Molecularly defined circuits for cardiovascular and cardiopulmonary control

The sympathetic and parasympathetic nervous systems powerfully regulate internal organs1, but the molecular and functional diversity of their constituent neurons and circuits remains largely unknown. Here we use retrograde neuronal tracing, single-cell RNA sequencing, optogenetics, and physiological experiments to dissect the cardiac parasympathetic control circuit in mice. We show that cardiac-innervating neurons in the brainstem nucleus ambiguus (Amb) are comprised of two molecularly, anatomically, and functionally distinct subtypes. One we call ACV (ambiguus cardiovascular) neurons (~35 neurons per Amb), define the classical cardiac parasympathetic circuit. They selectively innervate a subset of cardiac parasympathetic ganglion neurons and mediate the baroreceptor reflex, slowing heart rate and atrioventricular node conduction in response to increased blood pressure. The other, ACP (ambiguus cardiopulmonary) neurons (~15 neurons per Amb) innervate cardiac ganglion neurons intermingled with and functionally indistinguishable from those innervated by ACV neurons, but surprisingly also innervate most or all lung parasympathetic ganglion neurons; clonal labeling shows individual ACP neurons innervate both organs. ACP neurons mediate the dive reflex, the simultaneous bradycardia and bronchoconstriction that follows water immersion. Thus, parasympathetic control of the heart is organized into two parallel circuits, one that selectively controls cardiac function (ACV circuit) and another that coordinates cardiac and pulmonary function (ACP circuit). This new understanding of cardiac control has implications for treating cardiac and pulmonary diseases and for elucidating the control and coordination circuits of other organs.

The sympathetic and parasympathetic nervous systems regulate the activities of internal organs 1 , but the molecular and functional diversity of their constituent neurons and circuits remains largely unknown. Here we use retrograde neuronal tracing, single-cell RNA sequencing, optogenetics and physiological experiments to dissect the cardiac parasympathetic control circuit in mice. We show that cardiac-innervating neurons in the brainstem nucleus ambiguus (Amb) are comprised of two molecularly, anatomically and functionally distinct subtypes. The first, which we call ambiguus cardiovascular (ACV) neurons (approximately 35 neurons per Amb), define the classical cardiac parasympathetic circuit. They selectively innervate a subset of cardiac parasympathetic ganglion neurons and mediate the baroreceptor reflex, slowing heart rate and atrioventricular node conduction in response to increased blood pressure. The other, ambiguus cardiopulmonary (ACP) neurons (approximately 15 neurons per Amb) innervate cardiac ganglion neurons intermingled with and functionally indistinguishable from those innervated by ACV neurons. ACP neurons also innervate most or all lung parasympathetic ganglion neurons-clonal labelling shows that individual ACP neurons innervate both organs. ACP neurons mediate the dive reflex, the simultaneous bradycardia and bronchoconstriction that follows water immersion. Thus, parasympathetic control of the heart is organized into two parallel circuits, one that selectively controls cardiac function (ACV circuit) and another that coordinates cardiac and pulmonary function (ACP circuit). This new understanding of cardiac control has implications for treating cardiac and pulmonary diseases and for elucidating the control and coordination circuits of other organs. The autonomic nervous system is a vast network of neurons under involuntary control that is essential for maintaining physiological homeostasis throughout the body. In 1921, Langley divided the autonomic nervous system into the opposing sympathetic and parasympathetic nervous systems, which generally act antagonistically within a given organ through release of noradrenaline (typically mediating 'fight or flight' responses) and acetylcholine (typically mediating 'rest and digest' responses), respectively 1 . This provided a foundation for modern physiology and medicine that has remained mostly intact a century later 2 . Despite the prominent profile of the parasympathetic and sympathetic nervous systems, the molecular, cellular and functional diversity of their constituent neurons and circuits remains largely unknown, most notably at the central nervous system level 2,3 .
One critical organ controlled by the autonomic nervous system is the heart. The parasympathetic and sympathetic nervous systems powerfully influence heart rate, rhythm and contractility to meet internal and external demands 4 . Classical physiological studies have shown that the sympathetic nervous system exerts control of heart rate on a time scale of seconds, whereas the parasympathetic nervous system exerts control on an even faster, beat-to-beat timescale and is thus thought to be responsible for the tight coordination between cardiac and pulmonary function 5 . Cardiac parasympathetic outflow originates from brainstem preganglionic neurons, residing primarily in the nucleus ambiguus (Amb) in the medulla 4 . A minority of cardiac parasympathetic neurons are located in the dorsal motor nucleus of vagus, which controls ventricular inotropy and excitability, but does not control heart rate 6 . Amb neurons innervate cholinergic neurons in parasympathetic cardiac ganglia located on the surface of the heart, and these ganglion neurons in turn innervate the sinoatrial (SA) node, the atrioventricular (AV) node and myocardial tissue. Acetylcholine release from these ganglion neurons reduces heart rate, AV conduction velocity, and myocardial contractility by engaging cardiac muscarinic receptors. Altered sympathetic-parasympathetic signalling balance to the heart has been implicated in a wide range of cardiovascular Article diseases [7][8][9] , and is molecularly targeted by common cardiac therapies such as β-blockers and atropine, but the neuronal cell types within the cardiac parasympathetic and sympathetic nervous systems have not been molecularly defined.

