Comparative Ion Channel Transcriptomes of NK1R and Somatostatin Neurons in the PreBötzinger Complex of the Ventrolateral Medulla Authors: Hemalatha Bhagavan, Aguan D Wei, Luiz M Oliveira, Jan-Marino Ramirez Published: 2026 ABSTRACT AbstractThe preBötzinger complex is among the few neural circuits where selective elimination of a defined neuronal subpopulation is sufficient to destabilize a core autonomic function and can fatally impair breathing. Within this circuitry, neurons expressing the neurokinin-1 receptor (Tacr1/NK1R) and somatostatin (Sst) are critical subpopulations; Tacr1+ neurons respond to the neuropeptide substance P and are necessary for maintaining inspiratory rhythms, and ablation of Sst+ neurons results in apneas. To further dissect and analyze the specific roles of Tacr1+ and Sst+ cell types, we conducted a comprehensive transcriptomic analysis using single-nucleus RNA sequencing of the sampled preBötC tissue from neonatal C57BL/6 J mice. As respiratory rhythmogenesis is an inherently electrophysiological process, we focused on the ion channel transcriptomes of Tacr1+ and Sst+ populations as a first step towards resolving the molecular underpinnings of their distinct contributions to breathing. A balanced random forest classifier distinguished Tacr1+ from Sst+ neurons with higher accuracy, indicating a distinct and relatively homogeneous ion channel identity in Tacr1+ neurons. Differential expression analyses identified coordinated upregulation of Trpc5, Kcnc2, and Cacna2d2 genes in Tacr1+ neurons. Tacr1+ neurons further exhibited elevated expression of the NALCN channelosome subunits and selective enrichment of Htr2c and Adra1a neuromodulatory receptor genes. Together, these findings define a molecularly distinct ion channel composition in Tacr1+ neurons and imply convergent substance P-linked mechanisms: TRPC5-mediated ICAN and NALCN-mediated sodium leak conductance supporting rhythmic inspiratory activity, providing a molecular framework for targeted interrogation of respiratory circuit function.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s12035-026-06162-7.Keywords: PreBötzinger, Single-cell transcriptomics, Ion channels, Neurokinin-1 receptor, Tachykinin receptor 1 Introduction Breathing is vital and generated by neuronal networks within the lower brainstem. Investigations into identifying the neuronal network(s) underlying the generation of the respiratory rhythm are an area of active research, and a better understanding of how breathing is regulated at the level of specific identifiable neuronal types would provide a crucial step for devising targeted prevention therapies in disorders associated with breathing disturbances or sudden death due to failure to breathe [1–5]. The preBötzinger complex (preBötC) within the ventral respiratory column (VRC) is the core region responsible for generating the inspiratory rhythm [6–10]. Ablation of this region leads to cessation of breathing [11–14]. The inspiratory rhythm activity generated by the preBötC is critically dependent upon the embryonic expression of the homeobox transcription factor Dbx1 gene in V0 precursor cells of the developing medulla between E10.5 and 11.5 [15, 16]. In the absence of Dbx1 expression, the preBötC fails to generate rhythmic bursting activity, disrupting the core neural circuits required for respiration in the developing embryo. This also leads to a complete loss of somatostatin (Sst)- and the neurokinin 1 receptor (NK1R/Tacr1)-expressing neurons in the preBötC [15–24]. In concordance, silencing this Sst+ population led to apnea in adult rats [11], and specific and near-complete bilateral destruction of preBötC NK1R neurons resulted in ataxic breathing [14]. These findings emphasize the essential role of the Sst+ and Tacr1+ neuronal populations in maintaining respiratory rhythm. To gain insights into the specific role of Sst+ and Tacr1+ expressing neuronal populations in inspiratory rhythm generation, researchers have integrated pharmacological, genetic, and viral tracing approaches with electrophysiology [11, 25–34]. Here, we further characterize these populations by profiling ion channel gene expression at the single-nucleus level. To this end, we performed single-nucleus RNA sequencing (snRNA-seq) on micro-dissected brainstem tissue centered on the preBötC from seven neonatal C57BL/6 J mice (P5–P10), yielding a dataset from which we identified 465 Sst+ and Tacr1+ nuclei for downstream analysis. Our findings reveal distinct ion channel transcriptomic signatures for these two populations, supporting the notion that they serve specialized roles in respiratory rhythmogenesis. We examined whether Sst+ and Tacr1+ neuronal populations could be distinguished based on their ion channel transcriptomes, despite sharing a common Dbx1 progenitor lineage. Using a random forest classifier, we assessed classification performance and found that Tacr1+ neurons were identified with significantly greater accuracy than Sst+ neurons. We then performed differential expression (DE) analysis to highlight key transcriptomic features and found enrichment of ion channel genes in Tacr1+ neurons, including potassium channels such as Kcnc2, the voltage-gated calcium channel auxiliary subunit Cacna2d2, and the TRP channel Trpc5. We next evaluated components of the NALCN sodium leak channel complex and observed differences consistent with altered baseline excitability, alongside distinct neuromodulatory G protein-coupled receptor (GPCR) profiles enriched in the Tacr1+ population. Finally, we integrated these findings to demonstrate that Tacr1+ neurons exhibited clear transcriptomic specializations within the sampled preBötC/VRC [30], suggesting molecular mechanisms that may underlie their role in respiratory rhythm generation and modulation. Materials and Methods Animals Neonatal (P5–P10) C57BL6/J mice bred at Seattle Children’s Research Institute were used for sample collection in compliance with the guidelines of the Seattle Children’s Research Institute Animal Care and Use Committee and the National Institutes of Health. All animals had free access to water and food and were housed in a temperature-controlled (22 ± 1 °C) facility with a 12:12 h light/dark cycle. Tissue Processing Animals were deeply anesthetized with 3.0% isoflurane and decapitated, after which isolated brainstems were immediately placed in ice-cold artificial cerebrospinal fluid (aCSF with an osmolarity between 305–312 mOsm, containing (in mM): 118 NaCl, 3.0 KCl, 25 NaHCO3, 1 MgCl2, 1.5 CaCl2, and 30 d-glucose equilibrated with carbogen (95% O2, 5% CO2) at pH 7.4 ± 0.05. Brainstems were affixed on 5% agar blocks with cyanoacrylate glue and transferred to a vibratome (VT1000S, Leica Microsystems, Waukegan, IL, USA) with the rostroventral edge facing toward the blade for serial transverse sectioning along the rostral-caudal axis. Serial sectioning continued until the caudal edge of the facial nucleus was reached, after which a 400-µm slice was taken to expose the rostral edge of the preBötC [35–37]. A single 600-µm slice was then made, containing the entire preBötC. PreBötC-containing tissue was microdissected bilaterally from slices using a 33-gauge scalpel needle under visual guidance with a dissecting microscope. Boundaries for the dissection were identified using visible anatomical landmarks within the slices: the nucleus ambiguus (NA) and inferior olive (IO) were visualized, and tissue immediately ventral to the NA was collected, while tissue immediately surrounding the IO was avoided. As the dissection spanned the