Single-nucleus RNA-seq identifies one galanin neuronal subtype in mouse preoptic hypothalamus activated during recovery from sleep deprivation

SUMMARY

The preoptic area of the hypothalamus (POA) is essential for sleep regulation. However, the cellular makeup of the POA is heterogeneous, and the molecular identities of the sleep-promoting cells remain elusive. To address this question, this study compares mice during recovery sleep following sleep deprivation to mice allowed extended sleep. Single-nucleus RNA sequencing (single-nucleus RNA-seq) identifies one galanin inhibitory neuronal subtype that shows upregulation of rapid and delayed activity-regulated genes during recovery sleep. This cell type expresses higher levels of growth hormone receptor and lower levels of estrogen receptor compared to other galanin subtypes. single-nucleus RNA-seq also reveals cell-type-specific upregulation of purinergic receptor (P2ry14) and serotonin receptor (Htr2a) during recovery sleep in this neuronal subtype, suggesting possible mechanisms for sleep regulation. Studies with RNAscope validate the single-nucleus RNA-seq findings. Thus, the combined use of single-nucleus RNA-seq and activity-regulated genes identifies a neuronal subtype functionally involved in sleep regulation.

In brief

Using single-nucleus RNA-seq coupled with activity-regulated genes, Guo et al. identify one galanin neuronal subtype in the preoptic area activated during recovery sleep and show cell-type-specific increases in Htr2a and P2ry14. RNAscope studies using identified marker genes reveal the anatomical location of this galanin subtype as the ventrolateral preoptic area.

Graphical Abstract

INTRODUCTION

One of the unanswered questions in neurobiology is, “What is the mechanism for sleep homeostasis, i.e., the drive to sleep?”¹ Multiple mechanisms affecting sleep homeostasis have been proposed, including changing levels of somnogens,² Ca²⁺ signaling,³ double-stranded DNA breaks,⁴ and transcriptional and translational changes in specific sleep/wake circuits.^5,6 The preoptic area of the hypothalamus (POA) contains sleep-active GABAergic neurons that send inhibitory input to the arousal system to promote sleep (as reviewed by Saper et al.⁷). However, the exact makeup of this sleep-promoting signal and mechanism is not fully understood. There is evidence that the neuropeptide galanin plays an important role in homeostatic sleep promotion.^8–11 Galanin is primarily co-expressed with γ-aminobutyric acid (GABA) in inhibitory neurons.⁸ However, one study using single-cell RNA sequencing (single-cell RNA-seq) revealed multiple subpopulations of galanin neurons in POA, including both inhibitory and excitatory cell types.¹² Which subtypes of galanin neurons are involved in sleep regulation remains to be elucidated. Moreover, we do not know what transcriptional changes take place in these neurons under high sleep pressure. To address these questions, we used an unbiased discovery strategy based on single-nucleus RNA sequencing (single-nucleus RNA-seq). Extending and complementing the previous single-cell RNA-seq study that identified different subsets of neurons in POA,¹² we investigate the transcriptional changes in POA affected by sleep homeostasis at the single-nucleus level. single-nucleus RNA-seq can be applied to frozen tissues and allows more faithful capture of the transcriptional changes than single-cell RNA-seq.^13,14 Moreover, single-nucleus RNA-seq allows determination of which neuronal groups are activated by increased sleep pressure and understanding of differential expression of genes between mice with high and low sleep pressure in subsets of neurons, as well as in non-neuronal cells.

Homeostatic sleep pressure builds during wakefulness and dissipates as sleep progresses.¹⁵ One hallmark of sleep homeostasis is the more intense rebound sleep following extended wakefulness (i.e., sleep deprivation), characterized by increased slow-wave activity during non-rapid eye movement (NREM) sleep in mammals.¹⁶ Since sleep-promoting neurons in POA are activated under increased sleep pressure,¹⁷ we compared mice during rebound sleep after sleep deprivation (high sleep pressure) to mice after prolonged sleep (low sleep pressure); both groups were sacrificed at the same time of day to remove circadian influences. This is a different study design than previous studies evaluating transcriptional changes with sleep deprivation.^18–20 While previous studies have compared the transcriptome in animals awake and asleep, here we compared differences in the transcriptome in sleeping animals between those with high and low sleep pressure. In response to neuronal activation, neurons express many activity-regulated genes (ARGs), and different neuronal activity patterns induce different ARGs.²¹ A recently curated and comprehensive set of ARGs²² coupled with single-nucleus RNA-seq has successfully revealed cell-type-specific responses to neuronal activity in cortex using gene set enrichment analysis (GSEA).²³ We applied the same strategy here to unbiasedly map neurons activated by increased sleep pressure. Further, expression levels for all genes in these neurons and in all identified cell types were compared between mice under high and low sleep pressure. Lastly, multiplexed RNAscope fluorescent in situ hybridization was used to validate key findings and spatially characterize the identified cell groups.

Our single-nucleus RNA-seq results revealed ten major cell types in POA. With further characterization, we identified 43 inhibitory neuronal clusters and 25 excitatory neuronal clusters; these numbers are consistent with the previous single-cell RNA-seq study in POA.¹² Only one inhibitory subtype showed upregulation of ARGs under high sleep pressure, and this cell group is one of the three galanin-enriched inhibitory clusters. Molecular characterization of this galanin subpopulation revealed high expression of growth hormone receptor (Ghr) and low expression of estrogen receptor (Esr1) compared to other galanin cell groups. Spatial characterization of POA using the identified marker genes revealed that this galanin subpopulation was located in the ventrolateral preoptic area (VLPO). Differential gene expression analysis revealed specific upregulation of receptors for pyrimidine nucleotides and serotonin in only this specific inhibitory galanin subpopulation of VLPO that was activated under high sleep pressure. This provides insights into the cell-type-specific mechanism of sleep regulation by serotonin and pyrimidine nucleotides.

RESULTS

Generation of mice under high and low sleep pressure

A diagram describing the experimental setup and a flow chart explaining the process of selecting mice for the single-nucleus RNA-seq study are shown in Figures 1A and 1B, respectively. In brief, electroencephalographic/electromyographic (EEG/EMG) recording was performed in 2- to 3-month-old male and female C57BL/6J mice (n = 49 in total) for two 24-h baseline days (see STAR Methods). After the baseline recording, mice were randomly split into two groups. One group (n = 23) was allowed normal spontaneous sleep (SS) until sacrifice at the end of the light period (zeitgeber time 10.5 [ZT10.5]). The second group (n = 26) was subjected to 6 h of sleep deprivation (ZT4–10) by gentle handling²⁴ and allowed 30 min of recovery sleep (RS) before being sacrificed at the same time of day as the SS group (ZT10.5) (Figure 1A). On average, we observed no difference in plasma cortisol between the SS and RS mice (Figure 1C), suggesting no elevated stress in the RS condition. However, to further minimize potential influences of stress, two RS mice with the highest cortisol levels were excluded from the rest of the study. Five additional mice were removed due to poor EEG/EMG data quality, resulting in 20 SS mice and 22 RS mice for comparing delta power activities between conditions, as described below (Figure 1B). For each of these mice, relative delta power activity during NREM sleep in the last 30 min before sacrifice was calculated using the mean delta power activity in the last 4 h of the lights-on phase in the previous baseline day as a reference (lowest sleep pressure^15,16). As shown in Figure 1D, the relative delta power activities before sacrifice in the SS mice were close to or lower than 100% of baseline, indicating that the mice in the SS group were under low sleep pressure. The relative delta power activities before sacrifice in the RS mice were significantly higher than in the SS mice (mean relative delta power 137.0% vs. 88.1%; p = 1.1 × 10⁻⁸), indicating that mice in the RS group were collected under high sleep pressure, as designed. To maximize sleep pressure differences between mice in the two conditions, the ten RS mice (five male and five female) with highest relative delta power and ten SS mice (five male and five female) with lowest relative delta power were chosen for the final single-nucleus RNA-seq study (mean relative delta power 150.9% vs. 81.1%; p = 1.1 × 10⁻⁵; Figure 1F). In the chosen 20 mice, the cortisol levels remained non-elevated in the RS group compared to the SS group (one-sided p = 0.95) (Figure 1E).

