Single Neuronal and Glial Gene Expression in the Nucleus Tractus Solitarius in an Alcohol Withdrawal Time Series Reveals Novel Cellular Phenotypes and Networks

Alcohol withdrawal syndrome (AWS) is characterized by neuronal hyperexcitability, autonomic dysregulation, and severe negative emotion. The nucleus tractus solitarius (NTS) is largely responsible for the neurological processes underlying these symptoms as it is the main viscerosensory nucleus in the brain. The NTS receives visceral interoceptive inputs, inuences autonomic outputs, and has strong connections to the limbic system and hypothalamic-pituitary-adrenal axis to maintain homeostasis. Our prior analysis of single neuronal gene expression data from the NTS shows that neurons exist in heterogeneous transcriptional states that form distinct functional subphenotypes. Our working model conjectures that chronic ethanol induces a state of allostasis in which NTS neurons and peripheral tissues generally compensate for the persistent presence of alcohol, and that abrupt abstinence causes central and peripheral biological network decompensation resulting in the observed AWS. We collected single noradrenergic and glucagon-like peptide-1 (GLP-1) neurons and microglia from rat NTS and measured a subset of their transcriptome in an alcohol withdrawal time series. Inammatory subphenotypes predominate at certain time points, and GLP-1 subphenotypes demonstrated hyperexcitability post-withdrawal. We hypothesize such inammatory and anxiogenic signaling contributes to alcohol dependence via negative reinforcement. Targets to mitigate such dysregulation and treat dependence can be identied from this dataset. that


Introduction
Alcohol withdrawal syndrome (AWS) is characterized by negative physical and emotional symptoms. Physical symptoms are driven by autonomic dysregulation, γ-aminobutyric acid (GABA) hypoactivity, and increased glutamatergic signaling leading to dysphoria, nausea, diaphoresis, tachycardia, hypertension, seizures, and delirium tremens 1 . Fear and anxiety are the principle emotional symptoms. The negative reinforcement hypothesis of substance dependence postulates that these negative physical and emotional symptoms experienced in withdrawal motivate alcohol dependence [2][3][4][5] . We conjecture that peripheral network decompensation is a central facet of this model, and that neurovisceral feedback via the vagus nerve conveying peripheral information to the central nervous system contributes substantially to the severity of symptoms experienced 6 ( Fig. 1A-B). Investigation into the underlying mechanisms producing these symptoms may provide insight into targets that mitigate acute and protracted AWS severity and prevent relapse following abstinence. Such treatments may provide clinical utility for other substances of abuse with severe withdrawal syndromes such as opioids.
Neuroin ammatory processes have emerged as an important contributor to the severity of AWS symptoms, especially in the amygdala 6-14 . The amygdala is strongly implicated in threat detection and negative emotion, and in ammation here may drive some of the fear and anxiety experienced in AWS 6, 15,16 . The likely mechanism underlying this phenomenon is that in ammation causes neuronal hyperexcitability 17 .
The nucleus tractus solitarius (NTS) is another brain region that contributes to the symptoms of AWS and is implicated in alcohol dependence [18][19][20][21][22] . The NTS receives visceral inputs from the interoceptive vagal circuit, strongly in uences autonomic outputs, and has strong bidirectional connections to the amygdala (Fig. 1A). Connections to the paraventricular nucleus, ventrolateral medulla, and amygdala place the NTS in the center of the visceral-emotional neuraxis (Fig. 1B) 6,23,24 . Many of these connections use the neuropeptide glucagon-like peptide-1 (GLP-1) as a transmitter. Indeed, interoceptive vagal afferents synapse onto GLP-1 positive (+) NTS neurons that go on to form anxiogenic synapses [25][26][27][28][29] . Recently, GLP-1R activity in the NTS has been linked to alcohol-mediated behavior 30 . Additionally, the NTS houses noradrenergic (NE) neurons that also respond to vagal and higher-order inputs and principally function to maintain cardiovascular homeostasis (Fig. 1B). These NE + neurons also project to the amygdala where they contribute to emotional memory 24,31−33 . We conjecture a model in which GLP-1 and NE neurotransmission form parallel complementary pathways conveying the peripheral state via NTS to the limbic forebrain ( Fig. 1A-B).
