Chronic intermittent hypoxia elicits distinct transcriptomic responses among neurons and oligodendrocytes within the brainstem of mice

Abstract

Chronic intermittent hypoxia (CIH) is a prevalent condition characterized by recurrent episodes of oxygen deprivation, linked to respiratory and neurological disorders. Prolonged CIH is known to have adverse effects, including endothelial dysfunction, chronic inflammation, oxidative stress, and impaired neuronal function. These factors can contribute to serious comorbidities, including metabolic disorders and cardiovascular diseases. To investigate the molecular impact of CIH, we examined male C57BL/6J mice exposed to CIH for 21 days, comparing with normoxic controls. We used single-nucleus RNA sequencing to comprehensively examine the transcriptomic impact of CIH on key cell classes within the brainstem, specifically excitatory neurons, inhibitory neurons, and oligodendrocytes. These cell classes regulate essential physiological functions, including autonomic tone, cardiovascular control, and respiration. Through analysis of 10,995 nuclei isolated from pontine-medullary tissue, we identified seven major cell classes, further subdivided into 24 clusters. Our findings among these cell classes, revealed significant differential gene expression, underscoring their distinct responses to CIH. Notably, neurons exhibited transcriptional dysregulation of genes associated with synaptic transmission, and structural remodeling. In addition, we found dysregulated genes encoding ion channels and inflammatory response. Concurrently, oligodendrocytes exhibited dysregulated genes associated with oxidative phosphorylation and oxidative stress. Utilizing CellChat network analysis, we uncovered CIH-dependent altered patterns of diffusible intercellular signaling. These insights offer a comprehensive transcriptomic cellular atlas of the pons-medulla and provide a fundamental resource for the analysis of molecular adaptations triggered by CIH.

NEW & NOTEWORTHY This study on chronic intermittent hypoxia (CIH) from pons-medulla provides initial insights into the molecular effects on excitatory neurons, inhibitory neurons, and oligodendrocytes, highlighting our unbiased approach, in comparison with earlier studies focusing on single target genes. Our findings reveal that CIH affects cell classes distinctly, and the dysregulated genes in distinct cell classes are associated with synaptic transmission, ion channels, inflammation, oxidative stress, and intercellular signaling, advancing our understanding of CIH-induced molecular responses.

INTRODUCTION

Chronic intermittent hypoxia (CIH) is a prevalent condition characterized by recurring episodes of oxygen deprivation, commonly observed in respiratory disorders such as obstructive sleep apnea (OSA), apneas of prematurity, Rett syndrome, familial dysautonomia, mitochondrial diseases, and epilepsy (1–5). Prolonged exposure to CIH has been shown to impact neuronal function, as well as contribute to the development of endothelial dysfunction, chronic inflammation, and oxidative stress. These effects ultimately lead to various comorbidities including metabolic disorders, hypertension, and cardiovascular disease (6–13). Understanding the molecular mechanisms underlying the cellular response to CIH is crucial for elucidating the pathophysiology of associated disorders. Recent advancements in single-cell technology, including techniques like single-nucleus RNA sequencing (snRNA-Seq), have become a powerful tool for studying both gene expression patterns and cellular diversity at the individual cell level. This technology enables researchers to delve into gene dysregulation and cellular responses to chronic conditions (14–17).

In this study, we aimed to gain first insights into the impact of CIH on three major cell classes of the brainstem—excitatory neurons, inhibitory neurons, and oligodendrocytes. These major cell classes play pivotal roles in regulating essential physiological functions including autonomic tone, cardiovascular control, sleep-wake cycles, and respiration (9, 18–24). By using multiplexed single-nucleus RNA sequencing, we identified molecularly distinct cell classes, showed CIH-dependent transcriptional dysregulation, and predicted receptor-ligand interactions. We profiled gene expression in 10,995 nuclei derived from pontine-medullary tissues, comprising six samples of C57BL6/J mice. We identified 7 major cell classes of which 24 clusters correspond to excitatory neurons, inhibitory neurons, and oligodendrocytes. Our analysis revealed that there are substantial differences in the extent of gene expression changes across different cell classes. These differences underscore how distinct cell classes respond differentially to the effects of chronic intermittent hypoxia (CIH). We identify CIH-dysregulated gene expression in oligodendrocytes consistent with responses to oxidative phosphorylation and oxidative stress. In neurons, CIH-dysregulated gene expression indicated altered synaptic transmission and structural remodeling including dysregulated ion channels and inflammatory response. Using CellChat we found that cell classes and clusters altered their predicted networks of intercellular signaling. Our results provide important gene expression profiles of cell populations within the pons-medulla offering a foundation for subsequent investigations into the mechanisms underlying CIH-induced molecular adaptations, and their impact on physiological homeostasis.

MATERIALS AND METHODS

Sample Preparation

Twenty-one-day-old postnatal C57BL6/J male mice were weaned and group-housed with littermates in a controlled chamber with regulated oxygen conditions (Oxycycler, Huff Technologies Inc.), six animals from two different litters. For the CIH treatment, we used the protocol as described previously (25). Briefly, one chamber was dedicated to the chronic intermittent hypoxia (CIH) group, whereas the other chamber served as the normoxic control. The CIH group was subjected to intermittent episodes of hypoxia using a specific protocol. First, nitrogen (N₂) was continuously injected into the chamber for 60 s, reducing the percentage of inspired oxygen (O₂) within the chamber to a range of 4.5% to 5%. Subsequently, compressed air was injected for up to 5 min to restore the percentage of O₂ to 21% before the initiation of a new hypoxia cycle. This protocol was repeated 80 times per day over 8 h for 21 days, starting from P22 during the light phase (for mice it is a sleep phase), in a room with a 12:12-h light-dark cycle. The control mice were maintained in room air (constant 21% oxygen) and housed in the same apparatus as experimental animals, exposed to the same sound conditions (26). Throughout the experimental period, the mice were provided with food and water ad libitum. At the end of the experiment, animals were euthanized by isoflurane overdose for the dissection procedure. Dissected pons-medulla samples were immediately flash-frozen on dry ice and stored at −80°C until further subsequent use. Proper care and ethical guidelines were followed for all animal handling and sample collection under an Institutional Animal Care and Use Committee (IACUC) protocol approved by the Seattle Children’s Research Institute (SCRI).

