The neurobiology of interoception in health and disease

Study at a glance

• Authors: Lisa Quadt, Hugo D. Critchley, Sarah N. Garfinkel
• Year: 2018
• Journal: Annals of the New York Academy of Sciences
• DOI: 10.1111/nyas.13915
• Design: Narrative review (Special Issue: Health Neuroscience)

Key findings

• Frames interoception as an umbrella term spanning four levels — afferent signalling (neural + humoral), central encoding/integration, influence on other cognition/behaviour, and conscious psychological expression — and organises the field’s dimensional vocabulary (accuracy, sensibility, metacognitive awareness; plus HEP and an ‘executive’ dimension) around the Garfinkel/Critchley distinctions.
• Reviews the neurobiology: insula posterior→anterior gradient, NTS/parabrachial/PAG/thalamus subcortical relays, circumventricular organs for humoral signals, and the EPIC agranular-visceromotor / granular-sensory prediction loop; frames the whole system as interoceptive predictive processing (IPP) implementing allostasis.
• Applies IPP to six health conditions — sickness behaviours, fatigue, depression, autism spectrum conditions, anxiety disorders, and eating disorders — arguing interoceptive impairment takes condition-specific forms (hypervigilance, hyporeactivity, accuracy/sensibility discrepancy) rather than a single deficit.
• Carries the Garfinkel ITPE finding (autistic adults show low interoceptive accuracy paired with high sensibility, and the accuracy–sensibility discrepancy predicts trait anxiety in both autistic and control samples) and Van den Bergh’s account of symptoms as high-posterior-probability interoceptive inferences.
• Explicitly cautions that the heartbeat-tracking task’s generalizability is questionable — beliefs about heart rate influence it (Ring & Brener 1996), tracking and discrimination performance can diverge (Ring & Brener 2018), and its relationship to respiratory/gastric axes is scarce and inconsistent.

Limitations

• A narrative review, not new data: no sample, no meta-analysis, selective citation, and (in the authors’ own footnote) predictive processing is treated ‘as a model of neural functioning, parts of which are rather speculative, or are inferred from existing evidence.’
• Reproduces the authors’ own framework (Garfinkel/Critchley dimensions, EPIC, IPP) as the organizing lens, so the health/disease sections are read through a predictive-coding commitment the review does not defend against rivals.
• The disease evidence is mixed and the review says so — autism accuracy findings are ‘divergent’ (intact in one child sample, impaired in another; possibly driven by comorbid alexithymia not autism per se), depression accuracy is ‘more complex’ than a simple deficit, and eating-disorder accuracy is impaired in some anorexia studies and enhanced/normal in others.
• Uses the resting cardiac heartbeat tasks as accuracy measures throughout while simultaneously flagging their validity/generalizability problems — the same tension the wiki tracks on the validity debate.

Connections

• Methods: Heartbeat detection task, Event-related potentials
• Concepts: Interoception, Interoceptive taxonomy (Farb et al. Box 1), Interoceptive trait prediction error (ITPE), Interoceptive psychopathology, Sickness behaviors, Alexithymia, Insular cortex, Visceromotor areas (VMAs), Interoceptive inference, Predictive coding, Allostasis, computational psychosomatics, Theory of constructed emotion, Anxiety sensitivity
• Debates: inferential?, Is the heartbeat-counting task a valid measure of cardiac perception?, Is more interoceptive awareness better?
• Related studies: Khalsa et al. (2018), Seth & Friston (2016), Barrett (2017), Craig (2002), Bonaz et al. (2021), Jones et al. (2010)

Quadt, Critchley & Garfinkel (2018), a review in the Annals of the New York Academy of Sciences “Health Neuroscience” special issue. It is the Sussex group’s own statement of the interoception-and-mental-health programme, and for this wiki it does three jobs at once: it puts the Garfinkel/Critchley dimensional framework first-hand (the wiki had carried it secondhand through Interoceptive taxonomy (Farb et al. Box 1) and Interoceptive sensitivity), it consolidates the wiki’s scattered predictive-processing anatomy (insula, EPIC, Visceromotor areas (VMAs), Allostasis) into one IPP account, and it introduces the disease material the Khalsa roadmap gestured at but did not work through — sickness behaviours, fatigue, autism, and eating disorders in particular.