Targeted scRNA-seq of Amb Cardiac neurons
To molecularly define parasympathetic neurons that control the heart, we retrogradely labelled and transcriptionally profiled cardiac-innervating neurons from the mouse brainstem using single-cell RNA sequencing (scRNA-seq). Wild-type postnatal day 1 mice were injected with fluorescent retrograde neuronal tracer cholera toxin B into the pericardial space to label parasympathetic ganglia across the epicardial surface. After one day to allow tracer uptake into pre-ganglionic terminals and retrograde transport to the cell bodies in the brainstem, 50-60 fluorescently labelled cell bodies were observed in the Amb. The labelled cells (which we call Amb Cardiac neurons) were distributed bilaterally in Amb in a stereotyped pattern (Extended Data Fig. 1); similar distributions of Amb Cardiac neurons have been observed in adult rats 10 , cats 11 and dogs 12 . Because of the small number of Amb Cardiac neurons, we aspirated whole retrograde-labelled Amb neurons directly from acute brainstem slices (Fig. 1a). We did the same for an intermingled outgroup population of Amb neurons that do not innervate the heart, by retrograde labelling and aspirating neurons that innervate the cricothyroid laryngeal muscle (hereafter called Amb Laryngeal neurons, approximately 28 per Amb). Following sequencing, 151 Amb Cardiac neurons and 52 Amb Laryngeal neurons passed quality control (mean of 1.9 million reads, 7,761 genes detected per cell; Extended Data Fig. 2), suggesting that we achieved cellular saturation by sequencing more than the numbers of Amb Cardiac and Amb Laryngeal neurons per Amb.
Graph-based clustering of the expression profiles of the combined Amb Cardiac and Amb Laryngeal neuron datasets revealed three clusters of cells (Fig. 1b). One (cluster 3) corresponded largely to Amb Laryngeal neurons (Fig. 1c) and the other two corresponded to Amb Cardiac neurons. Comparison of the expression profiles of Amb Cardiac and Amb Laryngeal neurons identified more than 500 differentially expressed genes (Supplementary Tables 1, 2), including two known markers of Amb Laryngeal neurons (the neuropeptide genes Calca and Calcb 13 ), as well as Dlk1, a regulator of somatic fast motor neuron fate 14 (Fig. 1d). The approximately 390 genes selectively expressed in Amb Cardiac neurons (Supplementary Tables 2, 3) include genes encoding transcriptional (Tbx3) and splicing (Celf6) regulators, as well as genes that regulate synaptic connectivity or projection targets (Cntn5, Cdh8 and Efna5), electrophysiological properties (Kcna5), input signalling (Npy2r and Cnr1) and disease processes (Snca).
To determine whether the ascertained Amb Cardiac genetic signature includes genes that more generally define brainstem parasympathetic neurons, we used the Allen Brain Atlas 15 to examine expression of available Amb Cardiac -specific genes in other brainstem parasympathetic nuclei. This identified 26 Amb Cardiac -specific genes including Celf6, Kcna5, Tbx3 and Dgkb that are also expressed in parasympathetic neurons of the dorsal motor nucleus of vagus (10N), which controls thoracic and abdominal viscera, as well as parasympathetic neurons of the facial nerve controlling the lacrimal and salivary glands (Lac/ Sal) (Extended Data Fig. 3, Supplementary Table 3). These genes are not expressed in adjacent somatic cranial motor nuclei, suggesting that they are general markers of visceromotor neurons; we confirmed their expression in a recent adult scRNA-seq dataset from 10N 3 . We also identified Amb Cardiac -specific genes that were expressed in 10N but not Lac/Sal or were not expressed in brainstem parasympathetic nuclei other than Amb (Supplementary Table 3). Thus, our scRNA-seq analysis of Amb Cardiac neurons identified both general features of brainstem parasympathetic neurons and specific features of Amb Cardiac neurons.

Two types of Amb Cardiac neurons
Sub-clustering of Amb Cardiac neurons (Fig. 1b, c) without Amb Laryngeal neurons confirmed that they are not a homogeneous population, and identified two molecular subtypes (Fig. 2a) that did not split into further sub-clusters. We designated these subtypes ACP and ACV. Both ACP and ACV expressed similarly high levels of the cholinergic marker Slc18a3 and the cranial motor neuron transcription factor gene Isl1 (Fig. 2b), as expected for Amb neurons 16 , whereas more than 200 genes distinguished the two molecular types (Supplementary Tables 4, 5, 6). The differentially expressed genes included genes encoding neuropeptide and neurotransmitter receptors such as thyrotropin-releasing hormone receptor (Trhr), tachykinin receptor 3 (Tacr3), and melanocortin receptor 4 (Mc4r) that were selectively expressed in the ACP population, as well as growth hormone secretagogue receptor (Ghsr), serotonin receptor 3B (Htr3b), somatostatin receptors 2 and 5 (Sstr2 and Sstr5), purinergic receptor P2Y1 (P2ry1), and leptin receptor (Lepr) that were selectively expressed in ACV. We also found subtype-specific transcription factor genes, neuropeptide precursor genes, voltage-gated ion channel genes and genes known to regulate axon guidance and neuronal connectivity (Supplementary Table 6).
To localize the two molecular types of Amb Cardiac neurons in the brainstem, we retrogradely labelled Amb Cardiac neurons and performed immunostaining for the ACP marker calbindin (CALB1) and the ACV marker butyrylcholinesterase (BChE). As predicted by scRNA-seq, Amb Cardiac neurons consisted of one subpopulation that expressed high levels   Fig. 1 | scRNA-seq of Amb Cardiac neurons identifies a genetic signature of brainstem parasympathetic neurons. a, Strategy for isolating Amb Cardiac neurons. Fluorescent cholera toxin B (CTB) was injected into the pericardial space and 1-3 days later the retrograde-labelled cardiac-innervating neurons in the nucleus ambiguus (Amb Cardiac neurons) were aspirated from acute physiological brainstem slices and processed for scRNA-seq. In separate mice (not shown), CTB was injected into the cricothyroid laryngeal muscle to similarly visualize, aspirate and process control Amb Laryngeal neurons. b, t-Distributed stochastic neighbour embedding (t-SNE) plot comparing Amb neuron scRNA-seq expression profiles (dots). Note the three transcriptionally distinct neuronal clusters defined by nearest-neighbour analysis. c, t-SNE plot of the same Amb neurons as in b, but coloured by their retrograde label origin (purple, heart; coral, larynx) . Amb Laryngeal neurons largely comprise cluster 3, whereas Amb Cardiac neurons were largely split between clusters 1 and 2. d, Heat map showing log-transformed expression levels of selected genes differentially expressed between Amb Cardiac and Amb Laryngeal neurons. Gene symbols in black are expected pan-Amb genes. Mouse brain image and those in Figs. 3, 4, 6, 7 and Extended Data Figs. 5 and 7 are reproduced from ref. 35 . of calbindin and low levels of BChE (ACP) (Fig. 2c), and a second subpopulation with high levels of BChE and undetectable calbindin (ACV) (Fig. 2d), with only rare neurons (2-3 neurons per Amb) expressing neither marker. Of note, we observed a stereotyped topographic arrangement of the two subpopulations. ACP neurons (10-20 per nucleus) were located in the rostral Amb, below the compact formation of the nucleus, whereas ACV neurons (30-40 per nucleus) were located exclusively caudal to ACP, surrounding the loose formation of the nucleus (Fig. 2e, f). ACP and ACV marker expression and distributions were similar in adult mice (Extended Data Fig. 4, Supplementary Table 6).
Thus, Amb Cardiac neurons comprise two molecularly and anatomically distinct subtypes, ACP and ACV.