full 600-µm rostrocaudal thickness of the slices, the sampled tissue likely included a variable portion of the rostral ventral respiratory group (rVRG) in addition to the complete preBötC. Although our tissue dissections were centered on the preBötC, it is likely that the resulting dataset includes non-respiratory neurons due to the anatomical heterogeneity of the sampled tissue. All tissue collections were performed in oxygenated aCSF prechilled on ice and stored at − 80 °C until further processing. Proper care and ethical guidelines were followed for all animal handling and sample collection, under IACUC protocol 0058 approved by Seattle Children’s Research Institute (SCRI). Single Nuclei Preparation Single-nucleus suspensions were obtained as previously reported [38]. Briefly, we followed the 10X Genomics protocol (CG000366) with key modifications. First, Tween-20 was excluded from the working wash buffer. Second, the lysis buffer was supplemented with Nonidet-P40 Substitute (74,385, Sigma) at a final concentration of 0.5% and Tween-20 at 0.1%, per the manufacturer’s recommendation. Frozen tissues were mechanically dissociated in sterile 1.5 ml microcentrifuge tubes using a 1.5-ml pellet pestle in 1 × lysis buffer and incubated for 10 min to release free nuclei. The suspension was then centrifuged at 500 rcf for 5 min at 4 °C. The resulting nuclear pellet was washed, resuspended in a wash buffer, and passed through a 70-μm filter (pluriStrainer Mini 43–10,070−40). These modifications facilitated the successful isolation of nuclei for downstream analysis. All steps were conducted on ice. Single Nuclei RNA Sequencing We used seven brainstem samples of C57BL6/J mice between P5 and P10. They were micro-dissected and pooled to obtain the sampled preBötC region. Pooled nuclei were processed in two technical replicates. Barcoded cDNA libraries were prepared using the Chromium Single Cell 3′ Library and Gel Bead Kit v2 (10X Genomics). Sequencing was performed on an Illumina NovaSeq 6000 (S1 flow cell) to a total depth of 190 million reads (Northwest Genomics Center, University of Washington, Seattle, WA, USA). Raw sequencing reads were processed to count matrices with Cell Ranger (version 2.2.1) with default parameters, using the mm10 mouse genome as the reference. All statistical analyses were performed using R (version 4.4.1). Quality Control, Preprocessing, Normalization, Clustering, and Annotation Expression matrix files were analyzed using the Seurat v5.0 package [39]. To eliminate ambient RNA contamination, SoupX [40] was applied. We excluded nuclei by the criteria of nFeature_RNA > 800, nCount_RNA > 1000, and percent.mt < 15%. Cell cycle scoring was implemented to check the effects of cell cycle heterogeneity. The data were normalized using the sctransform (SCT) method with glmGamPoi, incorporating the proportion of mitochondrial and cell cycle genes as a variable to regress out [41, 42]. Principal components were computed, and key dimensions were selected based on association with genes of interest (Slc17a6, Slc17a7, Chat, Gad1, Gad2, Slc6a5, Slc32a1, Syt1, Mag, Mog, Aldh1l1) within the loading matrix. The top dimensions corresponding to genes of interest were then used to construct the nearest neighbor graphs, while adjusting for correlations between average expression and dispersion. Uniform Manifold Approximation and Projection (UMAP) was performed using the key dimensions from principal components (PCs) with a modularity optimization Louvain algorithm. Clustering was optimized using Seurat with a default resolution parameter, yielding a dataset of 7872 nuclei from the sampled preBötC. Single-cell recognition (SingleR) was used to annotate clusters by assigning cell type labels to individual cells based on their transcriptomic profiles and comparing them with built-in reference gene expression data from a mouse database. Using correlation and statistical methods, SingleR determines the most likely cell identity for each cell, providing accurate and automated annotations. We subsetted neurons from the sampled preBötC. From the neuronal cluster of 4751 nuclei, Tacr1+ and Sst+ neurons were isolated by subsetting cells with normalized gene expression of Tacr1 and Sst genes greater than 0.5. Cell type designations follow the gene symbol (no italics, first letter capitalized) with a superscript +, indicating a population defined by expression of that marker gene (e.g., Tacr1+, Sst+). Ion Channel and Receptor Analysis For all ion channel analyses described in this study, we curated a reference list of 229 ion channel genes, of which 192 were detected in the dataset across the Sst+ and Tacr1+ populations. Additionally, 62 receptor genes were curated, all of which were detected in the dataset. The complete annotated gene list, including family classifications, is provided in Table S1. Random Forest Analysis For the random forest analysis, two sequential filtering steps were applied to the original 192 detected ion channel genes, which led to 118 ion channel genes. First, we retained ion channels expressed in at least 10% of all cells combined (pct_total > 10%), thereby removing very lowly expressed or near-absent channels from the dataset. Second, we applied a minimum detection threshold of > 3.2% in Sst+ cells. This prevented bias toward artificially improved classification performance in the larger group, Tacr1+ cells (n = 313). To address the unequal group sizes (Sst+: 152 cells; Tacr1+: 313 cells) and reduce sampling bias, we used a balanced iterative resampling approach. We ran 200 iterations that randomly drew an equal number of cells from each population (matching the smaller group, that is, the Sst+ population), trained a random forest model on the balanced subset using 500 trees, and recorded model performance with cell type as the outcome variable and the 118 filtered ion channel genes as features. Random forest models [43, 44] were implemented using the RandomForest package in R. At each node, the best split was selected using Gini impurity as the splitting criterion, with the number of features randomly sampled at each split set to the default value of √p, where p is the total number of features. Bootstrap samples for each tree were drawn with replacement from the balanced subset, with the remaining out-of-bag samples used as an internal test set for that tree. Model performance was evaluated using out-of-bag (OOB) error estimates, and per-class accuracy was extracted from the OOB confusion matrix. Feature importance was assessed using mean decrease in accuracy, aggregated across all 200 iterations. To further evaluate discriminability, ROC curves were computed for each iteration using the pROC package, with predicted class probabilities for Tacr1+ cells as the classifier score. Mean AUC and standard deviation were calculated across iterations as a summary measure of overall classification performance. To validate that OOB accuracy estimates generalized beyond the balanced training subset, a stratified held-out evaluation was additionally performed. In each iteration, the 161 Tacr1+ cells not selected for training were withheld entirely from forest construction and evaluated using out-of-sample prediction via the trained model. Held-out Tacr1+ accuracy was computed as the proportion of correctly classified cells and compared with the OOB accuracy of the training subset across all 200 iterations. Unlike OOB accuracy, which aggregates bootstrap votes across 500 trees for each training cell and thereby dampens iteration-to-iteration variance, held-out