To classify cell types in the mouse PBN according to their transcriptional profiles, we harvested brain tissue from 10 adult male and female C57BL/6J mice, excised the PBN along its rostral-caudal extension (Figure 1A), and prepared single-cell suspensions for high-throughput scRNA-Seq using a commercial droplet-based assay (10× Genomics). After preprocessing the data to remove low-quality cells from the analysis (Rossi et al., 2021; Stuart et al., 2019), we retained a total of 39,649 single-cell barcodes that were sequenced to a median depth of 47,177 reads per cell. A total of 17,038 genes were detected, with a median of 1740 genes per cell (Figure 1—figure supplement 1).

To identify resident cell types of PBN tissue, cells were clustered on principal components and visualized in uniform manifold approximation and projection (UMAP) space (Mcinnes et al., 2018; Figure 1B). We then applied a likelihood ratio test to identify features that were differentially expressed between clusters (Hafemeister and Satija, 2019; Macosko et al., 2015; McDavid et al., 2013), and classified cells according to the specificity of canonical marker genes within each cluster (Materials and methods, Figure 1B–E, and Supplementary file 1). Low-resolution clustering identified seven transcriptionally distinct populations of neurons, glia, and stromal cells within the PBN (Figure 1B–E). Of these, neurons represented the largest proportion of cells at 57.2% (Figure 1C). Oligodendrocytes, marked by Mag and Opalin, were detected at 24.1% of total cells, and oligodendrocyte precursor cells (OPCs, 4.3% of total cells), were distinguished by their expression of Pdgfra and Gpr17 (Figure 1C–D). The large number of oligodendrocytes and precursors is not surprising because the PBN is bisected by the scp. We detected a population of astrocytes representing 8.1% of cells that were labeled by robust expression of Agt and Slc4a4, as well as a smaller population of microglia (2.6%) that were specifically labeled by Cx3cr1 and Tmem119 (Figure 1C–D). Additionally, we identified two distinct populations of cells marked by stromal markers; one of these populations was characterized by its selective expression of Tagln and Acta2 (3.2%), while another rare population of vascular leptomeningeal cells was marked by Dcn and Slc6a13 (0.26% of total cells). Although known canonical markers were used for the biased identification of broad classes of resident cells within PBN tissue (Figure 1B and D), each of these cellular classes was marked by a robust profile of unique transcriptional features (Figure 1E and Supplementary file 1).

Next, we sought to identify distinct subclasses of neurons within the PBN. Subclasses of cells within a terminally differentiated cell-type share a similar transcriptional landscape, and as a result, statistically discriminating subclasses through clustering analysis require the presence of a small set of high-confidence, high-variance features. To enable high-resolution subclustering of PBN neurons, we first applied a more stringent quality threshold to the neuronal population isolated in the initial analysis (Figure 2—figure supplement 1A–E), resulting in a smaller set comprised 7635 neurons (Figure 2A–B). These cells were sequenced to a much higher median of 99,583 reads per cell with each cell containing a median of 3189 genes and 7823 transcripts (Figure 2—figure supplement 1F–O). We used a clustering approach like that applied for all cells and discriminated 21 unique subclusters of neurons (N1-N21) according to their expression of differential feature sets (Materials and methods, Figure 2, and Figure 2—figure supplement 1P–Q). We first classified neuronal subclusters as glutamatergic or GABAergic and designated N1-N19 as glutamatergic with N1, N2 being enriched in vesicular transporter Slc17a7 (Vglut1), while N3-N19 are enriched in vesicular transporter Slc17a6 (Vglut2); the latter account for about 90% of all neurons sequenced (Figure 2C–E). The remaining two subclusters N20, 21 are GABAergic based on expression of Gad1, Gad2, and Slc32a1 (Vgat) (Figure 2C–E).