ACP and ACV target distinct heart neurons
To identify the cardiac projection targets of ACP neurons, we used a genetic strategy combined with an adeno-associated virus (AAV) reporter to specifically label this cell type. We injected the rostral Amb of Calb1 cre mice with a Cre-dependent AAV encoding fluorescent protein eYFP (Fig. 3a), which labelled ACP neurons but not ACV neurons (Extended Data Fig. 5c-e) and did not drive expression in 10N or the nodose-jugular complex (Extended Data Fig. 5g-i). Three weeks after AAV injection, eYFP-positive fibres were found innervating cardiac ganglionated plexuses (GPs) on the left atrial surface surrounding the pulmonary veins (Extended Data Fig. 6a, b), the GPs known to mediate cholinergic effects on the heart 17 . When either left or right ACP neurons were labelled, most eYFP-positive fibres innervated a GP around the base of the left pulmonary veins as well as a GP around the inferior pulmonary veins (Extended Data Fig. 6a, b). Within innervated GPs, eYFP-positive fibres coursed toward specific cholinergic ganglion neurons where they terminated in cholinergic, calbindin-positive synapses surrounding each target neuron. Fibres from a single side of the brainstem provided all the cholinergic input for a given target neuron (Fig. 3b)  . c, Immunostaining for the ACP marker calbindin and the ACV marker BChE in rostral Amb. Rostral Amb neurons (arrowheads) labelled by intrapericardial CTB stained positive for calbindin but expressed low levels of BChE, indicating that they were ACP neurons. d, Immunostaining for the ACP marker calbindin and the ACV marker BChE in caudal Amb of same sample as in c. Caudal Amb neurons (arrowheads) labelled by intrapericardial CTB expressed high levels of BChE but stained negative for calbindin, indicating that they were ACV neurons. Scale bars, 20 μm. e, Map of ACP and ACV neurons in Amb. Sagittal schematic view showing soma of all neurons (circles) in a representative single postnatal day 2 Amb that was retrograde labelled with CTB from the heart (CTB>heart) and stained for ACP marker calbindin (dark blue circles) and ACV marker BChE (light blue circles). A minority of retrograde-labelled cardiac neurons (green circles) did not stain for calbindin or BChE. f, Quantification of absolute numbers of ACP neurons (dark blue), ACV neurons (light blue), and double-negative cardiac neurons (green) per Amb. Data are mean ± s.d., n = 3 mice. showing two cardiac ganglion neurons (dashed outlines), one (target of ACP (tACP)) with eYFP-positive ACP input that was also positive for calbindin (ACP marker) and vesicular acetylcholine transporter (VAChT). An adjacent cardiac ganglion neuron (putative target of ACV [(tACV)]) received eYFP − calbindin − VAChT + pre-ganglionic input (red puncta), indicating that it was not innervated by an ACP, but probably by an ACV neuron. Scale bar, 10 μm. c, Schematic of typical ACP and ACV innervation pattern of individual cardiac ganglion neurons (grey circles) within a GP. ACP fibres (dark blue) provide all cholinergic input for tACP neurons. ACV fibres (light blue) provide all innervation for tACV neurons, which are intermingled with tACP neurons. d, Immunostaining of ACV terminals (labelled as in b with eYFP but in a Ghsr cre mouse) in two cardiac ganglion neurons (dashed outlines), one (tACV) with eYFP + calbindin − VAChT + ACV input. An adjacent cardiac ganglion neuron '(tACP)' received eYFP − calbindin + VAChT + input, so was not innervated by an ACV but probably by an ACP neuron. Scale bar, 10 μm.

AAV-DIO-eYFP
Article and right inputs. Thus, although both left and right ACP neurons generally innervate the same subset of cardiac GPs, they innervate a private subset of cholinergic ganglion neurons within these GPs. A similar strategy was used to label ACV neurons in the left or right caudal Amb and identify their cardiac targets (Fig. 3a), substituting a Ghsr cre driver specific for ACV neurons relative to ACP neurons (Extended Data Fig. 5a-d, f, j-l). Unlike ACP neurons, ACV neurons in the right and left Amb showed different GP targeting patterns. eYFP-positive ACV fibres from left Amb innervated all GPs, whereas right ACV neurons primarily innervated a GP at the base of the right pulmonary veins (Extended Data Fig. 6a-c), a GP rarely innervated by ACP neurons (Extended Data Fig. 6a, b, d). eYFP-positive fibres of ACV neurons terminated in cholinergic, calbindin-negative synapses that surrounded each cholinergic target neuron, providing its entire cholinergic input (Fig. 3d), similar to ACP neuron targets. Notably, although ACP and ACV target neurons were intermingled, ACP and ACV neurons almost never innervated the same neuron (Fig. 3c, d).
We conclude that both ACP and ACV neurons innervate cholinergic cardiac GPs. Left and right ACP neurons innervate the same subset of GPs, whereas left and right ACV neurons innervate different sets of GPs. Although each GP receives a mixture of innervation from different sides and different Amb Cardiac cell types, individual ganglion neurons receive all of their input from a single side and a single cell type, implying a mechanism that ensures individual ganglion neurons receive only one input. Such a mechanism could be mediated by classical axon guidance cues, repulsive cues or cell adhesion molecules (Supplementary Table 6), and may serve to avoid interference between the two pathways.

ACP and ACV both slow the SA and AV nodes
To determine the effects of ACP neuronal activation on the heart, we optogenetically activated this cell type in transgenic mice. We expressed the red-shifted channelrhodopsin bReaChES 18 in ACP but not ACV neurons using a Cre-dependent AAV encoding bReaChES delivered into rostral Amb of Calb1 cre mice (Extended Data Fig. 5c-e) along with a fibre-optic cannula. Two to four weeks later, we stimulated bReaChES-expressing ACP neurons with a laser while recording the electrocardiogram (ECG) and respiration under isoflurane anaesthesia. Photostimulation of left ACP neurons (Fig. 4a) caused an immediate, approximately 50% reduction in sinus rate with simultaneous prolonging of the P-R interval, occasionally leading to second-degree AV block ( Fig. 4b, d, e). Upon cessation of stimulation, the heart rapidly returned to normal sinus rhythm. Both SA and AV node effects of left-sided ACP stimulation were abolished by pre-treatment with muscarinic receptor antagonist atropine (Fig. 4d, e), consistent with ACP neurons mediating these effects through cholinergic ganglion neurons that activate cardiac muscarinic receptors, and not through withdrawal in sympathetic tone. Photostimulation of right ACP neurons (Fig. 4f) resulted in a similar atropine-sensitive bradycardia, but no first-or second-degree AV block (Fig. 4g, i, j). Thus, left ACP neurons slow sinus rate and AV node conduction velocity, whereas right ACP neurons slow only sinus rate.
To determine the effects of ACV neuronal activation on the heart, we used the same optogenetic approach in Ghsr cre mice. Photostimulation of left ACV neurons resulted in a ~40% reduction in heart rate with simultaneous AV block ( Fig. 4c-e). Photostimulation of right ACV neurons reduced heart rate by ~50% but with no first-or second-degree AV block ( Fig. 4h-j). Cardiac effects of stimulation of left and right ACV were both fully blocked by atropine, as were those of ACP neurons (Fig. 4d, e, i, j). Thus, the effects of ACV neuron activation on the heart were similar to ACP: left ACV stimulation slowed sinus rate and AV node conduction velocity, whereas right ACV stimulation slowed only sinus rate. Similar cardiac responses and left-right asymmetry have been observed with vagus nerve stimulation in other mammals 19 . In addition to these cardiac responses, Amb photostimulation in Calb1 cre and Ghsr cre mice also resulted in apnea or reduction in respiratory rate, however this respiratory effect was not altered by atropine (Extended Data Fig. 7a-d) and thus was attributed as an artefact of opsin expression in interneurons of the overlapping pre-Bӧtzinger complex breathing control region (Extended Data Fig. 7e-h).
We conclude that, despite their molecular and anatomical differences, optogenetic activation of ACP or ACV neurons results in almost identical inhibitory effects on cardiac SA and AV node function, and because the effects are blocked by atropine they are probably mediated through their projections to cholinergic ganglion neurons.

