# Vagal Sensory Neurons: The Body Reports to the Brain

> To learn what a vagal sensory population does, ask three separate questions: what stimulus the cells answer, what channel turns that stimulus into current, and what output changes when you drive or remove them. Light on one set of vagal sensory axons stops an anaesthetized mouse breathing, and light on another set in the same nerve makes it breathe fast and shallow. Evidence for one does not by itself answer the others, and a genetic label is a handle on a population rather than an answer about any of them. You work the questions through the airway, the lung, the artery and the gut, name what vagal afferents report from each organ, separate airway protection from lung inflation, and design experiments that tell volume from nutrient content. Going deeper holds the ganglion origins, the atlas counts, the trial evidence and two worked problems, including what the arterial baroreflex result does and does not settle.

Course: Sensory Systems (Proprioception and interoception). Level: advanced university and graduate. About 29 minutes.
Lesson page (link this one when citing): https://stemhop.com/lessons/vagal-sensory-neurons/

Builds on: Piezo Channels and the Physics of Force · Sensory Neurons by Their Genes: the Atlas Era · Five Ways to Turn a Stimulus into a Current

## What you'll learn

- State the three questions that establish what a vagal sensory population does, explain why evidence for one does not by itself answer the others, and say why a genetically labelled population is a handle rather than an answer about a stimulus, a channel or an output.
- Describe the P2ry1, Npy2r, GLP1R, GPR65 and Oxtr populations, and say what activation, ablation and receptor-deletion experiments each do and do not show.
- Separate airway protection from lung inflation, and explain what Piezo2 loss does to the adult inflation reflex and why the arterial baroreflex needed Piezo1 and Piezo2 deleted together.
- Explain how the brainstem processes vagal input into an organ map, and read an organ-selectivity count from an imaging experiment with the limits of its denominator.
- In Going deeper, place vagal cell bodies in the nodose and jugular ganglia with their developmental origins, and use an explicitly chosen saturating model to show how a partial loss of sensory input could stay inside the published error bars, and what about the model that conclusion depends on.

**Light on one set of vagal sensory axons stops an anaesthetized mouse breathing. Light on another set, in the same nerve, makes it breathe fast and shallow. Two labelled subsets of one nerve, opposite breathing patterns — so what does each set report, and how would you know?**

Opposite outputs from two labelled subsets demonstrate functional diversity. What naturally activates each subset is a separate measurement, and the harder question is what it takes to establish what any one of those lines does.

Three words first, because the whole lesson leans on them. An **afferent** is a sensory fibre that carries signals from the body toward the central nervous system. **Interoception** is the sense of the body’s own internal state: lung volume, blood pressure, how full the gut is. And a **genetic label**, usually a Cre line, is a mouse in which the promoter of one gene drives [Cre recombinase](https://stemhop.com/methods/cre-lox/), which then switches on a tool — a light-gated channel, a cell-killing toxin, a calcium indicator — in the cells that expressed that gene. The tool has to be supplied separately, as a floxed allele or an injected virus, and which cells end up labelled depends on that choice as much as on the promoter.

Two more, because every experiment below is one of them. Driving a population and seeing an output appear is a test of **sufficiency**: these cells can produce it. Removing or silencing the population and seeing the output fail is a test of **necessity**: in that assay, at that age, with that driver.

> **The one idea to carry through this lesson.** Ask three separate questions of any vagal population, and keep their answers apart: **(1) what stimulus do the cells answer?**, **(2) what channel turns that stimulus into current?**, and **(3) what output changes when you drive or remove the cells?** Each needs its own evidence, and evidence for one does not by itself answer the others. A population can be useful and well studied while some of the three stay open, and a single preparation can sometimes address more than one. A genetic label is a handle on a population: its name may well identify a receptor or channel those cells express, but labelling alone establishes neither the natural stimulus, nor the protein’s causal role in transduction, nor the output. And every answer holds only for the organ, age, assay and anaesthetic state in which it was measured.

One scope note: the frame is not a claim that the vagus is a set of fixed labelled lines arriving in the brain unchanged. The brainstem mixes and sharpens what arrives, as the imaging later shows.

This is mouse work unless a sentence says human, and it starts where [hypothalamus and autonomic control](https://stemhop.com/lessons/hypothalamus-autonomic/) and [sensory coding](https://stemhop.com/lessons/sensory-coding/) stop: at the peripheral afferent.

