceptive system in the ARC Laboratory). 2-month-old mice on a mixed C57BL6/129SVJ background were used for all experiments. Animals were housed in groups in cages under conditions of controlled temperature (22°C) with a 12:12 h light dark cycle and fed a standard chow diet with ad libitum access to water.
Single-cell real time qPCR and single-cell whole transcriptome amplification Brain slices were placed on the stage of an upright, infrared-differential interference contrast microscope (Olympus BX50WI) mounted on a Gibraltar X-Y table (Burleigh) and visualized with a 40X water immersion objective using infrared microscopy. We collected cytoplasm containing total RNA from individual cholinergic neurons in slices via aspiration into glass pipette as described in our previous work [16,18,35]. The initial reverse transcription (RT) reaction was conducted after pressure ejection of the single cell samples into a microcentrifuge tube with REPLI-g WTA single cell kit (Qiagen). Samples were incubated for 10 min at 42°C with 2 μl gDNA wipeout buffer prior to addition of 7 μl RT mix to synthesize first strand cDNA (RT mix: 1 μl oligodT primer, 4 μl RT buffer, 1 μl random primer, 1 μl RT enzyme mix). The tubes were incubated at 42°C for 1 hr, and at 95°C for 3 min and then incubated at 24°C for another 30 min with 10 μl ligation mix (8 μl ligase buffer, 2 μl ligase Mix). The reaction was stopped by incubating at 95°C for 5 min. Samples were incubated for another 2 hrs at 30°C after adding the amplification mix (29 μl buffer, 1 μl REPLI-g SensiPhi DNA polymerase) and at 65°C for 5 min. Once reverse-transcribed cDNAs had been made, we purified cDNA using column based Fragment DNA Purification Kit (Cat # 17287, Intron Biotechnology Inc). And then, the quality and quantity of single cell cDNA were determined using a NanoDrop ND-1000 spectrophotometer (Thermo Scientific).
Single-cell qPCR was performed in sealed 96 well plates with SYBR Green master Mix in a Light Cycler. Single-cell qPCR reactions were prepared in a final volume of 20 μl containing 2 μl of single cell whole cDNA (10 ng) and 10 μl of SYBR Green master mix in the presence of primers at 0.5 μM. We first prepared standard curves for housekeeping genes (e.g. 18S ribosomal RNA (18S rRNA)) and target genes using a cDNA mixture made of a small amount (2 μl) of each sample. Cycle number was plotted against the normalized fluorescence intensity to visualize the PCR amplification and determine the amplification efficiencies.
To ensure the reliability and integrity of the single-cell qPCR and single-cell WTA assay, we used several types of controls; (1) we tested several reference genes such as β-actin, glyceraldehyde 3-phosphate dehydrogenase (GADPH), 18S rRNA, or green fluorescent protein (GFP) and found that the 18S rRNA gene was most stable across cells, (2) all reactions were performed in triplicates, (3) the melting curves of each of the samples were checked to make sure that only one product was formed from the PCR reaction, (4) the PCR products were analyzed on agarose gels to further verify that the product size was correct and that there was only a single product, and (5) the amplification efficiencies of most primers tested were > 98% (S2 Table).

Gene clustering and statistical analysis
Cycle Threshold (C T ) values were extracted and melting curves of the C T values were visually inspected for a quality control. The data was imported, normalized, and visualized using R 3.2.2 [36] and HTqPCR [37] package. ΔC T normalization was carried out using C T values of 18S rRNA as a housekeeping gene. Normalized C T values were visualized as heatmaps and hierarchical clustering results were added on top as dendrograms. Pearson correlation distances between samples or genes were used in the clustering analysis for column-dendrograms and row-dendrograms, respectively.

The ARC contains cholinergic neurons
We crossed ChAT-IRES-Cre mice with tdTomato reporter mice to generate ChAT-IRES-Cre:: tdTomato mice [34]. Fig 1A shows an example of tdTomato-positive neurons in the ARC from ChAT-IRES-Cre::tdTomato mice. Immunostaining with an anti-ChAT antibody revealed that approximately 90% of ChAT-positive neurons were positive to tdTomato (Fig 1B; n = 7 animals), consistent with the prior study showing that there exist cholinergic neurons in the ARC [17]. We investigated whether individual neurons in this specific population are molecularly and neurochemically identical by analyzing the expression of our hand-picked 60 genes involved in neuronal activity, peptide processing, neurotransmitter synthesis, and signal transduction pathways (S1 Table).