Figure 1. Generation of mice under high and low sleep pressure.

(A) Experimental setup for the study. 2- to 3-month-old male and female C57BL/6J mice randomly split into two groups. One group (SS) was allowed normal spontaneous sleep until sacrifice near the end of the light period (ZT10.5). The second group (RS) was subjected to 6 h of sleep deprivation (ZT4–10) by gentle handling and allowed 30 min of recovery sleep before being sacrificed at the same time of day as the SS group (ZT10.5).

(B) Flow chart explaining the process of selection of mice for the single-nucleus RNA-seq study.

(C) Levels of plasma cortisol in all mice started in the experiment (n = 23 SS and n = 26 RS). Cortisol levels were not increased in RS mice compared to SS mice (one-sided Wilcoxon p = 0.6). To further minimize potential influences of stress, two mice in the RS group with cortisol >150 ng/mL were removed from the study.

(D) Levels of relative delta power (normalized to baseline) for mice after exclusions due to high cortisol or poor EEG quality, resulting in 20 SS and 22 RS mice remaining in the study. Mice during RS showed significantly higher delta power compared to mice during SS (mean relative delta power 137.0 vs. 88.1; two-sided Wilcoxon p = 1.1 × 10⁻⁸).

(E and F) Cortisol levels (E) and relative delta power (F) in the 20 mice used for the single-nucleus RNA-seq experiments that maximize sleep pressure differences between RS and SS mice (mean relative delta power 150.9 vs. 81.1; two-sided Wilcoxon p = 1.1 × 10⁻⁵). In the chosen 20 mice, cortisol remained not elevated in the RS mice compared to the SS mice (one-sided Wilcoxon p = 0.95).

Characterizing cell types and subtypes in POA using single-nucleus RNA-seq

The POA of the hypothalamus was dissected from frozen brain tissues between bregma +0.5 and −0.5 (region of dissection illustrated in Figure 2A; for other details see STAR Methods). Following pre-processing steps to remove low-quality nuclei and one outlier sample (see STAR Methods), our final single-nucleus RNA-seq dataset contained a total of 158,960 nuclei with a mean nuclei number of 8,366 ± 1,224 (mean ± SD) per animal. We used two rounds of clustering to first, separate the major neuronal (inhibitory and excitatory neurons) and non-neuronal cell types (glia, epithelial, and endothelial cells) and second, to further identify the cellular subgroups among inhibitory and excitatory neurons. In both clustering analyses, we leveraged bootstrapping approaches and repeated evaluation of the Jaccard index to optimize clustering parameters and identify stable and robust cell clusters²⁵ (see STAR Methods).

Figure 2. Characterization of preoptic area of hypothalamus using single-nucleus RNA-seq.

(A) Diagrams depicting dissection of preoptic area of hypothalamus (POA). Four consecutive brain slides (4 ×250 μm) were obtained between bregma +0.5 mm and −0.5 mm. Blue circles indicate the area of dissection used for single-nucleus RNA-seq. Landmarks used to identify POA are shown on the diagrams, including anterior commissure (ac), corpus callosum (cc), and lateral ventricle (lv).

(B) Uniform manifold approximation and projection plot of the 158,960 nuclei used for analysis reviewed ten broad categories of cell-type identities.

(C) Dot plot showing the expression of cell-type markers of the major cell types.

(D) Bar plot showing the proportions of major cell types in RS and SS samples, revealing overall even distribution of nuclei collected from mice under the two conditions. Number of biological replicates: n = 9 SS and n = 10 RS. One female SS sample was removed due to major differences in cellular makeup compared to the other samples (see STAR Methods).

See also Figures S1–S4.

The first round of clustering identified 55 cell clusters (Figure S1), which are annotated into ten major cell types (Figures 2B and 2C) based on the expression of known marker genes,^12,26 including inhibitory neurons (based on Gad1/Gad2 and Slc32a1), excitatory neurons (Slc17a6 and Slc17a8), astrocytes (Slc1a2 and Slc39a12), oligodendrocytes (Mbp and St18), oligodendrocyte precursor cells (OPCs; based on Vcan and Pdgfra), microglia (Selplg and C1qa), endothelial cells (Flt1), ependymal cells (Ccdc170 and Dnah12), vascular and leptomeningeal cells (VLMCs; based on Slc6a13), and arachnoid barrier cells (ABCs; based on Slc47a1). Based on comparisons of mean cell proportions of the cell types, we observed a comparable distribution of nuclei collected from mice in the RS and SS groups (Figure 2D). (The number of nuclei identified from each animal in each cluster is provided in Data S1). Given our goal of studying expression of ARG sets to identify cell clusters activated under high sleep pressure (RS) compared to low sleep pressure (SS), we also evaluated whether expression of ARGs (see gene panel in Data S2) influenced clustering outcomes. Cellular partitioning into clusters was not different before and after removing ARGs (Figure S2), indicating that these genes did not determine cell clusters.

In the second round of clustering, we identified 43 inhibitory neuronal subgroups (Figure S3) and 25 excitatory neuronal subgroups (Figure S4). These clusters are annotated with the initial letter “i” or “e” for inhibitory or excitatory neurons, respectively, followed by a number indicating the size of the clusters (with 0 indicating the largest cluster), with two exceptions. First, two clusters expressing choline acetyltransferase (ChAT), which is a marker for cholinergic neurons, were grouped with the inhibitory clusters, as in the previous single-cell RNA-seq study of POA,¹² and annotated with the initial letters “chat” to distinguish them from other inhibitory clusters. Second, two clusters expressing marker genes for both inhibitory neurons (Gad1) and excitatory neurons (Slc17a6 or Slc17a8) were grouped with the excitatory clusters and were annotated with the initial letter “h” to distinguish them from the other excitatory neurons. Expression profiles of selective marker genes are plotted for inhibitory clusters and excitatory clusters in Figures S3B and S4B, respectively.

Three inhibitory clusters were significantly enriched for Gal expression (i11, i12, and i14) but differed in their expression of hormonal receptors (Figure 3A). While prolactin receptors (Prlr) and androgen receptors (Ar) are highly expressed in all three galanin clusters, clusters i11 and i12 have high levels of estrogen receptors (Esr1), whereas i14 expresses high levels of growth hormone receptors (Ghr). Other marker genes for the three clusters are Stxbp6 for i11, Bcl11a for i12, and Gpr149 and Amigo2 for i14 (Figure S3). All three clusters (i11, i12, and i14) also express a second neuropeptide, Nts. Three other clusters (i20, e2, and e12) expressed low levels of Gal, although galanin was not significantly enriched as a marker gene in these clusters. All six of these galanin-positive clusters identified in our single-nucleus RNA-seq study had a corresponding single-cell RNA-seq cluster in the prior study from Moffitt et al.¹² (see correlation matrix in Figure 3B). The one other excitatory galanin cluster identified in the Moffitt et al.¹² study (sc.e22) did not show correlation with any of the galanin clusters identified here; this may be due to differences in POA dissection between our study and this prior study¹² (Figure S5).

