Molecular ontology of the parabrachial nucleus

Abstract Diverse neurons in the parabrachial nucleus (PB) communicate with widespread brain regions. Despite evidence linking them to a variety of homeostatic functions, it remains difficult to determine which PB neurons influence which functions because their subpopulations intermingle extensively. An improved framework for identifying these intermingled subpopulations would help advance our understanding of neural circuit functions linked to this region. Here, we present the foundation of a developmental‐genetic ontology that classifies PB neurons based on their intrinsic, molecular features. By combining transcription factor labeling with Cre fate‐mapping, we find that the PB is a blend of two, developmentally distinct macropopulations of glutamatergic neurons. Neurons in the first macropopulation express Lmx1b (and, to a lesser extent, Lmx1a) and are mutually exclusive with those in a second macropopulation, which derive from precursors expressing Atoh1. This second, Atoh1‐derived macropopulation includes many Foxp2‐expressing neurons, but Foxp2 also identifies a subset of Lmx1b‐expressing neurons in the Kölliker–Fuse nucleus (KF) and a population of GABAergic neurons ventrolateral to the PB (“caudal KF”). Immediately ventral to the PB, Phox2b‐expressing glutamatergic neurons (some coexpressing Lmx1b) occupy the KF, supratrigeminal nucleus, and reticular formation. We show that this molecular framework organizes subsidiary patterns of adult gene expression (including Satb2, Calca, Grp, and Pdyn) and predicts output projections to the amygdala (Lmx1b), hypothalamus (Atoh1), and hindbrain (Phox2b/Lmx1b). Using this molecular ontology to organize, interpret, and communicate PB‐related information could accelerate the translation of experimental findings from animal models to human patients.

In the dorsolateral corner of this region, a diverse constellation of neurons communicates with every major region of the central nervous system, from cerebral cortex to spinal cord Huang et al., 2020;Moga et al., 1990;Saper & Loewy, 1980). Herrick first identified this region as the "superior secondary gustatory nucleus" in fish (1905), and subsequent work identified taste-relay neurons in rats (Norgren & Leonard, 1971). These neurons surround the brachium conjunctivum (superior cerebellar peduncle) and are referred to as the "parabrachial" nucleus (PB).
One reason these neurons remain enigmatic is that, rather than forming discrete layers or nuclei (as in the cerebral cortex or thalamus), most PB subpopulations form an intermingled, three-dimensional web.
In addition to species differences, an unavoidable challenge when applying cytoarchitectural criteria is that this requires interpreting Nissl-stained tissue, an inherently subjective activity (Swanson, 2000). Distinctions can be subtle, and even experts cannot distinguish functionally diverse neurons if they have the same Nissl-stained appearance. A more ideal framework for classifying diverse, intermingled neurons would incorporate observer-independent, molecular information as a core feature. Using gene expression to classify neurons can produce results resembling cytoarchitectural analysis, while also distinguishing new subpopulations (Ortiz et al., 2020). Gene expression predicts the output connectivity of PB neurons (Huang et al., 2020;Huang et al., 2021), and basing a neuronal ontology on molecular and connectomic features could offer a more accessible, universal language for interpreting and communicating experimental findings (Bota & Swanson, 2008;Hamilton et al., 2012;Larson & Martone, 2009;Zeng & Sanes, 2017).
As a first step, we sought a framework of developmental-genetic information that identifies all PB neurons. Motivated by novel activity and connectivity patterns in rats (Geerling & Loewy, 2006bGeerling et al., 2010), this project began with the observation that two transcription factors have complimentary distributions in this region Miller et al., 2012;Shin et al., 2011). Here, we combine Cre fate-mapping, mRNA labeling, immunolabeling, and axonal tracing to replicate and extend these observations in mice.
Our new findings reveal that "the" PB does not have a single developmental origin. Its glutamatergic neurons are a blend of two, mutually exclusive macropopulations, defined by the embryonic transcription factors Lmx1b and Atoh1. We show that this framework clarifies the identity and connectivity of further PB subpopulations and challenges previous ideas about the composition and origin of the Kölliker-Fuse nucleus (KF). Together, these findings lay a developmental-genetic foundation for a molecular ontology that investigators can use to identify and target PB neurons.

Animals
We used a total of n = 68 male and female mice, aged 7-23 weeks and weighing 17-31 g. All mice were group-housed in a temperature-and humidity-controlled room on a 12/12-h light/dark cycle and had ad libitum access to standard rodent chow and water.
In addition to C57B6/J mice (Jackson Laboratories), we used a variety of knockin-Cre and Cre-reporter mice. Detailed information about each strain is provided in Table 1. For all mRNA and protein labeling that did not require a Cre-reporter, we replicated labeling in at least three, 8−12-week-old C57B6/J mice. All mRNA and protein labeling in Cre-reporter mice was replicated in at least three mice with a hemizygous Cre allele and a hemizygous Cre-reporter allele. Stereotaxic injections, histologic procedures, and confocal microscopy in n = 4 Harlan Sprague-Dawley rats were performed as described previously (Geerling & Loewy, 2007;Geerling et al., 2011;Shin et al., 2011) and in accordance with the guidelines of the Institutional Animal Care and Use Committee at Washington University in Saint Louis. All experiments in mice were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee and at the University of Iowa.

Stereotaxic injections
Mice were anesthetized with isoflurane and placed in a Kopf 1900 stereotactic frame. After midline incision, the skin was retracted to expose the skull and locate bregma. We injected the cholera toxin Bsubunit (CTb, 0.1% in distilled water; List, lot #10331A1) in Atoh1-Cre;R26-lsl-L10GFP mice. We made nanoliter injections through a finetipped micropipette (20-30 µm inner diameter) using controlled puffs of compressed air, typically 0.5-1 per second, with a target rate of

Perfusions and tissue sections
All mice were deeply anesthetized with ketamine-xylazine (i.p. 150-15 mg/kg) and then perfused transcardially with phosphate-buffered saline (PBS, prepared from 10x stock; P7059, Sigma), followed by 10% formalin-PBS (SF100, Fisher Scientific). After perfusion, the brain was removed and fixed overnight in 10% formalin-PBS at 4 • C, then submerged in 30% sucrose-PBS at 4 • C for an additional day. Each brain was sectioned into 40 µm-thick axial (coronal) slices using a freezing microtome. Three adjacent (1-in-3) tissue series were collected from each brain in separate tubes (labeled "A," "B," and "C") containing a cryoprotectant solution of 35% (v/v in PBS) ethylene glycol (102466, Sigma-Aldrich) and 25% glycerol (G22025, RPI). These tubes were stored at −20 • C until further processing. Each brain, therefore, yielded three sets of axial tissue sections that allowed us to study up to three separate combinations of molecular markers in the PB region from each mouse. For experimental replication of each endpoint, we analyzed brain tissue from at least three separate mice.