Given their function in autonomic homeostasis and emotion, NE + and GLP-1 + neurons in the NTS likely contribute to the negative physical and emotional symptoms of AWS and negative reinforcement motivating alcohol dependence in overlapping and distinct ways. Further, in ammatory glial-neuronal signaling in the NTS during AWS may also contribute to the severity of physical and emotional withdrawal symptoms 6, 9,10 . Local in ammatory signaling in the NTS contributes to the development of hypertension in rats implicating paracrine cytokine involvement in AWS 34,35 . Moreover, the antineuroin ammatory molecule ibudilast is in clinical trials to reduce alcohol craving and AWS severity 36 .
Here, we measured how the functional states of single NE neurons, GLP-1 neurons, and microglia in the NTS change over the course of alcohol withdrawal. Single-cell approaches allow for the identi cation of cellular subphenotypes-morphologically indistinguishable cells anatomically localized that use the same primary neurotransmitter yet have distinct transcriptomic pro les. Our previous work has demonstrated the heterogeneity of single-cells, and the functional importance of subphenotypes that may be missed in tissue-level approaches 37,38 . Changes in transcription during alcohol withdrawal revealed a pattern suggesting peak dysregulation and in ammation at the 32 hour (hr) withdrawal (wd) time point.
Additionally, the expression of GABA A receptor (R) subunits genes was downregulated in protracted withdrawal, measured as the 176 hr wd time point, suggesting hyperexcitability of anxiogenic GLP-1 + neurons.

Results
We used laser capture microdissection (LCM) to gather single cells from rat NTS in control, chronic ethanol (EtOH), 8 hour (hr) withdrawal (wd), 32 hr wd, or 176 hr wd treatments (Sup. Figure 1A). We subsequently used micro uidic RT-qPCR to measure a subset of the transcriptome in these single-cells (Sup. Figure 1B) 39 . Following strict quality control protocols, a total of 229 10-cell pooled samples (700 NE neurons, 650 GLP-1 neurons, and 940 microglia) and 65 gene transcripts were used for data analysis (Fig. 1C, Sup. Table 1-2). We targeted gene transcripts involved in in ammatory glial-neuronal signaling and GABA A R subunits. Cellular phenotype selection was validated by the expression of the cell type markers NeuN, Cd34, and Cx3cr1 (Fig. 1D). Neurons were selected based on NEUN and TH immuno uorescence. TH-neurons demonstrated signi cantly elevated levels of the Gcg transcript that codes for the GLP-1 precursor protein suggesting these samples constitute GLP-1 enriched neurons (Fig. 1D). A dimensionality reduction analysis (linear discriminate analysis) further demonstrated the differences between the three cell types gathered. Microglia differed from neurons along the x-axis, NE neurons and GLP-1 enriched neurons differed from each other along the y-axis (Fig. 1E).
The expression of the neurotransmitter precursor genes Th and Gcg for NE and GLP-1 neurons, respectively, is plotted across treatment time points in Fig. 2A and B. We nd that expression of these transcripts is inversely correlated-at time points in which Th expression is relatively high in NE neurons, Gcg expression is relatively low in GLP-1 neurons, and vice versa ( Fig. 2A-B). This suggests a push-pull dynamic mechanism in the genetic regulation of neurotransmission by these neurons. Notably, Gcg expression was induced in the three withdrawal time points, but suppressed in control and EtOH treatments suggesting GLP-1 neurotransmission is pathologically elevated during the withdrawal process. At the 176 hr wd time point, bimodal distribution of Th expression in NE neurons and trimodal Gcg expression in GLP-1 neurons is observed ( Fig. 2A-B). High and low Th-expressing NE neurons and Gcg-expressing GLP-1 neurons from this time point were separated into heat maps organized by Euclidian distance clustering of gene expression ( Fig. 2C-D). Th expression in NE neurons and Gcg expression in GLP-1 neurons is moderately predictive of cellular subphenotypes that loosely organize coexpression gene clusters. However, Th-expression and Gcg-expression alone, which is to say neurotransmitter expression, is not the best determinate of cellular subsphenotypes-a nding we have observed previously in other neuronal nuclei 38 .