Figure 3. Neuronal clusters expressing galanin identified using single-nucleus RNA-seq and comparison to the galanin clusters identified by Moffitt et al. using single-cell RNA-seq.

(A) Molecular marker gene expression profile of the Gal-expressing neuronal clusters.

(B) Correlation matrix showing similarities between galanin clusters identified in the current single-nucleus RNA-seq study of POA and the single-cell RNA-seq clusters from Moffitt et al.¹²

(C) Correlation matrix showing similarities between galanin clusters identified in the current single-nucleus RNA-seq study of POA and the MERFISH clusters from Moffitt et al.¹²

Pearson correlation was performed based on Z-score transformed averaged expression profiles for all clusters. Pearson correlation p values were corrected for multiple comparison using the Benjamini-Hochberg method. Only correlations with FDR < 0.05 are shown. See also Figures S3, S4, and S11; Tables S2 and S3.

Differentially expressed genes between recovery sleep and spontaneous sleep in cell types and neuronal subtypes of POA

To characterize the effects of homeostatic sleep pressure on gene expression regulation in POA, we performed differential gene expression analysis comparing the RS and SS groups for the ten major cell types, as well as the inhibitory and excitatory neuronal subtypes, using a negative binomial mixed-effects model (see STAR Methods). Throughout the text, we describe genes that show positive expression differences between RS and SS as “genes upregulated under RS” and genes that show negative expression differences between RS and SS as “genes upregulated under SS.” The results are summarized for each cell type in volcano plots (Figure S6) and in Data S3. Higher numbers of differentially regulated genes (false discovery rate [FDR] < 0.05) were identified in the four most abundant cell types: inhibitory neurons, excitatory neurons, oligodendrocytes, and astrocytes. For each, more genes were upregulated under SS (low sleep pressure) than were upregulated under RS (high sleep pressure). Specifically, among inhibitory neurons, increased expression was observed for 534 genes under low sleep pressure and 97 genes under high sleep pressure. For excitatory neurons, 165 and 71 genes had increased expression under low and high sleep pressure, respectively. In astrocytes, 674 and 224 genes had increased expression under low and high sleep pressure, respectively. Finally, among oligodendrocytes, increased expression was seen for 174 genes under low sleep pressure and 128 genes under high sleep pressure. Few genes were identified with FDR < 0.05 in the remaining less abundant cell types (e.g., OPCs, microglia, endothelial cells, ependymal cells, VLMCs, and ABCs) (Figure S6).

Identifying cell clusters responding to sleep pressure using activity-regulated gene sets

To identify cell clusters activated under high or low sleep pressure, we utilized a curated set of ARGs and performed GSEA. The ARG list contains three sets of genes—rapid primary response genes (rPRGs), delayed primary response genes (dPRGs), and secondary response genes (SRGs)—as well as a set of constantly expressed genes as negative controls.²² Distinct durations of neuronal activities induce different categories of ARGs^21,22; rPRGs (such as Fos and Egr1) respond to both brief (1–5 min) and sustained (minutes to hours) activities, dPRGs (such as Fosl2 and Nr4a3) respond to only sustained (minutes to hours) activities, and SRGs (such as Kcnh8 and Brinp1) respond to long-lasting (hours) activities. Gene expression changes between RS and SS conditions were examined for all cell types and neuronal subgroups using a negative binomial mixed model (see Data S4 and S5 for differential gene expression results in inhibitory and excitatory neuronal subgroups, respectively). Genes were then ranked based on −log₁₀ transformed p values (multiplied by the sign of the fold change to show direction of change), and statistical over-representation of the ARG sets was tested against the ranked gene list (see Figure 4 for enrichment plots). Positive enrichment indicates upregulation in RS, and negative enrichment indicates upregulation in SS.

Figure 4. Identification of cell clusters responding to sleep pressure using activity-regulated gene sets.

(A) Gene set enrichment analysis (GSEA) of ARGs in all inhibitory neuronal subtypes. Heatmap plots the FDR of the GSEA (−log₁₀ transformed) with positive numbers (shown in red) indicate significant upregulation, and negative numbers (shown in blue) indicate downregulation. FDR < 0.05 is considered significant.

(B) Enrichment plot showing positive enrichment of rPRG and dPRG in i14, and negative enrichment of rPRG in i2 and i5. Genes were ranked based on differential expression (−log₁₀ p value multiplied by sign of the log₂ fold change) between RS and SS mice with genes showing the most positive changes ranked at the top of the list (left in the plot) and genes showing the most negative changes ranked at the bottom of the list (right in the plot).

(C) Volcano plot showing expression changes of ARGs between RS and SS in i14, i2, and i5. The −log₁₀ converted p value from NEBULA was plotted against the log₂(fold change [FC]) between RS and SS. Genes from the rPRG and dPRG gene sets were upregulated in RS in i14, whereas genes from the rPRG gene set were downregulated in RS in i2 and i5.

rPRG, rapid PRG; dPRG, delayed PRG; PRG, primary response genes; SRG, secondary response genes; Control, unconditionally expressed genes. Number of biological replicates: n = 9 SS and n = 10 RS (after removing one female SS sample; see STAR Methods). See also Figure S7.

As shown in Figure 4A, only one inhibitory cluster, i14, showed significant positive enrichment of the ARGs in the RS condition (high sleep pressure). This cluster (i14) is one of the three Gal-enriched inhibitory clusters identified, and cells in i14 showed positive enrichment of both rPRGs (adjusted GSEA p = 5.21 × 10⁻⁷) and dPRGs (adjusted GSEA p = 2.42 × 10⁻⁹) (Figure 4B). We next examined which specific genes in the rPRG and dPRG sets were induced during RS in i14. Six rPRGs (Egr1, Fos, Fosb, Npas4, Nr4a1, and Dusp1) and four dPRGs (Fosl2, Hcrtr2, Nr4a3, and Dusp4) were upregulated in RS at an FDR of <0.05 (Figure 4C). No enrichment of the SRG set was identified, suggesting that i14 was not activated during the 6 h of sleep deprivation (see limitations of the study for further discussion). Although dPRGs are defined by being induced by sustained activities, it has been shown that an intermediately sustained stimulus of only 7 min was able to induce some genes in this gene set,²² which agrees with what we observed here. In summary, our results indicate that the Gal-enriched cluster i14 was activated during RS when sleep pressure was high. Gene expression changes within only the galanin-enriched clusters are provided in Data S6 to facilitate comparisons of i14 against other galanin clusters.

Two inhibitory Tac1-enriched cell clusters, i2 and i5, showed significant upregulation of rPRGs with low sleep pressure (adjusted GSEA p = 1.03 × 10⁻³ and p = 2.93 × 10⁻⁶, respectively; Figure 4B). Among these enriched genes, the strongest association was seen for Egr1 in i5 (Figure 4C). No enrichment of dPRGs or SRGs was identified, suggesting that i2 and i5 are activated by short-term stimulations specific to the SS condition.

No enrichment of any ARG sets with an FDR of <0.05 was identified in any of the excitatory subtypes or any of the non-neuronal cell types (Figure S7), suggesting that sleep-pressure-dependent activity changes occur selectively in inhibitory neurons. Also, no ARGs showed any sex-specific changes in any of the inhibitory and excitatory clusters (Data S4 and S5) tested using an interaction model (see STAR Methods).