Immunofluorescence
We removed tissue sections from cryoprotectant and selected sections of interest (typically, a series of nine tissue sections containing the full PB region or CTb injection site; otherwise, a full-brain series).
After rinsing the sections in PBS, we incubated them with primary antisera (

In situ hybridization
We used RNAscope probes to label a variety of mRNA transcripts (see Table 3 for detailed information on each probe  We counted the symbols in each layer by first locking other layers, then using "Document Info > Objects" in Illustrator, and then entered each count into a Microsoft Excel spreadsheet for further analysis.
We also used Illustrator to make drawings, arrange images, and add lettering for figure layouts. Scale bars were traced in Illustrator atop calibrated lines from cellSens to produce clean white or black lines in each figure.
We applied the Abercrombie correction factor to compensate for overcounting (Guillery, 2002). This correction factor requires knowing the section thickness and the diameter of the key counting element, which was the cell nucleus in our analyses. We, therefore, measured the diameter of FoxP2-, Lmx1b-, and Phox2b-immunoreactive nuclei, as well as the nuclear-void diameter in L10GFP-expressing neurons (in Atoh1-Cre;R26-lsl-L10GFP mice). For each cell type, 30 nuclei were selected at random and measured from each case (∼3 per marker, per section), across all Atoh1-Cre;R26-lsl-L10GFP cases, every Lmx1a-Cre;R26-lsl-tdTomato case, and all C57B6/J cases that were analyzed for Phox2b, Lmx1b, and/or FoxP2 counts. Average total nuclear diameters in all groups ranged 9.0−9.5 µm. As a rough estimate of the total number of neurons in the mouse PB, we added together the individual counts of each cell population, then subtracted the numbers of double-and triple-labeled neurons. Finally, since we worked with tissue subsamples (1-in-3 tissue series), we multiplied the overall, Abercrombie-corrected average sum of PB neurons by 3 to estimate the total number.
In this article, we use the term "mutual exclusivity" to mean that markers are found in largely separate populations that are at least 98−99% distinct from one another. Most histologic techniques for labeling protein or mRNA expression do not produce 100% mutually exclusive labeling, and a lack of perfect mutual exclusivity between a pair of mRNA transcripts, immunolabeled proteins, or fluorescent protein reporters may involve background thresholding as much as actual coexpression. In this study, we did not include faint or ambiguous labeling within the range of background tissue fluorescence. We analyzed most cells and cell populations in relatively thick (40 µm) tissue sections at digital magnifications spanning ∼10−100 µm (not submicron or subcellular distributions), which is optimal for comprehensively analyzing macropopulation-level features throughout the full, three-dimensional extent of the PB complex, which encompasses ∼1−2 mm 3 . Wherever we found clear-cut examples of colocalized markers, we describe our observations in detail.

Nomenclature
For rat and mouse genes, we used MGI nomenclature. For rat and mouse proteins and Cre-reporters, we used common abbreviations from the published literature. For neuroanatomical structures and cell populations, where possible, we use and refer to nomenclature defined in peer-reviewed neuroanatomical literature. In some instances, we use or refer to nomenclature derived from rodent brain atlases (Dong, 2008;Paxinos & Franklin, 2013;Paxinos & Watson, 2007;Swanson, 1992). For the PB region specifically, mutual inconsistencies among existing systems of nomenclature (both between atlases and within the current literature) is a core topic addressed by the experimental results and analysis in this study.
Initially, we did not plan to focus attention on the region ventrolateral to the rostral PB that is referred to as the "Kölliker-Fuse nucleus," but neurons in this location appeared as an emergent property in several of our results. These eponyms are historically inaccurate, and both the location and neuronal composition of the KF are defined inconsistently (Petrovicky, 1989), but we use the term "KF" to indicate a small region that is ventral to the rostral, ventrolateral tip of the PB at levels approximately 4.8−5.0 mm caudal to bregma (Paxinos & Franklin, 2013). This location borders a small white matter tract, labeled "ventral spinocerebellar tract" in rodent brain atlases (Dong, 2008;Paxinos & Franklin, 2013), which extends past the cerebellum rostrally, alongside the lateral PB, and then ventrally, alongside the KF. As described below, novel molecular features in this location identify and distinguish intermixed populations of neurons. The objective of this study is to maximize reproducibility by focusing on observer-independent, molecular properties that unambiguously identify populations of neurons, rather than boundary lines that are inferred from Nissl cytoarchitecture. Consequently, and by design, no results or conclusions in this study were based on or inferred from Nissl-stained tissue.

Lmx1b and FoxP2 in rats
Our investigation began in rats after identifying a small subpopulation of neurons activated by sodium deprivation (Geerling & Loewy, 2007). These sodium-deprivation-activated neurons intermingled with other PB neurons in a distribution that did not conform to existing, cytoarchitectural boundaries. Seeking a molecular marker that would allow us to better classify and then target these neurons, we examined regional expression of transcription factors and identified FoxP2 as a useful marker for these and several other PB subpopulations (Gasparini et al., 2021;Geerling et al., 2016;Geerling et al., 2011;Geerling et al., 2017;Shin et al., 2011;Verstegen et al., 2017). Virtually all PB neurons activated by sodium deprivation contained FoxP2 (Gasparini et al., 2021;Geerling et al., 2011), and many of these distributed within and alongside the superior cerebellar peduncle, in a location described as the inner portion of the external lateral PB subnucleus (PBeL; Herbert & . Therefore, we predicted that sodium-deprivationactivated FoxP2 neurons, like other PBeL neurons, project axons to the bed nucleus of the stria terminalis and amygdala (Geerling & Loewy, 2006. To our surprise, retrograde tracer injections in each target region exclusively labeled PB neurons that did not contain FoxP2 ( Figure 1a-f).
Seeking additional markers to fill this gap in the FoxP2 distribution led us to another transcription factor expressed in this region, Lmx1b (Asbreuk et al., 2002;Dai et al., 2008). Immunolabeling Lmx1b Lmx1b CTb F I G U R E 1 Lmx1b and FoxP2 in the rat parabrachial nucleus (PB). Neurons containing cholera toxin b (CTb, green) retrograde labeling after CTb injections into the central nucleus of the amygdala (CeA; a-c) or bed nucleus of the stria terminalis (BST; d-f) did not contain the transcription factor FoxP2 (blue). Neurons containing the transcription factor Lmx1b (red) filled a gap in the FoxP2 distribution (g-i) and were retrogradely labeled after CTb injection into the BST (j-l). Similarly, CTb injection into the CeA (m, rat case #5133) produced retrograde labeling predominantly in Lmx1b-containing neurons (n). Approximate level caudal to bregma (in mm) is shown at bottom-left in (m, n). Scale bar is 100 µm and applies to panels (a-l). Abbreviations: mesV, mesencephalic tract and nucleus of the trigeminal nerve; scp, superior cerebellar peduncle pressure (Miller et al., 2012). These mutually exclusive distributions suggested the possibility of using transcription factors to classify all PB neurons. To explore this possibility using a richer set of genetic tools, we began studying the PB in mice.