Heat maps establishing well-de ned data-driven cellular subphenotypes in neurons and microglia were generated using Euclidean distance clustering of z-scores of the -ΔΔC t values for each sample and gene in the dataset (Figs. 3,4,5). Co-expression gene clusters are labeled with numbers and cellular subphenotype groupings are labeled with letters. GLP-1 neurons had the same cellular subphenotypes with the same gene clusters across all treatments while NE neurons and microglia had two co-expression con gurations comprising the identi ed subphenotypes. The proportion of cells constituting a subphenotype in addition to gene cluster expression levels shifted with the treatment. GABA A R subunit genes clustered together in every con guration, and their expression was largely indicative of subphenotype groupings.
Two prominent subphenotypes, C and D, emerged in NE neurons in withdrawal time points.
Subphenotype C highly expressed gene cluster 4 which is rich in in ammatory ligands and receptors including Crh, Il1b, and Ptgs2. This was labeled the 'in ammatory' gene cluster (see below). Subphenotype C suppressed gene cluster 5, which includes Th and GABA A R subunits, and 6. Subphenotype D, conversely, had the opposite expression pattern in these gene clusters. At 8 hr wd, subphenotype C was predominant, but subphenotype D made up a higher proportion of cells at 32 and 176 hr wd. Another subphenotype, E, emerged in the 32 hr wd treatment that had moderately high expression of gene clusters 4 and 5. Subphenotypes D and E were further split into D1, D2, E1, and E2 based on medium or high expression, respectively, of gene cluster 6 which includes Cd200, cFos, and Mif. NE neurons from control and EtOH treatments shared gene clusters distinct from the withdrawal treatments. GABA A R subunits were co-expressed consistent with all sampled cell types. GABA A R subunit genes showed elevated expression in subphenotype B in control and EtOH treatments, moderate expression in subphenotype D and E in 8 and 32 hr wd treatments, and returned to high expression in 176 hr wd subphenotype D. In subphenotype E, only found at 32 hr wd, all assayed genes were at least moderately expressed suggesting the regulatory expression mechanisms that control this transcriptomic pro le are stimulated to this phase of the withdrawal process.
Co-expression genes clusters in GLP-1 neurons were consistent throughout the time series. Subphenotype A highly expressed 'in ammatory' gene cluster 1 rich in cytokine and chemokine ligands and receptors including Crh, Il1b, and Ptgs2, while suppressing 'GABA A R' gene cluster 2. Subphenotype B had the opposite pattern. GLP-1 co-expression clusters 1 and 2 were surprisingly similar to NE co-expression clusters 4 and 5, again suggesting the mechanisms of regulatory constraint are shared between these phenotypes. Interestingly, GLP-1 subphenotype B emerged only in the EtOH treatment. At 8 hr wd, subphenotypes A and B suppressed expression of their high-expressing gene clusters, 1 and 2 respectively, compared to control. At 32 and 176 hr wd, subphenotype A gene cluster 1 was more highly expressed than in the control condition. Subphenotype B gene demonstrated a steady decrease in expression of gene cluster 2, and by the 176 hr wd time had only moderate expression of 'GABA A R' gene cluster 2. Concurrently, expression of gene cluster 3 for subphenotype B consistently increases throughout the withdrawal process and by 176 hr wd is the most prominently upregulated gene cluster. Tnf did not group into any gene cluster and is isolated in GLP-1 neuronal heat map to display this clearly ( Fig. 4).