The baroreflex selectively activates ACV
We sought to identify physiological stimuli that activate ACP and ACV neurons. Amb Cardiac neurons are known 20 to mediate the baroreceptor reflex, a classic cardiovascular reflex essential for homeostatic maintenance of arterial pressure and end-organ perfusion 21 . Aortic arch baroreceptors convert increases in blood pressure to an afferent neuronal signal that results in activation of Amb Cardiac neurons, which counteract the increases in blood pressure by decreasing heart rate 22 .
To determine whether ACP and ACV neurons are activated by the baroreceptor reflex, we induced the reflex with phenylephrine, a peripherally acting α 1 -adrenergic receptor agonist that causes vasoconstriction and , there was a rapid onset second-degree AV block (P waves (red dots) that did not produce QRS complex (large inflection)). Also during stimulation, the P-P interval was increased, indicating a reduction in sinus rate. c, ECG trace during optogenetic stimulation of left ACV neurons in a Ghsr cre mouse. Note the second-degree AV block and lengthening of the P-P interval, as for left ACP stimulation. d, Quantification of effect on heart rate (HR, from P-P interval) in b, c (n = 5 mice per genotype) before (−) or after (+) addition of the muscarinic antagonist atropine (Calb1, P = 0.0009; Ghsr, P = 0.04). e, Quantification of effect on AV node conduction in b, c. Both cell types increased P-R (PR) interval (first-degree AV block) and caused second-degree AV block in some mice (red circled data points), and effects were abolished by atropine (Calb1, P = 0.01; Ghsr, P = 0.04). f-j, Strategy (f) and ECG trace in a Calb1 cre mouse (g), a Ghsr cre mouse (h), and change in HR (i) and change in PR interval (j) in Calb1 cre and Ghsr cre mice following optogenetic activation of right ACP or ACV neurons (n = 5 mice per genotype), as for left ACP and ACV neurons in a-e (Calb1 HR, P = 0.002; Ghsr HR, P = 0.005; Calb1 PR, P = 0.1; Ghsr PR, P = 0.2). Data are mean ± s.d. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant by paired two-tailed t-test.
reflex activation of Amb Cardiac neurons leading to bradycardia 20 . Awake wild-type mice were administered phenylephrine, and 150 min later animals were euthanized and ACP and ACV neurons were immunostained for c-Fos, a neuronal activity marker (Fig. 5a). In vehicle-treated control mice, neither ACP or ACV neurons expressed c-Fos, indicating that ACP and ACV activity is off or low under baseline conditions (Fig. 5b, Extended Data Fig. 8a, b). However, in phenylephrine-treated mice, we observed robust activation of ACV neurons (Fig. 5b, d), with around 70% staining positive for c-Fos. The effect was selective for ACV because ACP neurons remained c-Fos-negative under these conditions (Fig. 5b, c, Extended Data Fig. 9). Thus, the baroreceptor reflex selectively recruits ACV neurons.

ACP neurons also innervate and control the lung
A clue to ACP neuron function came from our discovery that they also innervate another organ. The lung receives innervation from the same branch of the vagus nerve as the heart 23 , and pulmonary and cardiac function are known to be tightly coordinated 24 . In parallel studies of lung innervation, we observed calbindin-positive innervation of pulmonary ganglia. To determine whether ACP neurons project to lung, we immunostained pulmonary ganglia in Calb1 cre mice after labelling ACP neurons with eYFP (Fig. 6a). In addition to innervating cardiac ganglia, we found that ACP eYFP-positive, calbindin-positive terminals also innervated most pulmonary cholinergic ganglia (75%) that surround proximal airways of the lung (Fig. 6b, g). Similar to cardiac ganglia, ACP fibres entered pulmonary ganglia and surrounded individual cholinergic ganglion neurons (Fig. 6c). In contrast to cardiac ganglia, almost all cholinergic pulmonary ganglion neurons were innervated by calbindin-positive fibres (Fig. 6c). In similar experiments with ACV neurons, we observed only rare pulmonary ganglia (3%) that contained eYFP-positive, calbindin-negative ACV fibres ( Fig. 6d-g). Thus, ACP neurons innervate cholinergic pulmonary ganglia and are the major, if not exclusive, source of cholinergic input.
To determine whether single ACP neurons project to both heart and lung, we used sparse-labelling approaches to clonally label ACP neurons in individual mice. Multiple mouse lines and AAV dose optimizations were tested to lower the efficiency of genetic labelling to increase the likelihood of labelling just a single ACP neuron (Methods). The number of GFP-labelled ACP neurons was determined by immunostaining and counting each ACP neuron in every brainstem section spanning Amb. In rare cases (3 out of more than 50 mice), a single ACP neuron was labelled with GFP in an individual mouse (Fig. 6m-o, Extended Data  Fig. 10). In all three of these clones (two left ACP neurons, one right ACP neuron) the single, labelled ACP neuron projected to both heart and lung, terminating in cholinergic calbindin-positive synapses that surrounded cholinergic ganglion neurons in both organs (Fig. 6m-o,  Extended Data Fig. 10). In all three cases, the ACP clone innervated only the heart and contralateral lung, innervating a variable fraction (3-100%) of the neurons in each targeted cardiac or lung ganglion and totalling 10-30 innervated ganglion neurons in the lung and 12-16 innervated ganglion neurons in the heart (Supplementary Table 7). Thus, single ACP neurons innervate both organs, sending dual projections to heart and lung.

