Respiratory interoception

The breath has been load-bearing on this wiki since the first ingest without ever being the subject of a page. It is the attentional anchor of mindfulness-meditation, the provocation in interoceptive-exposure, the channel Farb et al. name when they say training is domain-specific rather than general, and the system Weng et al. (2021) use as their case study for intervening on interoception at all.

This page exists mainly to record why that choice of case study is not neutral.

The exceptionality claim

Respiration is unique compared with other systems (e.g., gastrointestinal) insofar as conscious regulation can immediately impact respiratory processes and… respiratory processes can affect emotion and cognition. — Weng et al. (2021)

Two properties, and it is having both that makes breathing exceptional:

1. It is voluntarily drivable, immediately. You cannot decide to change your gastric motility, your bladder filling, your immune signalling or (much) your heart rate. You can change your breathing in one cycle. Interoception in every other channel is something the brain reads; here it is something the person can also write.
2. Driving it changes state. The change propagates — to sympathetic outflow via cardiopulmonary stretch receptors and the baroreflex (slow-breathing), and to cognition and emotion via the respiratory entrainment described below.

The methodological consequence is the one the wiki should carry: most of the field’s intervention evidence is respiratory evidence, and respiration is the channel least like the others. A framework validated on breathing generalizes to the gut and bladder only if the voluntary-control property was doing no work — and in the behavioural arm (slow-breathing) that property is the entire mechanism. Weng et al. concede the point obliquely, noting that “not all systems can be as readily consciously manipulated as the respiration system” and falling back, for those systems, on recognizing sensations rather than driving them.

The rhythm entrains the brain

Breathing is not only a signal to be sensed; its rhythm organizes activity elsewhere. The rhythm generator is the preBötzinger complex, a compact bilateral medullary nucleus that drives the inspiratory muscles and coordinates expiratory movement. Beyond driving muscle, breathing-related activity initiated there reaches suprapontine structures — hippocampus, prefrontal cortex, amygdala — and influences cognitive and emotional function (Tort et al. 2018; Zelano et al. 2016, on nasal respiration entraining limbic oscillations).

The same phasing shows up one synapse into the interoceptive pathway, and is exploitable: NTS relay neurons receive pulmonary-stretch and baroreceptor afference primarily during inhalation, and are proposed to receive facilitatory input during exhalation — which is why RAVANS gates stimulation to the exhalation window (see bioelectronic-medicine).

So the respiratory cycle is a timing structure for interoception generally, not merely one more channel. That is a further reason it is a bad representative case and a good intervention target.

Engelen, Solcà & Tallon-Baudry (2023) widen this claim from the same primary literature: respiration-locked activity appears not only in the limbic system but across somatosensory, motor, visual and auditory cortex and in default/salience/dorsal-attention networks at rest, and its phase alignment across regions is tighter (“rather good”) than the gastric rhythm’s travelling-wave organization. In their scaffolding hypothesis, the breath is a candidate carrier wave imposing shared windows of cortical excitability — the rhythm-based sibling of everything this page argues about the breath as a signal. See cardiac-cycle-effects for the same logic on the cardiac side.

The afferent hardware (Prescott & Liberles 2022)

Beneath the psychophysics and the entrainment sits the sensory apparatus, and the Liberles-lab review supplies it. Airway stretch sensation and the Hering-Breuer inspiratory reflex (lung inflation → apnea) are carried by vagal slowly-adapting stretch receptors whose molecular sensor is PIEZO2 (Nonomura et al. 2017) — global knockout is lethal at birth from respiratory distress, so this is the one interoceptive channel with a mechanotransduction sensor pinned to a lethal loss-of-function phenotype. Airway protective sensing (cough, laryngeal closure, apnea, air hunger) is not one broadly-tuned irritant sense but a set of labelled lines — Prescott’s P2RY1 water-sensing neurons, TRPV1/TRPA1 chemonociceptors, the MRGPRC11 itch receptor, NPY1R/NPY2R subsets. See interoceptive-sensors for the general sensor inventory; the point for this page is that the breath’s status as an exceptional channel now has a receptor-level anatomy under it, not just a voluntary-control argument.