Phox2b ventral to the PB
The ventral gradient of lighter Lmx1b expression ( Figure 2) and Foxp2 mRNA in the mouse PB. Diaminobenzidine (DAB) in situ hybridization revealed separate distributions of Lmx1b mRNA (a-d) and Foxp2 mRNA (e-h) expression across four rostral-to-caudal sections through the PB region in mice. Approximate level caudal to bregma is shown at the bottom-right of each panel (in mm). Translucent highlights identify the scp, brainstem surface, and fourth ventricular surface. Arrowheads in panels (b, c, f, g) highlight dense Lmx1b and absent Foxp2 in a region of the mouse PB that is homologous to the "external lateral" subnucleus in rats (Fulwiler & Saper, 1984). Scale bars in (a) and (e)   and Lmx1b (Figures 9 and 10). Lmx1b labeling here (ventral to the PB) was lighter and always colocalized with Phox2b, while Lmx1b labeling in the PB was more intense and rarely colocalized with Phox2b.
Across all rostrocaudal levels, we found Lmx1b immunofluorescence in approximately half the neurons that contained Phox2b (range 44-57%, n = 3 mice), but given the light and variable intensity of Lmx1b labeling in the supratrigeminal region, this may be an underestimate.

Diverse Kölliker-Fuse subpopulations
Rostrally and laterally, the Phox2b population merged seamlessly through the KF (Figures 9a and 10a). This was unexpected because a previous study in rats found Phox2b in very few KF neurons (Kang et al., 2007). Including this substantial Phox2b-immunoreactive population, the KF contained the greatest transcription factor diversity in this region of the brainstem.

Unidentified neurons
To determine whether the PB region includes additional neurons that were not identified by our adult transcription factor markers (Lmx1b,

Lmx1a
First, we examined adult expression of Lmx1a, the paralog of Lmx1b.
Like Lmx1b, Lmx1a skewed ventrally (through PBeL and KF) and F I G U R E 1 1 Lmx1a mRNA labeling. DAB in situ hybridization revealed the distribution of Lmx1a expression at four, successive rostral-to-caudal sections through the PB region. Approximate level caudal to bregma is shown at bottom right (in mm). Black arrowhead (a) highlights the KF region. Translucent highlights identify the scp and boundaries between the brainstem surface and cerebellum. Mauve arrowheads (b, c) highlight dense Lmx1a labeling in a region homologous to the "external lateral" subnucleus in rats (Fulwiler & Saper, 1984). Scales bar in (a) is 200 µm and applies to all panels. Abbreviation: mcp, middle cerebellar peduncle caudally (through the superior cerebellar peduncle and medial PB).
Unlike Lmx1b, however, we did not find Lmx1a labeling in the LC. Nor was there any Lmx1a labeling in the supratrigeminal nucleus or principal sensory trigeminal nucleus. We found Lmx1a expression in several regions that lacked Lmx1b, including the cerebellar flocculus, superior vestibular nucleus, and dorsal cochlear nucleus (not shown), as well as a population of small, dense cells along the brainstem surface and atop the middle cerebellar peduncle (Figure 11a,b). Also, the choroid plexus had more extensive Lmx1a labeling than Lmx1b (not shown).