Generation of Single Nuclei Suspensions and Fluorescent-Activated Nuclei Sorting

Single nuclei suspensions were prepared using the 10X Genomics protocol (CG000366) with the following modifications: 1) Tween 20 was excluded from the working wash buffer. 2) The lysis buffer was modified by increasing the Nonident-P40 Substitute (Sigma, 74385) final concentration to 0.5% and included Tween-20 (0.1%) as recommended by the manufacturer’s protocol. Frozen tissues were mechanically dissociated in sterile 1.5 mL microcentrifuge tubes, using a 1.5 mL pellet pestle in 1X lysis buffer and incubated in the same solution in total for 10 min to lyse cellular plasma membranes, releasing free nuclei. The suspension was then centrifuged at 500 rcf for 5 min at 4°C. The resulting nuclear pellets were washed and resuspended in a wash buffer. The suspension was passed through a 70 μm filter (pluriStrainer Mini 43–10070-40). To stain nuclei, 2 μM Hoechst dye (Abcam, ab228550) was added, and incubated for 10 min at 4°C. After incubation, nuclei were centrifuged at 500 rcf for 5 min and washed with 1× wash buffer. Finally, nuclei suspension was filtered through a 40 μm filter (pluriStrainer Mini 43–10040-40) and directly used for fluorescence-activated nuclei sorting (FANS) using the BD FACSAria III Cell Sorter to enrich single nuclei and remove debris. Log scale forward and side scatter were used to identify the nuclei population. All the steps mentioned above were performed on ice and a wash buffer was used to collect sorted nuclei. These modifications ensured the successful isolation of nuclei for downstream analysis.

Single Nuclei Capturing, Library Construction, and Sequencing

The Chromium Next GEM Single Cell 3′ Kit v3.1 (10X Genomics, Pleasanton, CA) was used to encapsulate single nuclei, cDNA preparation, and library construction according to the manufacturer’s protocol. The sequencing of three samples was performed in two batches aiming for 9,000 nuclei from each sample from two litters. The size and concentration of libraries were determined using a High-Sensitivity D5000 ScreenTape or 4200 TapeStation (Agilent Technologies, Santa Clara, CA). Subsequently, the libraries were sequenced on an Illumina HiSeq 4000 using 2 × 150 chemistry, following the standard 10X Genomics v2 paired-end configuration by the Northwest Genomics Center at the University of Washington, Seattle. Expression matrices were generated using 10X Genomics Cell Ranger v3.1.0, where the sequences were aligned to the mm10-2020-A reference transcriptome (Mus musculus).

Data Quality Control, Preprocessing, and Normalization

Expression matrix files were analyzed using the Seurat v4.0 package for R (27). The matrix files were individually processed using standard Seurat preprocessing cutoffs. SoupX (28) and DoubletFinder (29) were used to find and remove ambient RNA contamination and doublets, respectively. We further removed nuclei with a nFeature count below 1,500 and percent mitochondria above 2.0. Data were normalized using sctransform (SCT) with the method glmGamPoi and included the proportion of mitochondrial reads as a variable to regress (30, 31).

Data Integration and Clustering

To combine the data for the downstream analysis, we followed the vignettes for scRNA-Seq integration (27) with the selection method “versus” and 2,000 variable features. Principal components (npcs = 30) were computed, followed by calculating the nearest neighbor graphs while accounting for the correlation between average expression and dispersion. By using a modularity optimization Louvain algorithm, the first 20 principal components (PCs) were used to perform Uniform Manifold Approximation and Projection (UMAP) to embed the data set in low dimensionality for visualization. For optimal clustering, we used Seurat with a resolution parameter set at 0.6, resulting in a data set of 14,966 nuclei. For the unbiased survey of nuclei clusters, we used singleR, which enabled us to estimate cell class frequencies. Only the nuclei that matched the cell class identity were retained; resulting in a total of 10,995 nuclei used for all subsequent analyses. The excluded nuclei also include neuronal nuclei that coexpress markers for glutamatergic, GABAergic, or cholinergic cell types. Through anatomical staining, we quantified neurons that coexpress staining for Chat, vGlut2 (Slc17a6), and Gad2 cell classes. We found that the percentage of neuronal cell classes that coexpress vGlut2⁺ and Chat⁺ or vGlut2⁺ and Gad2⁺ varied based on anatomical location (Supplemental Fig. S1). These variations, reflective of gene patterns that coexpress, could have implications that are not fully captured in single-cell analysis. Hereafter we use the terms Slc17a6/vGlut2, Slc17a7/vGlut1, Slc32a1/vGat, Slc6a5/Glyt2 interchangeably.

Neuronal Clustering

Unsupervised clustering with a resolution of 0.6 yielded 27 distinct clusters for both the normoxic and CIH conditions (Supplemental Fig. S3A). The neuronal cluster (Supplemental Fig. S2B) was further parsed based on average relative expression levels of excitatory and inhibitory markers. Specifically, excitatory neurons were identified by the expression of Slc17a7 the gene encoding vGlut1 and Slc17a6 the gene encoding vGlut2 cell classes, whereas inhibitory neurons were characterized by vGat cell class by Gad1, Gad2, Slc32a1, and Glyt2 cell class by Slc6a5 expression (Supplemental Fig. S3A). After initial clustering, clusters 0 and 1 were subset and processed using standard Seurat preprocessing using 2000 nfeatures. This subsetting was performed for identifying and labeling excitatory and inhibitory neurons present within clusters 0 and 1. This revealed a total of 11 additional clusters with distinct excitatory and inhibitory cell classes (Supplemental Fig. S3C). Clusters 3 and 5 showed expression of both excitatory and inhibitory markers were excluded, resulting in eight clusters. These additional clusters were denoted with the label “sc.” and were mapped back to the original UMAP space (Supplemental Fig. S3E) and the names were modified according to their identity used in further analysis.