accuracy was computed as a single proportion across 161 withheld cells whose composition varied across iterations. This difference in estimator structure means that sampling variance across Tacr1+ cell draws is more directly visible in the held-out distribution, which is expected to be wider than the OOB distribution despite comparable means. All results, including accuracy distributions, confusion matrices, ROC curves, and feature importance scores, were derived from a single forest per iteration to ensure internal consistency. Differential Expression Analysis Seurat’s “FindMarkers” function was used to identify DE genes between population comparisons. The Wilcoxon rank-sum test with Benjamini–Hochberg correction was applied to obtain adjusted p values. Genes with an FDR-adjusted p value of < 0.05 and an absolute log2 fold change of > 0.5 were considered differentially expressed. To complement the analysis, Fisher’s exact tests were performed on binary detection rates for the candidate ion channel genes identified from the differential expression results. For each gene, detected cell counts per population were derived from the proportion of expressing cells reported by FindMarkers, and contingency tables were constructed for each gene and corrected with Benjamini–Hochberg. Effect sizes were calculated as Cohen’s d using the difference in mean normalized expression between Tacr1+ and Sst+ neurons divided by the pooled standard deviation. Pseudobulk Analysis To evaluate the directional consistency of ion channel gene enrichment across analytical approaches, an exploratory pseudobulk analysis was performed. Raw counts for the 118 filtered ion channel genes were aggregated by cell type and technical replicate using the “AggregateExpression” function in Seurat, yielding four pseudobulk samples (Tacr1+ and Sst+ across two technical replicates). Aggregated counts were normalized to counts per million (CPM) to account for differences in the number of cells contributing to each pseudobulk sample, followed by log2 transformation (log2(CPM + 1)). For each gene, log2 fold change was calculated as the difference in mean log2CPM between Tacr1+ and Sst+ groups across replicates. No formal statistical testing was performed, as the dataset comprised two technical replicates rather than independent biological replicates, and the analysis was used solely to assess the directionality of expression differences for genes identified as discriminatory by random forest (RF) classification and to confirm consistency with Fisher’s exact enrichment results. Module Scoring of NALCN Channelosome Gene Signature Module scores for the NALCN channelosome were calculated using the “AddModuleScore” function in Seurat [45], which computes the average expression of a defined gene set relative to randomly sampled background genes of similar expression levels. The gene set comprised transcripts encoding the four NALCN channelosome subunits: Nalcn (pore-forming), Unc79, Unc80, and Fam155a (auxiliary subunits). For the plotting, values below the 5th and above the 95th percentile were cut to reduce the influence of outliers on the color scale. Histology Adult mice (n = 3) were deeply anesthetized with 4% isoflurane delivered in 96% oxygen. Transcardial perfusion was performed via the ascending aorta with 20 ml of phosphate-buffered saline (PBS; pH 7.4), followed by 20 ml of 4% paraformaldehyde in 0.1 M phosphate buffer (PB; pH 7.4; Electron Microscopy Sciences, Fort Washington, PA, USA). Brains were carefully extracted and post-fixed in the same fixative at 4 °C for 4 h, then dehydrated in 20% sucrose for 8 h. Coronal Sects. (25 μm thick) were obtained using a cryostat and preserved at − 20 °C in cryoprotectant solution (20% glycerol and 30% ethylene glycol in 50% PB 0.1 M, pH 7.4). The slices were rinsed in PBS, mounted sequentially in rostrocaudal order, air-dried, and kept at − 80 °C. In situ hybridization was performed using the RNAscope Multiplex Fluorescent Reagent Kit (Advanced Cell Diagnostics). Brain sections were first defrosted in room air, fixed in 4% PFA for 15 min, and dehydrated sequentially in increasing alcohol concentrations (50, 70, and 100%, 5 min each). The slides were air-dried, and a hydrophobic barrier was drawn around the sections. The slides were then incubated in 0.1 M PBS for 5 min, followed by treatment with RNAscope hydrogen peroxide (322,381) for 10 min at room temperature. After three 2-min PBS washes, sections were incubated with Protease IV (322,381) for 30 min at room temperature and then rinsed again in PBS. Tissue was subsequently hybridized with Trpc5 (476,241) and Tacr1 (428,781) probes (purchased from Advanced Cells Diagnostics (ACD)) for 2 h at 40 °C, followed by a brief PBS wash. Signal amplification was carried out using RNAscope Multiplex Detection Reagents (323,110), including sequential incubations with FL v2 Amp1 and Amp2 for 30 min at 40 °C, with 5-min PBS washes between steps. Sections were then incubated with RNAscope Multiplex FL for 30 min at 40 °C. For HRP-C1 and HRP-C2 signal development, tissues were incubated with the corresponding RNAscope Multiplex FL v2 HRP reagents for 15 min at 40 °C, followed by PBS washes. Signal detection was achieved using TSA Plus fluorescein (TS-0002000) or Cyanine 3 amplification reagent (FP1170; 1:800) for 30 min at 40 °C. Finally, sections were treated with RNAscope Multiplex FL v2 HRP blocker for 15 min at 40 °C, rinsed, mounted with Fluoromount (Thermo Fisher), and coverslipped. Cell Counting, Imaging, and Data Analysis All sections were digitized using a VS120-S6-W Virtual Slide Scanner (Olympus). To minimize bias, imaging and quantification were conducted by a single experimenter blinded to experimental conditions. Cell counts were performed using ImageJ (version 1.41; National Institutes of Health, Bethesda, MD), and schematic illustrations were generated using Canvas software (ACD Systems, Victoria, Canada, version 9.0). A one-in-two series of 25-μm sections was analyzed per animal, resulting in 50 μm spacing between sampled sections. Data are presented as mean ± SEM. Anatomical alignment for the VRC was performed using a standardized reference section, in accordance with prior work from our laboratory and the Paxinos and Franklin mouse atlas [46]. Results Transcriptomic Analysis of the Tacr1+ and Sst+ Neuronal Population To define the molecular architecture of Sst+ and Tacr1+ neurons at single-cell resolution, we performed snRNA-seq on brainstem tissue sampled from the preBötC region. Tissue from seven C57BL/6 J mice (P5–P10) was micro-dissected and pooled for sequencing. Nuclei were isolated using a modified 10X Genomics protocol, and libraries were prepared using the Chromium Single Cell 3′ Kit v3.1. After quality control and filtering (see “Materials and Methods”; Fig. S1a), Tacr1+ and Sst+ neurons were identified by subsetting nuclei expressing Tacr1 or Sst genes (normalized expression > 0.5); these populations are referred to hereafter as Tacr1+ and Sst+ neurons, respectively. It is important to note that Tacr1+ and Sst+ neurons constitute relatively small subsets within the broader neuronal landscape of the sampled preBötC, comprising 8.34% (n = 396) and 3.26% (n = 155) of total neurons (n = 4751), respectively, with only a small fraction (0.32%; n = 15) co-expressing both Tacr1 and Sst (Fig. S1d). This proportional representation is broadly consistent with prior anatomical studies identifying NK1R-expressing neurons as a specialized minority population within the preBötC, estimated to comprise approximately 