### One worked case, because it is the cleanest

GPR65-labelled vagal afferents answer food arriving in the mouse intestine. That answers question 1. Driving them with light suppresses gastric contractions in anaesthetized mice. That answers question 3, for one output. The evidence summarized here does not establish the complete molecular route from intestinal food to current in these afferents, so question 2 is only partly answered — which does not stop this being a well-characterized population.

**Predict.** A population that reports food in the intestine, and that can stop the stomach contracting, looks like a satiety line. In awake mice, what should driving those same afferents do to how much the animal eats?

> **Check your prediction.** In the assays that Bai and colleagues (Cell, 2019) tested, it did not measurably change feeding, while activating intestinal stretch afferents labelled by Oxtr suppressed it strongly. One population, three answers: yes to the stimulus, yes to one output, no to another. Responding to a nutrient, controlling the stomach and ending a meal are three different measurements, and nothing requires one population to pass all three.

The rest of the lesson works these questions through the organs: the airway, the lung, the artery and the gut.

## One nerve, many labelled lines: genetic access to vagal afferents

Vagal sensory cell bodies sit in two ganglia at the base of the skull, the nodose and the jugular, and single-cell atlases of them sort the neurons into many types. A cluster is a hypothesis about function until genetic access, physiology or behaviour ties it to one. One note on notation, since this lesson turns on the difference between a label and a molecule: *P2ry1* in lower case is the mouse gene and the Cre line named after it, while P2RY1 in capitals is the protein.

Chang and colleagues (Cell, 2015) used a Cre driver to put the light-gated channel channelrhodopsin-2 into P2ry1-labelled or Npy2r-labelled vagal neurons, exposed the vagus in mice under deep isoflurane, and shone light on it. P2ry1 activation suppressed breathing. Npy2r activation produced rapid shallow breathing. This is sufficiency in anaesthetized mice: it says what the cells can do when driven, not what drives them in life.

Light-evoked nerve recordings then measured a different property from the breathing assay, in the same light-driven axons. The P2ry1 axons conducted largely as fast **A fibres**, which are myelinated and conduct quickly, and the Npy2r axons largely as slow **C fibres**, which are unmyelinated and conduct slowly. Two populations with different predominant conduction classes are physiologically different, which is independent support that these are different cells. It is not proof that the two sets are wholly separate, and it says nothing about what stimulus drives either in life: “largely” is the published word, and it is the right one.

*Figure (AI-generated illustration).* Chang and colleagues put channelrhodopsin-2 into P2ry1-labelled or Npy2r-labelled vagal neurons in mice and shone light on the exposed vagus under deep isoflurane. P2ry1 activation suppressed breathing and Npy2r activation produced rapid shallow breathing. This shows what the cells can do when driven, not what drives them in life. Image: two matching panels of an exposed nerve under a thin light fibre. On the left, gold P2ry1-labelled axons, largely fast A fibres, are lit and breathing is suppressed. On the right, gold Npy2r-labelled axons, largely slow C fibres, are lit and breathing is rapid and shallow. An inset shows a light-gated channel block in a membrane band.

Now question 1 in the gut. Williams and colleagues (Cell, 2016) imaged labelled vagal neurons in mice with the gut still connected. GLP1R-labelled neurons responded to stretch of the stomach and intestine. GPR65-labelled neurons innervated villi near the junction of stomach and small intestine; their food response and the gastric effect are the opening case. The authors suggested that most villous nutrient detection in their preparation was independent of GLP1R.

Three years later the same handle produced what looks like a contradiction, and the resolution is worth stating before the detail: the same label turned out to mark different endings than the imaging had implied. Bai and colleagues (Cell, 2019) labelled the line with a Cre-dependent virus delivered in adulthood, and found Glp1r-positive stretch endings in the stomach rather than the intestine, while the same labelling did reach intestinal mucosal endings. Williams’s preparation used a reporter carried from development, and the cells it labelled responded to both gastric and intestinal distension. So one paper reports intestinal stretch responses and the other finds no intestinal stretch endings, each faithfully.