Identification of Cluster Markers and Differentially Expressed Genes

The FindMarker function from the Seurat was used (test.use = “roc”) to identify specific markers for the clusters. Marker selection was based on the computed AUC scores > 0.6 [Area under the receiver operating characteristic (ROC) curve], pct.2 < 0.1 (percentage of cells that express feature in ident.2), and with log2 fold change > 0.38, which corresponds to a > 30% enrichment in expression level were selected. To identify the differentially expressed genes between normoxia and CIH, the Seurat FindMarker function was used with the Wilcoxon pairwise test and Benjamini–Hochberg correction to obtain adjusted P values. Genes detected with an FDR (false discovery rate): adjusted P value of <0.05 and a log2-fold change of > 0.05 were reported as significant and were used for plotting. such as volcano plots, and further analysis. For the differential expression (DE) analysis, the Seurat FindMarker function had shifted avg_log2FC values < 1, and the significance values were determined from a P value of <0.5 to a P value of <0.05 in the analysis and ribosomal RNA was removed before this analysis. Gene ontology (GO) enrichment analysis was performed using Metascape (https://metascape.org/gp/index.html#/main/step1) (32).

Transcription Factors Analysis

TF enrichment analysis (TFEA) was performed on differentially expressed genes (FDR: adjusted P value of <0.05 and a log2-fold change of >0.03) and annotated TF targets with TF-target gene set libraries, using ChEA3 (33), a web-based TFEA tool. From the results, the integrated rank scores lower than 0.02 were assigned as significant and used for plotting.

Ion Channel and Inflammatory Response Analysis

We compiled a list of ion channels (Supplemental Table S2) and compared this list with the differentially expressed genes (DEGs) found through the FindMarker function. The significant ion channels were selected between conditions for cell classes and across clusters. Augmentation of inflammatory genes was assessed by comparing the DEGs with the gene list obtained from informatics.jax.org/vocab/gene_ontology (GO:0006954), which comprises curated genes responsive to inflammation. Statistical significance was evaluated using the Wilcoxon rank-sum test.

Enrichment Score Analysis

Enrichment scores were calculated for ion channels and inflammatory response (upregulated DEGs). The AddModuleScore function was used to determine the average expression level for the module score at the single-cell level (27). In brief, the module score is derived by categorizing genes based on their average expression levels. Following this, the average expression of each gene is normalized by subtracting the cumulative expression of control gene sets (100 randomly selected per bin). This provides a comprehensive measure of gene module activity.

Cell Signaling and Signaling Pathways

Cell signaling analysis was performed using the CellChat package for R (34), following the default settings. CellChatDB mouse was used for the analysis.

In Situ Hybridization

For in situ hybridization, we used the RNAscope Multiplex Fluorescent Reagent kit from Advanced Cell Diagnostics. The procedure involved several steps: brain slices were incubated in 0.1 M PBS for 5 min, followed by incubation with RNAscope Hydrogen Peroxide (No. 322381) for 10 min at room temperature. Subsequently, three sequential washes were performed with PBS (3× PBS) for 2 min each. This was followed by treatment with Protease IV (No. 322381) for 30 min at room temperature, then a 2-min incubation in 0.1 M PBS. The vGlut2/Slc17a6 probes were incubated with the tissue for 2 h at 40°C, followed by a 2-min PBS wash. Hybridization was carried out using RNAscope Multiplex Detection Reagents (No. 323110), including FL v2 Amp 1, v2 Amp 2, and v2 Amp 3, for 30 min each at 40°C, with each step followed by a 5-min PBS wash. Further steps involved incubation with RNAscope Multiplex FL for 30 min at 40°C, followed by a 5-min PBS wash. The HRP-C1 signal was developed by incubating the tissue with RNAscope Multiplex FL v2HRP-C1 for 15 min at 40°C, followed by a 5-min PBS wash. In addition, incubation with TSA Plus fluorescein (Cy3 amplification reagent, 1:250) was carried out for 30 min at 40°C, followed by a 5-min PBS wash. Finally, the procedure included incubation with RNAscope Multiplex FL v2 HRP blocker for 15 min at 40°C and a 2-min PBS wash. To complete the process, all slides were covered with fluoromount (00–4958-02; Thermo Fisher), and the coverslips were secured in place using nail polish.

Cell Counting, Imaging, and Data Analysis

We used an Olympus VS120-S6-W Virtual Slide Scanner to scan all sections and the images were captured using a Nikon DS-Fi3 color camera. To mitigate potential bias, a researcher who was unaware of the experimental conditions conducted the photomicrography and cell counting. Cell counting was performed using ImageJ software (v. 1.41; National Institutes of Health, Bethesda, MD), and line drawings were created using Canvas software (ACD Systems, Victoria, Canada, v. 9.0). For each mouse, we analyzed a one-in-two series of 25-µm brain sections, with 50 µm between sections. We defined the analyzed area based on previous reports and identified neurons by their large cell bodies while excluding dendrites and axons. Bilateral counts were averaged, and results are reported as means ± SE. Section alignment relative to specific brain regions was determined according to established references and the Paxinos and Franklin mouse atlas (35).