10% of preBötC neurons [30]. This indicates that the current enrichment strategy is consistent with preserving physiologically relevant representation of Tacr1+ and Sst+ neuronal populations. Dimensionality reduction using UMAP (resolution = 0.8) showed distinct clustering within this dataset (Fig. S1b). To restrict the analysis to preBötC neurons (Dbx1-derived), we excluded cluster 5, identified as cholinergic motor neurons of the NA based on selective Chat expression, likely co-dissected during preBötC tissue enrichment [15] (Fig. S1c). We also excluded cells co-expressing Sst and Tacr1 genes (Fig. S1d–f) as co-expressing cells could confound population classification for our methods, as these cells represented only 0.32% of the dataset (n = 15), too few for reliable transcriptomic comparison. In total, 465 nuclei were retained for downstream analysis, with a mean of 1768 detected genes per nucleus (Fig. 1a, b). Based on exclusive marker expression, 152 nuclei were classified as Sst+ and 313 as the Tacr1+ population (Fig. 1b). Diversity of Sst+ and Tacr1+ preBötC/VRC neurons. a Violin plots showing quality control metrics for Sst+ and Tacr1+ populations, including the distributions of detected genes (nFeature RNA), total UMI counts (nCount_RNA), and mitochondrial transcript percentage (percent.mt). b UMAP projection of Sst+ (salmon) and Tacr1+ (teal), with a bar chart indicating the number of nuclei per population. c UMAP projection of neurotransmitter identity showing GABAergic markers Gad1, Gad2, and Slc32a1; mixed GABAergic/glycinergic markers in dark blue, the glycinergic marker Slc6a5 in teal, the glutamatergic marker Slc17a6 in red, and co-expressed transmitter identity in gray. The bar chart indicates the number of nuclei per neurotransmitter identity. d Stacked bar plots showing the proportion of neurotransmitter composition in Sst+ and Tacr1+ populations. e UMAP feature plots showing expression of canonical neurotransmitter marker genes across the dataset: Slc17a6 (glutamatergic), Slc6a5 (glycinergic), Gad2 and Gad1 (GABAergic), and Slc32a1 (vesicular inhibitory transporter). Cells expressing each marker are highlighted in magenta To assess neurotransmitter identity, we examined canonical excitatory and inhibitory markers. Projection of genes Slc17a6 (vGlut2), Slc6a5 (GlyT2), Gad1, Gad2, and Slc32a1 (vGat) onto the UMAP embedding revealed that both Sst+ and Tacr1+ populations comprise glutamatergic (21%), glycinergic (25%), and GABAergic (51%) phenotypes in mice, with a small fraction (2%) co-expressing multiple neurotransmitter markers (Fig. 1c–g). Despite some overlap, clear differences in neurotransmitter composition were observed between the populations. Tacr1+ neurons contained higher proportions of glycinergic and glutamatergic cells and a lower proportion of GABAergic cells compared to the Sst+ population. All 465 nuclei were retained for subsequent ion channel and receptor analyses. Ion Channel Transcriptomic Distinctions Between Tacr1+ and Sst+ Populations Ion channels are fundamental determinants of intrinsic neuronal excitability, shaping firing patterns and responses to neuromodulation. Given the known functional differences between Sst+ and Tacr1+ neurons in respiratory rhythm generation, we asked whether distinct ion channel expression profiles could define population-specific excitability. To address this, we first assessed whether neurons could be classified based solely on ion channel transcriptomic profiles using a RF classifier [43]. Across 200 balanced iterations (n = 152 nuclei per group), both populations were classified above chance. Tacr1+ neurons were identified with a mean OOB accuracy of 61.1% (SD = 3.37%), whereas Sst+ neurons showed lower accuracy (55.6 ± 3.32%), yielding an overall mean accuracy of 58.4% (SD = 2.64%; Fig. 2a). Modest accuracy and substantial overlap between populations indicate that ion channel expression differences are graded rather than categorical. Random Forest classification of Sst+ and Tacr1+ preBötC/VRC neurons. a Distribution of classification accuracy across 200 balanced random forest iterations (n = 152 nuclei per class). Density plots show classification accuracy for Sst+ neurons (red), Tacr1+ neurons (teal), and overall out-of-bag (OOB) accuracy (gray). Dashed lines indicate mean accuracy for each group. Both populations were classified above chance, with Tacr1+ neurons classified with significantly greater accuracy than Sst+ neurons across iterations. b Confusion matrix summarizing classifier performance across all iterations. Rows indicate the true class and columns indicate the predicted class. Color intensity reflects classification accuracy (%). Tacr1+ neurons were correctly classified at a higher rate (61.1%, n = 93) than Sst+ neurons (44.3%, n = 67), indicating an asymmetry in classification performance. c Top ion channel genes contributing to classification performance, ranked by mean decrease in accuracy across iterations. Bars indicate the relative importance of each gene in the random forest model, with error bars representing variability across iterations. Colors indicate ion channel family/functional class. Red, TRP channels; cyan, potassium channels; purple, calcium channels; black/grey, others; green, NALCN channelosome Despite this overlap, we found that classification performance was significantly asymmetric (Δ = 5.49%, Cohen’s d = 1.39, p < 0.001; Fig. 2a), indicating that Tacr1+ neurons are more readily distinguishable in ion channel space, with this asymmetry being highly consistent across iterations. Receiver operating characteristic (ROC) analysis [47, 48] confirmed above-chance discriminability across all iterations, with a mean AUC of 0.615 ± 0.015 (Fig. S2a). The low variance across iterations (Tacr1+ SD = 3.37%, Sst+ SD = 3.22%) further indicates that this asymmetry is a stable population-level feature rather than a sampling artifact. Consistent with this, confusion matrix analysis (Fig. 2b) showed that Sst+ neurons were more frequently misclassified as Tacr1+ (44.4%) than the reverse (38.9%), reinforcing the relative distinctness of the Tacr1+ population. To confirm that OOB accuracy estimates were not dependent on randomly selected subsets of 152 Tacr1+ cells, a stratified held-out evaluation was performed in each iteration. The remaining 161 Tacr1+ cells, withheld from training in each iteration, were classified using out-of-sample prediction. This yielded a held-out accuracy distribution that closely overlapped with the OOB estimate (Fig. S2b), confirming that classification performance generalizes to unseen Tacr1+ cells and is not a sampling artifact of the balanced subset. Key genes contributing to classification included Trpc5, Kcnd2, Cacna2d2, Cacna1c, and Kcnb2 (Fig. 2c). We next performed DE analysis focusing on ion channel genes (see “Materials and Methods”). Although individual gene-level effect sizes were modest (log2FC = 0.44–0.91; Cohen’s d = 0.27–0.34), a coordinated enrichment pattern emerged in Tacr1+ neurons (Fig. 3a; Fig. S2c). Enriched genes included potassium channel subunits (Kcnc2, Kcnab1), the voltage-gated calcium channel auxiliary subunit Cacna2d2, and the TRP channel Trpc5. While not all genes passed Benjamini–Hochberg correction at padj < 0.05 (Fig. S2c), population-level enrichment was supported by Fisher’s exact test with correction. Notably, Trpc5 showed the strongest enrichment (OR = 2.29, CI [1.40, 3.87], padj = 0.0022), followed by Dpp10 (OR = 2.37, CI [1.24, 4.51], padj = 0.0058), Cacna2d2 (OR = 1.97, CI [1.27, 3.10], padj = 