Article
Cholinergic pulmonary ganglia mediate bronchoconstriction 25 , and previous studies have shown that chemical stimulation of rostral Amb with an excitatory amino acid causes bradycardia and a simultaneous increase of about 15% in total lung resistance 26 , indicating bronchoconstriction. To determine whether ACP neurons mediate bronchoconstriction, ACP neurons in the right Amb were optogenetically activated in isoflurane-anaesthetized, mechanically ventilated, paralyzed Calb1 cre mice while respiratory mechanics and ECG were recorded (Fig. 6h). We found that immediately following optogenetic stimulation, the previously described bradycardia (Fig. 4) was accompanied by a 5-12% increase in total lung resistance (Fig. 6i, k). Similar to the cardiac effect, ACP-mediated bronchoconstriction was abrogated by atropine administration (Fig. 6k). By contrast, optogenetic activation of ACV neurons had little effect on total lung resistance (1-2%) (Fig. 6j, k), despite driving a similar reduction (around 50%) in heart rate as ACP neurons (Fig. 6l).
We conclude that ACP neurons, but not ACV neurons innervate and control the lung in addition to the heart, and they mediate these pulmonary effects through collateral projections to cholinergic ganglion neurons along proximal airways.

The dive reflex activates ACP neurons
To identify a physiological function for ACP neurons, we considered cholinergic reflexes involving coordinated cardiopulmonary responses. The mammalian dive reflex is a powerful, evolutionarily ancient reflex that conserves vital oxygen stores while submerged underwater 27 . Upon nasal water immersion, afferent signals to the brainstem rapidly trigger apnea, bradycardia, peripheral vasoconstriction and bronchoconstriction. Amb mediates the dive-induced bradycardia and bronchoconstriction [26][27][28][29] , the two functions that we showed to be under the control of ACP neurons. To determine whether ACP neurons are activated by the dive reflex, we induced this reflex in isoflurane-anaesthetized wild-type mice by exposing them to 10 brief (10 s) nasal immersions in thermoneutral (approximately 30 °C) water over 30 min (Fig. 5e). ECG monitoring confirmed dive reflex activation of bradycardia during nasal immersion (Extended Data Fig. 8e). ACP neuronal activity was then assessed 120 min later by c-Fos immunostaining, as above in the baroreflex experiments. We found robust induction of ACP neurons by nasal immersion (around 70% of ACP neurons were c-Fos-positive versus around 10% in control, non-immersion mice) (Fig. 5f, g, Extended Data Fig. 8c). Only a small fraction of ACV neurons were induced under the same conditions (around 20% of ACV neurons were c-Fos-positive versus around 10% in the control) ( Fig. 5f, h, Extended Data Fig. 8d). Thus, opposite to the baroreceptor reflex, the dive reflex preferentially activates ACP neurons.

Discussion
We have molecularly, anatomically and functionally characterized the parasympathetic brain neurons that control cardiac function in mice, and defined two parallel circuits (Fig. 7). One is the classical cardiovascular control circuit that is activated by increases in blood pressure and mediates the baroreceptor reflex. The central neurons of this circuit comprise approximately 35 neurons in the right and left Amb (around 70 in total) that we designated ACV. They are selectively activated by increases in blood pressure and directly project to and activate cholinergic cardiac ganglion neurons, which slow the SA node rate and AV node conduction velocity to homeostatically maintain blood pressure. The other is a newly identified circuit that controls both cardiac and pulmonary function. It is mediated by a smaller and more rostral population of Amb neurons (approximately 15 neurons in right and left Amb, around 30 in total) that we designated ACP. Similar to ACV neurons, ACP neurons project to and regulate cholinergic cardiac ganglion neurons that slow the SA and AV nodes, although they do this through a distinct subset of cardiac ganglion neurons intermingled with those innervated by ACV. However, the most surprising and important feature of ACP neurons is that they additionally project to and activate cholinergic pulmonary ganglion neurons that drive bronchoconstriction. Remarkably, single ACP neurons innervate both heart and lung, a rare example of a neuron that sends efferent projections to multiple organs. ACP neurons are specifically activated by the dive reflex, and thereby mediate the coordinated reduction in both cardiac and pulmonary function to conserve myocardial oxygen consumption and possibly protect the lungs while redistributing air from conducting to diffusing air spaces during water immersion. The ACV circuit might be deployed during other states requiring a cardiac-specific parasympathetic output 4 , such as expiration (respiratory sinus arrhythmia), detection of noxious stimuli or ischemia in cardiac ventricles (Bezold-Jarisch reflex), non-REM sleep, and various emotional states. The baroreceptor reflex may be the ancestral function of the ACV circuit, given the widespread conservation of this reflex in both terrestrial and aquatic vertebrates 30 . The dive reflex is widely conserved in terrestrial vertebrates 31 , suggesting that ACP neurons arose after ACV neurons, but soon after the transition to life on land. Indeed, this may be the ancestral function of ACP neurons because the dive reflex would have been especially valuable in early terrestrial intermediates moving between sea and land. A priority will be to identify other physiological states that engage the ACV and ACP cell types in addition to the canonical functions described here.
The identification of ACV and ACP could inform novel therapeutic approaches. Whereas the cardiac sympathetic system can be therapeutically targeted by β-blockers with relative cardioselectivity, the cardiac parasympathetic system has resisted such targeting owing to the challenges of developing muscarinic subtype-specific compounds 32 . Impaired parasympathetic signalling to the heart is independently associated with sudden cardiac death in patients with a history of myocardial infarction 7 , patients with heart failure 8 and in otherwise healthy adults 9 . Hyperactivity or impairment in this pathway is also thought to drive baroreceptor reflex dysfunction in certain neurological disorders 21 . Thus, the molecular identification of ACV neurons and their distinct receptor profile could enable novel therapeutics that specifically target the pathological baroreceptor reflex circuit in cardiovascular and neurological disease. Another priority is to investigate whether the ACP circuit contributes to reflex bronchoconstriction 33 , pathological cardiorespiratory reflexes such as the Cushing reflex 24 , or to asthma, a disease in which airway and cardiac vagal hyperreactivity can co-occur 34 , and to target ACP neurons accordingly.

Article
Finally, the discovery of the ACV and ACP circuits provides a cellular-resolution glimpse into how the central autonomic nervous system is organized to coordinate intra-organ as well as inter-organ physiology. The approach described here could be applied to molecularly and functionally identify the central parasympathetic and sympathetic neurons and circuits that control all organs, an effort that will be aided by the general parasympathetic gene signature described here. Elucidating these circuits could allow precision control of specific organs, reflexes or combinations of organs in health and disease.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04760-8.