We sought to further define these 21 neuronal subclusters and identified two major clades that could be distinguished by their expression of transcription factors (Figure 2F–G). One major clade (N4-N12) is represented by expression of Lhx2, Lhx9, Meis2, and Nrf1. This clade probably descends from neurons that express Atoh1 during development (Karthik et al., 2022). This group also includes nuclear receptors, Nr2f1, Nr2f2, Nr4a; zinc-finger protein, Tshz1; homeobox-containing transcription factors, Barx2 and Evx2; and forkhead transcription factor, Foxp2. A subgroup of this clade is represented by Nr4a2 and Foxp2, which are expressed prominently in five of the eight subclusters. The expression pattern of Foxp2 in the PBN and its axonal projections via the ventral pathway (VP) to the hypothalamus has been described in detail (Huang et al., 2021a).

Another major clade that includes N13-N18 (Figure 2F–G) is represented by expression of a group of homeobox-containing transcription factors Lmx1a, Lmx1b, Pou2f2, Pou6f2, En1, Tlx3, Onecut2, and Satb2, along with a member of the retinoic acid family (Rorb), the AP2 factor (Tfap2b), and co-repressor (Tle4). The two major clades represented by Atoh1 (N4-N12) and Lmx1 (N13-N18) descendants are largely non-overlapping populations (Karthik et al., 2022). They have distinct axonal projection patterns, with the Atoh1 clade following a VP to the forebrain and the Lmx1 clade following the central tegmental tract (CTT).

The two neuron subclusters that express Slc17a7 (N1 and N2) have distinct expression of transcription factors from each other and from the two major clades. Likewise, the two subclusters that express Slc32a1 (N20, N21) also express distinct sets of transcription factors. As will be shown below, the two major clades (N4-N12 and N13-N18) represent neurons that reside within the PBN. The remaining two subclusters contain cells that mostly border the PBN and express Phox2b (N19) or Nfib and Lhx2 (N3) (Figure 2F–G).

Within these two major clades, we also found that the 21 neuronal subclusters could be further delineated based on preferential expression of specific genes (Figure 3A). Although our analysis identified multiple known subclasses of PBN neurons, we also identified novel neuronal subclusters and their corresponding transcriptional markers. Thus, this analysis provides a representation of PBN neuronal diversity at a higher resolution than has been previously appreciated (Figure 3A–B). A summary file listing the average expression of each gene per cluster, fraction of cells expressing each gene within a cluster, and the differential expression p-value per cluster is provided for all genes in the dataset (Supplementary file 2).

Neuropeptides are valuable markers for selected subpopulations of neurons in many brain regions. Some of these neuropeptide mRNAs are robustly expressed in one or two subclusters, e.g., Calca, Ghrh, Nps, Npy, Pdyn, Pnoc, while others are expressed in multiple subclusters, e.g., Adcyap1, Cck, Crh, Gal, Gpr, Nmb, Nts, Penk, Tac1, Vgf (Table 1, Supplementary file 3). Like the neuropeptide group, the genes encoding G-protein-coupled receptors (GPCRs) are genetically useful because they are likely to be expressed in neurons that receive aminergic or neuropeptide input; thus, making them desirable for circuit mapping (Supplementary file 4). Indeed, Cre-driver lines of mice have been generated for many neuropeptide and GPCR genes. GPCRs are also of interest because they are viable targets for drugs that could potentially modify the function of select neurons and circuitry in the PBN. The expression of GPCR genes is generally much lower than that of neuropeptides, making them more difficult to detect by immunohistochemistry or in situ hybridization. Consequently, scRNA-Seq data provide another resource for identifying the GPCRs expressed by distinct PBN neurons. GPCRs with relatively restricted expression are listed in Table 1.

To determine if different neuron subclusters are in distinct PBN subdivisions, we performed fluorescent in situ hybridization on coronal sections of the PBN spanning the rostral-caudal axis from Bregma –5.0 to –5.4 mm. Representative probes for each neuronal subcluster (N1-N21) were chosen based on the distinguishing genes in each subcluster (Figure 4A). Two sets of 12 probes were used in HiPlex experiments and a composite image locating each subcluster within the PBN was generated based on the Allen Mouse Brain Atlas (AMBA) (Dong, 2008; Lein et al., 2007) as a guide (Figure 4B). The tissue punch for scRNA-Seq was centered on the PBN, but neighboring regions were also included; consequently, only 13 glutamatergic of the 19 glutamatergic UMAP subclusters are within the PBN (Figure 4C). An example of the results obtained with three probes (Calca, Brs3, and Phox2b) is shown in Figure 4—figure supplement 1. PowerPoint summaries and individual TIFF images of all the HiPlex probes for the 5 Bregma levels of the PBN examined are available at Zenodo (DOI: 10.5281/zenodo.6707404). Qualitative scoring of the expression of each probe within a specific PBN subregion (Figure 5A) or neighboring regions (Figure 5—figure supplements 1 and 2) is based on the number of transcripts (signal intensity) and number of cells (Figure 5B). The HiPlex data were supplemented with RNAScope experiments to further investigate differential gene expression within specific subclusters.