Spatial characterization of the Gal cluster activated during recovery sleep

To validate and spatially characterize the identified Gal-enriched cluster i14, we performed in situ hybridization using the RNAscope HiPlex v2 assay (Advanced Cell Diagnostics). Three coronal sections of POA were sampled at bregma 0.23 mm, 0.13 mm, and 0.03 mm for each mouse (total n = 4), and an area of 2.2 mm (height) × 2.5 mm (width) was scanned for each brain section. Image analysis was performed using QuPath.²⁷ Regions of interest (ROIs) were drawn for each brain section corresponding to Allen Brain Atlas-defined POA subregions (atlas.brain-map.org; also see Figure S8). Quantification of the signal levels for each ROI was performed using the H-score method (see STAR Methods). In agreement with previous studies,^8,28 galanin-positive cells are most concentrated in three POA regions, namely ventrolateral preoptic nucleus (VLPO), medial preoptic nucleus (MPN), and medial preoptic area (MPO) (Figure 5A). Comparing expression levels of hormone receptors (Prlr, Esr1, and Ghr) between VLPO, MPO, and MPN revealed that galanin neurons in VLPO expressed higher levels of Ghr and lower levels of Esr1, matching the marker gene expression pattern of i14 (Figures 5B, 5C, and S9). Comparing expression of Egr1 and Fos between animals in the RS and SS conditions revealed increased expression of Egr1 and Fos in mice under high sleep pressure in VLPO (Figures 6 and S10) but not in MPN and MPO (Figure S11). In VLPO of mice under high sleep pressure, 72.7% of the Egr1-positive cells, 86.2% of the Fos-positive cells, and 68.7% of the Ghr-positive cells co-expressed Gal. These results validate the finding from the single-nucleus RNA-seq study that the galanin neurons that co-express Ghr and have low expression of Esr1 were activated during RS and that these neurons are located in VLPO.

Figure 5. Spatial characterization of galanin neurons in POA.

(A) Representative in situ hybridization image of POA with overlaying staining of Gad1 (inhibitory marker), Gal, Ghr, Prlr, and Esr1. DAPI indicates nuclear staining. The boxed areas indicating VLPO (blue box), MPN (pink box), and MPO (green box) are shown in (B). Scale bar, 200 μm.

(B) Overlaying as well as individual channel images for a zoomed-in area are shown for VLPO (top), MPN (middle), and MPO (bottom). Arrow indicates inhibitory galanin neurons. Galanin inhibitory neurons in VLPO co-express Ghr and Prlr, whereas in MPN/MPO co-express Prlr and Esr1. Scale bars, 20 μm; scale bars in zoomed images, 5 μm.

(C) Ghr level is higher in VLPO compared to MPN and MPO, whereas Esr1 level is higher in MPN and MPO compared to VLPO. Prlr is not significantly different between areas. Data points represent ROIs on each brain slide as mean ± SEM (n = 12 total data points from n = 4 biological replicates). p values were calculated using two-sided Wilcoxon rank-sum test (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; ns, not significant).

See also Figure S9.

Figure 6. VLPO galanin inhibitory neurons are activated during recovery sleep under high sleep pressure.

(A) Representative in situ hybridization images of POA under RS or SS conditions with overlaying staining of Gad1 (inhibitory marker), Gal, Ghr, Fos, and Egr1. DAPI indicates nuclear staining. Individual channel images are shown for a zoomed-in area. Arrow indicates inhibitory galanin neurons. VLPO galanin inhibitory neurons co-express Fos and Egr1 under RS but not under SS. Scale bars, 20 μm; scale bars in zoomed images, 5 μm.

(B) Fos and Egr1 levels in VLPO galanin inhibitory neurons are higher under RS compared SS. Data points represent ROIs on each brain slide as mean ± SEM (n = 12 total data points from n = 4 biological replicates). p values were calculated using two-sided Wilcoxon rank-sum test (**p < 0.01).

See also Figures S10 and S11.

To further provide spatial information for our single-nucleus RNA-seq defined clusters, we utilized published multiplexed error robust fluorescence in situ hybridization (MERFISH) data for POA.¹² Our single-nucleus RNA-seq clusters correlated strongly with the MERFISH clusters (Figure S12). Specifically, for galanin clusters (see Figure 3C; Tables S2 and S3), our single-nucleus RNA-seq cluster i14 is correlated with merf_I7 (r = 0.75; adjusted p =3 × 10⁻²⁶), which, judged by the staining images, is the only MERFISH galanin cluster that is located near and around VLPO. The single-nucleus RNA-seq clusters i11 and i12 are correlated with merf_I12 and merf_I14, which are located near and around MPO/MPN, consistent with our RNAscope results.

Pathways regulated by sleep homeostasis in different cell types and subtypes of POA

To characterize the effect of sleep pressure on biological processes and metabolic pathways in POA cell clusters, for all major subtypes we ran GSEA using gene ontology biological processes (GO-BP) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (see STAR Methods and results in Data S7). To improve interpretability, redundant pathways were first removed from enriched pathways (GSEA nominal p < 0.05) in each cell type (see STAR Methods), after which independent pathways from each cell type were combined and duplicate pathways among the full set of results were removed. This resulted in 37 pathways upregulated in SS (Figure 7A) and 16 pathways upregulated in RS (Figure 7B). Interestingly, most of the RS-enriched pathways are cell-type specific, such as “neuroactive ligand-receptor interaction” (e.g., P2ry14 and Htr2a) in inhibitory neurons, “branched amino acid catabolic process” (e.g., Ivd, Bcat2, Dbt, and Aldh6a1) in astrocytes, “dorsal/ventral pattern formation” (e.g., Gia2) in oligodendrocytes, immune-response-related pathways in microglia (such as “cytokine-mediated signaling pathway” [e.g., Ccr5]), energy metabolic pathways in endothelial cells (such as “oxidative phosphorylation” [e.g., Ndufb3]), and extracellular matrix pathways (e.g., Lama2 and Lama4) in VLMCs and ABCs. On the other hand, pathways induced by SS were often shared among multiple cell types, such as cholesterol biosynthetic process (e.g., Srebf2, Lss, and Hmgcr), regulation of translation (e.g., Syncrip and Elavl1), synaptic transmission (Pcdhb11, Kif1b, and Cdh2), axon development (e.g., Wnt7a, Sptan1, Auts2, and Sema6d), and nervous system development (e.g., Zfhx3, Nfasc, and Adgrb2). There are also pathways showing opposite changes in different cell types, such as immunity-related pathways (e.g., “neutrophil mediated immunity”) that were upregulated during SS in inhibitory and excitatory neurons (e.g., Sptan1 and Pygb) but had increased expression during RS in microglia and endothelial cells (e.g., Trpm2 and Atp6ap2).

Figure 7. Pathways differentially regulated by homeostatic sleep pressure in major cell types of POA.

GSEA of GO-BP and KEGG pathways combined. Genes were ranked based on differential expression (−log₁₀[p value] with sign showing direction of change) between RS and SS. Heatmap plots the −log₁₀ transformed p value of the GSEA, with positive numbers (shown in red) indicating upregulation in RS and negative numbers (shown in blue) indicating upregulation in SS.

(A) Pathways upregulated in RS compared to SS. Most changes are cell-type specific.

(B) Pathways upregulated in SS compared to RS. Most pathways are shared between multiple cell types.