It contradicts nothing in the wiki. It is a consonant hub: a Critchley/Garfinkel review sitting in the same predictive-coding/constructionist family as Seth & Friston, Barrett and the roadmap. Its value is consolidation and the new clinical breadth, not a new claim.

Interoception as an umbrella, in four levels

The review’s opening move is definitional and worth recording because it is tidier than the wiki’s other definitions. Interoception encompasses: (1) afferent signalling — body-to-brain over distinct neural and humoral (immune, endocrine) channels; (2) central encoding, representation, integration; (3) the influence of that information on other perceptions, cognitions and behaviours (which “does not necessitate (or preclude)” awareness — the cardiac-timing effects on fear/decision/memory live here); and (4) the psychological expression as consciously accessible sensations and feelings. The humoral channel is the part the wiki’s mostly-neural accounts underweight: circulating cytokines and hormones reach the brain through the circumventricular organs (area postrema, OVLT, subfornical organ) and the NTS, and this is the door through which the immune system talks to interoception — the anatomical basis for sickness behaviours.

The dimensional framework, first-hand

This is the paper’s clearest contribution to the wiki’s taxonomy problem, because it is Garfinkel and Critchley stating their own distinctions rather than the wiki reconstructing them from Farb et al.’s pushback:

• Interoceptive accuracy — objective performance on a behavioural test (heartbeat tracking).
• Interoceptive sensibility — subjective belief about one’s own ability (questionnaires like the BPQ, or rated confidence).
• Metacognitive interoceptive awareness — insight into one’s own performance, derived from confidence–accuracy correspondence; the review says this is “a most appropriate use of the word ‘awareness’.”
• Plus a first dimension for the afferent signal itself (indexed by neuroimaging change or the heartbeat-evoked potential, HEP, whose amplitude tracks detection ability), a second for its influence on other processing, and a further executive dimension (flexibly attending to / switching between interoceptive and exteroceptive representations).

The review is candid about the heartbeat-tracking task it leans on: most work depends on it, but beliefs about heart rate influence it (Ring & Brener 1996), tracking and discrimination scores can diverge (Ring & Brener 2018), the two tasks tap different processes, and generalization to respiratory/gastric axes is “scarce and inconsistent.” From an IPP standpoint some modalities may be precision-weighted more than others, so “the assumption that the heartbeat-tracking task can serve as a valid proxy … needs to be treated with caution.” That is a pro-caution voice for Is the heartbeat-counting task a valid measure of cardiac perception? from inside the group that uses the task most.

The neurobiology and IPP, in brief

The anatomy is the wiki’s existing picture, restated cleanly: posterior/mid insula as primary viscerosensory cortex projecting rostrally to the AIC, which re-represents and integrates; AIC forming a functional unit with ACC, amygdala and VMPFC/OFC; subcortically, the NTS as the brainstem convergence point for vagal and spinal lamina-1 afferents, projecting via parabrachial nucleus, PAG and posterior ventromedial thalamus to hypothalamus, amygdala, insula and ACC. The computational overlay is interoceptive predictive processing (IPP) and its cortical implementation, EPIC (Barrett & Simmons): agranular visceromotor cortices (caudal VMPFC/OFC, mid/anterior cingulate, AIC) issue descending predictions; granular/dysgranular mid- and posterior insula compute prediction error; the agranular regions are normally “relatively insensitive” to that error, which is what lets predictions stay stable and enables abstract, future-directed (allostatic) control rather than reactive homeostasis. See Visceromotor areas (VMAs), computational psychosomatics.