The subregions of the PBN were derived from cytoarchitectural characteristics based on Nissl staining of rodent tissue sections and supplemented with anterograde and retrograde labeling approaches. We referenced the AMBA because it is widely used and includes a wealth of in situ hybridization and connectivity data, which was useful as a guide for determining the general location PBN subregions. The nomenclature in the AMBA is not entirely consistent with the atlas (Paxinos and Franklin, 2019) nor that developed for the rat (Fulwiler and Saper, 1984). For example, the internal lateral PBN does not exist in the AMBA, (it is called the superior lateral region, PBls), it does not divide the external lateral region into ‘inner’ and ‘outer’ regions but includes a lateral ventral region instead, and it does not show the lateral crescent. The boundaries of the PBN extend beyond those shown in the AMBA; neurons with PBN characteristics reside in regions that extend rostral to Bregma -5.0 and dorsally into what the AMBA indicates is the pedunculopontine nucleus (PPN) and the NLL (as described later). There is also lack of agreement on whether the KF is part of the PBN. It is unfortunate that the order of letters in the abbreviations used for subregion of the PBN does not follow common usage, e.g., the external lateral PBN is abbreviated PBle in AMBA and LPBE in Paxinos and Franklin. All atlases were generated without consideration of cellular phenotype (protein and gene expression); thus, we did not expect that neuronal subclusters identified in this study would fit nicely within atlas boundaries, or that they are the defining characteristics of those subregions. For all of the probes shown in Figure 5A, the general theme is that genes are predominantly expressed in one location with scattered expression in several other subregions.

Before discussing the glutamatergic subclusters within PBN, we first comment on distinguishing feature probes that were not, or rarely, detected in the PBN. Clusters N1 and N2 express Slc17a7 encoding Vglut1. While we detected a few Slc17a1-expressing cells in the lateral PBN (Figure 5—figure supplements 1 and 2), consistent with the AMBA, the majority of N1 cells are likely due to contamination from the cerebellar granular layer included in the tissue punch. N2 represents the MEV that expresses both Slc17a7 and Piezo2. We did not detect strong signals within the PBN for clusters N3 (probes Mylk, Nfib), N6 (Crh), or N18 (Shisal2b, Sostdc1). Most of these probes labeled cells outside the PBN (Figure 5—figure supplement 1).

Expression of fluorescent proteins from AAV injected into the PBN of Cre-driver lines of mice provides an independent means of locating the cell bodies and visualizing the axonal projections. We injected AAV1-Ef1a-DIO-YFP and AAV1-Ef1a-DIO-synaptophysin:mCherry into the PBN of 21 Cre-driver lines to visualize cell bodies and processes within the PBN (YFP) and synapses (mCherry) throughout the entire brain. The viruses were bilaterally injected into PBN of four to five mice; after at least 3 weeks for viral expression, the mice were perfused for histology, and a preliminary analysis of viral expression in the PBN region was performed. Brains with the most precise viral placement were used to generate the TIFF stacks of the entire brain, available on Zenodo (DOI: 10.5281/zenodo.6707404). The Cre-expressing cells in the PBN were categorized by their location within large PBN subdivisions and surrounding regions (Figure 6A–E). In addition to the 21 Cre-driver lines described here, PBN expression from additional Cre-driver lines has been reported (Table 3), not including those described in the Allen Institute Connectome project. Slc17a6^Cre (Vglut2), which is expressed in most of the subclusters, reveals the overall distribution of glutamatergic projections from the PBN (Huang et al., 2019). As expected, the location(s) of fluorescent cell bodies in the PBN of the Cre-driver lines of mice is consistent with in situ hybridization results, although the promoter in the virus and multiple viral particles per cell can provide more robust expression than the endogenous gene, e.g., many GPCRs where the in situ signal in the AMBA is undetectable.