0.0038), Kcnc2 (OR = 1.77, CI [1.17, 2.68], padj = 0.0058), and Kcnab1 (OR = 1.78, CI [1.17, 2.68], padj = 0.0058; Fig. 3b). These results indicate that enrichment reflects both increased detection frequency and elevated/reduced expression levels in Tacr1+ neurons (Fig. 3c). Tacr1+ neuronal population is enriched for specific ion channel transcripts in the preBötC/VRC. a Differential ion channel gene expression between Tacr1+ and Sst+ neurons. Each point represents an ion channel gene plotted by log2 fold change vs. effect size (Cohen’s d). Positive values indicate higher expression in Tacr1+ neurons, whereas negative values indicate higher expression in Sst+ neurons. Point size reflects statistical significance − log10 adjusted p value). Labeled transcripts represent the enriched candidates. b Forest plot showing enrichment of ion channel genes in Tacr1+ neurons based on detection rates using Fisher’s exact test (Benjamini–Hochberg corrected). Points represent odds ratios for gene detection in Tacr1+ relative to Sst+ neurons, with horizontal bars indicating 95% confidence intervals. Dashed vertical line marks OR = 1. c Violin plots show normalized expression levels of Trpc5, Cacna2d2, and Kcnd2 across nuclei from Sst+ and Tacr1+ neuronal populations. d Pseudobulk log2 fold change (LFC) of the top 20 random forest discriminatory ion channel genes across two technical replicates following CPM normalization. Bars extending right (teal) indicate higher expression in Tacr1+ neurons; bars extending left (coral) indicate higher expression in Sst+ neurons. Genes significant by Fisher’s exact test are highlighted in pink italic labels. The shaded band indicates LFC < 0.1, where direction is uncertain given the n = 2 technical replicates. e Dot plot showing average expression and detection frequency of ion channel genes across Sst+ and Tacr1+ populations. Dot size reflects the percentage of nuclei expressing each gene, and dot color indicates average expression level. Genes are ordered to contrast the enriched (Trpc5, Cacna2d2, Dpp10, Kcnc2, Kcnab1) and depleted (Kcnq5, Kcnd2, Cacna1c, Kcnh7, Kcnn3) ion channel sets in Tacr1+ neurons As an exploratory consistency check, pseudobulk log2 fold changes were computed across two technical replicates following CPM normalization. The analysis confirmed directional consistency with Fisher’s exact results. All five significantly enriched ion channel genes (Trpc5, Cacna2d2, Kcnc2, Dpp10, Kcnab1) showed higher expression in Tacr1+ neurons across both technical replicates (Fig. 3d). The analysis further revealed the directional contribution of genes selected by the random forest classifier. Of the top 20 ion channel genes by RF importance, 12 showed higher expression in Tacr1+ neurons, and four genes showed higher expression in Sst+ neurons (Fig. 3d), indicating that random forest classification of Tacr1+ neurons is driven by both enrichment of ion channel genes and relative depletion of a subset of channels compared to Sst+ neurons (Fig. 3c). To validate these findings, we performed in situ hybridization on the top classifier Trpc5 gene to examine Trpc5 and Tacr1 gene co-expression in adult mice (x̄ = 724 neurons per animal; n = 3) and investigated Bötzinger complex (BötC), preBötC, and nucleus paragigantocellularis (nPGi) nuclei (Fig. 4a,b). Section alignment related to the BötC (two sections; Bregma level − 6.65 to − 6.8 mm) and preBötC (three sections; Bregma level − 6.8 to − 7.05 mm) was relative to a reference section, as previously described [20, 21, 46, 49–54]. In the BötC, 179.7 ± 12.1 neurons were double-positive (Tacr1+/Trpc5+), compared to 44.3 ± 3.8 Tacr1+/Trpc5− and 21.7 ± 2.3 Trpc5+/Tacr1− neurons. In the preBötC, 136.3 ± 6.4 neurons co-expressed both markers, with 81.3 ± 4.4 Tacr1+/Trpc5− and 56.0 ± 6.9 Trpc5+/Tacr1− neurons. In the nPGi, 87.3 ± 4.4 neurons were double-positive, with 60.7 ± 6.4 Tacr1+/Trpc5− and 56.7 ± 4.1 Trpc5+/Tacr1− neurons. Quantification of cell density normalized to anatomical area revealed that double-positive neuron density was markedly higher in the BötC (1077 ± 101 cells/mm2) and preBötC (939 ± 26 cells/mm2) compared to the nPGi (261 ± 8 cells/mm2), providing anatomical validation of the transcriptomic enrichment observed by snRNA-seq (Fig. 4b). Tacr1+ neurons exhibit regional co-expression of Trpc5 across the ventral respiratory column. a Representative coronal sections showing fluorescence in situ hybridization (FISH) for Tacr1 (red) and Trpc5 (green) with DAPI (blue). Low-magnification images (left) outline anatomical boundaries, with dashed lines indicating regions of interest and boxed areas corresponding to higher-magnification views. High-magnification merged images (middle) reveal the co-localization of Trpc5 transcripts within Tacr1+ neurons. Single-channel images (right) show Tacr1 and Trpc5 gene signals separately. Dashed outlines highlight subregions analyzed, including the preBötzinger complex (preBötC). Scale bars as indicated. b Cell density (cells/mm2) across three brainstem regions: preBötC (red circles), BötC (blue triangles), and nPGi (teal squares). Left, mutually exclusive subpopulations: neurons expressing Tacr1, Tacr1+/Trpc5⁻; Trpc5, Tacr1−/Trpc5+; and double-positive neurons co-expressing both markers, Tacr1+/Trpc5+. Right panel (separated by a dashed line), aggregate densities: total Tacr1 (double-positive and Tacr1) and total all (Tacr1, Trpc5, and double-positive). All values represent cell density normalized to anatomical section area. Individual animal data points are shown with group means ± SEM. c Volcano plot showing ion channel differentially expressed genes (DEGs) between cluster 5, identified as a putative nucleus ambiguus motor neuron contaminant based on Chat gene, and the remaining Tacr1+ preBötC population. Dashed vertical lines indicate log2FC thresholds of ± 1. Dashed horizontal line indicates the significance threshold (padj < 0.05). Genes enriched in the Tacr1+ preBötC population are shown in teal; genes enriched in Tacr1+ + Chat+ populations are shown in red; nonsignificant (ns) in gray. Note the absence of Trpc5+ cells in NA (Fig. 4a), consistent with this DEG analysis. Bold gene labels indicate genes significant by Fisher’s exact test and RF top features from Figs. 2 and 3 Ion Channel Transcriptomes Distinguish Tacr1+ PreBötC Interneurons from Chat+/Tacr1+ Motor Neurons As Random Forest classification identified Tacr1+ neurons as transcriptionally homogeneous, we next asked whether this reflects intrinsic specialization of preBötC interneurons or a feature shared across Tacr1+ expressing populations. We therefore compared ion channel transcriptomes of Tacr1+ preBötC interneurons within the double-positive Chat+-Tacr1+ subpopulation (cluster 5; n = 85) (Fig. S1c), which likely corresponds to cholinergic motor neurons of the nucleus ambiguus. DE analysis revealed clear divergence, with eight ion channel genes contributing to Tacr1+ preBötC classification significantly depleted in the Chat+-Tacr1+ population. Consistent with this, in situ hybridization showed no co-localization of Trpc5 and Tacr1 genes in the nucleus ambiguus (Fig. 4a), confirming the absence of the Trpc5 gene in Tacr1+ motor neurons. Together, these findings indicate that the ion channel signature underlying Tacr1+ classification is circuit-specific and selectively enriched in rhythmogenic preBötC interneurons. This suggests that Dbx1-derived Tacr1+ neurons acquire a molecularly distinct ion channel identity not shared by cholinergic Tacr1+ neurons outside the rhythmogenic circuit. Enhanced