(C) P2ry14 and Htr2a are the top upregulated genes in all inhibitory neurons combined, with fold changes between RS and SS of 1.13 (GSEA p = 1.64 × 10⁻³; FDR = 0.045) for Htr2a and 1.06 (GSEA p = 4.77 × 10⁻⁴; FDR = 0.023) for P2ry14.

(D) P2ry14 and Htr2a are only upregulated in i14 among all inhibitory subpopulations in POA. Heatmap shows the differential expressions of P2ry14 and Htr2a between RS and SS (−log₁₀ transformed FDR-adjusted p value from NEBULA with sign showing direction of change). Red cells indicate upregulation in RS, and blue cells indicate downregulation in RS compared to SS.

(E) P2ry14 and Htr2a are the top upregulated genes in i14, with fold changes between RS and SS being 1.49 (NEBULA p = 3.30 × 10⁻⁷; FDR = 3.76 × 10⁻⁴) for Htr2a and 1.22 (NEBULA p = 1.60 × 10⁻⁹; FDR = 6.39 × 10⁻⁶) for P2ry14.

Number of biological replicates: n = 9 SS and n = 10 RS (after removing one female SS sample; see STAR Methods). See also Figure S6.

Since “neuroactive ligand-receptor interaction” is the only pathway upregulated during recovery sleep in POA inhibitory neurons, we next explored the expression changes between RS and SS for the individual genes in this pathway. The “neuroactive ligand-receptor interaction” pathway contains a collection of cell-signaling ligands and receptors. Two genes (P2ry14 and Htr2a) were significantly upregulated (FDR < 0.05) in inhibitory neurons during RS (Figure 7C). We next examined the expression changes between RS and SS for P2ry14 and Htr2a in all inhibitory subtypes. Cluster i14 is the only inhibitory subpopulation that showed changes in these two genes (Figure 7D). When examining expression changes of all the genes in the “neuroactive ligand-receptor interaction” pathway in i14, P2ry14 and Htr2a are also the top two most significantly upregulated genes under RS (Figure 7E), indicating that the changes in P2ry14 and Htr2a during RS in POA inhibitory neurons are led by the changes in the galanin cluster i14. These genes may play an important role in sleep regulation in POA.

DISCUSSION

We combined single-nucleus RNA-seq and a panel of ARGs to map, in an unbiased manner, cell types activated during RS after sleep deprivation in POA. Transcriptional changes were compared between animals allowed to sleep undisturbed toward the end of the light phase (low sleep pressure) and animals sleep deprived for 6 h and allowed 30 min of RS (high sleep pressure). Differences in sleep pressure were confirmed by delta activity during NREM sleep. Molecular characterizations of the cell types using single-nucleus RNA-seq revealed 43 inhibitory neuronal clusters and 25 excitatory neuronal clusters in POA. Only one neuronal cell cluster showed increased expression in both rapid and delayed ARGs during RS compared to normal sleep; molecular annotation of this cell cluster (i14) revealed it as one of the three galanin-enriched inhibitory neuronal clusters. Compared to the other two galanin-enriched clusters, i14 had increased expression in Ghr (growth hormone receptor) and decreased expression in Esr1 (estrogen receptor). An in situ hybridization experiment validated this result and demonstrated that neurons in i14 were located in VLPO.

The current single-nucleus RNA-seq study is extensively compared to a previous single-cell RNA-seq study of POA performed by Moffitt et al.¹² Our single-nucleus RNA-seq study identified similar numbers of neuronal clusters and revealed overall cellular classification highly correlated with that of the single-cell RNA-seq study. More specifically, the galanin cluster i14 correlated most strongly with Moffitt’s single-cell RNA-seq cluster i8 and the MERFISH cluster I7. The MERFISH cluster I7 is the only galanin neuron cluster that is located around the VLPO area in Moffitt’s study, agreeing with our in situ hybridization results. We previously reported that the pooled galanin neurons isolated using laser-capture microdissection from VLPO²⁹ showed the best transcriptional alignment with the single-cell RNA-seq cluster i8 from Moffitt et al. Together, these data indicate a general agreement in the molecular and spatial characterization of i14 despite differences in the applied techniques.¹²

Our study revealed that i14 expressed higher levels of Ghr compared to the other galanin inhibitory subtypes. The link between growth hormone and sleep has long been explored (see review by Van Cauter and Plat³⁰). Growth hormone release is known to be tightly related to slow-wave sleep, and sleep restriction suppresses growth hormone secretion.^31,32 In patients with fatal familial insomnia, the normal nocturnal elevation of serum growth hormone is absent.³³ The amount and time course of the level of growth hormone in plasma is positively correlated with delta-wave activity.³⁴ More intense growth hormone release is associated with RS after sleep deprivation³⁵ and pharmacologically increased slow-wave sleep.^30,34 The age-associated decline in slow-wave sleep occurs in parallel with the age-associated decline in growth hormone release.³⁶ Combined, these data suggest that growth hormone signaling is associated with homeostatic sleep regulation. On the other hand, growth hormone promotes sleep. Growth hormone secretion is regulated by the opposite actions of growth hormone-releasing hormone (GHRH) and somatostatin (SRIF).³⁷ In rats, injection of GHRH increases NREM sleep as well as the number of c-Fos-positive inhibitory neurons in VLPO.³⁸ In the same study, the authors showed that the injection of octreotide (a SRIF analog) and a GHRH antagonist decreased NREM sleep and decreased the number of c-Fos-positive inhibitory neurons in VLPO.³⁸ Therefore, there is a bidirectional relationship between growth hormone and sleep, and our results suggest that growth hormone may regulate sleep via the sleep-promoting Ghr-positive galanin neurons in VLPO.

We also identified pathways regulated by homeostatic sleep pressure. The only upregulated pathway in POA inhibitory neurons under high sleep pressure was “neuroactive ligand-receptor interaction.” This change was driven by the changes in galanin cluster i14. Two genes, Htr2a and P2ry14, showed the most significant changes in i14. Htr2a encodes a receptor for serotonin (5-HT2a). Serotonin is a known regulator of sleep, although its exact role in sleep and wake control has been a topic of debate for decades.³⁹ Recent studies using modern optogenetic and chemogenetic tools demonstrated that the serotonergic system promotes sleep.^40,41 Additionally, ablation of serotonergic raphe nucleus in mice reduced sleep and impaired homeostatic response to sleep deprivation,⁴⁰ and knocking out 5-HT2a decreased the amount of NREM sleep in mice and disabled the accumulation of sleep pressure during sleep deprivation.⁴² These studies support the role of serotonin as a modulator for sleep. Earlier work showing the serotonergic raphe neurons being wake active is the main evidence supporting the wake-promoting role of serotonin.⁴³ However, given serotonin’s involvement in homeostatic sleep regulation, it has been suggested that the activity of the serotonergic raphe nucleus during wakefulness is related to the accumulation of sleep pressure.³⁹ Indeed, the 5-HT release during wakefulness is believed to be involved in homeostatic regulation of slow-wave sleep via VLPO.⁴⁴ Our findings support the role of serotonin in homeostatic sleep regulation and identify a specific neuronal group involved in this regulatory system. The second gene that was specifically upregulated in i14 during RS was P2ry14, which encodes a G-protein-coupled P2Y purinergic receptor that is selectively activated by uridine diphosphate (UDP) and UDP sugars such as UDP-glucose.⁴⁵ Uridine is a precursor for UDP and UDP-glucose.⁴⁶ Uridine has long been recognized as a sleep-promoting substance,^47,48 and the level of uridine in the brain is regulated by sleep homeostasis, with increasing secretion following spontaneous and enforced wakefulness.⁴⁹ Importantly, bilateral lesion of the lateral preoptic area abolishes the slow-wave sleep-promoting effect of uridine, suggesting that uridine promotes sleep via the sleep-promoting neurons in POA.⁵⁰ Overall, our finding of the specific upregulation of P2ry14 and Htr2a in the only galanin subpopulation that is activated during RS provides additional insights into the cell-type-specific mechanism of sleep regulation by serotonin and pyrimidine nucleotides.