**Predict.** Which of the three questions is each result answering, and which claim should a student carry forward?

> **Check your prediction.** The 2016 imaging answers question 1: it shows what the cells *labelled by that line* responded to in that preparation. The 2019 anatomy answers none of the three. It is a fact about the handle — which endings the adult labelling reaches. Carry forward that the labelled cells include gastric stretch afferents, and treat the intestinal stretch finding as reporter-dependent and unresolved for this line. Bai and colleagues suggested broader developmental Glp1r expression as the explanation, though the comparison does not establish it uniquely (Going deeper). GLP1R and GPR65 are labels, and neither paper shows that the receptor named is the molecule that senses stretch or food.

## Two different populations, one outcome: breathing stops

**Film transcript: Two Ways to Stop Breathing**

### Film transcript: Two Ways to Stop Breathing

Two findings both end with a mouse not breathing. Are they the same line, used twice? Start with the opposite results. Light on P2ry1 afferents suppressed breathing in anaesthetized mice. Light on Npy2r afferents produced rapid shallow breathing. Opposite outputs, in the same nerve.

This is sufficiency. It says what each population can do when driven, not what drives it in life. Nerve recordings added an independent check. One population was mostly fast, the other mostly slow. A fibres are myelinated and conduct quickly. C fibres are unmyelinated and conduct slowly. Conduction speed was measured apart from the breathing assay. Most of one population conducted fast, most of the other slow. That is a physiological difference, not proof of complete separation.

Now the two routes to one output. First, inflation. When the lungs fill, further inspiration is cut short. This is the Hering-Breuer inflation reflex, described more than a century ago. Cut both vagi and the reflex goes. Second, protection. Water or acid at the larynx triggers protective reflexes that keep the airway clear. Driving P2ry1 afferents with light recruited the whole program, not apnoea alone.

Ablating those cells impaired swallowing to laryngeal water and acid. That is necessity, in that assay. So these cells can recruit airway defence, and one tested piece of it needs them. If they were also the lung's stretch detector, what would inflation imaging show?

If one population did both jobs, the inflation responders should sit inside the labelled set. They did not. Most stretch-responsive neurons were PIEZO2-positive and P2RY1-negative. That imaging was done with central connections severed, under isoflurane, with the airway inflated in the living animal. So two populations, two challenges, one shared output. The table separates them.

The takeaway is general. An output alone does not identify which pathway produced it. Measure stimulus and output separately.

From the film

- Light on P2ry1 afferents suppressed breathing and light on Npy2r afferents produced rapid shallow breathing in anaesthetized mice; this shows what each population can do when driven, not what drives it in life.
- Airway protection and lung inflation are separate routes to a similar output, and most stretch-responsive neurons were Piezo2 positive and P2ry1 negative.
- An output alone does not identify which pathway produced it, so measure stimulus and output separately.

Here is the first place the three questions pay off. Two quite different findings both end with a mouse not breathing, and anyone holding only question 3, the output, will merge them.

**Inflation.** When the lungs fill, further inspiration is cut short. This is the **Hering-Breuer inflation reflex**, described more than a century ago and the classic reason to think the vagus reports lung volume: cut both vagi and the reflex goes. It is a volume report, and the stimulus is stretch.

**Protection.** Water or acid at the larynx triggers protective reflexes that keep the airway clear. The challenge here is an irritant, not stretch. That is a fact about the reflex, not yet about any labelled cell: which afferents sense the irritant is a separate measurement, not made below.

Both end with inspiration halted. So: one population used two ways, or two populations?

Prescott and colleagues (Cell, 2020) drove P2ry1-labelled afferents with light and recruited a coordinated defence rather than one response: **apnoea**, a pause in breathing, together with closure of the vocal folds, swallowing and expiratory reflexes. That is sufficiency. They then ablated the P2ry1 cells genetically, and swallowing evoked by water and acid at the larynx was impaired. That is necessity, in that assay, at that age, with that driver. Anatomy fits an airway role: in the Chang study, labelled P2ry1 terminals contacted pulmonary neuroepithelial bodies in fixed adult lungs, which are clusters of secretory neuroendocrine cells within the airway epithelium.

So activating these cells can recruit airway defence, and they are needed for one tested piece of it — not the same as owning every protective response, since the loss of the ATP receptors P2X2 and P2X3 impaired swallowing too (Going deeper). Now ask about inflation.

**Predict.** If these same cells were also the lung’s stretch detectors, what should a ganglion imaging experiment show when you inflate the airway — which cells light up, and what label should they carry?