Retrograde labelling of Amb Laryngeal neurons
CTB-488 or CTB-555 was injected bilaterally into the cricothyroid muscles of P1 wild type mice. Mice were deeply anaesthetized by hypothermia and analgesia was administered using buprenorphine (0.025-0.05 mg kg −1 , subcutaneously). A ventral incision was made in the skin overlying the larynx and the fat pads and sternohyoid muscles were dissected away to reveal the cricothyroid muscles. A fine glass micropipette filled with CTB conjugate was inserted into the muscle and 10 nl of solution was injected over 1 min to fill each cricothyroid muscle. The skin was sealed with VetBond tissue adhesive (3M). Mice recovered on a heating pad until the righting reflex was regained and then returned to the litter. 1-3 days after injection, ~28 neurons were labelled in the Amb external formation, exclusively in the rostral compartment below the compact formation. This distribution in the mouse Amb is similar to the observed distribution in rats 38 .
Isolation of Amb Cardiac and Amb Laryngeal neurons for scRNA-seq P1 wild type C57BL/6NCrl mice (n = 28 mice) were injected with 1 μl CTB-488 or CTB-555 into the pericardial space, and 16 other mice of the same age and genotype were injected with 10 nl CTB-488 or CTB-555 into the cricothyroid muscle. P2-P4 mice were anaesthetized with saturating vapours of isoflurane and rapidly decapitated. The brain was immediately placed in cold (4 °C) physiological artificial cerebrospinal fluid 39 (aCSF) containing (in mM): 124 NaCl, 3 KCl, 1.5 CaCl 2 , 1 MgSO 4 , 25 NaHCO 3 , 0.5 NaH 2 PO 4 and 30 d-glucose, equilibrated with 95% O 2 and 5% CO 2 (pH 7.4). Brains were embedded in 4% low melting point agarose (Invitrogen 16520050) and 220 μm sagittal slices were cut on a vibratome (Leica Biosystems) while the aCSF bath was maintained at 4 °C with continuous equilibration. Slices were transferred to an electrophysiology rig equipped with an upright fluorescent microscope (Leica) and maintained for up to 4 h in the same aCSF solution at 22 °C (neonatal medullary slices of the same region remain healthy for at least 4 h, as measured by maintenance of a physiological breathing rhythm 40 ). Borosilicate capillaries (Sutter Instrument B150-86-10) pulled to a large bore diameter (1/4-1/3 soma diameter) were lowered onto fluorescent Amb neurons near the surface of the slice and a seal was obtained on a labelled neuron. Gentle suction was applied by syringe to aspirate the soma out of the brain slice. Immediately after extraction and while still attached to the pipette, each aspirated cell was inspected under alternating fluorescent and brightfield illumination using the 40× objective to ensure that no other cells or visible debris were attached to the neuron or the pipette. The neuron was then immediately expelled into 4 μl lysis buffer 41 , snap frozen and stored at −80 °C until cDNA synthesis.
scRNA-seq cDNA synthesis was performed using the plate-based SmartSeq2 protocol 41 , with 20 cycles of amplification during the PCR step. Amplified cDNA was purified twice with 0.7× AMPure beads (Fisher A63881). cDNA quality and concentration were then assessed by capillary electrophoresis on a Fragment Analyzer (AATI) before sequencing library preparation. Illumina sequencing libraries for cDNA from single cells were prepared as previously described 41 . In brief, cDNA libraries were prepared using the Nextera XT Library Sample Preparation kit (Illumina, FC-131-1096). Nextera tagmentation DNA buffer (Illumina) and Tn5 enzyme (Illumina) were added, and the sample was incubated at 55 °C for 10 min. The reaction was neutralized by adding Neutralize Tagment Buffer (Illumina) and centrifuging at room temperature at 3,220g for 5 min. Samples were then indexed via PCR by adding i5 indexing primer, i7 indexing primer, and Nextera NPM mix (Illumina). Following cDNA sequencing library preparation, wells of each library plate were pooled using a Mosquito liquid handler (TTP Labtech), then purified twice using 0.7× AMPure beads. Libraries were sequenced on a NextSeq 500 (Illumina) using 75 bp paired-end sequencing.
scRNA-seq data analysis cDNA sequencing reads from the the obtained Amb Cardiac (191) and Amb Laryngeal (77) neurons were pruned for low nucleotide quality scores and adapter sequences using Skewer 42 (version 0.2.2), and aligned to the mm10 genome using STAR 43 (version 2.6.1d) in two-pass mapping mode, in which the first pass identifies novel splice junctions and the second pass aligns reads after rebuilding the genome index with the novel junctions. Neurons with less than 500,000 aligned reads or more than 1,800 counts of the glial gene Apoe, indicating significant glial mRNA contamination, were removed from the analysis before clustering, resulting in 151 Amb Cardiac neurons and 52 Amb Laryngeal neurons with high quality transcriptomes (mean: 1.9 million reads and 7,761 genes detected per cell) (Extended Data Fig. 2a-c).
Using Seurat v3.1.4 (https://satijalab.org/seurat/), read counts per gene were normalized across cells, scaled per 10,000 and converted to log scale using the NormalizeData function. These values were converted to z-scores using the ScaleData command and highly variable genes were selected with the FindVariableFeatures function (dispersion function: LogVMR). Principal components were calculated for these selected genes using the RunPCA command. Clusters of cells with similar expression profiles were detected using the Louvain method for community detection including only biologically meaningful principal components (see below) to construct the shared nearest-neighbour map, as implemented in the FindClusters function (resolution: 0.4). Differentially expressed genes for each defined cluster were identified using the FindMarkers command in Seurat using the Wilcoxon rank sum test.
Although all neurons that passed quality control expressed high levels of neuronal-specific genes (Extended Data Fig. 2d), they also contained low levels of glial transcripts (Extended Data Fig. 2e) presumably from attached glial mRNA, processes, or cells that were not visible under a 40x objective. To exclude such non-neuronal genes from the clustering, principal components containing non-neuronal genes were excluded. The three identified neuronal clusters had similarly low levels of glial gene expression (Extended Data Fig. 2e), indicating that these genes did not drive clustering. In addition, all differentially expressed genes between neuronal types (Supplementary Tables 1-6) examined in the Allen Brain Atlas or localized by in situ labelling in this paper were expressed in Amb neurons and not glia, further demonstrating that the Amb Cardiac , Amb Laryngeal , ACP and ACV cell types were separated based on Amb neuronal genes. Although scRNA-seq was performed at an early postnatal time point at which cardiac parasympathetic innervation is occurring 44 , ACP and ACV marker expression was conserved between neonatal and adult time points as assayed by immunostaining (Extended Data Fig. 4) and the Allen Brain Atlas (Supplementary Table 6); however, there may be subtle changes in gene expression not detected by these methods.
Adult mice (> 6 weeks of age) were anaesthetized with 3% isoflurane (for induction, and 1-2% for maintenance) for AAV injections. Anesthetized mice were placed in a stereotactic instrument (David Kopf Instruments, Model 940), with body temperature maintained at 37 °C using a feedback-controlled heating pad (Physitemp, TCAT-2LV). Mice were pre-treated with analgesic (carprofen 5 mg kg −1 subcutaneously and buprenorphine SR 0.5-1.0 mg kg −1 subcutaneously). To target ACP neurons in the rostral Amb, 400 nl AAV vector was injected into Calb1 cre mice at the following stereotactic coordinates: 2.4 mm caudal to lambda, ± 1.33 mm lateral to lambda, 4.9 mm ventral to the brain surface. To target ACV neurons in the caudal Amb, 500 nl AAV vector was injected into Ghsr cre mice at 2.95 mm caudal to lambda, ± 1.3 mm lateral to lambda, 4.8 mm ventral to the brain surface. Immediately following AAV injection, a fibre-optic cannula (Thorlabs CFM12L05-10) was implanted 350 μm (ACP) or 500 μm (ACV) above the injection site and secured to the skull with dental cement (Parkell C&B Metabond). Mice in which the AAV transgene was not expressed in Amb neurons were excluded from analysis (with the exception of Extended Data Fig. 7e-h, where we specifically characterized the non-Amb neurons). Mice recovered for 2-4 weeks before optogenetic and projection mapping experiments.