RESULTS

Gene Expression Profiling of Pons-Medulla

Neural networks in the pons-medulla play crucial roles in coordinating diverse physiological and emotional states, including breathing and cardiorespiratory coupling (6–13). We conducted snRNA-Seq to investigate gene expression changes in the pons-medulla associated with chronic intermittent hypoxia (CIH), focusing on inhibitory neurons, excitatory neurons, and oligodendrocytes (Fig. 1A). Three pairs of C57BL6/J mice were exposed to a cycle of 5%–4.5% oxygen (hypoxia) for 1 min followed by 21% oxygen (normoxia) for 5 min, repeated for 8 h per day over a period of 21 days (CIH). The control group was exposed to a constant 21% normoxic condition in the same room as the experimental group. A modified 10X Genomics nuclei isolation protocol was used to obtain a single nuclei suspension, followed by debris removal using FANS. The 10X Genomics droplet-based platform (v3.1) was then used to generate single-cell bar-coded libraries for sequencing by Illumina-based next-generation sequencing. After performing the necessary preprocessing steps, removing ambient RNA and doublets, a total of 10,995 nuclei were retained. Single-cell transcriptomes were visualized using Uniform Manifold Approximation and Projection (UMAP) res 0.6 (Supplemental Fig. S2). For unbiased annotation of single-nuclei identity, we utilized SingleR (Single Cell Recognition), a computational tool based on reference mapping with MouseRNASeq (36). This analysis revealed and retained major cell classes such as Neurons, Oligodendrocytes, Microglia, Astrocytes, Endothelial, and Fibroblast with neurons being the most abundant cluster (Supplemental Fig. S2B). These major cell classes showed similar expression profiles with well-established anatomical/molecular markers. Oligodendrocyte nuclei showed positive labeling for Mag and Mog, whereas neurons exhibited positive labeling for Syt1, Snap25, and Rbfox3. Astrocytes were positive for Aldh1l1 and microglia expressed Tmem119 (Supplemental Fig. S2C). The neuronal clusters were further distinguished based on the average relative expression levels of markers for excitatory neurons: Slc17a7/vGlut1 and Slc17a6/vGlut2, and markers for inhibitory neurons: Gad1, Gad2, Slc32a1/vGat, and Slc6a5/Glyt2 (Fig. 3, B, C, and D). For the neuronal clusters, clusters Ex4, 5, 9, and 10 demonstrated markers characteristic of vGlut1 and vGlut2 neurons. Cluster “Endo” was colabeled with neuronal markers for inhibitory neurons whereas cluster “Fibro” was colabeled with markers for excitatory neurons.

Figure 1.

Exploring the repercussions of chronic intermittent hypoxia within the pons-medulla region at single-nuclei resolution. A: illustration outlining the procedural and analytical process. 1) three pairs of 21-day-old postnatal C57BL6/J male mice were group housed, one group was exposed to normoxia (21% oxygen), and the other group was repetitively treated with 1 min hypoxia (oxygen dropping to 5%) and 5 min normoxia (21% oxygen) for 8 h during sleep-duration for 21 days, 2) pons-medulla dissection, nuclei dissociation, and nuclei sorting using FACSAria III Cell Sorter, 3) capturing single nuclei, library preparation, sequencing using 10X Genomics and Illumina technologies, and data analysis conducted using the R package Seurat. B: shown is the proportion of different major cell classes identified by mapping with MouseRNASeq captured in the UMAP clustering. C: visualization of the UMAP clustering. Colors indicate distinct cell classes. D: UMAP visualization of single-nuclei transcriptome. Clusters were identified by Louvain algorithm using the top 2,000 genes with the highest variance (PCs = 30/20, and resolution = 0.6/0.8). E: dot plot representing the percentage of nuclei expressing genes for each cluster using FindMarkers, test.use = “roc”. The saturation of color represents the average normalized expression level (scaled and centered). F: gene markers showing distinct anatomical loci of expression within the pons-medulla retrieved from ISH data: Allen Brain Atlas. UMAP, uniform manifold approximation and projection. [Image A created with a licensed version of BioRender.com; image F was used with permission from Allen Brain Atlas; https://mouse.brain-map.org/search/index.]

Figure 3.

Chronic intermittent hypoxia (CIH) dysregulated ion channels. A: volcano plot of dysregulated ion channel genes due to CIH from differential gene expression (DEG) analysis of all neuronal cell classes. Displaying dysregulated ion channels from the differential expression analysis of neurons. B: dot plot showing the percentage of nuclei and the magnitude of expression difference for differentially expressed ion channel genes, categorized by major cell classes, and comparing the conditions of normoxia and CIH. The saturation of color represents the average normalized expression level (scaled and centered). Starred genes were upregulated also in inhibitory cells C: dot plot showing the percentage of nuclei expressing ion channels in the CIH condition by cell clusters. Note, dotted outlines for two clusters: Ex5 showed significant downregulation of Scn9a and Ex6 showed significant downregulation of Kcnq3 (see materials and methods). D: clusters Ex5 and Ex6 in UMAP space shown with cluster marker genes for Ex5 (Cdh23, Slc17a7) and Ex6 (Il16, Slc17a7).

Marker genes for each cluster were determined using the Seurat FindMarker function by “roc” (receiver operating characteristic). The scaled average expression for the selected three marker genes classifying the cell types for each cluster was visualized as a dot plot (Fig. 1E). This allowed us to pinpoint gene expression patterns associated with distinct cellular identities. We further cross-referenced our cell class expression with the Allen Brain Atlas In-situ hybridization data (ISH Data:: Allen Brain Atlas: Mouse Brain; brain-map.org) to pinpoint anatomical localization for these cells. We identified cluster “A” as Monoaminergic neurons, positive for Ddc, Tph2, Slc6a4, Th, and Slc18a2. “Ex7” was identified as excitatory neurons of the inferior olivary nucleus (IO), being positive for vGlut2, Ano2, and Foxp2 (Fig. 1F). “Ex5” maps to excitatory neurons in the tegmental reticular nucleus, based on the expression of vGlut1 and the cluster-specific marker Cdh23. Kit is expressed in the inferior olivary nucleus and observed in “In3” and “Ex7” (Fig. 1F). “Ex9” was identified by the marker Kcnq4; Kcnq4 in the brainstem is specifically expressed in neurons involved in auditory/vestibular processing, including the lateral vestibular nucleus and the spinal trigeminal consistent with previous reports (37, 38). Cholinergic neurons were underrepresented in our whole pons and medulla transcriptome analysis. “Ex2, Ex13” identifies excitatory Vglut2⁺/Sst⁺ neurons, which may include neurons important for respiratory circuitry, as Sst+ interneurons in the preBӧtzinger locus are reported to mark a critical premotor interneuronal pool, which interposes between the inspiratory rhythm-generating circuitry and the hypoglossal motor nucleus (39–43). Markers for the oligodendrocytes did not identify specific local anatomical regions within the pons-medulla, however, Galnt6 expression was selectively abundant throughout the entire brainstem and thalamus. Several markers for each defined cluster are listed in Supplemental Table S1. All clusters are grouped and depicted into eight major cell classes: excitatory neurons, inhibitory neurons, monoaminergic neurons, oligodendrocytes, astrocytes, microglia, fibroblasts, and endothelial cells (Fig. 1C). These analyses provided an initial molecular framework for characterizing the phenotypic diversity of neuronal clusters, as well as other major tissue classes within the mouse pons-medulla.