NALCN Channelosome Subunits Assembly in Tacr1+ Neurons Given the essential role of NALCN-mediated sodium leak currents in respiratory rhythm generation [31, 34, 55–58], we next examined whether Sst+ and Tacr1+ neurons differ in expression of all subunits required for assembly of the functional NALCN channel complex [56, 57, 59]. The NALCN channelosome consists of the pore-forming subunit encoded by Nalcn, the core auxiliary subunit Fam155a, and the auxiliary subunits Unc79 and Unc80, all of which are required for functional channel expression [57, 58]. To quantify coordinated expression of the NALCN channelosome, we derived single-cell module scores using “AddModuleScore” (Nalcn, Unc80, Unc79, and Fam155a) as implemented in Seurat. Tacr1+ neurons exhibited significantly higher module scores than Sst+ neurons (Wilcoxon rank-sum test, p = 9.45 × 10−3; Fig. 5a–c). Consistent with this, 47.0% of Tacr1+ neurons expressed the full set of NALCN channelosome genes compared to 37.5% of Sst+ neurons, whereas partial complex expression was similar between groups (Tacr1+: 48.2%; Sst+: 50.7%) (Fig. 4d). Cells lacking detectable expression of any NALCN-related subunits were rare in both populations but were slightly more frequent among Sst+ neurons (11.8 vs. 4.8% in Tacr1+ neurons) (Fig. 4d). These results were not confounded by sequencing depth, as NALCN module scores showed no correlation with total UMI counts (r = 0.027, p = 0.57; Fig. S2d). Together, these findings indicate that while both populations express components of the NALCN channelosome, a larger fraction of Tacr1+ neurons co-express the full complement of genes required for functional channel assembly. NALCN channelosome expression and neuromodulatory receptor enrichment in Sst+ and Tacr1+ preBötC/VRC neurons. a Dot plot showing the percentage of nuclei and the magnitude of expression difference for genes associated with the NALCN channelosome, including Nalcn, Unc79, Unc80, and Fam155a across populations. The saturation of color represents the average normalized expression level (scaled and centered). b UMAP projections showing expression patterns of the NALCN channelosome module score across nuclei. For visualization, expression values below the 5th percentile and above the 95th percentile were clipped to reduce the influence of outliers on the color scale. c Boxplots show the distribution of NALCN channelosome module scores across nuclei. Tacr1+ neurons exhibit significantly higher module scores compared to Sst+ neurons (Wilcoxon rank-sum test, p = 9.45 × 10−3). d Proportion of cells expressing the full NALCN channelosome (Nalcn, Unc79, Unc80, Fam155a), partial components (missing one or more of the complex genes), or none of the complex genes in Sst+ and Tacr1+ populations. Tacr1+ neurons show a higher proportion of cells expressing the full set of genes encoding the functional NALCN channelosome. e Volcano plot showing expression of neuromodulatory GPCR transcripts, comparing Tacr1+ versus Sst+ neurons. Dashed lines indicate thresholds for significance and fold change Given that NALCN activity is strongly regulated by GPCR signaling [55, 60, 61]. We next asked whether Sst+ and Tacr1+ neurons differ in neuromodulatory receptor expression. DE analysis identified selective enrichment of 5-HT2C receptor (Htr2c; padj = 0.004) and α1A adrenergic receptor (Adra1a; padj = 0.019) in Tacr1+ neurons (Fig. 5f), suggesting enhanced serotonergic and noradrenergic modulation in this population. Together, these data reveal distinct molecular specializations between Tacr1+ and Sst+ neurons in the preBötC, providing testable hypotheses for differential neuromodulatory control of respiratory rhythm-generating circuits. Discussion Over the past two decades, Sst+ and Tacr1+ neuronal populations in the ventrolateral medulla have been central to canonical models of respiratory rhythm generation and modulation [6, 15, 16, 18, 19, 21, 28, 33, 62–65]. Lesion and genetic fate-mapping studies have demonstrated that normal rhythmic breathing critically depends on these genetically defined neuronal populations localized within the preBötC [15, 16, 19, 21, 22, 25, 33, 34, 66]. Building on this foundation, our single‑nucleus transcriptomic profiling of Sst+ and Tacr1+ neuronal populations in the sampled preBötC of the mouse medulla provides a molecular framework for this rhythmogenic circuitry. The data reveal a population-specific repertoire of ion channels and neuromodulatory receptors, providing additional insight into how these historically defined populations differentially contribute to respiratory control. Across both DE and RF classification, Tacr1+ and Sst+ neuronal populations occupy distinct molecular states within the respiratory network, distinguished by their ion channel repertoires. In particular, RF classification revealed that ion channel expression is more homogeneous across Tacr1+ neurons relative to Sst+ neurons. Given prior evidence that Substance P (SP)/NK1R signaling in preBötC increases respiratory rhythm frequency and promotes stable bursting activity, we hypothesize that the relatively homogeneous ion channel transcriptomic profile of Tacr1+ neurons may support a greater propensity for stable bursting activity, although this remains to be tested experimentally (Fig. 2) [9, 28, 30, 31, 67–71]. Notably, this identity was consistent across 200 iterations, as indicated by the low variance of RF importance scores, supporting a stable population-level feature rather than a sampling artifact (Fig. 2). Among the top discriminatory features, Tacr1+ neurons showed enrichment of ion channels and auxiliary subunits associated with enhanced excitability and high-frequency firing capacity, including Trpc5, Kcnc2, and Cacna2d2 (Fig. 2c). To validate these findings, XGBoost (gradient-boosted decision tree ensemble) classification was performed using identical balanced subsampling parameters. It identified the same top discriminatory ion channel genes, supporting the robustness of the molecular signature across independent classification frameworks (Fig. S2e). Pseudobulk log2 fold change analysis further revealed that Kcnd2 (Kv4.2), Cacna1c (L-type Cav1.2), Kcnh7 (ERG-family), and Kcnn3 (SK3) were expressed at higher levels in Sst+ neurons, indicating that their contribution to RF classification algorithms reflects depletion relative to Tacr1+ neurons rather than enrichment. While these expression differences could influence membrane excitability, their functional consequences remain to be experimentally determined (Fig. 3d,e) [72]. In contrast to Tacr1+ neurons, the lower classification accuracy suggests greater ion channel transcriptomic heterogeneity in Sst+ neurons (Fig. 2a,b). This interpretation is consistent with findings from de Sousa Abreu et al. (2022), who demonstrated that activation of a preBötC Sst+ neuron produces probabilistic, state-dependent effects on breathing pattern. Together, these observations raise the possibility that Sst+ neurons comprise a functionally heterogeneous population, with diversity in ion channel transcriptomes contributing to their state-dependent role within the preBötC network and their interaction with other behaviors [8, 65]. Given that both Sst+ and Tacr1+ neuronal populations are required for normal respiratory rhythm generation, our findings suggest that these populations may achieve complementary physiological functions through distinct ion channel transcriptomes, with Tacr1+ neurons exhibiting a more homogeneous molecular signature and Sst+ neurons displaying greater transcriptomic