In non-neuronal cell types, most of the pathways upregulated during RS are also cell-type specific. For example, in astrocytes, the only upregulated pathway was “branched amino acid catabolic process,” which involves the astrocytic control of glutamate synthesis and the nitrogen cycle between astrocytes and neurons.^51,52 In mature and precursor oligodendrocytes, respectively, “dorsal/ventral pattern formation” and “positive regulation of receptor-mediated endocytosis” were found to be upregulated during RS, which may relate to the proposed role of sleep and wake in myelination regulation and OPC differentiation.^53,54 Additional changes include immune-response pathways in microglia, energy pathways in endothelial cells, and extracellular matrix pathways in other vascular cell types that are part of blood-brain barrier, i.e., VLMCs and ABCs. These pathways generally align with the function of the cell types. It will be interesting to test in future studies whether the distinct responses of the different cell types to sleep deprivation and RS are essential to sleep regulation in POA. In contrast, pathways showing down-regulated changes (or upregulation following spontaneous sleep) are mostly shared among cell types. Many of these pathways, such as cholesterol synthesis, protein translation, and metabolic pathways, have been revealed to be regulated by sleep in previous bulk transcriptomic studies,^19,20 including our own.^18,29,55 This suggests possible systemic mechanisms controlling multiple cell types during undisturbed sleep.

Our study examines neuronal activities in POA during sleep toward the end of the light phase, when the sleep pressure is naturally low. The data reveal that VLPO galanin neurons are differentially activated during sleep under different sleep pressures. Although the sleep-promoting role of VLPO galanin neurons has been demonstrated in many studies,^8–11,56 there is also literature suggesting that these neurons actually promote wakefulness.⁵⁷ Chung et al. found a surprising wake-promoting effect for optogenetic activation of POA galanin neurons.⁵⁷ However, Kroeger et al. proposed that this paradoxical result could be explained by optogenetic stimulation being at too high a frequency, resulting in depolarization block.⁵⁶ Our current study also did not replicate the findings of other sleep-promoting cells in POA that expressed Tac1, Cck, Crh, or Pdyn.⁵⁷ On the contrary, our results indicate that some Tac1 subtypes in POA are activated under low sleep pressure. Indeed, one study showed that activating Tac1 neurons in POA promoted wakefulness.⁵⁸

In conclusion, using single-nucleus RNA-seq we identified and molecularly characterized one subtype of galanin cells (i14) in POA that was activated during RS following sleep deprivation. In situ hybridization results demonstrated that this galanin subtype was located in VLPO. This galanin cluster (i14) expressed higher levels of Ghr and lower levels of Esr1 compared to other galanin cells, suggesting a role of growth hormone signaling in sleep regulation. Our data also identified cell-type-specific upregulation of Htr2a and P2ry14 in i14, providing additional insights regarding potential signaling mechanisms involving serotonin and pyrimidine nucleotides for homeostatic sleep regulation.

Limitations of the study

This study examined gene expression changes during 30 min of RS following 6 h of sleep deprivation and revealed that one galanin subpopulation (i14) showed increased expression in selective groups of ARGs, i.e., rapid and delayed PRGs (which respond to stimuli lasting minutes to hours) but not SRGs (which respond to stimuli lasting for hours). Based on this finding, our results suggest that i14 is more likely to be activated during RS. However, the ARG panel was established based on studies performed on the visual cortex,²² and the question as to whether the experience-dependent transcriptional regulation is cell-type specific remains. As far as we know, there is no study looking at the cell-type-specific transcriptional response to different duration or patterns of stimulations in hypothalamus. Based on studies performed in other brain regions, there are likely cell-type-specific experience-regulated transcriptional changes, particularly for late-response genes.^59,60 Despite this limitation, our findings are in agreement with a number of previous studies, e.g., the number of c-Fos-positive neurons in VLPO is not elevated in animals collected after sleep deprivation.⁶¹ Similarly, Szymusiak et al. found that the VLPO neuron-firing rates were elevated during RS but not during sleep deprivation.⁶² However, there are also studies reporting an increased number of activated neurons in POA/VLPO responding to both extended wakefulness and RS.^9,17 Therefore, without examining the temporal (e.g., hourly) changes of the ARGs during extended wakefulness with sleep deprivation, we cannot rule out the possibility that i14 is activated under accumulated sleep pressure after extended sleep deprivation (e.g., 4 h). This temporal study design could also identify how many hours of extended wakefulness are needed to activate these neurons, and thus represents an important next step based on the results presented here. We also identified cell-type-specific upregulation of Htr2a and P2ry14 in i14 during RS. However, just because these transcriptional changes occur does not necessarily mean that these are part of the mechanism responsible for sleep homeostasis. Future studies are needed to directly test the roles of Htr2a and P2ry14 in homeostatic sleep regulation.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Allan Pack (pack@pennmedicine.upenn.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Single-nucleus RNA-seq data have been deposited at GEO (https://www.ncbi.nlm.nih.gov/geo/) and are made publicly available. Accession numbers are listed in the key resources table.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Critical commercial assays
Minute™ Single Nucleus Isolation Kit for Neuronal Tissues/Cells	Invent Biotechnologies, Inc.	Cat # BN-020
Deposited data
Raw and process data	This paper	GEO: GSE243489
Experimental models: Organisms/strains
Mouse: C57BL/6J	The Jackson Laboratory	JAX:000664
Software and algorithms
Cellranger (v6.1.2)	10x genomics	https://www.10xgenomics.com/support/software/cell-ranger/
Seurat (v4.0.1)	Hao, Y. et al.72	https://github.com/satijalab/seurat
DIEM (v2.4.0)	Alvarez, M. et al.70	https://github.com/marcalva/diem
DoubletFinder (v2.0.3)	McGinnis, C. S. et al.71	https://github.com/chris-mcginnis-ucsf/DoubletFinder
Nebula (v1.3.0)	He, L. et al. et al.73	https://github.com/lhe17/nebula
Fgsea (v1.24.0)	Korotkevich, G. et al.74	https://github.com/ctlab/fgsea
QuPath (v0.1.2)	Courtney, J. M. et al.27	https://github.com/qupath/qupath
RNAscope™ HiPlex Image Registration Software (v2.1)	ACD	https://acdbio.com/

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

Experiments were performed on male and female C57BL/6 mice (The Jackson Laboratory; JAX:000664) at 2–3 months of age housed individually in a 12 h/12 h light/dark room with food and water available ad libitum. All experiments were approved by the Institution of Animal Care and Use Committee of the University of Pennsylvania and were carried out in accordance with all National Institutes of Health guidelines.