ARC cholinergic neurons are neurochemically heterogeneous
We collected cytoplasm containing total RNA from individual cholinergic neurons in hypothalamic slices (Fig 1C). Single-cell qRT-PCR analysis revealed that cholinergic neurons in the ARC were neurochemically phenotypically distinct. Approximately two-thirds of cholinergic neurons in the ARC expressed tyrosine hydroxylase (Th) mRNA (n = 19 out of 26 neurons; Fig  2A and 2B). A large subpopulation of these TH-positive cholinergic neurons had the GABA synthesizing enzymes, including glutamic acid decarboxylase 65 (Gad2) and 67 (Gad1) (n = 18 and 13 out of 19 neurons, respectively) and vesicular GABA transporter (Slc32a1) transcript (n = 12 out of 19 neurons ; Fig 2A and 2B). These findings are consistent with the recent study showing co-expression of dopamine and GABA in the ARC [38]. Interestingly, these cholinergic neurons expressed vesicular glutamate transporters in particular type 2 (Slc17a6, n = 11 and 19 neurons, respectively; Fig 2A and 2B).
We also found that 40% of ARC cholinergic neurons expressed the Pomc gene (n = 11 out of 26 neurons, Fig 2A and 2B), whereas no cholinergic neuron had Npy mRNA (n = 0 out of 14 neurons, data not shown). POMC processing enzymes such as prohormone convertase 2 (Pcsk1) and N-acetyltransferase-1 (Nat1) transcripts were detected in most cholinergic POMC neurons (n = 9 and 7 out of 11 neurons, respectively). These are consistent with the prior study showing co-expression of POMC and ChAT in the ARC [17]. However, other POMC processing enzymes, including prohormone convertase 2 (Pcsk2), carboxypeptidase E (Cpe) and peptidyl α-amidating monooxygenase (Pam) transcripts were expressed at extremely low levels below the detection threshold among these POMC-expressing cholinergic neurons (n = 1, 1, and 1 out of 11 neurons, respectively, Fig 2A and 2B). We also noted that cholinergic non-POMC neurons contained peptide processing enzymes such as Pcsk1, Pcsk2, Cpe, and Pam transcripts, suggesting that they have an ability to process neuropeptides that are not derived from POMC. We next performed immunocytochemistry with anti-POMC and anti-Th  antibodies to examine if these cholinergic neurons indeed contain POMC and Th. We found that a subpopulation of cholinergic neurons was immune-positive to these antibodies (Fig 3A  and 3B), further suggesting that some cholinergic neurons in the ARC contain POMC and Th.

Gene clustering of leptin receptor-expressing cholinergic neurons in the ARC
We then examined whether genes with similar function would be more closely related than others using hierarchical clustering analysis. This shows that the Lepr-expressing neurons were subdivided into three groups of gene signatures and each cluster was represented by a genes of similar function (Fig 4). The first group of genes (G1 gene group) showed co-expression of peptide-processing enzymes, Cpe, Pcsk2, and Pam transcripts with the Chat gene (Fig  4). Within the second cluster (G2 gene group), The Th gene was co-expressed with GABAsynthesizing enzymes, Gad2 and Gad1. On the other hand, Pomc, Pcsk1, and Nat1 (an enzyme responsible for N-acetylation of α-MSH [40,41]) were localized in the third cluster (G3 gene group) (Fig 4).