> **Check your prediction.** If one population did both jobs, the inflation-responsive neurons should be the P2RY1-positive ones. Prescott and colleagues imaged the ganglion with central connections severed, under isoflurane, and inflated the airway in vivo. Most stretch-responsive neurons were **PIEZO2-positive and P2RY1-negative** — the other population. The merge fails.

|  | Protection afferents | Stretch afferents |
| --- | --- | --- |
| Challenge used to evoke the reflex | laryngeal water and acid; P2ry1-cell ablation tests their contribution to the reflex, not direct water or acid transduction | lung inflation; inflation-responsive afferents were imaged |
| Handle | P2ry1 | PIEZO2-positive, P2RY1-negative |
| Output when driven (question 3) | apnoea plus vocal-fold closure, swallowing, expiratory reflexes | apnoea, from driving the broader Piezo2-labelled population — a different experiment from the imaging |
| Cell-ablation result | P2ry1-cell ablation impairs water- and acid-evoked swallowing | no matched ablation of this exact PIEZO2-positive, P2RY1-negative subset is described here |
| Channel-loss result | P2X2/P2X3 double loss impairs some laryngeal swallowing responses; this is not a P2ry1-cell-specific deletion (Going deeper) | adult sensory-neuron Piezo2 deletion impairs the inflation reflex (next section) |

Read across the table and the general point falls out: the same output can be reached from two different stimuli through two different populations, so **an output alone does not identify which pathway produced it, or what naturally activates it.** This is one clear example of why questions 1 and 3 need separate evidence. Note also that the last two rows are two different interventions — removing cells, and removing a channel from a much broader set of cells — so they are not two readings of one experiment.

## What transduces force in the lung and the artery

**Film transcript: Which Channel, and What the Test Can Prove**

### Film transcript: Which Channel, and What the Test Can Prove

Stretch has to become electrical current somewhere. Which channel does that, and how would a deletion show it? This is the second question. Deleting a channel tests whether the reflex needs it. Showing that it converts force to current needs recordings. Force can open a mechanosensitive channel. Inward current depolarizes the terminal, and if it reaches threshold the afferent fires. Deleting Piezo2 in adult mouse sensory neurons impaired the lung inflation reflex. And driving Piezo2-positive vagal neurons with light caused apnoea. That uses Piezo2 as a handle. The artery was harder. The baroreflex is the reflex by which a rise in blood pressure slows the heart.

Channels were deleted from nodose and petrosal neurons. Blood pressure was raised with a drug in anaesthetized mice. Deleting Piezo one alone: no statistically detectable difference against littermate controls. Deleting Piezo two alone: no statistically detectable difference either.

Deleting both essentially abolished the drug-evoked baroreflex. The magnitude of its sensitivity fell from about five to under one beat per minute per millimetre of mercury. in conscious double knockouts, blood pressure in the active period was higher than in controls. The tempting reading is that every baroreceptor neuron carries both channels. Predict. Does that follow from these three rows?

It does not follow. Either picture could fit, if the input that remains can hold the reflex up. One possible explanation is a nonlinear relation between sensory input and the reflex. This curve is a hypothesis, not a measurement. On this chosen curve, halving the input costs one point two three. An illustrative cutoff is one point four one. Neither number tells you what the real experiment could detect. Losing all of it goes to zero on the curve, and the measured drop is easy to see.

So the experiment shows a collective channel requirement. It does not measure summed current, or prove coexpression inside a cell. Separating them needs identified baroreceptors. Measure channel expression and mechanosensitive current cell by cell, under controlled mechanical stimulation. The channels are required collectively in the targeted population. The result does not show that each neuron carries both.

From the film

- Deleting Piezo2 in adult mouse sensory neurons impaired the lung inflation reflex.
- For the baroreflex, deleting Piezo1 or Piezo2 alone showed no statistically detectable difference, while deleting both essentially abolished the drug-evoked reflex.
- That shows the channels are required collectively in the targeted population, not that each neuron carries both.

Question 2 asks what turns stretch into current, and deleting a channel is how it is usually approached. Be careful about what a deletion buys: losing a reflex when a channel is gone shows the channel is *required* in the cells targeted. Showing that the same protein is the thing that converts force into current also takes recordings of mechanosensitive current and evidence that the channel is itself gated by force.