Optogenetic stimulation with ECG
Mice injected with AAV6-CAG-DIO-bReaChES-TS-eYFP and recovered as above were anaesthetized with isoflurane (3% induction, 1-2% maintenance) and body temperature maintained at 37 °C. Single lead ECG was recorded using needle electrodes (ADInstruments MLA1213), an ADInstruments Octal Bio Amp, and an ADInstruments PowerLab data acquisition system. Respiration was simultaneously recorded using a spirometer (ADInstruments). The implanted fibre-optic cannula was connected using a fibre-optic cable (Thorlabs M77L01) to a 577 nm laser (CNI Laser). Laser light was delivered using the following parameters: 10-15 mW power from the fibre tip, 10 ms pulse width, 40 Hz. To assess the role of muscarinic receptors, atropine (Sigma A0132, dissolved at 50 mg ml −1 in ethanol, then prepared as a 0.5 mg ml −1 working solution in PBS) was administered (10 mg kg −1 intraperitoneally) and optogenetic stimulation was repeated 20 min later. To calculate the percent change in heart rate, the minimum sinus rate (60 divided by the P-P interval in seconds) during stimulation was subtracted from the sinus rate immediately before stimulation and divided by the pre-stimulation sinus rate. To calculate the percent change in P-R interval, the P-R interval immediately before stimulation was subtracted from the maximum P-R interval during stimulation and divided by the pre-stimulation P-R interval.

Optogenetic stimulation with respiratory mechanics and ECG
Mice injected with AAV6-CAG-DIO-bReaChES-TS-eYFP and recovered for 3-4 weeks as above were anaesthetized with 3% isoflurane, tracheostomized with an 18G cannula, and attached to the Flexivent (SCIREQ) as previously described 45 . Core temperature was maintained at 36-37 °C with a heating pad, with the Flexivent delivering 1-2% isoflurane in pure oxygen. The mice were ventilated at 10 ml kg −1 at 150 breaths per min with a positive end-expiratory pressure (PEEP) set at 3 cm H 2 O. ECG was recorded as described above. Before the evaluation of respiratory mechanics, mice were paralyzed with vecuronium bromide (Sigma 76904, 0.1-0.2 mg kg −1 intraperitoneally) to stabilize the airways and eliminate breathing efforts. Vecuronium was used owing to its lack of vagolytic properties. The compliance, elastance, and overall resistance of the respiratory system (Rrs) were measured every 3 s with the Snapshot-150 manoeuvre before, after, and during optogenetic stimulation (10 s, 20 Hz), with the entire trial lasting ~50 s. The trial was repeated 20 min after atropine (10 mg kg −1 intraperitoneally) delivery. To calculate the relative change in Rrs, the difference between the maximum Rrs value obtained during optogenetic stimulation and the average baseline Rrs was divided by the average baseline Rrs, which was defined as the average of the Rrs values measured in the 10 s prior to optogenetic stimulation.

Induction of baroreceptor reflex and dive reflex
Wild-type juvenile (postnatal day 21) mice were used for all baroreceptor and dive reflex experiments because the dive reflex is more pronounced earlier in life 27 . To induce the baroreceptor reflex, awake mice in their home cage were either injected with phenylephrine (10 mg kg −1 intraperitoneally) (Tocris 2838, dissolved in PBS) to induce peripheral vasoconstriction, or with PBS vehicle as control. Mice remained in their home cages and were transcardially perfused 150 min following injection, to allow time for pressor effects and baroreflex activation (~20 min) 46 and c-Fos protein expression (~120 min) 47 , and brains were then fixed and immunostained for neuronal activity markers as described below. In control studies, we found that ACP neurons were not activated by phenylephrine injection 30, 60, 90 and 120 min before perfusion (Extended Data Fig. 9), the same result obtained at the 150 min time point (Fig. 5), consistent with prior studies showing rostral Amb neurons are not barosensitive 48 .
To induce the dive reflex, mice were first anaesthetized with isoflurane (3% induction, 2% maintenance) and maintained on a 37 °C heating pad while recording single-lead ECG, and then subjected to nasal immersion. Anaesthetized mice in the dive condition received a total of 10 nasal immersions in thermoneutral (~30 °C) water (every 3 min over a 30-min period). Immersions lasted between 5-10 s and were terminated when stable reflex bradycardia was observed on the ECG recording. Similar water temperature and dive immersion times have been used for dive reflex studies in awake rats 49 and mice 50 . Isoflurane control mice were anaesthetized and maintained identically without nasal immersions. All mice remained under anaesthesia continuously until transcardial perfusion 120 min after completion of the 30 min dive period, and brains were then fixed and immunostained for neuronal activity markers as described below.
Classical neuroanatomical and ablation studies localized the key neurons controlling the baroreceptor and dive reflexes to Amb Cardiac neurons 20,27 and excluded contributions from other brainstem nuclei, consistent with our optogenetic stimulation, activity mapping, and neuroanatomical data. However, despite optimization of the AAV vector and promoter (see Methods, above, AAV cloning, injections, and fibre-optic implantations) we were unable to achieve the high transduction efficiencies necessary to demonstrate a requirement for ACV and ACP in the baroreceptor and dive reflexes and formally exclude a contribution from the non-activated Amb Cardiac cell type, since activation of small subsets of ACP or ACV (20-40% unilaterally) is sufficient to reproduce reflex bradycardia (Fig. 4, Extended Data Fig. 5d), whereas our AAV transduction efficiencies for ACP and ACV were 60% and 20% (Extended Data Fig. 5d).