CIH Alters Cell Class-Specific Patterns of Gene Expression

To gain insights into how CIH alters gene expression patterns, we conducted a differential gene expression (DEG) analysis on the major cell classes of excitatory neurons, inhibitory neurons, and oligodendrocytes, compared between normoxia and CIH conditions. The results showed differences in the gene expression changes among major cell classes. Excitatory neurons exhibited the largest dysregulated gene count of 1,192, followed by oligodendrocytes with 207, and inhibitory neurons with 170 DEG (Fig. 2A).

Figure 2.

Patterns of gene expression of major cell classes due to chronic intermittent hypoxia (CIH). A: volcano plots showing the dysregulated genes for excitatory neurons, inhibitory neurons, and oligodendrocyte cell classes. Upregulated genes are displayed in deep-raspberry and downregulated genes are shown in blue. The number inside the arrows indicates the total number of significantly dysregulated genes for each specific cell class (see text for significance threshold used). B: Gene Ontology (GO) pathways predicted by Metascape, as referenced, reveal enrichment due to chronic intermittent hypoxia (CIH). These enrichments are based on the dysregulated genes for each cell class, with excitatory neurons represented in green, inhibitory neurons in yellow, and oligodendrocytes in deep raspberry. C: four-set Venn diagram showing ChEA3-predicted transcription factors (TFs) from the dysregulated genes, plotted comparing inhibitory-dysregulated TFs with excitatory-dysregulated TFs, inhibitory-dysregulated TFs with oligodendrocyte-dysregulated TFs, and excitatory-dysregulated TFs with oligodendrocyte-dysregulated TFs. Dotted circles on the Venn diagrams show the number/percentage of shared TFs between the compared sets.

To understand the functional implications of these differentially expressed genes (DEGs), we conducted a gene enrichment analysis for annotated biological pathways and processes using Metascape (32). We identified CIH upregulated gene expression in oligodendrocytes associated with oxidative phosphorylation, the production of reactive oxygen species (ROS), and other cellular responses to stress. A similar pattern of results was observed for inhibitory neurons in pathways associated with neurodegeneration, cellular responses to stress, and oxidative phosphorylation suggesting a common profile of upregulated gene expression between oligodendrocytes and inhibitory neurons in response to CIH. Excitatory neurons exhibited substantial dysregulation with both upregulated (44%) and downregulated (56%) genes. These dysregulated transcripts implicated significant alterations in multiple cellular processes and neuronal functions, including synaptic organization, synaptic transmission (trans-synaptic signaling, transmission across chemical synapses), regulation of membrane potential, and neuronal structural remodeling (synapse organization, cell-cell adhesion, axonogenesis) (Fig. 2B).

We next evaluated the transcription factors (TFs) that were differentially enriched by CIH. TFs were identified using ChEA3 (33) from the sets of differentially expressed genes shown in Fig. 2A for upregulated and downregulated genes. Integrated scaled rank scores below 0.02 were considered significant. Three pairwise comparisons were performed: inhibitory neurons versus oligodendrocytes, inhibitory neurons versus excitatory neurons, and excitatory neurons versus. oligodendrocytes. This TF analysis revealed a 485.7% increase in similar TF profiles between inhibitory neurons and oligodendrocytes, compared with inhibitory and excitatory neurons in upregulating TF genes (Fig. 2C). This result was consistent with our gene ontology results (Fig. 2B) suggesting that similar patterns of TFs underlying similar patterns of gene ontology responses between inhibitory neurons and oligodendrocytes in CIH.

A substantial number of CIH-dysregulated DEGs in excitatory neurons are involved with signaling, membrane potential, and regulation of transmembrane transport as suggested by the Metascape analysis (Fig. 2, A and B). We next checked for a list of ion channels (Supplemental Table S2) that were most significantly dysregulated in both excitatory and inhibitory neurons. At the major cell class level (excitatory neurons, inhibitory neurons, monoaminergic neurons, astrocytes, microglia, and oligodendrocytes), genes encoding multiple potassium channels (12 genes) were upregulated in CIH, followed by three calcium channels (Cacna1c, Cacna1a, Cacna1e), one sodium channel (Scn2a), and one transient receptor potential (TRP) channel (Trpm3). Interestingly, two of the three upregulated calcium channels are known to be enriched in presynaptic terminals and implicated in the control of synaptic transmission (Cacna1a or P/Q-type; Cacna1e or R-type) (44–46). A similar pattern was observed for ion channel genes significantly downregulated by CIH across all major cell types. These genes encoded multiple potassium channels (8 genes), followed by two sodium channels (Scn3a, Scn9a) and one TRP channel (Trpc5) (Fig. 3, A and B). Among the dysregulated sodium channel genes, Scn2a (NaV1.2) encodes a sodium channel enriched in excitatory cortical neurons and Scn9a (NaV1.7) encodes a sodium channel implicated in pain sensation (47, 48). Among the set of dysregulated genes encoding potassium channels, members representative of all major functional classes of potassium channels were observed (49, 50) including nonvoltage gated inward rectifiers (Kcnj3/GIRK1, Kcnj6/GIRK2, Kcnj9/GIRK3, Kcnj12/Kir2.2) and voltage-gated outward rectifier leak channels (Kcnk1/TWIK1, Kcnk3/TASK1, Kcnk10/TREK2, Kcnk12/THIK2), which contribute to resting K⁺ conductance. Additional dysregulated potassium channels genes encoded canonical voltage-gated potassium channels (Kcnb2/Kv2.2, Kcnc1/Kv3.1, Kcnc2/Kv3.2, Kcnc3/Kv3.3, Kcnd2/Kv4.2, Kcnd3/Kv4.3), KCNQ-class potassium channels (Kcnq3, Kcnq5), which underlie M-currents, Eag/Erg/Elk-class potassium channels (Kcnh1/Kv10.1/Eag, Kcnh8/Kv12.1/Elk3) potentially modulated by intracellular cAMP, and the broad class of potassium channels gated by cytosolic Ca²⁺ or Na⁺/Cl⁻ (Kcnn2/SK2, Kcnma1/Slo1/BK, Kcnt2/Slo2.1/SLICK), which may confer more specialized repolarizing influences.