Microglia shared co-expression clusters in subphenotypes A, B, and C for control, 32 hr wd, and 176 hr wd treatments. EtOH and 8 hr wd treatments shared co-expression clusters in subphenotypes D and E.
Subphenotype A was the exclusive expression pattern for the 32 hr wd treatment which upregulated 'in ammatory' gene cluster 1 including Crh, Il1b, and Ptgs2. This subphenotype A is most similar to the so-called M1 phenotype 40 . At 176 hr wd, subphenotype A made up a low proportion of samples while subphenotype C moderately expressed all genes assayed and more robustly than in the control treatment. Tnf did not group into a gene cluster for microglia either. High expression of GABA A Rs and a few other genes including Sod1, Cd200, Mapk1, and Stat3 characterized Subphenotype E in the EtOH and 8 hr wd treatments.
These data are further displayed in cellular diagrams in which expression for each gene in the corresponding subphenoytpe is designated by a box that is colored by a scale represent the average zscore of -ΔΔC t expression values (Figs. 6,7,8). The location of this box in the cellular cartoon corresponds to the protein function of that gene. These diagrams provide a higher-level display of the functional state of the subphenotype at that time point. Many dynamic processes can be observed in these gures. In brief, Fig. 6 shows a clear upregulation of GABA A R genes at 176 hr wd as compared to 8 hr wd in Group C and D NE neurons. Additionally, CD200 expression is one of the primary distinguishers of Group D1 vs. GroupD2. Figure 7 displaying GLP-1 neurons shows that GABA A R gene expression at 176 hr wd is decreased in both subphenotypes. At the 32 hr wd time point, Cxcl10, Cxcr1, Cxcr2, and Cxcr3 expression distinguish Group A1 most prominently from Group A2. Microglia displayed in Fig. 8 showed the most Tnf expression in groups D and E, subphenotypes only identi ed at the EtOH and 8 hr wd treatments. Cx3cr1, a mircoglia gene prominently involved in neuronal adhesion, showed increased expression at the 176 hr wd time point in all subphenotypes 41 . GLP-1 Group B neurons had the most increased Cx3cl1 expression suggesting this subphenotype interacts most with microglia at this time point.

Discussion
NTS neurons regulate emotion, autonomic homeostasis, and stress responses. Multiple neuronal nuclei, ligands, receptors, and signaling dynamics are involved in these complex functions including NE, GLP-1, CRH, and GABA ( Fig. 1A-B). Moreover, local glial-neuronal paracrine signaling via in ammatory cytokines like tumor necrosis factor-alpha (TNF-α) also play a role. We microdissected single Th + neurons, Thneurons, and microglia from the rat NTS as 10-cell pooled samples using LCM and measured their expression of 96 gene transcripts in an alcohol withdrawal time series (Sup. Figure 1). Time points were chosen based on rat alcohol metabolism 42 . 8 hr wd represents the start of acute AWS, 32 hr wd represents the end of acute AWS, and 176 hr wd represents a protracted withdrawal state. We found that neurons that stained Th + had signi cantly elevated Th expression and labeled them NE neurons (Fig. 1D). Th-neurons had signi cantly elevated expression of the GLP-1 precursor transcript Gcg, and were labeled as neuronal samples enriched with GLP-1 + neurons. Likewise, CD11β + cells expressed the microglial markers Cd34 and Cx3cr1 at signi cantly elevated levels and were labeled microglia. In a dimension reduction analysis (LDA), these three cell types formed distinct clusters with microglia separating out from neurons along the x-axis and NE and GLP-1 neurons separating along the y-axis (Fig. 1E).
Further analysis of Th and Gcg expression showed an inverse relationship with respect to time point with Gcg expression demonstrating elevated expression levels only during withdrawal ( Fig. 2A-B). However, expression of these neurotransmitter precursor genes did not organize the other genes assayed into distinct subphenotypes correlated to their expression levels ( Fig. 2C-D). A data-driven approach to cellular subphenotype organization identi ed stark subphenotypes unique to each cell type likely with discrete functions (Figs. [3][4][5]. Strikingly, these subphenotypes shared similarities in their expression of their in ammatory gene clusters (Sup. Table 4).