Brain and heart immunostaining
Mice were euthanized with CO 2 , transcardially perfused with 4% paraformaldehyde (PFA), and tissues were post-fixed in 4% PFA overnight at 4 °C. Brains and hearts were cryoprotected in 30% sucrose at 4 °C overnight. Cryoprotected tissue was embedded in optimal cutting temperature (OCT) compound and sectioned at 25 μm (Leica CM3050S cryostat). Sections were permeabilized in PBS + 0.3% Triton X-100, blocked for 1 h in block buffer (PBS + 0.3% Triton + 10% normal donkey serum), and incubated with primary antibodies in block buffer at 4 °C overnight. Slides were washed three times, incubated in secondary antibodies in block buffer for 1 h at room temperature, washed three times and a coverslip was applied with Prolong Gold antifade reagent. Primary antibodies used in this study were: rabbit anti-calbindin (Swant CB38, 1:2,000), chicken anti-calbindin (Novus Biologicals NBP2-50028, 1:1,000), goat anti-BchE (R&D Systems AF9024, 1:100), chicken anti-GFP (Aves Labs GFP-1010, 1:1,000), goat anti-VAChT (Millipore ABN100, 1:500), rabbit anti-c-Fos (Synaptic Systems 226 003, 1:5,000). Species-specific donkey secondary antibodies conjugated to Alexa Fluor 488, 568 or 647 were obtained from Life Technologies or Jackson ImmunoResearch and used at a 1:500 dilution. Stained neurons were counted manually from z-stacks acquired on a Zeiss LSM 780 confocal microscope. The presence or absence of calbindin was used to differentiate ACP and ACV neurons (punctate background calbindin staining outside of ACP and ACV neurons is probably calbindin + fibres coursing through medulla). ACP neurons were defined histologically as calbindin + cell bodies with low BChE expression (distinguishing them from calbindin + interneurons in the pre-Bӧtzinger complex, which express no BChE). ACV neurons were defined as BChE + calbindin − cell bodies. For smFISH, sections were processed with a RNAscope Multiplex Fluorescent Assay v2 kit (Advanced Cell Diagnostics) according to the manufacturer's instructions with the probe Mm-Ghsr-C2 (426141-C2) and immunostaining was performed afterwards as above. For c-Fos experiments, control and stimulus samples were processed and imaged together in the same immunostaining experiment to minimize variability.
For cardiac immunostaining, owing to anatomical variability in ganglia locations, ganglia were divided into GPs in four quadrants to compare innervation patterns across animals. To estimate the proportion of GP neurons innervated by a given cell type (Extended Data Fig. 6a), the observed proportion of innervated cells (innervated neurons/ total GP neurons) was divided by the labelling efficiency of the cell type (eYFP + cell bodies/total cell bodies of cell type on given side). For clonal labelling analysis of ACP terminals in the heart, every fourth section of cardiac ganglia was stained, and ganglion neuron counts were multiplied by four.

Clonal labelling of ACP neurons
For clonal labelling of ACP neurons, multiple AAV-based approaches were tested in parallel to reduce the efficiency of ACP neuron labelling and increase the likelihood of labelling just a single ACP neuron. To determine whether a single ACP neuron had been labelled, one month after AAV injection sagittal sections (25 μm) of the entire Amb including >150 μm of laterally surrounding tissue were immunostained for GFP and robust ACP and ACV markers (calbindin and BChE). Each tissue section (25 μm, ~15 sections per Amb) was optically sectioned (~4 μm) and systematically scanned for labelled ACP cell bodies (typical diameter: ~20 μm). We excluded any mouse (n = 10 mice excluded) in which any brainstem section contained a fold, damage, or inadequate staining, because a labelled cell in the affected area could have been missed. All ACP neurons were counted by examining each optical section from all tissue sections, and only mice in which all Amb sections were intact (n = 48 mice) and only a single ACP neuron was labelled with GFP (n = 3 of 48 mice) were included in the analysis. For clone 1 (Fig. 6m-o), 200 nl of AAV6-CAG-FLEX-ArchT-GFP (UNC Vector Core, 5 × 10 12 viral genomes per ml) (denoted AAV-FLEX-GFP) was injected into the rostral Amb of Calb1 2a-dgcre mice, which were then injected with 200 mg kg −1 trimethoprim (intraperitoenally) five days later and sacrificed one month after AAV injection. Of the 21 injected mice, only one contained a single labelled ACP neuron (in left Amb). Clone 2 (Extended Data Fig. 10a-c) was identified in a rare case in which injection of AAV6-CAG-DIO-bReaChES-TS-eYFP (denoted AAV-DIO-eYFP) in a Ghsr cre mouse labelled a single ACP neuron (in right Amb). For clone 3 (Extended Data Fig. 10d-f), 50 nl of AAV6-CAG-FLEX-ArchT-GFP was injected into the rostral Amb of Calb1 cre mice. Of the 37 injected mice, only one contained a single labelled ACP neuron (in left Amb).

Statistics and data collection
All results are presented as mean ± s.d. with all data points displayed. All statistical analyses were performed with GraphPad Prism. All statistical Extended Data Fig. 8 | c-Fos negative control studies and heart rate response to dive reflex. a, Immunostaining of ACP neurons in rostral Amb following vehicle injection (see Fig. 5). Note ACP neurons (calbindin + ; white arrowheads) are c-Fos negative. b, Immunostaining of ACV neurons in caudal Amb following vehicle injection. Note ACV neurons (BChE + ; white arrowheads) are c-Fos negative. c, Immunostaining of ACP neurons in rostral Amb following isoflurane anesthesia without nasal immersion. Note ACP neurons (calbindin + ; white arrowheads) are c-Fos negative. d, Immunostaining of ACV neurons in caudal Amb following isoflurane anesthesia without nasal immersion. Note ACV neurons (BChE + , white arrowheads) are c-Fos negative. Bars, 20 μm. e, Example heart rate trace recorded by ECG during dive reflex activation for Fig. 5 experiments. Isoflurane-anesthetized mouse underwent nasal immersion (arrow, start of dive) for 10 s. Bradycardia and AV block were observed during nasal immersion, and heart rate returned to baseline following cessation of immersion.