Where the wiki already had this material

• Panic. Respiratory challenge elicits sensations that trigger panic (Nardi et al. 2009), and hyperventilation provocation is the paradigm case of interoceptive-exposure. The cognitive-model-of-panic and anxiety-sensitivity pages are largely about misread respiratory and cardiac sensation.
• Contemplative training. Breath sensitivity is the one interoceptive measure that contemplative training does seem to improve (Daubenmier et al. 2013), against a null for domain-general heartbeat accuracy — see does-mindfulness-enhance-interoceptive-accuracy. Long-term mindfulness training is associated with reliably lower resting respiration rate (Wielgosz et al. 2016), and meditation lowers respiratory rate even when breathing changes are not instructed.
• Primary interoceptive cortex. The posterior insula is “sensitive to the respiratory rate and putatively considered primary interoceptive cortex” (Farb, Segal & Anderson 2013, now held first-hand). Its coupling to respiratory rate is gated by attention: strong when the breath is attended (r=0.63), absent when it is not (r=0.23), with the signal then tracking somatosensory cortex instead — so the breath’s status as primary-interoceptive-cortex input is itself an attentional achievement. The distinct claim that mindfulness training raises posterior-insula activation belongs to the separate SCAN 2013 companion ([[farb-2013-training-interoceptive-cortex|Farb, Segal & Anderson 2013, SCAN]], now held first-hand), where it is more precisely a practice-dose effect — daily-practice compliance, not MBSR-group membership, predicts greater posterior-insula IA activation (r=0.61) — while the group-level training effect sits in the anterior insula. See insular-cortex.
• Threat anticipation. Respiratory resistance can be parametrically manipulated with a device, and the anticipation versus the experience of respiratory threat are represented in ventrolateral versus lateral periaqueductal gray respectively (Faull et al. 2015) — one of the few places the wiki has an anticipation/experience dissociation with an anatomical address. Relevant to forecasting. (Faull is olivia-harrison under her former name; the same programme produced the FDT and the BLT.)

A measurement note

The wiki’s persistent complaint that interoception is measured almost entirely through the heart (see is-the-heartbeat-counting-task-valid) has a respiratory answer available and under-used. Breathing is measurable objectively and continuously without asking the participant anything, and it can be manipulated experimentally in a graded way. That it remains a secondary measure in a field bottlenecked on cardiac tasks is worth noting. The respiratory tasks named above — parametric resistance loading, breath-focus decoding, resting respiration rate — are alternatives already in use.

Answered, in part, by the Allen (2026) ingest. This paragraph stood for three ingests ending “none of them yet represented in this wiki by a primary source.” Allen (2026) is the first, and it delivers what the paragraph asked for — a continuous, objective respiratory measure requiring no report from the participant, with good trial-level reliability (Spearman–Brown .857) — while arriving at it from an unexpected direction. See respiratory-tracking-task.

The property this page calls exceptional is also what makes measurement possible

The argument above treats voluntary drivability as a liability: it makes respiration a poor representative of interoception generally, so respiratory intervention evidence may not generalize. Allen (2026) takes the same property as an enabling condition, and the two readings should be held together rather than reconciled.

His charge against the field is that it studies the perception of bodily state and neglects the control of it — and a control task can only be built on a channel a person can drive on demand. There is essentially one. You cannot ask a participant to track a sinusoidal target with their gastric motility.

So:

• This page’s claim: respiration is the least representative channel, because it is voluntarily drivable.
• Allen’s claim: respiration is the only channel on which interoceptive control is measurable at all, for the same reason.

Both are right, and together they imply that interoceptive control may not be a general construct but a fact about breathing. That is the uncomfortable position interoceptive-control exists to hold — parallel to the channel-specificity worry already recorded for accuracy on interoceptive-taxonomy.

One further asymmetry the control framing exposes: this page describes the breath as “something the person can also write,” and then, like the rest of the wiki, goes on to discuss only reading. Nothing above measures how well anyone writes.

The exceptionality claim, tested — and it is worse than exceptional, it is disconnected

The argument above says respiration may be unrepresentative of other interoceptive channels. Banellis et al. (2026) supply the first direct measurement, and the finding is stronger than unrepresentativeness: in 241 participants measured on both axes with matched psychophysics (RRST and HRDT), respiratory and cardiac interoceptive sensitivity, precision and metacognitive efficiency show no relationship whatever — moderate Bayesian evidence for the null, at power to detect r ≥ 0.179. Corroborated by a second respiratory instrument on a different feature: a breath-duration discrimination task likewise fails to correlate with cardiac interoception (Bodart et al. 2024).