METHOD DETAILS

Animal preparation and surgery

Sleep was monitored continuously for two baseline recording days and on the following experimental days, using EEG/EMG.^63,64 EEG electrodes were implanted under anesthesia at the following stereotactic coordinates relative to bregma: AP: 1.5 mm, ML: 1.7 mm, and AP: 1 mm, ML: −1.7mm. Two Teflon-coated EMG electrodes bared at the tips were sutured to the dorsal neck muscles. All leads from the electrodes were connected to a plastic socket connector (Plastics One), which was fixed to the skull with dental cement. Following surgery, animals were allowed to recover for 10 days before any studies were performed. Mice under high sleep pressure (recovery sleep, RS) were collected after 30 min of recovery sleep following 6 h of sleep deprivation. Sleep deprivation was performed using minimally required human interventions, by placing novel objects into cages, gentle tapping of cages, and, if required, lightly stroking the back of the mouse with a small paint brush. Mice under low sleep pressure were collected after prolonged spontaneous sleep (SS) under undisturbed condition toward the end of the light phase. The two groups of mice were collected at matching time of day (ZT10.5) to control for any circadian effects. The study started with a total of n = 26 RS mice (16 and 10 female and male mice, respectively) and n = 22 SS mice (12 and 10 female and male mice, respectively).

EEG/EMG

EEG and EMG signals were amplified using a Neurodata amplifier system (Models M15, Astro-Med. Inc, West Warwick, RI, USA) and recorded using the Grass Gamma software (Astro-Med). The quality of EEG/EMG signals was examined by an experienced experimenter and five mice (2 female RS and 3 female SS mice) were excluded from the study due to poor EEG/EMG data quality. The recordings of the remaining 42 mice were converted to the European Data Format and behavioral states (wake, NREM, and REM) and were scored in 4 s epochs using SleepLearning, an online machine learning-based tool that has been extensively validated against independent human scorers.⁶⁶ To evaluate the accuracy of SleepLearning in our dataset, recordings of 28 (67%) mice were randomly chosen and scored by experienced gold-standard human scorers (TechIndia). The overall epoch-to-epoch agreement between the machine learning and human scorers was 92.6%, with 97.3% agreement in wake scoring, 89.7% agreement in NREM, and 84.9% agreement in REM. Spectral power analysis was performed in 0.25 Hz bins using the Somnologica science software (Medcare), with a Hamming window function, for the 3 behavioral states. EEG spectra were expressed in pre-defined frequency bands, including delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), sigma (12–14 Hz), beta (14–30 Hz), low gamma (30–60 Hz), and high gamma (60–120 Hz). Relative delta power activity was calculated using the mean delta power of the last 4 h during lights-on (ZT9–12) on the baseline recording day as reference (100%).⁶⁴ The delta power during the last 0.5 h before sacrifice was compared between mice in RS and SS groups using a non-parametric Wilcoxon rank-sum test. As shown in Figure 1D, mice during RS had significantly higher delta power activities than mice following SS. To maximize sleep pressure differences among the mice in the two conditions (RS and SS), 10 RS mice (5 male and 5 female) with the highest delta power and 10 SS mice (5 male and 5 female) with the lowest delta power were chosen for inclusion in the snRNA-seq experiment (Figure 1B).

Blood cortisol measurement

At the end of the experiment, mice were decapitated following cervical dislocation and trunk blood was collected to measure levels of plasma corticosterone using ELISA (ALPCO).⁶⁵ Although mice in the RS condition did not show elevated cortisol levels compared to the mice in the SS condition on average (Figure 1C), two female RS mice with the highest cortisol levels were excluded from the study due to concern of potentially elevated stress levels in these mice.

Tissue collection and nuclei isolation

Brains were harvested following cervical dislocation and flash frozen on dry ice. Although most sub-regions of the preoptic area (POA) (e.g., MPO, VLPO, and MPN) are centered around a region spanning from bregma +0.2 to −0.2, to ensure a consistent tissue sampling between individual mice we dissected a larger area surrounding the preoptic region (bregma +0.5 to 0.5). Tissue dissection was performed on frozen brains in a cryostat, as previously described⁵⁵ and explained below. The starting position of the dissection region was carefully approached by serial removal of approximately eighty 50mm slides from the olfactory bulb (~bregma +4.5) to rostral of POA (bregma +0.5) guided by a combination of anatomical landmarks based on the Franklin-Paxinos atlas,⁶⁷ including anterior commissure (AC), corpus callosum (CC), and lateral ventricle (LV). Starting from bregma +0.5, tissues were dissected from four consecutive 250mm brain slides using a 3mm Harris Uni-Core micro sampling tool (GE Healthcare). The micro sampling tool was horizontally positioned to the center of the brain and vertically positioned immediately below AC as illustrated in Figure S5. The entire micro-punching process was performed within the cryostat, which was maintained at −20°C. The dissected POA regions were transferred into an RNase-free microcentrifuge tube (Invitrogen) that had been pre-cooled on dry ice. The dissected tissues were stored at −80°C until preparation of nuclei.

Nuclei were isolated from the dissected POA tissues using Minute Single Nucleus Isolation Kit for Neuronal Tissues/Cells (Invent Biotechnologies, Inc., BN-020) following manufacturer’s protocol with minor modifications. Specifically, Protector RNase inhibitor (Sigma, PN-3335399001) was added at a concentration of 0.2U/μL to all the buffers provided by the kit. The isolated nuclei were resuspended in 2%BSA solution (in 1xPBS with 0.2U/ul of Protector RNase inhibitor) to a concentration of 900–1200 nuclei/μL and were loaded onto the 10x Genomics microfluidic device at maximum capacity for a target capture of ~10,000 nuclei per sample.

QUANTIFICATION AND STATISTICAL ANALYSIS

snRNA-seq using 10x genomics and data analysis

Next-generation sequencing libraries were prepared using the 10x Genomics Chromium Single Cell 3′ Reagent kit v3 per manufacturer’s instructions. Libraries were uniquely indexed using the Chromium dual Index Kit, pooled, and sequenced on an Illumina NovaSeq 6000 sequencer in a paired-end, dual indexing run. Sequencing for each library targeted 20,000 mean reads per nuclei. Data were then processed using the Cellranger pipeline (10x genomics, v.6.1.2) for de-multiplexing and alignment of sequencing reads to the mm10 transcriptome with the “include introns” parameter and creating the feature-barcode matrices.

Data pre-processing was performed to remove low quality nuclei. Recent studies have shown that common quality control (QC) methods that remove cells based on the number of unique genes and total counts cannot model the complexity in scRNA-seq and snRNA-seq data, resulting in insufficient removal of low-quality cells or removal of biologically relevant cells.^68–70 Therefore, in the current study we used two published methods for QC, DIEM⁷⁰ and DoubletFinder,⁷¹ to evaluate empty droplets (containing mainly ambient RNAs released from lysed cells) and doublets (more than one cell apportioned into one droplet), respectively. DIEM is a semi-supervised machine learning tool specifically designed for snRNA-seq data, which requires homogenization of tissues to release nuclei and often leads to more background RNA contamination. DoubletFinder detects doublets based on cell-proximity to simulated artificial doublets in gene-expression space. DIEM and DoubletFinder were applied to calculate QC metrics indicating the likelihood of each nucleus being an empty droplet (“score.debris”) or a doublet (“DF.score”), respectively (Figure S13). Clusters with a median DF.score ≥0.2 or a median score.debris ≥0.5 were removed. One mouse sample (a female SS sample) was removed due to major differences in cellular makeup compared to the other samples; >60% of nuclei belonged to 4 clusters that were predominantly made up of nuclei from this one mouse. Therefore, the final biological replicates included in all the snRNA-seq data analysis were 9 SS (n =5 male and n = 4 female) and 10 RS (n = 5 male and n = 5 female) mice.