Atoh1 derivation identifies remaining PB neurons
After Lmx1a Cre fate-mapping failed to identify many neurons in the "NeuN-only" distribution, we next focused on neurons that derive from Atoh1-expressing precursors. Across several days of embryonic development, Atoh1-expressing precursors in the rhombic lip neuroepithelium produce (1) glutamatergic neurons that populate the PB and many other brainstem nuclei, followed by (2)  Adult cells do not express Atoh1, so we used Cre fate-mapping to identify Atoh1-derived neurons. Atoh1-Cre;R26-LSL-tdTomato mice had extensive tdTomato expression in the cerebellum and brainstem (n = 4, Figure 14a), similar to previous descriptions in other Cre-reporter strains (Machold & Fishell, 2005;Wang et al., 2005). We found ubiquitous tdTomato expression in cerebellar granule cells, and none in Purkinje cells. Fibrous labeling pervaded the molecular layer, arbor vitae, superior cerebellar peduncle, and middle cerebellar peduncle. Outside the cerebellum, we found fibrous labeling in a ventral majority of the trigeminal principal sensory nucleus, which contrasted a near-absence of labeling dorsally in this nucleus. The trigeminal motor nucleus and a round, central portion of the facial motor nucleus were both devoid of labeling, but the absence of labeling in these small regions contrasted a broader meshwork of labeled fibers in the surrounding brainstem. In the LC, tdTomato did not colocalize with tyrosine hydroxylase, but we found smaller, tdTomato-expressing neurons scattered through the dorsal LC, and Barrington's nucleus contained large, tdTomatoexpressing neurons. We also identified a small number of tdTomatoexpressing neurons that contained Phox2b caudal and medial to the caudal LC, near the medial vestibular nucleus (not shown).
The dense, fibrous background of axonal and dendritic tdTomato made it difficult to distinguish individual neurons and colocalize other molecular markers (Figure 14), so we crossed Atoh1-Cre mice to the L10GFP Cre-reporter strain described above. In the resulting progeny (Atoh1-Cre;R26-lsl-L10GFP, n = 6), the GFP reporter concentrated in cell bodies, rather than axons and dendrites, allowing us to more definitively distinguish Atoh1-derived neurons and colocalize molecular markers (Figures 15-17).
Within the PB, we did not find any cells containing both L10GFP and Lmx1b, indicating total mutual exclusivity between Atoh1-derived neurons and Lmx1b expression. At the dorsal edge of the principal sensory trigeminal nucleus, we occasionally found an L10GFP-expressing neuron with an Lmx1b-immunoreactive nucleus, but these rare cells F I G U R E 1 2 Lmx1a Cre fate-mapping. Cre fate-mapping for Lmx1a (tdTomato, pseudocolored ice blue), followed by immunofluorescence labeling for Lmx1b (red) and FoxP2 (green), shown at four rostral-to-caudal levels through the PB region. Approximate level caudal to bregma is shown at bottom-left in the first panel of each row (in mm). Scale bars (a, d, g, j) are 200 µm and apply to all other panels in the same row F I G U R E 1 3 Lmx1a-derived KF neurons and rostral-to-caudal counts across the PB region. Panels (a-g) show Cre fate-mapping for Lmx1a (tdTomato, pseudocolored ice blue) and immunofluorescence labeling for Lmx1b (red) and FoxP2 (green) in the KF region (ventral to the rostral PB). Panels (h-n) show the lack of Lmx1a Cre-reporter expression in "caudal KF" neurons that contain FoxP2 (ventrolateral to the caudal PB). Approximate level caudal to bregma (in mm) is shown at bottom-right in (g, n). All scale bars are 50 µm. Scale bar in (a) applies to (b-f) and scale bar in (h) applies to (i-m). Bottom graph (o) shows rostral-to-caudal counts of PB neurons expressing the tdTomato Cre-reporter for Lmx1a, FoxP2, or Lmx1b, as well as neurons containing both Lmx1b and Lmx1a Cre-reporter. Counts were averaged at each level (n = 3 mice), with variance represented by a standard deviation envelope. Approximate bregma levels are shown on the x-axis F I G U R E 1 4 Atoh1 Cre fate-mapping with tdTomato. Cre fate-mapping for Atoh1 identified neurons in the cerebellum, PB, and several other brainstem regions, plus extensive labeling in fibrous processes. (a) Fluorescent reporter expression for Atoh1-Cre (tdTomato, pseudocolored coral-red) across five rostral-to-caudal tissue sections spanning the PB region. At each level, the approximate distance caudal to bregma is shown at bottom-right. (b-d) Cre fate-mapping for Atoh1 followed by immunofluorescence labeling for Lmx1b (ice blue). (e-h) Immunofluorescence labeling for Phox2b (green) and ChAT (blue) at a mid-level through the PB region (approximately bregma −5.1 mm). All scale bars are 200 µm. Scale bar in (b) also applies to panels (c, d). Scale bar in (h) also applies to panels (e-g) F I G U R E 1 5 Atoh1 Cre fate-mapping with L10GFP. The L10GFP Cre-reporter identified Atoh1-derived cell bodies. Panels (a-f) show L10GFP expression across six rostral-to-caudal levels of the PB region in an Atoh1-Cre;R26-lsl-L10GFP mouse after immunofluorescence labeling for Lmx1b (red) and FoxP2 (magenta). Nuclear FoxP2 immunofluorescence overlapping cytoplasmic L10GFP appears white. Arrowheads in panels (a, b) highlight the KF region. Arrowheads in (d) highlight a prominent, dorsal cluster of L10GFP-expressing neurons. Translucent outlines in each panel show the location of a small white matter tract alongside the lateral PB and KF region, which is labeled "ventral spinocerebellar tract" in current brain atlases. Approximate distance caudal to bregma is provided at the bottom-right of each panel (in mm). All scale bars are 200 µm represented just 0.002% of the 12,606 L10GFP-expressing neurons counted across this region (n = 3 mice). We also found mutual exclusivity between L10GFP and Phox2b in the PB and LC, although sparse Atoh1-derived neurons located caudal and ventral to the LC did contain Phox2b along the medial vestibular nucleus and facial nerve genu.
In contrast to their mutual exclusivity with Lmx1b, many Atoh1derived PB neurons contained FoxP2. As in the tdTomato strain, double-labeled neurons intermingled with other Atoh1-derived PB neurons that lacked FoxP2 (Figures 15 and 16). At caudal levels, these intermingled populations (Atoh1-derived neurons with and without F I G U R E 1 6 Atoh1 Cre fate-mapping with L10GFP: magnified color separations and plots. Panels (a-g) show immunofluorescence labeling for FoxP2 (magenta) and Lmx1b (red) after fate-mapping for Atoh1-Cre at a mid-rostral level of the lateral PB. (h-n) Immunofluorescence labeling for FoxP2 and Lmx1b after L10GFP fate-mapping for Atoh1-Cre at a mid-caudal level of the PB, centered over the "head" and "waist" of the scp. Approximate bregma levels are shown at bottom-right in (g, n). All scale bars are 50 µm. Scale bar in (a) applies to (b-f) and scale bar in (h) applies to (i-m). (o, p) Rostral-to-caudal plots show the distribution of Atoh1-derived neurons across the PB region, including large subsets with and without FoxP2. Arrowheads in (p) highlight a dorsal cluster of L10GFP-expressing neurons that lack FoxP2. Throughout the PB, Lmx1b and L10GFP were mutually exclusive (no L10GFP-expressing PB neurons contained Lmx1b). Approximate bregma levels are shown at bottom-right in (g, n). Abbreviation: cKF, "caudal KF" population F I G U R E 1 7 Cre fate-mapping for Atoh1 in KF and "caudal KF" with counts of Atoh1-derived neurons across the PB region. Immunofluorescence labeling for Lmx1b (red) and FoxP2 (blue) after Cre fate-mapping for Atoh1 (green) in the KF region (a-g), and in the "caudal KF" (h-n). Approximate distance from bregma is shown at bottom-right in (g, n). All scale bars are 50 µm. Scale bar in (a) also applies to (b-f). Scale bar in (h) also applies to (i-m). (o) Counts of Atoh1-derived, FoxP2, Lmx1b, and Atoh1+FoxP2 double-labeled neurons across six rostro-caudal levels. Counts were averaged at each rostrocaudal level (n = 3 mice), with variance represented by a standard deviation envelope. Approximate distance caudal to bregma is shown on the x-axis. (p) Venn diagram of transcription factor markers that identify neuronal populations in the PB region FoxP2) extended around and through the superior cerebellar peduncle and intermingled extensively with Lmx1b-immunoreactive neurons (Figure 15e,f). At middle and rostral levels, Atoh1-derived neurons filled the lateral PB except for PBeL (Figure 15a-d). At rostral levels, their distribution extended up to the dorsal nucleus of the lateral lemniscus and down to the PB-KF border (Figure 15a).
At mid-levels of the PB, we again found a prominent, dorsal cluster of Atoh1-derived neurons resembling the rat "internal lateral" subnucleus (arrowheads in Figures 15d and 16p). These neurons lacked FoxP2 and bordered a thin white matter tract that is labeled "ventral spinocerebellar tract" in current atlases (Dong, 2008;Paxinos & Franklin, 2013). Outside this one cluster, Atoh1-derived neurons with and without FoxP2 intermingled extensively. FoxP2 colocalization was most prevalent near PBeL (Figures 15b-d and 16a-g), less prevalent dorsally, and least prevalent dorsomedially, near the head and waist of the superior cerebellar peduncle (Figure 16h-n). Also intermingling with these subsets were fewer neurons containing FoxP2 without L10GFP; these neurons did not form any discrete clusters except for the "caudal KF" population ventrolateral to the PB.
At rostral levels, none of the KF populations described above (neurons containing Lmx1b, Lmx1b+FoxP2, Lmx1b+Phox2b, or Phox2b) expressed L10GFP. Nor did we find L10GFP expression in the contiguous "lateral crescent" neurons alongside PBeL. Also, L10GFP was not expressed in any "caudal KF" neurons ( Figure 17). Thus, we did not find any evidence to support previous claims that KF neurons derive from Atoh1-expressing precursors in the rhombic lip (Gray, 2008; Heijden & Zoghbi, 2018).