Nonomura and colleagues (Nature, 2017) deleted Piezo2 conditionally in sensory neurons of adult mice, which impaired the inflation reflex. Driving Piezo2-positive vagal neurons with light caused apnoea, which is a sufficiency test on a labelled population rather than a statement about the pore. Newborn lethality after earlier or broader Piezo2 deletion is a separate experiment (Going deeper). For the pore-level evidence that Piezo channels are themselves mechanically gated, see [Piezo channels and the physics of force](https://stemhop.com/lessons/mechanotransduction-channels/).

Arteries add a twist, and the reflex needs naming first. The **baroreflex** is the reflex by which a rise in blood pressure slows the heart; its sensitivity is the change in heart rate per unit rise in pressure, in beats per minute per mmHg. Zeng and colleagues (Science, 2018) used Phox2b-Cre to delete channels from the nodose ganglion and the neighbouring petrosal ganglion, which belongs to the glossopharyngeal nerve, and raised blood pressure with a drug in anaesthetized mice while measuring heart rate against pressure. Deleting Piezo1 alone produced no detectable change against littermate controls. Deleting Piezo2 alone produced no detectable change either. Deleting both abolished the baroreflex, and in conscious double knockouts carrying implanted pressure sensors that report continuously, active-period blood pressure was higher than in controls.

**Predict.** The tempting reading of those three rows is that every baroreceptor neuron carries both channels. Does that follow, and if not, what does the experiment actually constrain?

> **Check your prediction.** It does not follow. What the experiment establishes is which gene deletions abolish this reflex in the targeted population: a collective requirement for Piezo1 and Piezo2. It does not quantify the population’s mechanosensitive current, and it does not say whether both channels act in each baroreceptor. One cell carrying both channels and two separate groups each carrying one are both *compatible* with the three rows, but neither picture predicts them on its own — each needs extra assumptions about how much input the reflex can lose before it moves. A hypothetical nonlinear relation between sensory input and reflex gain is one such assumption, and the worked version under Going deeper shows how much the conclusion depends on the curve you choose. Separating the pictures would mean identifying baroreceptors and testing, cell by cell, which channels they express and which they need — with validated single-channel deletions and controlled mechanical stimulation, since a partial deletion on its own can leave the population reflex unchanged either way.

## A map of the organs in the brainstem, sharpened by inhibition

Vagal sensory axons primarily target the nucleus of the solitary tract (NTS) in the medulla, the first relay where visceral afferent signals are received and distributed to the circuits that run breathing, circulation and digestion.

Ran and colleagues (Nature, 2022) imaged the caudal NTS with a nuclear calcium indicator in anaesthetized mice and reported organ-biased ensembles: responsive neurons were arranged in space by organ. Cutting the vagus below the diaphragm on both sides removed the gastric distension response, which confirmed the route.

If the map were simply the peripheral lines arriving in order, it would need no maintenance. It is not.

**Predict.** Inhibition is common in the brainstem, and these cells sit among inhibitory neighbours, so some of what a cell does not respond to may be something it is being stopped from responding to. Block GABA-A receptors in the caudal NTS with bicuculline while imaging organ responses. If inhibition is holding the organ map apart, what happens to each cell’s tuning?

> **Check your prediction.** Tuning broadened across organs: many cells now answered both stomach and duodenum stretch, or both stomach stretch and laryngeal water. So the separation depends on inhibition rather than on anatomy alone. Within one organ, tuning was already broad even without the blocker: most larynx-responsive cells answered water, high salt and citric acid alike. The map is processed, not a set of fixed lines read off the periphery.

## From afferent to behaviour: stretch, satiety and what the trial evidence covers

Bai and colleagues (Cell, 2019) worked in awake mice at least six weeks old at surgery, which is question 3 with behaviour as the output. Activating Oxtr-positive intestinal stretch afferents strongly suppressed feeding; their terminals are intraganglionic laminar endings, flattened arborizations associated with the nerve clusters (myenteric ganglia) between the circular and longitudinal muscle layers of the gut wall. That location is compatible with sensing gut-wall tension; anatomy alone neither establishes mechanosensitivity nor requires every stretch detector to sit there. Activating Glp1r-positive afferents, whose stretch endings Bai’s adult labelling found in the stomach rather than the intestine, suppressed feeding more weakly. Activating Gpr65-positive afferents did not measurably change feeding in the assays tested.