We also mapped the expression of Calca in mice with tdTomato targeted to the Calca locus (Calca^tdT, Figure 6F). Fluorescence from Calca^tdT mice reveals that in addition to robust expression in PBle, Calca-expressing cells are scattered throughout the lateral and medial PBN, which are especially prominent in rostral sections in agreement with the in situ experiments above. This approach also reveals tdTomato fluorescence in the caudal KF, LC, trigeminal, and facial nuclei, in agreement with the AMBA and (Huang et al., 2021b). Note that a genetic cross of Calca^Cre mice with Cre-dependent reporter line, e.g., Gt(ROSA)26Sor^lsl-tdTomato (Ai14) results in widespread fluorescence throughout the brain, presumably due to developmental expression (Carter et al., 2013).

For each of the 21 Cre-driver lines and the Calca^tdT line that we analyzed, images were taken from every third coronal, 35-μm section, stitched together and registered to generate a TIFF stack that can be manipulated to view expression of YFP and mCherry throughout the brain using ImageJ. All images are available to view and download from Zenodo; an example of Calca neuron expression is shown in Video 1.

Cell populations in the PBN can be conceptually divided into two major groups based on overall projection patterns (Figure 7A). However, many of the lines did not neatly fit into one group, so the projections listed represent trends for the two pathways. The relatively dense projections from neurons in the PBle region generally follow the CTT, whereas the cells in the dorsal regions (PBld/ls/lc) follow the VP as described by Geerling and colleagues (Huang et al., 2021a, Huang et al., 2019; Karthik et al., 2022). There is also a smaller periventricular pathway that passes through the periaqueductal gray (PAG, not shown) and a weak descending pathway to the hindbrain. None of the Cre-driver lines tested show projections exclusively along the periventricular or descending pathways. Some PBN neurons, e.g., Tac1 neurons project along the descending pathway into the medulla and affect escape-like behaviors and breathing (Arthurs et al. manuscript submitted; Barik et al., 2018). Qualitative scoring of synaptophysin expression in a selection of brain regions is shown (Figure 7B) based on relative fluorescence levels (strong, medium, weak, Figure 7C). Most neurons that reside in the PBmm are mixture cells that are also expressed in the lateral PBN; consequently, their axons join those from the lateral PBN.

For Cre-driver lines that have most of their expression in the PBle (Calca, Crhr1, Nts, Chat), or strongly in the PBle along with other regions (Crh, Adcyap1, Adcyap1r1, Tac1, Mc4r, Oprm1), the axon terminals with the brightest signal are in the BNST and CEA. Although expression is generally spread throughout these subregions, the oval and ventral regions of the BNST and the capsular region of the CEA have the strongest fluorescence. It is likely that the PBle cells are responsible for most of the oval/ventral BNST and capsular CEA staining density, while neurons outside of PBle project to adjacent regions. Although there is some variation, the next brightest regions are the IC, SI, VPMpc, and PSTN. There are no lines with cell bodies in the PBle that do not have at least some synaptophysin signal in the above listed regions. Other areas that are often innervated are the LS, CM, DMH, LHA, RE, IMD, PAG, DR, RN, and NTS. A complete list of abbreviations used here is shown in Figure 4D. Many of the Cre-driver lines that have most of their expression in PBle still have cells scattered throughout other subregions in the PBN, which may explain the expression in areas that have been associated with the ventral tract, such as the DMH and LHA.

There is overlap of axonal projections to both pathways, which likely occurs because none of the neuronal subclusters is restricted to one subregion of the PBN. Cre-driver lines that mainly have expression in the dorsal PBN regions (Pdyn, Tacr1, Brs3, Cbln4, Ptger3) also have axons that tend to travel through and target ventral brain regions such as the VTA, LHA, DMH, PVH, and MEPO. Cre-driver lines that are categorized into the dorsal group often have fewer cells and weaker projections as a result. Some Cre-driver lines have strong cellular expression across most of the lateral PBN. For these (Adcyap1, Adcyap1r1, Oprm1, Crh), there is robust expression of the AAV-driven fluorescent proteins in areas associated with the CTT such as the BNST/CEA and areas associated with the VP such as the MEPO. Overall, their projections are a combination of areas seen in the other groups.