We found two clusters that showed specific ion channel genes significantly downregulated by CIH through DEG analysis. Scn9a was significantly downregulated in the excitatory neuronal cluster “Ex5,” which is defined by the expression of the cadherin-related gene marker Cdh23 (blue box in Fig. 3C) and the predominant expression of Slc17a6/vGlut1. The Allen Brain Atlas identifies the labeling of Cdh23⁺/vGlut1⁺ neurons within the tegmental pontine reticular nucleus (Fig. 3D), which may be a candidate anatomical location for neurons identified by this DEG analysis. The second subcluster showed significant downregulation of Kcnq3 in the excitatory cluster “Ex6,” defined by the marker gene Il16 and expression of vGlut1 (Fig. 3D). The GENSAT Brain Atlas project identifies a discrete pool of Il16⁺ neurons within the pons (GENSAT Brain Atlas of gene expression in EGFP Transgenic Mice; data not shown) (51, 52), which may correspond to neurons of cluster Ex6, identified by our DEG analysis.

Overall, our analysis indicates that CIH treatment evokes gene expression responses consistent with alterations in response to oxidative stress, synaptic transmission, and remodeling. In addition, we find the dysregulation of specific sets of ion channels at the level of major cell classes as well as more focal significant downregulation of specific ion channel genes within two distinct excitatory neuronal subclusters. The Allen Brain Atlas and GENSAT Brain Atlas provide candidate anatomical locations within the pons corresponding to these specific neuronal subclusters.

CIH Alters Patterns of Cell-Cell Communication Pathways

Examining the impact of CIH on gene expression across major cell classes and its subsequent effect on pathway enrichment led us to investigate how the clusters influence one another. To address this question, we used CellChat (34). This algorithm leverages gene expression data from individual cells and utilizes mass action models to infer cell-state-specific signaling interactions. It combines these models with differential expression analysis and statistical tests on distinct cell groups to identify potential receptor-ligand communication pairs. Through this approach, we identified the specific network of clusters underlying pairwise signaling interactions under normoxic and CIH conditions within this predicted signaling network.

An analysis of global communication patterns linking clusters in the context of outgoing and incoming signaling under both normoxia and CIH conditions reveals the presence of three distinct patterns for outgoing and incoming signaling under normoxia (Fig. 4A). These patterns are conserved within neurons, oligodendrocytes, and other cell types in their interactions. However, the communication pattern for outgoing signals under CIH conditions undergoes a change, giving rise to a new pattern originating from clusters of neurons and oligodendrocytes (Fig. 4A). Quantitative analysis of intercellular communication revealed a notable reduction in both the number of interaction pairs and interaction strength under CIH (Fig. 4B) and among the clusters (Fig. 4C).

Figure 4.

Cell-cell communication changes due to chronic intermittent hypoxia (CIH). A: river plots showing inferred patterns of signaling between cell clusters by CellChat (34). Two plots within normoxia and CIH show inferred outgoing patterns for the secreted cell clusters and incoming patterns for the target cell clusters. The width of flow lines reflects the relative contribution of the respective cell cluster or signaling pathway to each pattern. B: total interaction strength between cluster groups under normoxia and CIH. C: heatmap illustrating the differential magnitude of signaling interactions in CIH compared with normoxia, as a function of pairwise cell cluster interactions. The top-colored bar plot reflects the summation of column values displayed in the heatmap, indicating incoming signaling. The right-colored bar plot shows the summation of row values, indicating outgoing signaling. In the heatmap, red (or blue) represents increased (or decreased) signaling in CIH relative to normoxia. The relative value in the bar provides a measure of the interaction strength from source to target. Differential signaling interactions are displayed by the number of interactions (left plot) and by interaction strength (right plot). D: an overview of the overall information flow within each signaling pathway. Relative information flow is quantified as the ratio of communication probabilities in the CIH relative to normoxia when both conditions are combined as a function of individual signaling pathways. E: outgoing and incoming signal strengths of individual signaling pathways between cell clusters are presented, depicted by paired plots under normoxia (left) and CIH (right). Note that CIH most impacts TGFβ, CSF, ACTIVIN, VEGF, VISFATIN, and ANGPT signaling pathways (highlighted by outlined boxes). F: circular plots visualizing the pairwise connections between all cell clusters for selected signaling pathways under normoxia (TGFβ) and CIH (TGFβ, VEGF, ANGPT, VISFATIN). Note that CIH decreases TGFβ signaling to microglia (MG cluster), whereas CIH elevates convergent VEGF, ANGPT, and VISFATIN signaling from neuronal and oligodendrocyte clusters to endothelia (end cluster).

Through a comparative analysis of the information flow and interaction strength of each signaling pathway, we identify changes in their activity, including deactivation, reduction, activation, or increased activity, when observing alterations in information flow between normoxia and CIH. The signaling pathways identified are shown in Fig. 4, D and E, suggesting that angiopoietin (ANGPT), visfatin (VISFATIN), and vascular endothelial growth factor (VEGF) signaling pathways are the most prominently activated by CIH, whereas activin (ACTIVIN), angiopoietin-like (ANGPTL), transforming growth factor-β (TGFb), and chemokine ligand (CCL) pathways are downregulated. We then identified potential clusters involved in specific ligand/cognate receptor communication pairs. These results highlighted the enhancement of specific neuronal clusters, including excitatory neuronal clusters (Ex6, Ex9), and inhibitory neuronal clusters (In3, In4, In6), as well as oligodendrocytes (O3, O5) and astrocytes (Ast), which exhibited outgoing angiopoietin (ANGTP) and vascular endothelial growth factor (VEGF) signaling (Fig. 4E). This suggests that exposure to CIH has a selective impact on the expression of specific diffusible ligand-receptor signaling pathways in these clusters.