diversity. Our results from DE and RF classification identified a consistent set of ion channel genes associated with Tacr1+ neurons, supporting the presence of a molecularly distinguishable ion channel signature in this population (Figs. 2 and 3). As many ion channels are broadly expressed throughout the nervous system, their significance in the present study lies in the coordinated enrichment within Tacr1+ neurons relative to Sst+ neurons. Rather than functioning as isolated molecular markers, these genes define a distinct ion channel transcriptomic signature that is predicted to shape the integrated conductance landscape of Tacr1+ neurons. Such coordinated expression may establish intrinsic electrophysiological properties by collectively regulating membrane depolarization, action potential kinetics, calcium influx, and responsiveness to neuromodulatory inputs. Among these features is Kcnc2, which encodes the Kv3.2 subunit. Kv3 channels are characterized by high activation thresholds and rapid deactivation kinetics, enabling fast action potential repolarization and short refractory periods [73, 74]. Although it might be emphasized that the cells characterized in the present study comprise not only respiratory rhythmic neurons but also those likely involved in other behaviors in the context of respiratory rhythm generation, such properties may support high-frequency and temporally precise firing within the preBötC network. We also observed enrichment of Trpc5 in Tacr1+ neurons. TRPC5 is a calcium-activated and calcium-permeable, Gq/PLC-coupled channel associated with slow depolarizing currents and sustained excitability in central neurons [75–79]. In vivo studies have implicated TRPC5 in autonomic regulation, including modulation of heart rate and reflexive bradycardia [80–83], suggesting that its expression in Tacr1+ neurons may have broader implications for cardiorespiratory modulation. Notably, the co-enrichment of Trpc5 and Tacr1 genes raises the possibility of receptor–effector coupling within the same neuronal population. SP acting through the NK1/Tacr1 receptor signals via Gq/PLCβ and may recruit and modulate TRPC5 (receptor-operated cation channels), generating a sustained inward cation current that promotes membrane depolarization and neuronal excitability [75, 77]. This pathway is broadly consistent with the calcium-activated non-selective cation current (ICAN) first described in preBötC pacemaker neurons under synaptic blockade by Peña and Ramirez (2004). The potential contribution of the ICAN in respiratory rhythm generation, its role in the generation of the inspiratory burst, as well as the molecular identity of ICAN have been studied extensively, yet many questions remain [84–88]. Our transcriptomic data raise the possibility that TRPC5 may contribute to conductance in a subset of Tacr1+ neurons. A neuronal preBötC population with these properties may contribute to the SP-induced intrinsic bursting observed under TTX conditions. This hypothesis remains speculative, and we did not directly address it in the present study [31, 89, 90]. We further identified the enrichment of Cacna2d2, encoding the α2δ auxiliary subunit of P/Q-type, N-type, and L-type voltage-gated calcium channels, which is known to regulate trafficking and surface expression of Cav channels and may influence activity-dependent Ca2+ influx and synaptic signaling properties within Tacr1+ neurons. Collectively, these findings suggest that Tacr1+ neurons possess a distinct ion channel transcriptomic profile that may contribute to mechanisms regulating respiratory network function [28, 30, 91, 92]. Although our analysis focused on ion channel genes that distinguished Tacr1+ from Sst+ neurons, it is equally noteworthy that many ion channel and receptor genes exhibited comparable expression between these populations. Such conserved expression likely reflects a shared molecular framework required for fundamental respiratory network function, including maintenance of intrinsic excitability, synaptic integration, and responsiveness to homeostatic feedback such as changes in CO2 and pH. Within this common molecular background, the differentially enriched ion channels identified here may provide the additional specialization that distinguishes the physiological properties of Tacr1+ and Sst+ neurons [93–96]. Mechanistically, TACR1 signaling has been demonstrated to modulate NALCN-mediated sodium leak currents, generating a persistent inward current that can contribute to depolarization of the resting membrane potential and inspiratory bursting [34, 66]. NALCN is a voltage-independent cation channel highly expressed in the preBötC (Yen et al., 2017). Knockout studies show that conditional deletion of NALCN results in recurring apneas and neonatal death within 24 h after birth (Lu et al., 2009). Consistent with these observations, enrichment of genes encoding the complete NALCN channelosome in Tacr1+ neurons raises the possibility that substance P (SP) enhances tonic depolarizing drive through NALCN modulation in this neuronal population (Fig. 5a–d) [26, 31, 34, 55]. Together with the enrichment of Trpc5 described above, our findings further raise the possibility that SP engages two complementary depolarizing mechanisms in Tacr1+ neurons: tonic NALCN-mediated sodium leak and receptor-operated TRPC5-mediated cation influx. These mechanisms are not mutually exclusive and may act synergistically to lower burst threshold and sustain rhythmic activity across different neuromodulatory states. We also found significant expression of transcripts encoding the Htr2c serotonin and alpha-1A adrenergic receptors (Fig. 5f), suggesting additional potential pathways through which neuromodulatory signals may influence NALCN activity and the excitability of Tacr1+ neurons [97]. Overall, our study identifies a molecular signature associated with Tacr1+ neurons’ excitability, characterized by enrichment of K⁺ channels, TRP channels, and neuromodulatory receptors. Finally, we would like to acknowledge various limitations and caveats. First, our dataset is derived from neonatal mice (P5–P10), which will likely not fully capture the properties of mature circuit states. To our knowledge, comparable transcriptomic studies examining Tacr1+ and Sst+ neuronal populations in the adult preBötC have not yet been reported, and previous molecular profiling studies of the preBötC have similarly relied on neonatal preparations [36, 98]. Hence, extending this analysis to adult animals will be an important direction for future investigation to determine how these molecular signatures are maintained or modified during maturation [99]. However, the shared Dbx1 lineage of Sst+ and Tacr1+ populations makes systematic developmental bias unlikely, and the convergence of results across multiple analytical approaches supports a cell-type-specific interpretation. Second, single-nucleus RNA sequencing is subject to dropout (transcriptional absence cannot be differentiated from true absence), which may obscure low-abundance transcripts. Nevertheless, the consistent identification of key ion channel genes across 200 independent subsampling iterations, together with concordant differential expression and enrichment analyses, argues against dropout-driven artifacts. Third, given the anatomical constraints of tissue microdissection under microscopic guidance, our sampling region was centered on the preBötC but likely extended beyond its strict anatomical boundaries [35, 37]. As dissections spanned the full 600-µm rostrocaudal thickness of the