The cells that reside in the PBmm do not appear to have a separate innervation profile. There was no line examined that had its expression limited to the PBmm exclusively. However, in one example (Tac1), the cells were transduced in PBmm on only one side, allowing for a comparison of projections between hemispheres. The result was a slightly brighter synaptophysin signal in the common projection targets of IC, BNST, and CEA, and a much brighter signal in the cortical amygdala on the side with the PBmm expression. In another example (Phox2b), most of the cells transduced were in the PBmm, with some cells in the PBlc and PBls. This line showed a projection to the septohippocampal (SH) nucleus, which likely originated from cells in PBls rather than PBmm because we did not see the same SH projection from the Tac1 injection that was heavily expressed in the PBmm (compare Phox2b and Tac1 files on Zenodo).

The Satb2-expressing neurons mainly reside in PBlv, scp, and PBmm. Although they are only partially expressed in PBle, their projection pathway still largely follows the CTT with connections to areas typically associated with PBle such as BNST and CEA. However, the axon terminals from Calca^Cre-driver line are concentrated in the oval BNST and lateral CEA, whereas Satb2^Cre-driven projections surround those regions (Figure 8A–B).

Ntsr1 population of neurons has the VMH as its most prominent projection target (Figure 8C). A population of VMH-projecting PBN neurons was identified by Fulwiler and Saper, 1984; Fulwiler and Saper, 1985, who assigned it to the PBls subnucleus and found that about 80–90% of these neurons stained immunohistochemically for CCK. Ntsr1 neurons are a subpopulation of the Cck cells in the far rostral PBN (Figure 5—figure supplement 3). The location of Ntsr1 neurons is distinctly different from populations such as the Tacr1 neurons in the PBls defined by the AMBA. All Cre-driver lines that have projections to the VMH (Crhr1, Nts, Adcyap1, Adcyap1r1, Oprm1) have some cell bodies in the same region as Ntsr1.

Many of the genes of interest are widely expressed in regions surrounding the PBN, which made restricting the transduction of cells within the PBN nearly impossible. Gad2 and Slc32a1, the two GABAergic genes, are only sparsely expressed within the PBN compared to adjacent areas; consequently, we were unable to determine whether they function as interneurons or as projection neurons. Gad2 is expressed more widely and without Slc32a1 in some PBN glutamatergic cells, e.g., Calca neurons, so using that Cre-driver line to examine inhibitory projections outside of the PBN could be misleading.

All experiments were approved by the Institutional Animals Care and Use Committee at the University of Washington (Protocol #2183–02). Animals were group-housed with littermates on a 12-hr light cycle at ~22°C with food and water available ad libitum. Male and female mice from the same litter were used, but no comparisons were done between sex. Refer to Table 3 for sources of all Cre-driver lines of mice. Most of them have been described previously and/or deposited at Jackson labs, where details can be found. All mice were on a C57BL/6J genetic background. We have not documented that each of the Cre-driver lines tested is restricted to neurons that normally express that gene in the adult.

The generation of Calca^tdTomato mice was described (Jarvie et al., 2021). The generation of four new lines is described below. The Cbln4, Ntsr1, and Tacr1 Cre-driver lines were made by inserting IRES-Cre:GFP just beyond the termination codon of these genes, whereas the Pdyn line was made by inserting Cre:GFP just 5’ of the initiation codon. The 5’ and 3’ arms (5–8 kb) were amplified from C57BL/6 BAC clones by PCR using Q5 polymerase (New England Biolabs) and inserted into a cloning vector that contains frt-flanked SV-Neo for positive selection and Pgk-DTA and HSV-TK genes for negative selection (Jarvie et al., 2021). The linearized constructs were electroporated in G4 ES cells (129× C57B/L6). About 80 G418-resistant clones were picked and expanded for Southern blot analysis using a ³²P-labeled probe located just beyond either the 5’ or 3’ arm. Single inserts were established by southern blot with a Neo gene probe. Correctly targeted clones were injected into blastocysts and transferred to recipient female mice. Germline transmission of the targeted allele was determined by three-primer PCR (two primers from gene on interest that flanking Cre insertion region and one reverse primer in IRES or Cre). Mice were bred with Gt(Rosa26)-FLPo (Jax: 07844) mice to remove the SV-Neo gene and then bred with C57BL/6 mice for at least six generations.