Gene cluster 4 in NE neurons, gene cluster 1 in GLP-1 neurons, and gene cluster 1 in microglia constituted these 'in ammatory' clusters (Sup. Table 4). 18 genes were shared across all of these co-expression clusters and only 5 genes were unique to a single cluster suggesting similar mechanisms across cell types that regulate their expression. In NE neurons, subphenotype C highly expressed this in ammatory cluster while subphenotype E had moderate in ammatory co-expression cluster elevation. At 8 hr wd, NE subphenotype C was 62.5% of the samples (5/8) and at 32 hr wd C and E combined to 62.2% of the samples (23/37). By 176 hr wd, subphenotype C was only 29% of NE neuron samples (5/17). This suggests that this subphenotype of NE neurons experiences a marked increase in in ammation during acute AWS, but that this subphenotype is not involved in protracted withdrawal symptoms such as lowgrade anxiety 12 . This increase in local paracrine in ammation likely increases the excitability for this NE neuron subphenotype 17 . This is consistent with the observation of hypersympathetic activity in acute, but not prolonged, AWS 1 .
In GLP-1 neurons, subphenotype A highly expressed the 'in ammatory' gene cluster (gene cluster 1). The We expected neuroin ammatory markers to be increased at the 176 hr wd time point, especially in microglia, but these data suggest that the NTS experiences in ammation during acute withdrawal only (8 hr and 32 hr time points) and recovers by the 176 hr time point. We observe this in all three cell types assayed and speculate that compensatory endogenous anti-in ammatory signaling may be driving this observation, though we cannot substantiate this claim with the genes measured in this study.
The 176 hr wd time point is meant to measure long term changes in gene expression that occur in protracted withdrawal. At this time point, some similarities across cell types were observed in the subphenotypes that highly expressed GABAR subunits as was observed at other time points. NE neuron cluster 5, GLP-1 neuron cluster 2 and microglia cluster 2 contained the majority of the GABAR subunit genes, and the makeup of this 'GABAR' co-expression cluster was not as consistent as the in ammatory cluster across cell types-16 genes are shared across all cell types and 16 genes are unique to a single cell type within its respective GABAR cluster (Sup. Table 4). GLP-1 neurons in subphenotype B upregulates this co-expression cluster in the control treatment, but the relative level of expression of this cluster decreases throughout the time series within this subphenotype (Fig. 4). At the 176 hr wd time point, this GABAR cluster is only moderately expressed suggesting long term changes to this neuronal subphenotype as a result of alcohol dependence and withdrawal. The decrease in expression of inhibitory GABAR gene transcripts, along with the concurrent upregulation of co-expression cluster 3, which contains Gcg, suggests that this GLP-1 neuronal subphenotype experiences long term functional changes such that its neurotransmission increases. Literature indicates that GLP-1 signaling from the NTS to the amygdala and other nuclei is anxiogenic 29 . Taken together, these data suggest that this GLP-1 neuronal subphenotype is not primarily involved the acute withdrawal process characterized by in ammation, but rather experiences GABAR subunit downregulation over a longer process potentially leading to increased anxiety and susceptibility to stress in protracted AWS.