CIH Elicits Inflammatory Response in Excitatory Neurons

We evaluated the inflammatory response (GO:0006954) (jax.org) for major cell classes and used upregulated genes to score inflammatory enriched populations (Figs. 5 and 6). We observe that excitatory neurons show a highly significant enrichment (p - 1.42e-94), with Il-16 and Camk4 as the most prominently upregulated genes. There was also a significant enrichment in inhibitory neurons (p - 1.57e-12) and oligodendrocytes (p - 2.69e-15). Gene enrichment analysis using Metascape identifies positive inflammatory responses for excitatory neurons. No significant difference was found for microglia; P = 0.07, which may be attributed to the low nuclei count (Figs. 5 and 6).

Figure 5.

Chronic intermittent hypoxia (CIH) dysregulated inflammatory response. A: volcano plots show dysregulated inflammatory response genes for excitatory neurons, inhibitory neurons, and oligodendrocytes. Upregulated genes are displayed in deep raspberry, and downregulated genes are indicated in blue. The numbers within the arrows are the total count of significantly dysregulated genes for each specific cell class (see text for significance threshold used). B: Gene Ontology (GO) pathways predicted by Metascape, as referenced, reveal enrichment attributed to CIH in excitatory neurons. C: violin plot showing the expression of the cytokine, Il16 (log normalized).

Figure 6.

Impaired cell classes by inflammatory response due to chronic intermittent hypoxia (CIH). A: violin plot showing inflammatory enrichment scores for the major class classes. CIH-positive cells are displayed in deep raspberry, and normoxia is in gray. Wilcoxon rank-sum test. B: microglia exhibit a trend toward activation by CIH. Top: UMAP displays microglial nuclei under normoxia (green) and CIH (purple). Bottom: violin plot displays expression level (log normalized) for Trem2, an activated microglial marker. Wilcoxon rank-sum test applied. UMAP, uniform manifold approximation and projection.

DISCUSSION

The pontomedullary region of the brain is essential for survival and critical for coordinating various physiological functions including breathing regulation, the processing of chemosensory information, and cardiorespiratory coupling. In this study, we used single-nucleus RNA sequencing (snRNA-Seq) to investigate the gene expression changes in the pons-medulla due to chronic intermittent hypoxia (CIH), using 10,995 nuclei. The focus was on the major cell classes of inhibitory neurons, excitatory neurons, and oligodendrocytes (Fig. 1). This gene expression profiling of the pons-medulla allowed us to define the molecular diversity of major cell classes as well as clusters for normoxia and CIH (Fig. 1). We found that excitatory neurons, inhibitory neurons, and oligodendrocytes revealed distinct gene expression adaptations to CIH (Figs. 2, 3, 5 and 6), thereby enhancing our understanding of the cellular mechanisms involved. In addition, we elucidated the regulatory roles of ligand secretion within specific subclusters (Ex6, Ex9, In3, In4, In6, O3, O5, and Ast) (Fig. 4).

Functional Implications of Enriched Genes

Gene enrichment analysis was conducted using Metascape to understand the functional implications of differentially expressed genes (DEGs) under CIH. This analysis highlighted that CIH triggered gene expression changes consistent with responses to altered synaptic transmission, cellular remodeling, and oxidative stress within the pons-medulla, as well as shows the distinct sets of gene expression in responses to CIH, by different major cell classes (Fig. 2).

In oligodendrocytes, the results point to significantly upregulated genes that are associated with responses to reactive oxygen species (ROS), oxidative phosphorylation, and heme oxygenase 1 (Hmox1) (Fig. 2B). Upregulation of ROS indicates that it is likely produced and sensed by oligodendrocytes within the pons-medulla in response to CIH. ROS production is associated with intermittent hypoxia (13, 53–55) and is known to be deleterious if generated in excessive amounts. Buffering ROS levels specifically in oligodendrocytes could thus be a potential strategy for mitigating cellular damage due to CIH. The impact on oxidative phosphorylation suggests a heightened vulnerability perhaps due to high constitutive metabolic demand for ATP production; possibly related to extensive myelination in supporting and insulating neurons by forming myelin sheaths around axons (56–58). Interestingly, the upregulation of heme oxygenase 1 (Hmox1) may provide a cytoprotective function in response to cellular stress.

Like oligodendrocytes, inhibitory neurons exhibited a similar pattern of gene expression based on gene enrichment and transcription factor analysis. In inhibitory neurons, CIH induced the expression of genes highly enriched for processes associated with neurodegeneration, oxidative phosphorylation, and cellular stress. The upregulation of metallothionein-associated genes suggests involvement in metal ion homeostasis and detoxification. Metallothionein 3 (MT3) is a small Zn²⁺ binding protein, which may play an important protective role in response to oxidative stress by scavenging excessive ROS production and regulating autophagy and apoptosis (59).