Discussion
In this study, we provide cellular evidence for molecular and neurochemical heterogeneity of cholinergic neurons in the ARC with an approach by combining single-cell real time quantitative PCR and single-cell whole transcriptome amplification methods. We found that the ARC contained cholinergic neurons. These cholinergic neurons were neurochemically distinct as they had the ability to synthesize and release diverse peptides, amines, and neurotransmitters. Approximately 70% of these cholinergic neurons expressed the Th gene. Th-positive cholinergic neurons contained Gad65/67 and vesicular GABA transporter transcripts, consistent with the recent study showing co-expression of dopamine and GABA in neurons in the ARC [38]. In addition, a subset of GABA-expressing cholinergic neurons had vesicular ACh as well as glutamate transporter transcripts. Gene clustering analysis further showed that membrane receptor transcripts clustered with their respective intracellular signaling and downstream targets.
The cholinoceptive system in the ARC plays a role in regulating food intake [11][12][13]. Activation of nicotinic acetylcholine receptors reduces energy intake via modulation of melanocortinergic neurons in the ARC [13,14]. Both POMC and NPY neurons in the ARC are innervated by cholinergic input [14], implying that endogenous ACh can influence the activity of these neurons. In fact, mice lacking the M3 muscarinic ACh receptor consume less food and have lower body weights [12,44]. The M3 receptor is expressed in the ARC and oppositely regulates Pomc and Agrp mRNA expression [44]. Hence, it appears that stimulation of hypothalamic nicotinic and muscarinic receptor-mediated components results in opposing physiological responses in part through the melanocortinergic system. It has been reported that a small subset of POMC neurons expressed ChAT and vAChT in rats [17]. Our single-cell qRT-PCR and immunohistochemical studies of ARC neurons also revealed the co-expression of POMC and ChAT in mice, suggesting that there are local cholinergic neurons in the ARC. Therefore, this novel population of neurons may be part of the neural circuitries that detect homeostatic need for food and control the drive to eat.
Most hypothalamic neurons contain multiple neuropeptides and/or neurotransmitters [45]. Within the ARC, Npy/AgRP neurons have GABA as a fast neurotransmitter and GABA released from Npy/AgRP neurons is involved in ghrelin-mediated feeding [46]. It has been also shown that AgRP, Npy, and GABA in these neurons play temporally distinct roles in driving food intake [47]. Of particular interest is that α-MSH and β-endorphin derived from the same precursor POMC have opposite effects on feeding. For instance, β-endorphin released from the same population of POMC neurons promotes feeding [48], while α-MSH release reduces food intake [49]. These findings suggest that, although the expression of distinct peptides and neurotransmitters in the same population appears to be a paradox, these transmitters and peptides could independently but cooperatively influence feeding. Likewise, multiple neurotransmitters and peptides expressed in cholinergic neurons can contribute to the control of feeding behavior.
Although single-cell gene expression profiling showed heterogeneity in gene expression within the same population, gene expression patterns in individual cholinergic cells appear to be highly regulated in a cell-specific manner. For instance, TRPC2, TRPC5 and TRPC7 channels that are a downstream target of leptin and insulin receptors [27,28] were detected in cholinergic neurons with leptin and insulin receptors. Apelin-sensitive KCNQ potassium channels [18] were found in apelin receptor-expressing neurons. The muscarinic ACh receptor type 1 was also co-detected with its downstream target KCNQ channels [39]. In line with these findings, the membrane receptors also clustered with their respective signaling pathways (i.e. apelin receptor-PLCβ, [18]; leptin receptor-PLCγ [27]; Insulin receptor-PI3K 110α [28], mAChR1 and 3-PLC-IP3R [42]). In addition, genes that are functionally linked to one another were clustered in the same group (i.e. AMPK α and UCP2 [32,43]. All UCP2-expressing cholinergic POMC neurons co-expressed K ATP channels, which require for glucose-sensing in POMC neurons [32]. The experimental methodology described in this study has some technical advantages over Fluorescence Activated Cell Sorting (FACS) technique. First, our single-cell whole transcriptome amplification of total cytoplasmic RNA from individual neurons can yield approximately 20 μg cDNA /μl. This is larger than traditional single cell RT-PCR protocol practiced previously [50] and is a sufficient amount to quantify transcript levels of thousands of genes for each cell. It should be noted that the amounts produced are sufficient for whole transcriptome profiling by methods like RNASeq. Second, this method has the ability to determine gene expression profiles of retrogradely and/or anterogradely labeled neurons as the small number of fluorescently labeled neurons is sufficient. Third, this approach may be useful for discovering therapeutic targets. In fact, many drugs are designed to disrupt receptor functions, specific enzyme isoforms, or their respective downstream targets in order to impact disease relevant molecular pathways only. These targeted therapies may require identification of specific disease biomarkers at a single cell level.
Our present study demonstrates a novel population of cholinergic neurons in the ARC. Individual cholinergic neurons were neurochemically heterogeneous as they expressed enzymes responsible for the synthesis and release of GABA, glutamate, catecholamines, POMC-derived peptides as well as other neuropeptides. Despite this heterogeneity, gene expression patterns in individual cholinergic cells appear to be highly regulated in a cellspecific manner. In fact, membrane receptor transcripts clustered with their respective intracellular signaling and downstream targets. Therefore, our present study has a potential in establishing links between the molecular identities at a neuronal level and their physiological functions within a neural network.
Supporting Information S1