Interim summary
Identifying the large population of Atoh1-derived PB neurons that lack FoxP2 completed a core set of developmental-genetic markers useful for classifying neurons in this region. Specifically, Lmx1 paralogs (primarily Lmx1b, but also Lmx1a) and Atoh1 define two, mutually exclusive macropopulations of glutamatergic neurons (Figure 17p). We designed our experiments to assess colocalization, rather than total numbers, but as a rough estimate, summing these two macropopulations suggests that there are approximately 20,000 PB neurons on each side of the mouse brainstem (21,472 ± 1636; n = 3 mice).

Subpopulations of Lmx1-and Atoh1-derived PB neurons
Transcription factors influence the repertoire of genes (including other transcription factors) that neurons transcribe, and this "transcriptome" shapes the pattern of synapses a neuron establishes with other neurons (Hirsch et al., 2021). Therefore, using transcription factors to classify PB neurons should help us interpret other patterns of gene expression and neural circuit connectivity. For example, under a cytoarchitectural framework, Foxp2 expression marked an indiscrete population of similar-appearing cells, strewn across virtually every PB subnucleus (Figure 2). In contrast, our new framework identifies two glutamatergic subpopulations (a large subset of Atoh1-derived PB neurons and a small subset of Lmx1-derived neurons; Figure 19) and a GABAergic subpopulation ("caudal KF"). Our molecular framework also revealed novel, diverse subsets of KF neurons (Phox2b-only, Phox2b+Lmx1b, FoxP2+Lmx1b, Lmx1b-only; Figure 10). Cataloging all the genes that distinguish subsidiary PB subpopulations is not within the scope of this study, where our goal is to provide a molecular framework for interpreting such data. Nonetheless, to show the predictive advantages of a molecular ontology, we will highlight two further subpopulations within each PB macropopulation.
Next, we examined the expression of Calca, which encodes the neuropeptide CGRP (calcitonin gene-related peptide). Calca-expressing PB neurons form a critical link in signaling pathways for conditioned taste aversion, malaise, and anorexia, plus alerting responses to aversive stimuli that include pain, itch, and hypercarbia (Carter et al., 2015;Chen et al., 2018;Kaur et al., 2017;Palmiter, 2018;Saper, 2016). We found Calca mRNA in the outer part of PBeL and in other neurons described previously (Huang et al., 2021), but not in any Atoh1-derived neurons (Figure 18h-j), consistent with the mutual exclusivity between CGRP and FoxP2 in this region (Huang et al., 2021). Instead, labeling Calca mRNA in Lmx1a Cre-reporter mice revealed extensive colocalization with tdTomato (Figure 18k-m), resembling the colocalization between CGRP and Lmx1b in both rats and mice (Huang et al., 2021;Miller et al., 2012). Therefore, like Satb2, Calca expression identifies a subset of the Lmx1 PB macropopulation ( Figure 19).
We then identified two markers within the Atoh1 macropopulation. Labeling Pdyn mRNA, which encodes the neuropeptide dynorphin, identified a distribution resembling previous reports in rats and mice (Geerling et al., 2016;Huang et al., 2020;Miller et al., 2012). Pdyn expression in the mouse PB identified neurons in regions resembling the "dorsal lateral" (Figure 18r) and "central lateral" subnuclei in rats (Fulwiler & Saper, 1984), plus a caudal subset of Pdynexpressing neurons in the "pre-LC" population that spans the medial PB and dorsal LC (see Gasparini et al., 2021), and all these neurons expressed L10GFP. Therefore, Pdyn-expressing PB neurons are a subset of the larger Atoh1 macropopulation (Figure 19).
Another neuropeptide, gastrin-releasing peptide (Grp), identified a separate subpopulation (Figure 18q). At mid-caudal levels, Grpexpressing neurons occupied a dorsomedial subregion, as reported F I G U R E 1 8 Subpopulations of Lmx1-and Atoh1-derived PB neurons. Subsidiary genes define separate subpopulations within the Atoh1-and Lmx1-derived PB macropopulations. (a-g) Immunofluorescence labeling for Satb2 (light blue) and Lmx1b (red) with L10GFP Cre-reporter for Atoh1 (green) at a mid-caudal level of the PB. (h-j) FISH labeling for Calca (red) with L10GFP Cre-reporter for Atoh1 (green) at a mid-level of the PB. (k-m) FISH labeling for Calca with tdTomato (pseudocolored ice blue) Cre-reporter for Lmx1a in PBeL. (n-t) FISH labeling for Grp mRNA (red) and Pdyn mRNA (magenta) with L10GFP Cre-reporter for Atoh1 (green) at a mid-caudal level of the PB. (u-w) FISH labeling for Grp mRNA (red) in a dorsal cluster of larger neurons that resemble the "internal lateral" subnucleus of the rat PB (Fulwiler & Saper, 1984), with Ubc mRNA (green) shown for neuroanatomical background. Approximate bregma levels shown at bottom-right in (g, j, m, t, w). All scale bars are 100 µm and apply to similar panels F I G U R E 1 9 Distribution of Atoh1 and Lmx1 subpopulations. The PB contains two neuronal macropopulations, which derive from embryonic precursors expressing Atoh1 (green) or Lmx1 (Lmx1b and Lmx1a; red). This figure illustrates the rostral-to-caudal distribution of subsidiary populations within each macropopulation. Some genetic markers highlight a subset of just one macropopulation, exemplified by Pdyn and Grp (Atoh1 macropopulation) and by Calca and Satb2 (Lmx1 macropopulation). In contrast, Foxp2 identifies separate subsets within each macropopulation. Foxp2 is expressed by an extensive subset of Atoh1-derived neurons (dark-green dots) and by a more restricted, ventral subset of the Lmx1 macropopulation that is located rostrally, in the KF (dark-red dots), separate from the caudal population of GABAergic FoxP2 neurons in the "caudal KF" (cKF, light blue dots). Approximate level caudal to bregma (in mm) is shown at bottom-left for each of the three PB levels illustrated  Figure 6H of Wada et al., 1990), surrounding the "head" and "waist" of the superior cerebellar peduncle (Figure 18p). At mid-levels of the PB, Grp identified a prominent, dorsal cluster (Figure 18u-w) resembling the rat "internal lateral" subnucleus (Fulwiler & Saper, 1984). Further rostrally, Grp-expressing neurons concentrated near PBeL and expressed Foxp2 but not Calca (not shown). Overall, Grp expression identifies another, distinct subset within the Atoh1 PB macropopulation (Figure 19).