Live, single-cell suspensions were prepared as described (Rossi et al., 2021; Rossi et al., 2019), and tissue was harvested approximately 6 hr after the onset of the dark cycle (Zeitgeber time ~18:00). Briefly, male and female mice were transcardially perfused with a cold artificial CSF solution containing N-methyl-D-glucamine (NMDG) (NMDG-aCSF), modified from Ting et al., 2018. All steps were performed under continuous oxygenation and CO₂ buffering using 95%/5% O₂/CO₂. Brains were rapidly dissected, and coronal slices (200 µm) spanning the PBN were prepared using a vibrating microtome (VT1200, Leica Biosystems). Slices were allowed to recover in NMDG-aCSF containing 500-nM TTX, 10-µM AP-V, and 10-µM DNQX (NMDG-aCSF-R) for 20 min, and the PBN was subsequently isolated using tissue punches (500–750 μm). The isolated tissue was enzymatically digested using 1 mg/mL pronase (Roche) for 50 min, and enzymatic digestion was quenched with 0.05% BSA. Cells were mechanically dissociated using a fire-polished glass pipet with an internal diameter of 200–300 µm, filtered through a 40 μm strainer, washed, and depleted of dead cellular fragments using a commercial kit (Miltenyi Biotec, Bergisch Gladbach, Germany). The remaining cells were resuspended in PBS containing 0.05% BSA about 1000 cells/µL.

Single-cell RNA libraries were generated using chromium single-cell 3’ v2 chemistry (10× Genomics, Pleasanton, CA) following the standard manufacturer protocol. A pool of ~17,000 cells harvested from 5 mice were loaded per reaction, with the first pool run on a single reaction and the second pool spread across three reactions (Figure 1—figure supplement 1H–I).

For each library, cDNA amplification was performed using 12 cycles, and indexing was performed using 11 cycles. Library size distribution and concentration were determined using High Sensitivity D5000 ScreenTape or 4200 TapeStation (Agilent Technologies, Santa Clara, CA) and Qubit HS DNA Assay (Invitrogen, Waltham, MA). Each reaction library was sequenced on two lanes of an Illumina HiSeq 4000 using 2×150 chemistry by Genewiz, Inc (South Plainfield, NJ) following the standard 10× Genomics v2 paired-end configuration. Approximately 875 million reads were generated per lane with a mean sequencing saturation of 82.6%. Sequences were aligned to the mm10-3.0.0 genome, and digital expression matrices were created using 10× Genomics Cell Ranger v3.1.0 with 128 GB of memory on 24 cores.

Clustering was performed using Seurat v3.1.2 (Stuart et al., 2019) and custom code in R v3.6.1 as described (Rossi et al., 2021). Briefly, to remove low-quality cells, total input cells were first filtered using a threshold for genes and fraction mitochondrial reads, and doublets were subsequently removed using DoubletDetection v2.5.4 under default parameters on Python 3.7 via Reticulate v1.14 (Gayoso and Shor, 2022). Cells containing ≤800 genes and ≥10% mitochondrial reads were removed from the primary analysis (Figure 1 and Figure 1—figure supplement 1). Within the analysis of all cells, one subcluster was unable to be mapped to any specific features and was excluded from analysis (Figure 1B, gray). Following these filters, a total of 39,649 cells were retained with a median of 1740 genes and 3366 transcripts represented across a median of 47,177 reads per cell (Figure 1 and Figure 1—figure supplement 1A–E). The neuronal cluster identified in the initial analysis of all cells (Figure 1B) was then isolated and subjected to a more stringent quality threshold where only cells containing ≤2% mitochondrial reads were retained to enable high-confidence, high-resolution subclustering (Figure 2—figure supplement 1C and H). Within the neurons, two subclusters were unable to be mapped to any specific features and were excluded from the analysis (Figure 2A, gray). This resulted in a final analysis of 7635 neurons containing a median of 3189 genes, 7823 transcripts, and 99,583 reads per cell (Figure 2 and Figure 2—figure supplement 1A–J).