Following the above preprocessing steps, the remaining nuclei were integrated and clustered using Seurat v4.⁷² Clustering was performed in two steps. The first round of clustering was used to separate the major cell types, including the neuronal and non-neuronal cell types. The second round of clustering was used to further identify the cellular subpopulations among inhibitory and excitatory neurons. The clustering results from Seurat are sensitive to a number of parameters, including the number of principle components (PC) to construct shared nearest neighbor (SNN) graph, the k-parameter that defines the number of nearest neighbors, and the resolution that is used in the final modularity optimization that defines clusters. To evaluate clustering stability using different values of each of the three parameters, we used a published bootstrapping strategy that repeatedly clusters a random subset of the full dataset and calculates a Jaccard index to evaluate clustering similarity before and after each re-clustering²⁵ (see Figures S14 and S15). After identifying a stable clustering solution, marker genes statistically enriched in each cluster were identified using the “FindAllMarker” command from Seurat (Wilcoxon rank-sum test FDR<0.05).

Differential gene expression analysis

To compare gene expressions in each cluster between RS and SS, the R package NEBULA⁷³ was used, which accounts for the hierarchical structure of the data by modeling both between-individual and within-individual effects using a negative binomial mixed model (NBMM). Removal of low-expression genes was performed using the recommended threshold (counts per cell ≤0.005) for each cell type, which resulted in slightly different numbers of genes being tested for differential expression across different cell types. The main comparisons were made between conditions (RS vs. SS) while controlling sex as a covariates. Sex-specific changes were examined by testing the interaction between sex and condition.

Gene set enrichment analysis

Gene set enrichment analysis was implemented in R software (www.r-project.org; fgsea, v.1.20.0) to estimate the statistical enrichment of gene sets against ranked gene lists.⁷⁴ Genes from the differential gene expression results obtained using NEBULA were ranked based on the –log₁₀ transformed p-value multiplied by the sign of the fold change (positive when up-regulated in RS and negative when up-regulated in SS). To identify cell groups showing activity regulated expression changes under the tested experimental conditions (RS vs. SS) we also tested enrichment of a curated set of 172 activity-regulated genes (ARGs) (Data S2). This curated set included 3 categories of ARGs – rapid (e.g., Fos and Egr1) and delayed (e.g., Bdnf and Fosl2) primary response genes (rPRGs and dPRGs, respectively) and secondary response genes (SRGs; e.g., Kcnh8 and Brinp1), as well as a control set of genes that are known not to show activity-regulated changes.²² To identify pathway enrichment of the differential gene expression results, the most recent GO-BP and KEGG databases were downloaded from Enrichr (https://maayanlab.cloud/Enrichr/#libraries) and combined. Independent pathways from the enrichment results were selected using the “collapsePathways” function from the fgsea package.⁷⁴ The method uses Bayesian network construction approaches and filters redundant pathways that do not provide new information given other pathways already present in the output.

RNAscope

Multiplex in-situ hybridization was performed using the RNAscope HiPlex v2 assay (Advanced Cell Diagnostics, Inc.) with 12 probes, including neurotransmitter/neuropeptides (Gad1, Slc17a6, Gal, and Tac1), hormone receptors (Prlr, Esr1, and Ghr), and activity markers (rPRGs [Egr1 and Fos], dPRGs [Hcrtr2], and SRGs [Kcnh8 and Brinp1]). Three coronal sections of POA were sampled at bregma 0.23mm, 0.13mm, and 0.03mm for each mouse (total n = 4) and an area of 2.2mm (height) x 2.5mm (width) was scanned for each brain section. Sections were imaged in four rounds with 3 probes in each round using a 20x resolution on a Zeiss Axio Scan.Z1 digital slide scanner. Images of the same tissue from multiple rounds were registered and combined together using the RNAscope HiPlex Image Registration software v2.1.0 using the DAPI signal as a reference, following the manufacturer’s protocol. Background subtraction was performed (ImageJ) to remove imaging artifacts such as autofluorescence and other noise. Image analysis was performed using QuPath.²⁷ Regions of interest (ROIs) were drawn for each brain section corresponding to Allen Brain atlas defined POA sub-regions (atlas.brain-map.org) (Figure S8). Table S1 shows the abbreviation used for each ROI. Quantification of the signal levels was performed for each cell as: negative (0 copies); 1+ (Minimum Copies/Cell: 1); 2+ (Minimum Copies/Cell: 4); 3+ (Minimum Copies/Cell: 10); or 4+ (Minimum Copies/Cell: 16). The total levels of expression for a given probe for each ROI was calculated as: H-Score = (1 × % Probe 1+ Cells) + (2 × % Probe 2+ Cells) + (3 × % Probe 3+ Cells) + (4 × % Probe 4+ Cells), following recommendations by Cell Diagnostics USA (Newark, CA, USA) for the evaluation of RNAscope assays.⁷⁵

Comparison to previous scRNA-seq study of POA

Clusters identified in the current snRNA-seq study of POA were compared to the clusters previously described from the scRNA-seq study by Moffitt et al.¹² Between cluster Pearson’s correlations were evaluated based on Z score transformed averaged expression profiles for all clusters. Correlation p-values were corrected for multiple comparisons using the Benjamini-Hochberg method⁷⁶ and correlations with an FDR<0.05 were considered significant. Correlation between our snRNA-seq clusters and the multiplexed error robust fluorescence in-situ hybridization (MERFISH) clusters¹² were performed similarly. For ease of comparison of our snRNA-seq clusters to Moffitt’s results, we annotated the previously-identified scRNA-seq clusters with a prefix of “sc.” and MERFISH clusters with a prefix of “merf_”. The correlation matrices obtained comparing our snRNA-seq clusters to Moffitt’s scRNA-seq clusters and Moffitt’s MERFISH clusters are shown in Figure S12. For each cluster identified in our study, the most strongly correlated scRNA-seq clusters and the most strongly correlated MERFISH clusters are listed in Table S2 (for inhibitory neurons) and Table S3 (for excitatory neurons). The spatial localizations of the matching MERFISH clusters were determined based on the Allen Mouse Brain atlas (mouse.brain-map.org) as well as the staining density plots from Moffitt’s publication¹² based on marker gene expressions. For clarity, Figure S5 illustrates the spatial sub-regions of POA included in the dissection area used for snRNA-seq in this study. Locations of POA sub-regions were based on the Allen Mouse Brain atlas (mouse.brain-map.org) and other publications.^8,56 Notably, the scRNA-seq study by Moffitt et al.¹² included a larger area, which includes anterior commissure (Ac) and bed nucleus of stria terminalis (BNST) regions above Ac where additional excitatory galanin cells were found that was absent from our snRNA-seq study.

Statistics

Two group comparisons were performed using non-parametric Wilcoxon rank-sum test, unless noted otherwise. Figures with error bars were reported as mean ± SEM. For figure panels that report p-values, statistical tests used to create the p-values have been stated in the figure legends. The number of replicates has been included in the proper figure legends.

Highlights.

• single-nucleus RNA-seq of the preoptic area (POA) reveals cell-type-specific changes during recovery sleep

• Only one galanin neuronal subtype in POA is activated, which is positive for Gad1 and Gal

• single-nucleus RNA-seq reveals higher Ghr and lower Esr1 expression levels in this neuronal subtype

• RNAscope in situ hybridization reveals the galanin subtype is located at the ventrolateral POA