*Figure (AI-generated illustration).* In awake mice, activating Oxtr-positive intestinal stretch afferents, whose intraganglionic laminar endings associate with myenteric ganglia between the muscle layers, strongly suppressed feeding. Activating Glp1r-positive afferents with gastric stretch endings suppressed feeding more weakly, while activating Gpr65-positive afferents, which innervate villi, did not measurably change feeding in the assays tested. The results do not exclude nutrient signalling through gut hormones. Image: a schematic intestinal wall with villi facing the lumen. A slate Gpr65 mucosal terminal branches inside a villus beneath its intact epithelial lining. A gold Oxtr intraganglionic laminar ending lies against a myenteric ganglion between circular and longitudinal muscle. An inset compares activation in awake mice: intestinal Oxtr afferents strongly suppressed feeding, gastric Glp1r afferents suppressed it more weakly, and Gpr65 afferents produced no measurable feeding change in the assays tested.

Does that make stretch the signal that ends a meal? Loading the intestine with a non-nutritive hyperosmolar solution (mannitol) inhibited food intake and the activity of agouti-related peptide (AgRP) neurons, the hypothalamic cells whose firing drives feeding. So nutrients were not required. The load was not a pure volume change either: a hyperosmolar load pulls fluid into the gut.

Nutrients also reach the vagus by a route these experiments did not test. Enteroendocrine cells in the gut lining release cholecystokinin and other hormones that act on afferent terminals, and both of Bai’s stretch populations express the cholecystokinin receptor Cckar. So a nutrient signal could reach an afferent without that afferent sensing the nutrient itself — which would be a second kind of stimulus arriving at the same cells. Whether cholecystokinin does act on these particular afferents was not tested here. For central feeding circuits see [motivation and reward](https://stemhop.com/lessons/motivation-and-reward/).

Evidence on ascending anatomy agrees across species: rat tracing and mouse tracing support parabrachial relays toward thalamus and insular cortex, and mouse insular activity tracked recovery of physiological state (Livneh and colleagues, Neuron, 2020). No study here proves one serial chain.

The human evidence is narrow, and it answers a different kind of question. The VNS-REHAB trial of implanted vagus-nerve stimulation after stroke (Dawson and colleagues, Lancet, 2021) favoured active over sham stimulation, each paired with rehabilitation, on a motor outcome. What it cannot do is identify a natural stimulus, a transduction mechanism, or the contribution of any selected cell type, because no population is labelled. It covers one device and one pairing with rehabilitation, and says nothing about “vagal tone” or ear devices.

### Design problem

Take the three activation results above as given. Design experiments that test whether volume or nutrient content ends a meal.

> **One way to do it.** Dissociate the stimuli. Give an equal-volume, non-nutritive load in one group and an iso-volumetric nutrient load in another, to awake animals with intake measured directly. Use activation for sufficiency, and use generic reversible silencing or loss of the labelled population for necessity. Allow for the indirect hormone route above, since a nutrient signal need not be sensed by the afferent itself. Add a route control, such as interrupting the subdiaphragmatic vagus. Read the outcomes in a fixed order. If the non-nutritive load cuts intake as much as the nutrient load, a nutrient-independent signal is sufficient in this assay. Crediting volume itself would need osmolality-matched loads and measured distension, because of the osmotic fluid shift. If silencing the stretch population raises intake of a large meal, that population is needed there. If neither moves intake in the way predicted, report the result and name which control would be run next. State at the end that mouse intake does not quantify a human meal.

## Going deeper

### Ganglia, atlases and the 18 versus 37 counts

The two ganglia have different origins: nodose neurons derive from epibranchial placodes (thickenings of embryonic head ectoderm), and jugular neurons from [neural crest](https://stemhop.com/lessons/neural-crest/). In the mouse atlas of Kupari and colleagues (Cell Reports, 2019), Phox2b marked the nodose population and Prdm12 the jugular one. Their preparation was acutely dissociated ganglia from mice about 4.5 to 5 weeks old, and they described the dissected tissue as merged jugular and nodose ganglia that could include petrosal neurons. A mouse “nodose” sample can therefore hold three kinds of cell.

That atlas, from Patrik Ernfors’s laboratory, classified 18 nodose types by [single-cell RNA](https://stemhop.com/methods/single-cell-rna-seq/) sequencing, with several functional assignments proposed from expression signatures rather than shown for every type. A later atlas from Stephen Liberles’s laboratory (Prescott and colleagues, Cell, 2020) reported 37 types from a combined vagal and glossopharyngeal sample of eight-week-old male mice. The two counts differ in scope and are not interchangeable.