Microglia also showed elevated GABAR expression at the 176 hr wd time point, but the pattern of increased GABAR expression was unexpected. Control microglia in subphenotype C show moderate expression of both cluster 1 (in ammatory) and cluster 2 (GABA A R) (Fig. 5). Expression of both of these clusters increases at the 176 hr wd time point. This suggests elevated in ammation, but not by distinct M1 phenotype microglia (subphenotype A), and also elevated GABAR expression. These ndings are best visualized in Fig. 8. Of note, there many genes in microglia cluster 2 that are not GABAR subunits. Moreover, microglial Tnf expression was signi cantly elevated in control, EtOH and 8 hr wd treatments compared to 176 hr wd independent of subphenotype (Sup. Table 3). Indeed, Tnf expression by microglia did not t neatly into a gene cluster. Cluster C has some cells that demonstrate high tnf in both control and 176 hr wd, where Cluster A showed a decrease in Tnf expression between these two time points. Cluster B, conversely, increased its expression of Tnf from control to 176 hr wd. This apparent absence of a pattern in microglia Tnf expression suggests that in microglia this gene that is central to neuroin ammation is constrained by a mechanism that is independent of other gene expression regulation. Further, the decrease in overall microglia Tnf expression at 176 hr wd as measured by an average of -ΔΔC t values and two-tailed heteroscedastic t-tests may be misleading. A single-cell analysis reveals that overall expression may not be the best indicator of in ammation. Rather, shifts in subphenotype proportion, and the number of cells showing a moderately increased Tnf expression, as seen in subphenotype C, may have more of a physiologic impact than total gene expression levels.
Cell diagrams in Figs. 6,7,8 average the expression of a gene within a subphenotype designated by color and display that color in a location on the cell cartoon that corresponds to the protein function. This method of data presentation allows for analysis of receptor-ligand interactions within and between subphenotypes. For example, Fig. 6 suggests that subphenotype C experiences an increase in both CCL-CCR and CXCL10-CXCR signaling at 32 hr wd as the ligand and receptor genes for these proteins are highly expressed. Figure 6 also provides clarity in subphenotype D upregulation of Mapk1 at 176 hr wd suggesting long term transcription is altered during protracted withdrawal in this subphenotype. Moreover, transcription factor genes cFos, Junb, NfkB, and Stat3 have increased expression in subphenotype D2 suggesting this subset of NE neurons undergoes substantial long-term changes in transcription following alcohol withdrawal. Microglia in subphenotype C upregulate IL1a, IL1b, and IL1r1 at 176 hr wd in subphenotype C as compared to control, while subphenotype B downregulate these genes at 176 hr wd compared to control (Fig. 8). This dynamic provides indirect evidence that subphenotype B provides an anti-in ammatory function that is most active in protractracted withdrawal. Moreover, it suggests that microglia subphenotype C, identi ed here as a microglia subset that can function in a multitude of processes whether in ammatory or anti-in ammatory based on their lack of a clear coexpression module pattern in control, is pushed towards an in ammatory state in protracted withdrawal. This dataset has allowed the identi cation of cellular subphenotypes and their gene expression dynamics in alcohol withdrawal through time. Analysis has revealed valuable insights in both neurotransmission signaling and local paracrine signaling processes that relate to what is observed clinically in the context of what is already established about such neurotransmission. The dataset is unique in that micro uid RT-qPCR, a method lower in throughput but more reliable than RNA-seq 46 , is combined with anatomic and staining speci city using LCM for single-cell selection in a time series. This allows for analysis of complex signaling dynamics at multiple levels, and the in uence of such signaling dynamics on both acute AWS and protracted withdrawal based on the clinical symptoms at that time point.
We have collected the data, validated the accuracy of the dataset, and identi ed cellular subphenotypes and their major signaling dynamics. However, signaling dynamics measured in our dataset can be further investigated and may identify clinical targets to treat acute or protracted AWS and potentially alcohol dependence itself. Future studies analyzing these signaling dynamics with the addition of female rats that also include other brain cell types such as astrocytes and endothelial cells are needed to further understand the underlying pathophysiology of AWS and dependence.
Lastly, these ndings are consistent with the hypothesis that neuroin ammation in the visceral-emotional neuraxis contributes to antireward which motivates alcohol, and opioid, dependence (Fig. 1A-B; Sup. Figure 7) 6 . In brief, this hypothesis suggests that neuroin ammation in the NTS and amygdala stimulates antireward which contributes to negative reinforcement. This study does not only provides evidence of neuroin ammation in the NTS in acute and protracted alcohol withdrawal, but also an understanding of the emergence of this neuroin ammation and its relation to neurotransmission and AWS. Improved understanding of such processes in alcohol withdrawal lends insights into targets that may mitigate in ammation, decrease antireward in AWS, and treat substance dependence.