slice, the sampled tissue likely included portions of the rVRG immediately caudal to the preBötC, as well as neighboring regions. This limitation is further compounded by the absence of distinct cytoarchitectural or visible landmarks defining the preBötC under a dissecting microscope, making precise anatomical isolation inherently difficult. Therefore, the resulting dataset likely contains non-preBötC, non-respiratory neurons, and the reported cell proportions should be interpreted as representative of the sampled tissue rather than those of the preBötC exclusively. Finally, our comparison between Tacr1+ and Sst+ neurons was performed at the population level and does not resolve molecular differences among excitatory, inhibitory, or Tacr1+/Sst+ co-expressing neuronal subsets. Although the combined Tacr1+ and Sst+ populations comprised glutamatergic (21%), glycinergic (25%), and GABAergic (51%) neurons, indicating substantial functional diversity, only a very small fraction of neurons co-expressed both Tacr1 and Sst (0.32%; n = 15). Consequently, the limited representation of these neuronal subpopulations precluded statistically robust transcriptomic analyses, and future studies with greater sampling depth will be required to determine whether Tacr1+/Sst+ co-expressing neurons represent a distinct molecular subtype or more closely resemble either Tacr1+ or Sst+ neurons. Notably, ion channel differential expression analysis of cluster 5, identified as a putative nucleus ambiguus motor neuron contaminant, revealed depletion of key Tacr1+ preBötC markers, including Trpc5 and Cacna2d2, and enrichment of distinct ion channel transcripts, including Hcn1 and Scn7a, further supporting the molecular specificity of the preBötC Tacr1+ ion channel signature (Fig. 4c). In summary, our study establishes a molecular framework distinguishing Tacr1+ from Sst+ neurons in the preBötC. Despite their shared developmental origin, Tacr1+ neurons exhibit a distinct ion channel and neuromodulatory receptor signature that may contribute to their specialized roles in respiratory rhythm generation and modulation. Supplementary Information Below is the link to the electronic supplementary material. Fig. S1 Quality control, clustering, and marker gene expression of preBötC nuclei. a Quality control metrics and clustering of preBötC nuclei. UMAP projection showing unsupervised clustering of preBötC nuclei. Scatter plots illustrate relationships between total UMI counts (nCount_RNA), number of detected genes (nFeature_RNA), and mitochondrial transcript percentage (percent.mt). Violin plots show the distribution of these quality metrics across clusters. b UMAP projection showing unsupervised clustering of Sst+ and Tacr1+ neuronal populations after subsetting from the preBötC cluster. c Feature plots showing expression of Chat, marking cholinergic neurons across clusters within the Sst+ and Tacr1+ neuronal populations. d UMAP projection showing co-expression of Sst and Tacr1 within the Sst+ and Tacr1+ neuronal populations. e For downstream analysis, cluster 5 (Chat+ cholinergic neurons) and all nuclei co-expressing Sst and Tacr1 were excluded from the dataset (PPTX 1.34 MB) Fig. S2 ROC curves for random forest classification, Differential expression analysis, NALCN channelosome module scores are independent of sequencing depth, and XGBoost classification of Tacr1+ neurons. a Receiver Operating Characteristic (ROC) curves showing classifier performance across 200 balanced random forest iterations distinguishing Tacr1+ and Sst+ neurons based on ion channel transcriptomic profiles. Light blue lines represent individual iterations, and the dark blue line indicates the mean ROC curve. The mean area under the curve (AUC) was 0.619 ± 0.015, indicating modest but consistent discriminability between the two neuronal populations. b Density distributions of Tacr1+ classification accuracy across 200 balanced random forest iterations. Dark green represents held-out test set accuracy and light green represents out-of-bag (OOB) training accuracy. Mean classification accuracy ranged between approximately 61–63% for both evaluation strategies, reflecting the overlapping transcriptomic profiles of the two populations. c Volcano plot showing differential expression of ion channel genes between Tacr1+ (positive) and Sst+ (negative) neurons. Positive log₂ fold change values indicate higher expression in Tacr1 neurons. Differential expression was assessed using the Wilcoxon rank-sum test with BH correction. Labeled transcripts (Trpc5, Cacna2d2, Dpp10, Kcnc2, Kcnab1) represent the most highly ranked candidates by adjusted p-value. No transcripts reached the corrected significance threshold, reflecting limited statistical power due to the modest number of nuclei and sparse transcript detection in snRNA-seq data. d Scatter plot of NALCN channelosome module scores versus total transcript counts (nCount_RNA) for Sst+ (salmon) and Tacr1+ (teal) neurons. Lines of best fit are shown for each population. Spearman correlation coefficients were near zero for both Sst+ (r = 0.028) and Tacr1+ (r = 0.006) neurons, indicating that differences in NALCN channelosome scores between populations are not attributable to technical variation in library size. e Bar plot showing log2 fold change (LFC) of the top 20 ion channel genes ranked by XGBoost mean gain, computed from single-cell normalized expression values (log2CPM + 1). Positive LFC indicates higher expression in Tacr1+ neurons; negative LFC indicates higher expression in Sst+ neurons. Bar opacity reflects LFC reliability: opaque (LFC ≥ 0.3), semi-transparent (0.1 ≤ LFC < 0.3), faded (LFC < 0.1, direction uncertain). Gene labels are colored by analytical overlap: red = significant in Fisher’s exact test and present in Random Forest top 20; teal = present in Random Forest top 20 only; gray = XGBoost-specific. Shaded band indicates LFC < 0.1 zone of uncertain direction. XGBoost was run using identical balanced iterative subsampling (n = 152 per class, 200 iterations) with fivefold cross-validation for performance estimation (PPTX 554 KB) Table S1 Curated reference list of ion channel and receptor genes used for analysis, including detection status in the Sst+ and Tacr1+ preBötC dataset (CSV 15.0 KB) Acknowledgements We sincerely thank the developers and maintainers of the High-Performance Core (HPC) at SCRI and the open-source software that supports this field. We also extend our thanks to Dr. Ryan Philips (Seattle Children’s Research Institute) for perceptive comments. Finally, we are grateful to the anonymous reviewers for their careful evaluation of the manuscript and for their insightful suggestions, which greatly strengthened this work. Author Contribution H.B., J.-M.R. and A.D.W. conceived, designed, and performed the research. H.B., L.M.O., analyzed, interpreted, and prepared figures (H.B. transcriptomics, L.M.O. Histology). H.B., A.D.W., and L.M.O., interpreted and drafted the manuscript. H.B., A.D.W., L.M.O., and J.-M.R. edited and revised the manuscript. J.-M.R. acquired funding. All authors approved the manuscript. Funding This study was funded by the following grants from the National Institutes of Health under the Heart, Lung, and Blood Institute: P01 HL14454 and Project 2 (awarded to JMR), R01 HL144801 (awarded to JMR), R01 HL151389 (awarded to JMR), and R01 HL126523 (awarded to JMR) to support this project. Data Availability The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request. Declarations Competing Interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.