Genetic segregation of axonal trajectories and output targets
Embryonic transcription factors establish patterns of neurogenesis and axon growth needed for adult circuit functions. Connectivity patterns that define and distinguish adult neurons often reflect their separate developmental lineages, which are determined in large part by embryonic transcription factors (Hirsch et al., 2021). Thus, we hypothesized that macropopulation identity determines the trajectory and targets of axons projecting from the PB to distal brain regions.
First, to test whether developmental-genetic identity predicts axonal trajectory, we compared axonal labeling in the midbrain, rostral to the PB, after Cre fate-mapping with tdTomato. In Lmx1a Crereporter mice, we found tdTomato labeling in fascicles of axons within the central tegmental tract (CTT), both in late-stage embryos (E17.5, n = 2; not shown) and in adult mice (n = 2, Figure 20a). In every case, this fibrous tdTomato labeling in the CTT traced back to an origin in the PB (not shown). In contrast, the CTT was unlabeled in Atoh1 Cre-reporter mice (n = 4). In these mice, we instead found fibrous tdTomato labeling in the superior cerebellar peduncle (arborizing extensively in the red nucleus) and in the medial lemniscus (arrowhead in Figure 20b), plus a contiguous band of axons lateral to the medial lemniscus, in-between the red nucleus and substantia nigra. Lmx1a Cre-reporter mice did not have any axonal labeling in this ventral pathway.
These dichotomous developmental patterns matched patterns of axonal labeling we identified after transducing adult PB neurons that express Calca (Lmx1 macropopulation) or Pdyn (Atoh1 macropopulation). Our previous injections of AAV8-DIO-synaptophysin-mCherry into the PB (Huang et al., 2020;Huang et al., 2021) produced axonal labeling in either the CTT (Calca cases; Figure 20c) or ventral pathway (Pdyn cases; Figure 20d). The strikingly similar results of developmental fate-mapping and adult Cre-conditional labeling support our hypothesis that macropopulation identity predicts the trajectory of PB axonal projections. Specifically, neurons in the Lmx1 PB macropopulation extend axons through the CTT (between the oculomotor nucleus and red nucleus), while neurons in the Atoh1 PB macropopulation extend axons through a ventral pathway (between the red nucleus and substantia nigra, alongside the medial lemniscus).
Next, to test whether developmental-genetic identity also predicts the distal targets of these axonal projections, we combined Cre fatemapping with retrograde axonal tracing. In Atoh1-Cre;R26-lsl-L10GFP mice, we injected a retrograde tracer (cholera toxin b subunit, CTb) into brain regions that receive heavy input from glutamatergic PB neurons (Huang et al., 2020). We then compared L10GFP expression (marking  (Li et al., 2005;Marfurt & Rajchert, 1991;Rokx et al., 1986;Stanek et al., 2016), and we found that glutamatergic neurons here project axons to the hindbrain reticular formation Huang et al., 2020). We also confirmed in mice that injecting CTb into the medullary reticular formation produces retrograde labeling in neurons along the ventral edge of the rostral PB (Huang et al., 2021).
To test whether these neurons contain Phox2b or Lmx1b, we made additional CTb injections into the medullary reticular formation (n = 4; Figure 21c), then immunolabeled Phox2b, Lmx1b, and CTb. In these cases, CTb-labeled neurons distributed along the ventral PB border.
Many contained Phox2b (with or without Lmx1b), and some contained Lmx1b without Phox2b (Figure 21c 1-c4). Therefore, glutamatergic neurons expressing Phox2b or Lmx1b immediately ventral to the PB project axons to the hindbrain reticular formation.

DISCUSSION
Our findings lay the foundation for a new ontology that uses intrinsic, molecular information to identify PB neurons. This molecular ontology offers an observer-independent alternative to taxonomies that are based on the position of a white matter tract that disperses PB neurons (Figure 22a,b). Molecular features blur some of the atlas boundaries inferred from Nissl cytoarchitecture and reveal an intermingled web of developmentally distinct macropopulations spanning the superior cerebellar peduncle, challenging its utility for subdividing the PB.
Beyond laying the groundwork for a new way to classify PB neurons, the information presented here opens opportunities to improve our understanding of interoceptive neural circuit functions.

Practical implications, limitations, and opportunities
Our findings represent the first comprehensive framework for interpreting neuron-type-specific molecular information in the PB region.
Adopting this developmental-genetic framework will help investigators organize transcriptomic data and focus future attempts to identify and study PB neurons.
Transcriptomic data from an adult brainstem should be sufficient to classify glutamatergic neurons that express Lmx1a/b and other adult markers, but adult neurons do not express Atoh1 (Machold et al., 2011), so identifying Atoh1-derived subpopulations that do not express Foxp2 may require a fate-mapping strategy like the one used here. Among these Atoh1-derived PB neurons that do not express Foxp2, we identified a large, Grp-expressing subset. This Grp subset likely contains further subpopulations, prominently including a dorsal cluster resembling the "internal lateral" subnucleus in rats (Fulwiler & Saper, 1984;Paxinos & Watson, 2007). Neurons in this cluster relay pain-related information from the spinal cord to the thalamus Bester et al., 1999;Bester et al., 1995;Bourgeais et al., 2001;Feil & Herbert, 1995;Kitamura et al., 1993;Krout & Loewy, 2000), but the term "internal lateral" was omitted from a major mouse brain atlas (Dong, 2008;Wang et al., 2020) and its location was added to a separate subnucleus ("superior lateral"), which had been distinguished in rats as sending axonal projections to the hypothalamus (Bester et al., 1997;Bester et al., 1995;Fulwiler & Saper, 1984). This new taxonomy--substituting "superior lateral" for "internal lateral" in mice--appeared in subsequent studies of neurons that relay noxious spinoparabrachial information to the thalamus (Barik et al., 2021;Deng et al., 2020). In contrast to many other findings that blur atlas boundaries, our results unambiguously support the cytoarchitectonic distinction of an "internal lateral" population (Fulwiler & Saper, 1984) by identifying these neurons as a Foxp2-negative subset of the Atoh1 macropopulation. Except for a rostral subset that express Foxp2, Grp-expressing neurons resemble the distribution of PB neurons that express the substance P receptor Tacr1 (Barik et al., 2021), which fail to develop in mice with Atoh1 deletion from the rostral rhombic lip (van der Heijden & Zoghbi, 2018). The Grp F I G U R E 2 1 Distal targets of axonal projections from separate PB macropopulations. Neurons in the PB region contained retrograde CTb labeling after tracer injections into the ventromedial hypothalamic nucleus (VMH), central nucleus of the amygdala (CeA), and intermediate reticular formation (IRt). (a-c) CTb injection sites (white). (a1-a4) Immunofluorescence labeling for CTb (white) after tracer injection into the VMH along with L10GFP Cre-reporter expression in Atoh1-derived neurons and Lmx1b immunofluorescence (red) at a level approximately 5.0 mm caudal to bregma. (b1-b4) Immunofluorescence labeling for CTb after tracer injection into the CeA, along with L10GFP Cre-reporter for Atoh1 and Lmx1b immunofluorescence (approximately 5.1 mm caudal to bregma). (c1-c4) Immunofluorescence labeling for CTb, Phox2b (blue), and Lmx1b along the ventral border of the PB. Light blue arrowheads indicate neurons that contain both CTb and Phox2b. Light red arrowheads indicate neurons that contain both CTb and Lmx1b. White arrows highlight neurons that contain CTb and both transcription factors. Approximate bregma levels are shown at bottom-left in (a4, b4, c4). Scale bars in (a-c) are 1 mm. Scale bars in (a4-c4) are 200 µm and apply to panels above F I G U R E 2 2 Summary. (a, b) Illustrated distribution of Atoh1 (green) and Lmx1 (red) PB neurons in late-embryonic (a) and adult (b) mice. Cerebellar axons project through this region and form the scp, dispersing PB neurons. (c) PB neurons in separate Atoh1 and Lmx1 macropopulations project axons through separate pathways to separate forebrain targets. Note that neurons in each macropopulation project to additional brain regions (see Huang et al., 2021). Hindbrain projections from ventral populations are not shown distribution also partly resembles the distribution of Penk-expressing PB neurons, a subset of which project axons to the thalamus (Hermanson & Blomqvist, 1997). We are testing several additional markers (unpublished) and expect that single-cell transcriptomic analysis will identify a larger suite of neuronal subpopulations within the Atoh1 macropopulation.
From a purely practical perspective, it is useful to classify neurons using developmental-genetic features, rather than cytoarchitecture, because the resulting information can guide genetically targeted experiments that evaluate hypotheses about specific subpopulations.
Until recently, studying the PB involved inserting an electrode, needle, micropipette, or cannula (to stimulate, inhibit, destroy, trace, or record neurons) and then using a cytoarchitectural taxonomy to correlate experimental results with histological location in each brain.
Limiting the immense potential of this approach were the lack of a framework for understanding ontological relationships between subpopulations and a lack of markers for remaining PB neurons. Now, learning that the PB contains two, genetically distinct macropopulations has immediate practical advantages. For example, investigators can rapidly focus the hunt for a subpopulation mediating a particular effect by separately ablating or stimulating each macropopulation to exclude one or the other half of all PB neurons. Even without a full and final menu of markers to distinguish every remaining subpopulation, combining our framework with existing information opens opportunities to access previously indistinguishable subsets by using intersectional genetic targeting methods (Branda & Dymecki, 2004;Fenno et al., 2014;Madisen et al., 2015;Weinholtz & Castle, 2021). As an example, retrograde axonal tracing identified two, distinct subsets of Cck-expressing PB neurons: one sends output to the cerebral cortex, and the other to the hypothalamus . Using intersectional methods to target Cck-expressing neurons derived from precursors expressing either Lmx1b (cortex-projecting) or Atoh1 (hypothalamus-projecting) should provide the selective access needed to isolate and test each subset. Intersectional methods will be particularly important for studying KF neurons (discussed below).