Prescott and colleagues used mice aged 8 to 24 weeks for the P2ry1 activation work.

### Where else vagal afferents arrive

Some gut afferents also reach the area postrema. Jugular afferents project predominantly to the paratrigeminal nucleus, in the spinal trigeminal tract, so the crest-derived arm of the nerve has its own central target.

### Before the nerve: an epithelial step and a candidate cell type

The P2X2/P2X3 result points to a step before the nerve: it fits ATP passing from the epithelium to the afferent, and the receptors deleted there were P2X2 and P2X3, not P2RY1. Read through the three questions, this is a fourth one — what happens before the afferent — and it does not change the finding that P2ry1 cells are needed for the swallowing response. Requiring both the cells and the receptors is compatible with one serial pathway, in which epithelial detection leads to ATP signalling onto P2ry1-labelled afferents; it does not by itself establish parallel routes.

One transcriptomic type, NP19, was a candidate for mediating these reflexes, and “candidate” is the right word: it is a cluster with a proposed job, not a population tied to a stimulus, a channel and an output.

The secretory neuroendocrine cells of pulmonary neuroepithelial bodies express Piezo2 themselves, and the bodies are also contacted by Piezo2-labelled sensory afferents. The loss-of-function evidence, though, is neuronal: it is deletion in adult sensory neurons that impairs the inflation reflex, so epithelial expression on its own does not assign the reflex to the neuroendocrine cells.

### Why the Glp1r-Cre line changed its answer

What a Cre line marks depends on the Cre-dependent tool as much as on the driver. A constitutive Cre driver paired with a reporter that is already present during development, and that stays switched on once recombined, can mark cells by their past driver activity. A Cre-dependent virus delivered in adulthood mainly captures Cre activity after it arrives, and an inducible Cre depends in addition on when it was induced. Recombination can also be incomplete, and can differ between target alleles, so labelling efficiency and specificity have to be validated.

Bai and colleagues suggested that broader developmental Glp1r expression, retained by the constitutive reporter used for the 2016 imaging, could explain why their own adult viral labelling found different endings. Differences in reporter strategy are part of the comparison, so it does not establish that explanation uniquely. Either way it is the marker-versus-gene problem in its most practical form: the handle is a fact about the mouse and its reporter, not about the stimulus.

### Piezo2 in newborns and the baroreflex numbers

A separate experiment deleted Piezo2 in the whole mouse, or from development in neural crest-derived sensory neurons with Wnt1-Cre, which caused respiratory distress and death in newborns. Deleting it from nodose neurons alone with Phox2b-Cre did not: those pups survived with normal blood oxygen, so the neonatal deaths were not caused by loss in the airway-innervating nodose population. The authors propose Piezo2 as an airway stretch sensor at birth as well, but that phenotype is a different experiment from the adult reflex defect.

Control mice had a baroreflex sensitivity of -5.0 ± 0.5 bpm/mmHg (mean ± SEM). Piezo1 and Piezo2 double knockouts had -0.6 ± 0.4 bpm/mmHg.

In conscious mice with telemetry, active-period mean arterial pressure was 112 ± 0.4 mmHg in double knockouts and 95 ± 0.5 mmHg in controls, a difference of 17 mmHg.

### Worked problem: one way a single deletion could stay invisible

This model problem sits behind the Predict box on the double knockout. The obvious use is to show how a partial loss of sensory input could hide inside the published error bars; the better use is to see how much that conclusion depends on choices nobody measured, because once you can break the argument you know what it is worth.

Nothing below was measured. Zeng and colleagues measured reflexes and afferent nerve activity, not a curve relating sensory input to reflex gain, and no experiment here tells you how much mechanosensitive current either single deletion removed. So we posit a curve outright, scaled so that full input reads the published control magnitude of 5.0 bpm/mmHg:

> **F(x) = 5 × (1 + 0.35²) × x² / (0.35² + x²)**, where *x* is a hypothetical fraction of control sensory input and *F* is the predicted **magnitude** of baroreflex sensitivity in bpm/mmHg.