The disease sections — the review’s real payload

Health and disease differ in reported symptoms; the review adopts Van den Bergh et al.’s (2017) account that interoceptive signals reach awareness mainly when they generate precise prediction error, and become symptoms when the highest-posterior-probability hypothesis encodes an aberrant, disease-related cause. Each condition is a different way for that inference to go wrong:

• Sickness behaviours — inflammation signalled to the brain (vagal, humoral, microglial) evokes a stereotyped fatigue/anhedonia/social-withdrawal/fever pattern, with the insula mediating the experiential side (right AI metabolism tracks lost social interest; AI–mid-cingulate connectivity predicts malaise). The bridge from inflammation to mood disorder.
• Fatigue — an SB and a chronic condition; frontostriatal (ventral striatum) reward machinery plus insula; reframed top-down as a Bayesian verdict of low allostatic self-efficacy (Stephan et al. 2016) — see computational psychosomatics.
• Depression — autonomic dysfunction + raised inflammation (IL-6, CRP); reduced dorsal-mid-insula reactivity and reduced HEP amplitude; the EPIC “locked-in” account (Barrett, Quigley & Hamilton 2016) in which over-predicted metabolic demand and downweighted error drive a self-maintaining cycle that enlists SBs to conserve energy. The accuracy story is not a simple deficit: the review flags that the more severely depressed can show control-level accuracy.
• Autism spectrum conditions — divergent accuracy findings, possibly driven by comorbid Alexithymia rather than autism per se; a signature discrepancy of low accuracy paired with elevated sensibility (an enlarged ITPE), read as aberrant sensory precision.
• Anxiety disorders — mixed accuracy findings reconciled by the ITPE: it is the accuracy–sensibility discrepancy, not accuracy alone, that predicts trait anxiety (Garfinkel et al. 2016).
• Eating disorders — impaired interoceptive self-report (EDI) predicts ED vulnerability longitudinally; cardiac-accuracy findings in anorexia/bulimia are inconsistent; reduced AI/dorsal-mid-insula activation during heartbeat/stomach attention.

All six feed Interoceptive psychopathology, where the transdiagnostic frame is held, and sharpen Is more interoceptive awareness better?: if the impairment is hypervigilant in anxiety and hyporeactive in depersonalization, no single “raise awareness” prescription can be right across the board.

The companion review, three years later

Bonaz et al. (2021) is this paper’s closest relative in the wiki, and Critchley is a senior author of both. The division of labour between them is worth stating, because together they cover the clinical field:

	Quadt et al. (2018)	Bonaz et al. (2021)
conditions	six, all psychiatric/neurodevelopmental	twenty-odd, across psychiatry, neurology and somatic medicine
mechanism	one — Van den Bergh’s inferential account of symptoms, worked through each condition	one architectural diagram (Stephan’s allostatic self-efficacy circuit), applied by assertion
measurement	flags the heartbeat task’s validity problem explicitly	uses the resting cardiac tasks throughout with no caveat
depth vs breadth	depth	breadth

The caution asymmetry is the item to keep. This review states that the assumption the heartbeat-tracking task is a valid proxy “needs to be treated with caution”; the 2021 review then establishes interoceptive deficits in MS, Parkinson’s, Alzheimer’s, Tourette’s, autism and chronic pain on exactly that task without repeating the warning. Same senior author, three years apart. See Is the heartbeat-counting task a valid measure of cardiac perception? and Hugo D. Critchley.

What the 2021 review adds that this one does not have: the lesion evidence (interoceptive dimensions dissociating under insula/ACC damage), the somatic-medicine conditions (IBS, Chronic pain, pelvic pain, Joint hypermobility), the Microbiota-gut-brain axis, early adversity as cross-cutting determinant, and Bioelectronic medicine as an intervention class.

Provenance and reading

First author Lisa Quadt (BSMS PhD-level researcher) does not get a wiki page under the co-author convention (single appearance); the load-bearing figures are the senior authors — Sarah N. Garfinkel (created this ingest; the accuracy/sensibility/awareness and ITPE constructs are hers) and Hugo D. Critchley (page already exists; this is his second first-hand paper in raw/ and his first as an author of a review the wiki reads directly). The review’s therapeutic suggestion — biofeedback to raise accuracy and so recontextualize sensations as non-threatening — is the clinical training logic, and inherits its open question. Declared no competing interests.