Comparison with previous neurodevelopmental literature
Our study is the first to identify the PB as a blend of two, developmentally distinct macropopulations. Previous descriptions of Lmx1a and Lmx1b here did not mention Atoh1 (Asbreuk et al., 2002;Dai et al., 2008;Miller et al., 2012;Sarropoulos et al., 2021;Zou et al., 2009), and initial studies reporting Atoh1-derived PB neurons did not mention Lmx1a or Lmx1b (Machold & Fishell, 2005;Rose et al., 2009;van der Heijden & Zoghbi, 2018;Wang et al., 2005), leaving the ontological relationships between these populations unclear. One study reported that many PB neurons fail to develop in Atoh1-null mice (Rose et al., 2009) and showed a loss of Slc17a6 (Vglut2) mRNA at a far-rostral level without showing middle or caudal levels of the PB or labeling markers that might have identified neurons in the Lmx1 macropopulation. Another study reported that all Atoh1-derived PB neurons contain CGRP (van der Heijden & Zoghbi, 2018), which is incompatible with the mutual exclusivity between Calca mRNA and Atoh1-derived neurons (Figure 18h-j). This previous result likely involved off-target immunoreactivity because there is little overlap between the distribution of Atoh1-derived neurons and that of Calca mRNA, or of CGRP labeling produced by antisera validated in knockout mice (Huang et al., 2021).

In the embryonic (E14) brainstem, gene-chip analysis identified
Foxp2 expression in Atoh1-derived neurons (Machold et al., 2011). In neonatal (P0) mice, another study identified FoxP2 immunoreactivity in Atoh1-derived neurons that appear to be located in the lateral PB but were labeled "KF" (Figures 3 and 4 of Gray, 2008). Other images from the same study showed Lmx1b immunoreactivity in what appears to be the external lateral PB (labeled "Pr5" in Figure 3E of Gray, 2008), plus a ventrolateral cluster of FoxP2-immunoreactive, Ptf1a-derived neurons resembling the "caudal KF" (Figure 4A of Gray, 2008). Besides discrepant nomenclature, these results are consistent with our conclusion that adult Foxp2 expression identifies three, developmentally distinct subpopulations here: (1) a small, rostral subset of Lmx1-derived glutamatergic neurons (within KF); (2) a large subset of Atoh1-derived glu-tamatergic neurons (within PB); and (3) a small, ventrolateral cluster of Ptf1a-derived GABAergic neurons ("caudal KF").
A functional role remains to be determined for many PB neurons, including many of the neurons identified here as expressing Grp.
Also, we do not yet know the ontological relationships of PB neurons expressing Oxtr1 (water intake), Htr2c (food, water, and salt intake), or Oprm1 (opioid-withdrawal; opioid-induced respiratory depression) relative to our developmental-genetic framework. While investigating these and other genetic markers (unpublished), we expect that transcriptomic analysis will identify additional, functionally distinct subpopulations within each PB macropopulation.
An interesting question is whether Lmx1 and Atoh1 derivation confer separate functional themes. For example, a broadly interoceptive theme may befit the Lmx1 macropopulation. Its neurons overlap axon terminal fields that deliver viscerosensory information from the nucleus of the solitary tract (Geerling & Loewy, 2006b;Rinaman, 2010) and an oral-sensory subregion of the spinal trigeminal nucleus (Dallel et al., 2004). We showed here that the Lmx1 macropopulation includes Satb2 gustatory-relay neurons, as well as Calca neurons that activate in response to visceral stimuli, and KF neurons, which receive chemosensory and other viscerosensory input and exert visceral motor effects on airway tone, breathing, and sympathetic function (Chamberlin & Saper, 1992Dutschmann et al., 2021;Dutschmann & Herbert, 2006;Yokota et al., 2015).
However, there may be exceptions to these broad functional themes. Caudal, Atoh1-derived neurons that express Pdyn and Foxp2 may have appetitive, rather than exteroceptive functions (Gasparini et al., 2021;Kim et al., 2020;Lee et al., 2019). And while we do not yet know the connectivity and function of the Atoh1-derived, Grpexpressing neurons that surround the caudal "head" of the superior cerebellar peduncle, this part of the PB may receive primarily interoceptive input (Feil & Herbert, 1995;. Also, previous investigators highlighted neurons in the lateral PB that receive pain-related inputs from both the viscera and the body surface and send output to both the amygdala and hypothalamus (Bernard & Besson, 1990;Bernard et al., 1994;Bourgeais et al., 2003;Chiang et al., 2020;Gauriau & Bernard, 2002). While our neuron-type-specific axonal tracing revealed a sharp, developmental distinction between PB neurons that send output to the amygdala (Lmx1) versus hypothalamus (Atoh1), additional work is needed to test whether interoceptive and exteroceptive input connections to this region segregate similarly. Intersectional genetic and viral targeting methods will help address this question to advance our understanding of the neural pathways transmitting specific channels of interoceptive and exteroceptive information to the brain. This information is necessary for understanding pain, alimentary function, and several other homeostatic functions.