Animals
Approval of protocols was given by Institutional Animal Care and Use Committee of Thomas Jefferson University. The study was carried out in compliance with ARRIVE guidelines and in accordance with all relevant guidelines and regulations. Ten young, male, Sprague Dawley rats (35-45 grams) ordered from Harlan Laboratory were housed individually in the Thomas Jefferson University Alcohol Research Center Animal Core Facility. Standard chow and water was given until rats weighed 120 grams. Rats were then fed an alcohol-free, maltose-dextrin substituted, Lieber DeCarli liquid diet (bioServe, Frenchtown, NJ) for three days 47 . Animals were then assigned to ve treatment groups: control, chronic alcohol exposure (EtOH), 8 hour (hr) withdrawal (wd), 32 hr wd, or 176 hr wd (Sup. Fig. 1). Animals received eight months of ethanol or control diet. Control diet animals received a quantity of the liquid diet that equaled the caloric intake of the matched alcohol-fed animal 24-hours prior. Withdrawal animals were withdrawn such that sacri ce by rapid decapitation was at the same circadian time for all conditions. Single-cell gene expression variance within an animal demonstrated similar or greater variance as that between animals (Sup. Fig. 2).
Single cell sampling and High-Throughput qRT-PCT 3230 single brain cells, 950 Th+ neurons, 1030 Th-neurons, and 1250 microglia, were collected from the NTS using LCM. Cells were grouped into 10-cell pools comprising 323 total samples analyzed. This pooling of cells increases the number of samples analyzed by the micro uidic RT-qPCR platform. cDNA from mRNA transcripts was generated by reverse transcription (SuperScript™ VILO™ cDNA Synthesis Kit; ThermoFisher). TaqMan PreAmp Master Mix was used for pre-ampli cation of cDNA (22 cycles) with forward and reverse PCR primers (96 pairs). The Biomark micro uidic qPCR platform (Fluidigm©) was used to measure expression levels of 96 genes. Four batches of probe-based qPCR measured the previously ampli ed 96 cDNA transcripts. Supplemental Table 1 lists primers used. Primer amplicon validation was performed on agarose gel electrophoresis. Following strict quality control protocols, a total of 229 10-cell pooled samples (70 NE neuron samples, 65 GLP-1 neuron samples, and 94 microglial samples) and 65 gene transcripts were used for data analysis.
The four micro uidic RT-qPCR batches run for this study were assessed for intra-and inter-batch experimental quality (Sup. Fig. 3-4). Technical replicates assessing intra-batch quality demonstrated high similarity with r values listed (Sup. Fig. 3). Inter-batch replicates demonstrated high batch similarity, though batch 4 sample 40 showed contamination (Sup. Fig. 4). A dilution series using standard rat brain RNA was also included in each batch for quantitative analysis (Sup. Table 2); However, the data normalization method explained below calculated relative expression and was used for all analysis in this study.

Data Normalization
A two-step median-centering -∆∆C t method was used for expression level normalization was explained elsewhere 48 . Brie y, a raw C t value was obtained for each gene and sample. Each individual C t value was normalized to the overall sample median ((Median sample expression) -C t gene = -∆C t sample ). The newly obtained -∆C t values were then median-centered to the gene across all samples (-∆C t sample -(Across sample -∆C t median) = -∆∆C t gene ). This yields a -∆∆C t value for each measurement allowing comparison of relative gene expression values across treatment groups and batches. This analysis was carried out in R version 3.5.2. The raw C t values are listed in Supplemental Table 2 without gene quality control. The normalized dataset with quality-control that was used for all analysis is also displayed in Supplemental Table 2       Legend with grey boxes in center of gure labels which boxes correspond to which gene. Box color represents expression (blue is low expression and yellow is high expression). The location of the box represents the localization or function of the protein product from that gene transcript.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.