Debate at a glance
- Summary: Brain–body coupling measures — gastric–brain phase synchrony, heartbeat-evoked potentials, cardiac–neural coherence — are proposed as objective interoceptive biomarkers, free of the report-dependence that dogs perceptual tasks. The proposal requires that a person have a coupling value: a stable quantity that could correlate with symptoms, treatment response, or trait. Levakov et al. (2023) tested this directly for the gastric case and found nothing — run-to-run reliability of gastric–brain synchrony at r ≈ 0, while both the gastric signal and the brain signal retested well. The phenomenon is robust at the group level, causally supported by vagotomy and stimulation work, and currently unmeasurable at the individual level. Whether that is a fixable property of current instruments or a deeper fact about coupling quantities is the open question.
- Status: open
Positions in the field
- Gidon Levakov & Galia Avidan: The phenomenon is real; the individual measure does not exist yet. Their own data establish the Gastric network under strict confound control and show it generalizes across samples — and then show its test–retest reliability is absent (r(21) = −.14 within the network, −.08 within grey matter, node-wise mean r = .003), while EGG peak frequency retests at r = .74 and functional connectivity is subject-identifiable. Their conclusion is explicitly methodological rather than sceptical: reliability bounds validity (Goodwin & Leech 2006), so relating gastric–brain coupling to physiology or behaviour ‘should be done with great caution’ until recording and analysis improve. Note the self-implication — the same first author’s 2021 NeuroImage paper used individual gastric–brain coupling to predict weight loss. (Levakov et al. (2023), Levakov et al. (2021) Neural correlates of future weight loss reveal a possible role for brain-gastric interactions, NeuroImage 224:117403)
- Ignacio Rebollo & Catherine Tallon-Baudry: Discovered it, and warn against exactly this use of it. Rebollo et al. (2018) established the gastric network; Rebollo & Tallon-Baudry (2022) located it in unimodal sensory and motor cortex and stated the caution Levakov et al. quote approvingly — that relating this coupling to individual differences in physiology or behaviour should be done cautiously. They shared code and spatial maps with the group whose paper trimmed their network’s extent by a factor of 17. This position holds that the finding is a fact about brains and stomachs in general and was never offered as a trait measure. (Rebollo et al. (2018) Stomach-brain synchrony reveals a novel, delayed-connectivity resting-state network in humans, eLife 7:e33321, Rebollo & Tallon-Baudry (2022) The sensory and motor components of the cortical hierarchy are coupled to the rhythm of the stomach during rest, J Neurosci 42:2205-2220)
- Nils Kroemer, Sophie Müller & the causal-manipulation camp: Coupling is causally real and modulable, therefore worth measuring better rather than abandoning. Müller et al. (2022) show auricular vagus nerve stimulation increases stomach–brain coupling via a vagal afferent pathway; Cao et al. (2022) show gastric–brain coherence in head-fixed rats (no motion confound possible) that varies with hunger state and is substantially reduced by bilateral vagotomy. On this view the reliability problem is an instrumentation problem — high-density EGG, imaging-guided electrodes, longer recordings — and the correct response is better measurement, not retreat from the construct. Nothing in this position denies the reliability data; it disputes what follows from it. (Müller et al. (2022) Vagus nerve stimulation increases stomach-brain coupling via a vagal afferent pathway, Brain Stimul 15:1279-1289, Cao et al. (2022) The vagus nerve mediates the stomach-brain coherence in rats, bioRxiv 2022.01.25.477693)
- Sahib S. Khalsa & the biomarker programme: Interoception needs objective markers, and coupling measures are the best candidate class. Khalsa & Lapidus (2016) argue interoception could supply the pragmatic biomarkers psychiatry lacks; the 2018 roadmap treats objective physiological indices as the way past the report-dependence of questionnaires and counting tasks. This position is the reason the reliability question matters — it is a programme that requires person-level measurement by construction. It has no answer to the reliability data yet; the roadmap predates them. (Khalsa et al. (2018), Khalsa & Lapidus (2016) Can interoception improve the pragmatic search for biomarkers in psychiatry?, Front Psychiatry 7:121)
- Frederike H. Petzschner and the computational sceptics: Predicted, and generalized beyond the stomach. Their review concedes that the computational-biomarker programme has almost no empirical test behind it, and names properties of interoceptive signals — sparse, multi-timescale, oscillatory — that analysis methods borrowed from vision handle badly. A coupling statistic estimated from ~45 gastric cycles is a concrete instance of the general worry. This position expects reliability failures to be the rule rather than the exception until estimation is matched to the signals’ actual statistics. (Petzschner et al. (2021))
Connections
- Related debates: Is the heartbeat-counting task a valid measure of cardiac perception?, Is interoception domain-general or organ-specific?, Is more interoceptive awareness better?