The exponent 2, the half-maximum at 0.35, and the assumption that a single deletion removes half the input are all choices made for illustration, not fitted values. The published sensitivity is negative, since heart rate falls as pressure rises, so the table gives magnitudes.

| hypothetical fraction of control input | predicted magnitude of baroreflex sensitivity (bpm/mmHg) |
| --- | --- |
| 1.0 | 5.00 |
| 0.6 | 4.19 |
| 0.5 | 3.77 |
| 0.3 | 2.38 |
| 0.1 | 0.42 |
| 0 | 0.00 |

(a) How big a difference would these error bars resolve? The published standard errors are ±0.5 for controls and ±0.4 for the double knockout. For two independent groups each with SEM 0.5, the standard error of their difference is √(0.5² + 0.5²) = 0.71, so an approximate two-standard-error cutoff is 1.41 bpm/mmHg, about 28% of the control value; using the double knockout’s 0.4 instead gives √(0.5² + 0.4²) = 0.64 and a cutoff of 1.28. The single-deletion groups’ own standard errors, sample sizes, test and power are not given here, so these are illustrations, not the real experiment’s detection limit.

(b) On the posited curve, halving the input costs 5.00 − 3.77 = 1.23 bpm/mmHg, just under that cutoff. Resist the temptation to call the single deletions undetectable. Treat the estimated difference as normally distributed about 1.23 bpm/mmHg with a fixed standard error of 0.71, and it exceeds 1.41 about 40% of the time — a normal-approximation illustration, not a power estimate for the published experiment. The model shows how a partial loss *could* be missed, not that it had to be.

(c) Now break it. Saturation on its own tells you nothing about the half-input effect, because the answer depends on where the curve bends. Put the half-maximum at half of control input instead of 0.35 and halving costs 5.00 − 3.13 = 1.88 bpm/mmHg, which clears both cutoffs. Or take F(x) = 10x² / (1 + x²), which is also saturating and also passes through 5.00 at full input: halving costs 5.00 − 2.00 = 3.00. Three saturating curves, three different answers.

(d) The measured fall is a different matter. From 5.0 to 0.6 is 4.4 bpm/mmHg, 88% of control, comfortably resolvable at these standard errors (t ≈ 6.9).

(e) So what survives? Only the collective requirement itself. The error-bar arithmetic above sets no limit on how much current either single deletion removed, so none of these curves decides between one cell carrying both channels and two groups each carrying one.

### The stroke trial in detail

The VNS-REHAB trial (Dawson and colleagues, Lancet, 2021) randomized 108 adults at least 9 months after an ischaemic stroke, with moderate-to-severe arm weakness, to implanted vagus-nerve stimulation or sham stimulation, each paired with rehabilitation.

The difference in Fugl-Meyer upper-extremity improvement, measured on the first day after six weeks of in-clinic therapy, was 2.6 points (95% confidence interval 1.0 to 4.2).

The trial also reported Wolf Motor Function Test scores and Fugl-Meyer scores 90 days later, both favouring stimulation.

### Worked problem: reading the NTS map

Ran’s platform recorded about 2,800 neurons at once. Stimuli included gastric balloon volumes of 150, 300, 600 and 900 microlitres and laryngeal citric acid at 25 mM, pH 2.6.

In Ran’s six-mouse experiment, 1,018 of 1,133 responsive neurons answered to only one tested organ.

> **single-organ fraction = 100 × (single-organ neurons) / (all responsive neurons)**

(a) 100 × 1,018 / 1,133 = 89.8%. (b) 1,133 − 1,018 = 115 neurons responded to more than one tested organ. (c) 100 × 115 / 1,133 = 10.2%. (d) The denominator is neurons responsive to this panel in an anaesthetized mouse, not all nucleus of the solitary tract (NTS) neurons and not awake humans. The 1,133 cells come from six mice, so they are not 1,133 independent units.

(e) In a different experiment in the same paper, 42.2% of glucose-responsive cells also responded to duodenal distension, while 31.0% of distension-responsive cells also responded to glucose. Call the overlap I, the glucose population G and the stretch population S. Then I = 0.422 G = 0.310 S, so S / G = 0.422 / 0.310 = 1.36. The two percentages are consistent: the stretch-responsive population is about 36 percent larger, which gives the smaller conditional fraction. Do not merge this with the 1,133-cell data set. The published percentages are rounded, so 1.36 is approximate.

*Interactive on the lesson page:* Two conditional percentages fix a size ratio

---

From STEMhop (https://stemhop.com), free lessons by Christopher Deppmann. Lesson text © the authors. Practice questions: https://stemhop.com/lessons/vagal-sensory-neurons/