Three things follow for this page.

The exceptionality claim is vindicated in the least useful way. This page argued respiration is a bad representative case because of voluntary drivability — a mechanistic story. The data say the channels are unrelated but say nothing about why, and the same null now holds between cardiac and gastric, thermosensory, nociceptive and affective-touch measures (Ferentzi et al. 2018; Crucianelli et al. 2022), none of which are voluntarily drivable. So respiration may not be exceptional at all; it may just be one more independent channel in a field that has no general one. See is-interoception-domain-general.

It cuts the other way too, and the page should say so. If respiratory findings do not generalize to the heart, then the wiki’s cardiac findings do not generalize to the breath — and the clinical literature this page hosts is mostly about respiratory phenomenology (cognitive-model-of-panic, hyperventilation provocation, air hunger) while having been studied through cardiac tasks. The under-use of respiratory measurement complained about below is not merely a missed opportunity; it means the panic literature’s interoceptive evidence is largely from the wrong organ.

The confidence exception. Respiratory confidence did correlate with cardiac confidence — and more strongly still with confidence about auditory tones (r = 0.642, the largest correlation in the study). Whatever binds the channels subjectively is not respiratory, not cardiac, and not bodily.

Anxiety attaches to the breath with the opposite sign to the heart

The decorrelation above is a null: the channels do not covary. Harrison et al. (2021) supply something stronger and more awkward — the same trait predicting the two channels with opposite sign.

The wiki’s panic thread is unambiguous that anxious and panic-disordered people detect their heartbeats better (Ehlers 1993; Zoellner & Craske 1999), and that better cardiac perception predicts worse outcome. On the FDT, moderate-trait-anxiety participants were less sensitive to an added inspiratory load (threshold 3.0 vs 4.0 filters, Z = −2.4, p = 0.01) — replicating Garfinkel et al. (2016a) and Tiller et al. (1987), so this is the respiratory literature’s own consensus rather than one result.

This page has been arguing for several ingests that “the panic literature’s interoceptive evidence is largely from the wrong organ.” That claim can now be stated more sharply: the panic literature’s central empirical fact points the other way in the organ the panic phenomenology is actually about. Air hunger, hyperventilation and suffocation fear are respiratory; the evidence that anxiety heightens interoceptive detection is cardiac; and where anyone has looked respiratorily, anxiety goes with duller detection.

Two readings, both live and neither settled here — organ-specificity, or an indictment of the counting task, on which anxiety has a documented route to manufacturing accuracy that the FDT’s physically-specified stimulus does not offer. See is-interoception-domain-general.

The breath as the channel where the theory finally got tested

One more thing this page should record, because it is the exceptionality claim paying off rather than costing.

Every computational account the wiki holds — interoceptive-inference, predictive-coding, active-inference — is about quantities that change trial by trial. Testing them needs an interoceptive stimulus that can be applied and withdrawn on a seconds timescale, safely, many times over, in graded amounts. There is essentially one such channel. Harrison et al. used it (see breathing-learning-task) to produce the wiki’s first model-based, trial-resolved human evidence for interoceptive prediction and prediction error.

So the property this page calls a liability for generalization is, for a third time, an enabling condition for measurement — after Allen’s control task and the RRST’s psychophysics. The pattern is now consistent enough to state as this page’s real thesis: almost everything the field can measure well about interoception, it can measure well only in the lungs, and the lungs are the channel it has the least reason to think representative.

A fourth instance, and the first that partially escapes the thesis. The IBL lineage — Paulus et al. (2012) through Kruschwitz (2019) and Walter (2020) — is the field’s instrument for interoceptive anticipation, and it is respiratory for exactly the reason above: a resistive load is the one aversive interoceptive stimulus that can be dosed, timed and repeated at will.

The escape is that Wilzok et al. (2023) took the paradigm out of the lungs. Running the same cues and scales with graded pinprick pain gave a second gradable aversive channel, and produced two results that only a two-channel design could give: the anticipation–experience gap correlates across channels (r = 0.57), and the pain literature’s overprediction asymmetry — people overpredict severe pain, and underpredicted pain hurts more — holds for inspiratory loading too. That second finding is the more interesting one for this page, because it is a property of aversive interoceptive expectation that was established in a channel this page’s thesis says nothing about, and then found in the breath. For once the generalization ran the right way.