- Related concepts: Gastric network, Interoceptive taxonomy (Farb et al. Box 1), computational psychosomatics, Interoceptive sensitivity, Interoception
Created with the Levakov et al. (2023) ingest. It is the wiki’s third measurement debate, and it targets the escape route the other two were leaving open.
Why it is a distinct debate and not a corollary
Is the heartbeat-counting task a valid measure of cardiac perception? asks whether one task works. Is interoception domain-general or organ-specific? asks whether the construct travels across organs. Both have the same implicit remedy running under them: if self-report is the problem, measure the body directly. Heartbeat-evoked potentials, gastric–brain synchrony, respiratory–neural coupling — physiological quantities requiring no judgment, no counting, no confidence rating.
This debate asks whether that remedy works, and the first direct test of it says no. Not because the physiology is unmeasurable — it is, well — but because the coupling between body and brain, which is where the interoceptive content is supposed to live, does not retest.
The finding, and how clean it is
| quantity, same participants, two runs | reliability |
|---|---|
| EGG peak frequency (months apart, different room and equipment) | r = .74 |
| resting-state functional connectivity | strongly within-subject identifiable (t(1033) = 10.89) |
| gastric–brain PLV-delta, within the gastric network | r(21) = −.14, p = .54 |
| gastric–brain PLV-delta, within grey matter | r(21) = −.08, p = .71 |
| node-wise PLV, Schaefer-100 | mean r = .003, ns |
The control conditions are what make this hard to dismiss. A null reliability finding is normally attributable to a noisy instrument; here both instruments demonstrably work in the same participants and the same sessions. The failure is specific to the derived quantity.
The honest brake: n = 23, because reliability required two usable runs and 36 of 59 subjects were excluded. A correlation estimated at n = 23 has wide confidence bounds, and “r = −.14” does not exclude a true r of .2–.3. What it does exclude is a coupling measure strong enough to be a biomarker.
What is agreed
- The gastric network exists. Nobody in this debate disputes it. It survives strict confound regression, generalizes to a new sample and an independent dataset, and has causal support from vagotomy and vagal stimulation.
- Its published extent was mostly artefact. 19.89% of grey matter before confound control, 1.16% after. This is accepted, including by the group whose maps shrank.
- Group-level and person-level claims come apart. Both camps grant that establishing a phenomenon says nothing about whether it can be scored in an individual.
The crux: is r ≈ 0 a fact about the instrument or about the quantity?
| Instrumentation (Kroemer, Müller, and by implication Levakov’s own prescription) | The quantity itself (the sceptical reading) | |
|---|---|---|
| Why reliability fails | Too few gastric cycles (~45 per 15-min run), electrode placement uncalibrated, no respiratory control, PLV-delta compounds noise in two terms | Brain–body coupling may be genuinely state-dependent — varying with hunger, arousal, posture, attention — so there is no stable trait value to recover |
| What would fix it | Longer recordings, high-density EGG, imaging-guided placement, better synchrony estimators | Nothing. You would stop looking for a trait and start modelling coupling as a state |
| Prediction | Reliability rises with recording length; a power analysis exists to be done | Reliability stays near zero however long you record, and coupling tracks manipulable states well |
The second column is not represented by any published author in this debate, which is itself worth noting — the field’s default assumption is that a stable coupling trait is there to be found. But Cao et al.’s hunger-state modulation is direct evidence that coupling moves with state, and a quantity that moves with state is a poor candidate for a trait even when measured perfectly.
These are cheaply distinguishable, and that is the useful thing about this debate: record long, record twice, and plot reliability against recording duration. Nobody has.
What generalizes beyond the stomach
The pattern the wiki now holds three times:
| level | reliable | not reliable |
|---|---|---|
| cardiac perception | — | counting accuracy as perception (Is the heartbeat-counting task a valid measure of cardiac perception?) |
| cross-channel | confidence (Banellis et al. (2026)) | cross-channel performance |
| brain–body coupling | the body signal, the brain signal | their coupling |
Three labs, three instruments, three levels. Each individual result is defensible as a local failure; together they describe a field whose components are measurable and whose interoceptive constructs are not. Whether this is one problem or three that rhyme is the reason this page is open rather than a note on another page.
The immediate consequence for reading this wiki: any claim of the form “group X shows altered brain–body coupling” is a group-level claim that cannot currently be decomposed into individual scores — and any claim of the form “coupling predicts outcome Y” is bounded by a reliability nobody has measured for their own instrument.
Why open
Because the reliability datum is one study at n = 23 in one channel; because the causal evidence for the phenomenon is genuinely strong and pulls the other way; because the discriminating experiment (reliability as a function of recording length) is straightforward and unrun; and because the state-versus-trait reading — the one that would make this a permanent rather than a technical limitation — has no advocate and no test.