Regression and integration of samples were performed using a regularized negative binomial regression (Hafemeister and Satija, 2019; Stuart et al., 2019) and canonical correlation analysis (Butler et al., 2018; Stuart et al., 2019) as described (Rossi et al., 2021; Figure 1—figure supplement 1F–G and Figure 2—figure supplement 1K–M). Principal components were calculated on all genes, and cells were plotted in UMAP space using iteratively tuned parameters to optimize for visualization (Mcinnes et al., 2018; Figure 1—figure supplement 1J and Figure 2—figure supplement 1O). Clustering was performed using the Louvain algorithm with multilevel refinement (Rodriguez and Laio, 2014) with a K parameter of 25 and resolution of 0.04 (all cells) (Figure 1), or a K parameter of 15 and resolution of 0.25 (neurons) (Figure 2). Feature discovery for cell-type assignment was performed on Pearson residuals using a likelihood ratio test for single-cell data as implemented in Seurat (Hafemeister and Satija, 2019; Macosko et al., 2015; McDavid et al., 2013).

Prepossessing, regression, integration, dimensionality reduction, clustering, and feature discovery were run on a Dell blade-based cluster at the University of North Carolina at Chapel Hill running Linux RedHat Enterprise 7.7. All other steps were performed on an Apple MacBook Pro running macOS 11.4.0. Raw and processed data for the scRNA-seq experiment have been deposited at the National Center for Biotechnology Information Gene Expression Omnibus (NCBI GEO, accession number GSE207708). The code used in this analysis is available at a Github repository affiliated with the Stuber Laboratory group (https://github.com/stuberlab/Pauli-Chen-Basiri-et-al-2022; Basiri, 2022).

The Calca^Cre mice were used to identify mRNAs that were selectively being translated in neurons that express the Calca gene by injecting a virus expressing a Cre-dependent hemagglutinin (HA)-tagged ribosomal protein 22 (AAV-DIO-Rpl22-HA) (Sanz et al., 2015) into the PBN of eight adult mice. After several weeks to allow incorporation of the tagged ribosomal protein into ribosomes, tissue punches (two pools of bilateral punches from four mice) were collected, total cell extract was prepared, and then polyribosomes were precipitated with an antibody against the HA-tagged ribosomes (Sanz et al., 2019; Sanz et al., 2009). For microarray analysis, 10 ng of total RNA was amplified and biotin-labeled using the Ovation Pico SL WTA system with the EncoreIL biotinylation module (NuGEN), and 750 ng of the labeled cDNA was hybridized to a MouseRef-8v2.0 gene expression BeadChip (Illumina). Signals were detected using the BeadArray Reader (Illumina) and analyzed using the GenomeStudio software (Illumina). Average normalization and the Illumina custom error model were applied to the analysis. Only transcripts with a differential score of >13 (p<0.05) were considered. The results shown are means of two biological replicates. Raw and normalized RiboTag data have been deposited in the NCBI GEO (accession number GSE207153). In some cases, the microarray included more than one probe for the same gene, and the absolute values could differ greatly; thus, the relative enrichment is more reliable.

Mice were anesthetized with isoflurane and placed on a robotic stereotaxic frame (Neurostar GmbH, Tübingen, Germany). AAV1-EF1a-DIO-YFP and AAV1-EF1a-DIO-synaptophysin:mCherry were injected bilaterally into the PBN (AP –4.8 mm, ML ±1.4 mm, DV 3.5 mm) at a rate of 0.1 µl/min for 2 min. At least 3 weeks after virus injection, mice were deeply anesthetized with sodium pentobarbital and phenytoin sodium (0.2 ml, i.p.) and intracardially perfused with ice-cold PBS followed by 4% PFA. Brains were post-fixed overnight in 4% PFA at 4°C, cryoprotected in 30% sucrose, frozen in OCT compound, and stored at –80°C. Coronal sections (35 µm) spanning the brain (Bregma 2.0 to –8.0 mm) were cut on a cryostat and collected in cryoprotectant for long-term storage at –20°C. The YFP and mCherry signals overlapped virtually in all neurons, indicative of co-infection by both AAV. The synaptophysin mCherry signal helps to distinguish axon terminals from fibers of passage. The two colors can be visualized separately or together in the TIFF stacks on Zenodo. Although AAV1 serotype has been shown to cross synapses in an anterograde direction, we never observed fluorescence in post-synaptic cells.