CIH-induced gene dysregulation in excitatory neurons was collectively associated with intricate and highly regulated mechanisms. These mechanisms are used by these neurons to transmit signals and modulate synaptic strength. Excitatory neurons play critical roles in brainstem neuronal networks, implicated in both respiratory rhythmogenesis (42, 60–68) and arousal (21, 24, 68). Central respiratory rhythmicity is an emergent property of dynamic signaling within discrete neuronal networks (69). Gene expression changes that impact intrinsic neuronal capacity for chemical neurotransmission and postsynaptic responses even modestly could thus be greatly amplified in the context of in vivo neural networks, perhaps in unpredictable ways. Further functional studies will be necessary to better understand how the CIH-induced changes in gene expression affect neuronal network functions. We know that in the preBötC CIH decreases neuronal excitability, which leads to irregularities in respiratory frequency and failed excitatory transmission to the hypoglossal nucleus (54). CIH has also been reported to enhance glycinergic synaptic transmission within the nucleus tractus solitarius (NTS) (70). We observed that excitatory neurons also showed a highly significant enrichment, with Il16 and Camk4 as the two most prominently upregulated genes by CIH. Il16 belongs to the category of molecules known as alarmins, also known as danger-associated molecular patterns (DAMPs). These are endogenous molecules released by damaged or stressed cells, triggering an immune response. (71) (Fig. 5). Camk4 encodes a multifunctional calcium/calmodulin-dependent serine/threonine kinase, which has been implicated in transcriptional responses to cerebral hypoxia (72). Gene enrichment analysis using Metascape identifies positive inflammatory responses primarily in excitatory neurons. No significant difference was found for microglia (Fig. 6), although a trend toward greater activated microglia was observed with the marker Trem2, for activated microglia. The lack of significant differences in microglia should not be overinterpreted. Microglia are very reactive and potential differences induced by CIH may, for example, be masked by a larger response to the tissue handling itself, but it is interesting to note that there are variations within the clusters from pons-medulla responding differently to CIH (Figs. 2 and 3, and Supplemental Fig. S4).

In our analysis, we observed a prevalence of downregulated genes within excitatory neurons. This downregulation extends to processes such as synapse organization and synaptic signaling. In addition, the downregulation of cell-cell adhesion molecules may potentially reduce synaptic plasticity, impacting cell communication, and signaling, which could potentially alter neuronal network connectivity and excitability. Prolonging the CIH treatment while carefully monitoring changes in excitatory neurons could provide valuable insights and identify potential interventions to address this altered neurophysiological state. Further experimental investigations and functional studies are warranted to elucidate the exact implications of these changes in neuronal function, and its contribution to overall brain physiology and pathology. Overall, the results focusing on major cell classes, indicate that CIH exerts a pronounced and differential impact on the gene expression of specific cell types within the pons-medulla region.

Other cell types, such as microglia, astrocytes, and endothelial cells, were present in smaller numbers and could not be included in the analysis to identify significant gene expression differences. Targeted enrichment of these populations could enhance the interpretability of the results. Within the neuronal population, we observed that clusters 0 and 1 exhibited a high degree of complexity with fewer variations. Applying iterative clustering using Scrattch.hicat generated clusters (data not shown) presented methodological challenges for validating these subtypes within the pons and medulla (https://github.com/AllenInstitute/scrattch.hicat) (73, 74). Additional snRNA-Seq analysis focused on specific anatomical regions, incorporating spatial location data will be necessary to better characterize region-specific neuronal populations within the pons-medulla (Supplemental Fig. S1).

Our study has several important limitations. As stated earlier, a more detailed spatial analysis of expression changes in functionally defined cells in the pons and medulla will be necessary to obtain deeper insights into the physiological relevance of the CIH-induced changes for regulating autonomic functions. Moreover, the present study characterized only one relatively small developmental time window, and we would expect that older or younger animals may exhibit different changes than those characterized here. Furthermore, the present study was only performed in males. Follow-up studies characterizing changes in females and assessing sex differences will be of great clinical interest.

Cell-Cell Communication in CIH

Our analysis of signaling pathways using CellChat under normoxic and chronic intermittent hypoxia (CIH) conditions provided insights into the diversity of cellular responses to CIH. By evaluating both conserved and context-specific pathways, we gained first insights into the molecular interactions occurring in CIH-exposed cells. The determination of information flow within the inferred networks suggested significant alterations in signaling patterns. Notably, ANGPT, VISFATIN, and VEGF emerged as key players enriched in CIH, whereas ACTIVIN, ANGPTL, TGFb, and CCL were downregulated, underscoring the dynamic nature of cellular signaling in response to hypoxia (Fig. 4D) (75, 76). The result indicates that ligand secretion by specific subclusters (Ex6, Ex9, In3, In4, In6, O3, O5, and Ast) may be involved in activating ANGPT, VISFATIN, and VEGF signaling pathways (Fig. 4E). Both ANGTP (angiopoietin) and VEGF (vascular endothelial growth factor) are activated upon HIF1α activation under hypoxia. These pathways may potentially play a role in vascular protection or hypertrophy underlying hypertension or vascular remodeling (75, 77–79). These findings suggest that CIH exposure not only alters cellular sensitivity to hypoxic conditions but also redefines the regulatory roles of cell signaling networks.

In conclusion, our study provides first insights into the gene expression changes occurring in the pons-medulla region under chronic intermittent hypoxia, offering valuable insights into the cellular responses and potential mechanisms underlying CIH-induced effects. This research contributes to our understanding of how CIH impacts cellular function, communication, and signaling within this crucial brain region involved in various physiological processes.

DATA AVAILABILITY

The raw data generated during this study have been uploaded to the National Center for Biotechnology Information Gene Expression Omnibus database (NCBI GEO), Accession No. GSE256102 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE256102).

SUPPLEMENTAL DATA

Supplemental Figs. S1–S4 and Supplemental Tables S1 and S2:

https://doi.org/10.6084/m9.figshare.24299041.v6.

GRANTS

This study was funded by the following grants from the National Institutes of Health under the National Heart, Lung, and Blood Institute: P01 HL144454 and Project 2 (awarded to J.-M.R.), R01 HL144801 (awarded to J.-M.R.), R01 HL151389 (awarded to J.-M.R.), and R01 HL126523 (awarded to J.-M.R.).

DISCLOSURES

J.-M.R. is an editor of Journal of Neurophysiology, an American Physiological Society-associated journal, and was not involved and did not have access to information regarding the peer review process or final disposition of this article. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

H.B. and A.D.W. conceived and designed research; H.B., A.D.W., and L.M.O. performed experiments; H.B., A.D.W., and L.M.O. analyzed data; H.B., A.D.W., and L.M.O. interpreted results of experiments; H.B. and L.M.O. prepared figures; H.B., A.D.W., K.A.A., and J.-M.R. drafted manuscript; H.B., A.D.W., K.A.A., and J.-M.R. edited and revised manuscript; H.B., A.D.W., L.M.O., K.A.A., and J.-M.R. approved final version of manuscript.