Novel insights into the Kölliker-Fuse nucleus
Our most unexpected finding was that intrinsic, molecular features highlight the KF as a diverse blend of overlapping populations. Having identified the colocalization of Lmx1b and FoxP2 here in rats (Miller et al., 2012) and having distinguished these rostral, glutamatergic neurons from GABAergic neurons in the "caudal KF" , we expected to find a homogenous KF population. Instead, without imposing any cytoarchitectonic boundaries or other constraints, molecular markers highlighted this region as a focally diverse cluster of intermingled neurons. Specifically, immunolabeling Lmx1b, Phox2b, and FoxP2 revealed that the KF contains at least four different populations of glutamatergic neurons (Figure 10a, d-g), separate from the "caudal KF" population of GABAergic, Foxp2-expressing neurons.
None of these populations were identified using cytoarchitectural criteria.
Also, while previous reports suggested that KF neurons are Atoh1derived (Gray, 2008;van der Heijden & Zoghbi, 2018), we found no more than sparse Atoh1 reporter expression ventrolateral to the PB.
Instead, most neurons in the KF region express Lmx1b (alone, or with Foxp2). This Lmx1b-expressing majority intermingles with a substantial minority expressing Phox2b (alone, or with Lmx1b). That is, KF neurons express either Lmx1b or Phox2b, both of which are absent from Atoh1-derived neurons. This information, combined with our discovery that many KF neurons derive from Lmx1a-expressing precursors, challenges previous claims that KF neurons derive from Atoh1expressing precursors in the rhombic lip (Gray, 2008;van der Heijden & Zoghbi, 2018). These claims may have resulted from mislocalizing the KF in mice, which skews rostrally relative to rats Yokota et al., 2015).
Discovering that "the" KF encompasses several populations opens opportunities to test which of these populations are responsible for specific respiratory, orofacial, and autonomic activities associated with this region (Chamberlin & Saper, 1992Dutschmann & Dick, 2012;Stanek et al., 2016;Varga et al., 2020). Specifically, intersectional genetic targeting methods should help distinguish the functions of KF neurons that express Lmx1b (with and without Foxp2) or Phox2b (with and without Lmx1b).
Foxp2-expressing neurons within the Lmx1 macropopulation are unique to the KF and caudally contiguous "lateral crescent" (Figure 19), so these neurons probably mediate a function exclusive to the KF region. In contrast, Phox2b-expressing KF neurons are contiguous with many similar neurons extending back through the supratrigeminal nucleus and hindbrain reticular formation. It is not yet clear whether the functional role of Phox2b-expressing KF neurons diverges from that of Phox2b-expressing supratrigeminal neurons or the contiguous "dA3" population of Phox2b-expressing interneurons in the hindbrain reticular formation (Gray, 2013;Hernandez-Miranda et al., 2017;Kang et al., 2007). Many supratrigeminal neurons send output to cranial motor neurons that pattern movements like chewing, licking, and swallowing (Dempsey et al., 2021;Takatoh et al., 2021;Travers & Norgren, 1983), and in mice, the Phox2b distribution overlaps the locations of neurons labeled by viral retrograde tracing from motor neurons that control orofacial movements (Takatoh et al., 2021). In contrast, very few Phox2b-expressing neurons project axons to autonomic and respiratory premotor neurons in the ventrolateral medulla in rats (Kang et al., 2007). These observations suggest that Phox2b-expressing KF neurons have cephalic premotor functions similar to premotor neurons in the supratrigeminal region and reticular formation and distinct from the respiratory activities typically associated with the KF.
Also relevant to human health is the role of this region in opioidinduced respiratory depression (Prkic et al., 2012;Varga et al., 2020).
Slice recordings identified opioid-sensitive neurons in or near the KF (Levitt et al., 2015;Levitt & Williams, 2018;Saunders & Levitt, 2020), yet expression of the mu opioid receptor (Oprm1) is prominent in Calcaexpressing and intermingled neurons located dorsal and caudal to the KF (Chamberlin et al., 1999;Huang et al., 2021). Intersectional genetic targeting methods should allow investigators to determine whether the Lmx1b neurons expressing Foxp2 (KF) or Calca (PB) contribute to opioid-induced respiratory depression.

CONCLUSION
Developmental-genetic markers identify the PB as a blend of two, mutually exclusive macropopulations. These two macropopulations, defined by Lmx1b and Atoh1, communicate with separate neural circuits. We also found that Lmx1b (with and without Foxp2) and Phox2b (with and without Lmx1b) identify neuronal subpopulations in the KF.
This new, developmental-genetic framework will help organize future transcriptomic and experimental work involving PB and KF neurons, and using a molecular ontology to identify and compare neurons across species may accelerate the translation of PB-related discoveries from experimental animals to human patients.

ACKNOWLEDGMENTS
We thank Yu-Qiang Ding for providing an aliquot of rabbit-anti-Lmx1b antibody (used in rats), Carmen Birchmeier for providing aliquots of guinea pig-anti-Lmx1b antiserum (used in mice), and Hideki Enomoto of Kobe University for providing an aliquot of guinea pig-anti-Phox2b antiserum (used in mice). We thank Kathleen Millen for sharing Lmx1a-Cre mice. We also thank Aislinn Williams for sharing Atoh1-Cre and Andy Russo for sharing (and Richard Palmiter for his permission to use) Calca-Cre mice from their breeding colonies. Finally, we thank Jadylin Tolda for proofreading the manuscript.

AUTHOR CONTRIBUTION
JG planned experiments and supervised the project. JG and RM performed histologic staining and confocal microscopy in rat brain tissue. CM performed confocal microscopy and image processing in rat All authors reviewed and discussed the results and contributed critical feedback and edits that were incorporated into the final manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

PEER REVIEW
The peer review history for this article is available at https://publons.