Sections were washed two times in PBS and incubated in a blocking solution (3% normal donkey serum and 0.2% Triton X-100 in PBS) for 1 hr at room temperature. Sections were incubated overnight at 4°C in blocking solution with primary antibodies including: chicken-anti-GFP (1:10,000) and rabbit-anti-dsRed (1:2000). After three washes in PBS, sections were incubated for 1 hr in PBS with secondary antibodies: Alexa Fluor 488 donkey anti-chicken and Alexa Fluor 594 donkey anti-rabbit, (1:500). Tissue was washed three times in PBS, mounted onto glass slides, and coverslipped with Fluoromount-G with DAPI (Southern Biotech).

Whole-slide fluorescent images were acquired using a Keyence BZ-X710 microscope and higher magnification images using an Olympus FV-1200 confocal microscope. Images were minimally processed using Fiji to enhance brightness and contrast for optimal representation of the data. For TIFF stacks, images were aligned using the BrainMaker workflow in NeuroInfo (MBF Bioscience).

Male mice were deeply anesthetized with sodium pentobarbital and phenytoin sodium (0.2 ml, i.p.), decapitated, and brains rapidly frozen on crushed dry ice. Coronal sections (15 µm) were cut on a cryostat, mounted onto SuperFrost Plus slides, and stored at –80°C. RNAscope HiPlex assay or RNAscope fluorescent multiplex assay was performed following the manufacturer’s protocols. For one experiment, two AAV-SaCas9 with guide RNAs directed against different parts of Slc17a6 (Hunker et al., 2020) were injected into the PBN 16 weeks prior to RNAScope for Calca and Slc17a6.

Images centered on the scp in the PBN were acquired in a 3×3 grid at 20× using a Keyence BZ-X710 microscope then stitched together using Fiji. Images of probe staining within the four-channel sets were subtracted from one another using Fiji’s image calculator function to remove background autofluorescence. The DAPI images from each of the four sets of images were registered using the HiPlex Image Registration Software (ACDBio) and then used to align all the probe images. This process was repeated for a second HiPlex experiment. Colors were assigned using the HiPlex Image Registration Software.

Registered HiPlex probe images were combined into five stacks for each Bregma level for both experiments. PBN and surrounding subregions of interest (ROIs) were drawn based on the AMBA designations and distinct probe locations. We recognize that the boundaries of the PBN in the AMBA may not accurately reflect the cytoarchitecture. It was used as a reference to allow readers to locate mRNA expression more easily. Using these ROIs, we generated a score based on an estimation of the number of transcripts present per cell and number of cells labeled per region. Projection regions were evaluated based on synaptophysin density by matching the brain sections in the TIFF stacks as closely to the AMBA as possible. The values shown in the tables are only meant to be a guide, and all the raw data are available for interested readers to evaluate on their own.

Using an RNA/DNA ratio of 2 for total brain (probably an underestimate for neurons), a DNA content of 6.4 pg/cell and about 80% of total RNA being ribosomal RNA, there is about 10 pg of ribosomal RNA per brain cell. Ten pg rRNA/cell corresponds to about 1 million ribosomes/cell. With 10 ribosomes per mRNA that equals 100,000 mRNAs/cell. At steady state, the amount of protein synthesized per mRNA equals rate of synthesis times the half-life divided by the natural log of 2 (0.67). The rate of synthesis equals the amount of mRNA times the translational efficiency (number of proteins made/min/mRNA). With a translational efficiency of 10, and a half-life of 700 min, one mRNA per cell can maintain about 10,000 proteins/cell at steady state.