Reading the interoception literature

The starting point§

Interoception is the modern construct that names the body's sense of itself — the felt visceral state through which the brain tracks the homeostatic condition of the organism. The term is a century-old coinage that sat largely unused for ninety years before its revival in 2002; what the term covers — heart rate, breath, hunger, thirst, gut state, temperature, pain — has of course been studied continuously, under labels like visceral afferent, vagal afferent, and homeostatic feedback. The construct's modern shape is post-2002, but its substance is older.

The field's principal definitional disagreement is between two formulations that have not been finally reconciled. A. D. (Bud) Craig's 2002 direct-readout model treats interoception as a distinct sensory modality with its own thalamocortical pathway, terminating in the right anterior insula, where homeostatic afferent signal is re-represented as conscious feeling. The post-2013 predictive-processing alternative, worked out by Anil Seth in 2013 and Lisa Feldman Barrett and W. Kyle Simmons in 2015, inverts the directionality: interoception is not direct readout but actively- inferred prediction about expected body state, with prediction errors driving learning. The first reads the signal bottom-up; the second reads it top-down.

What follows is a chronological reading of the literature with the conceptual moves in front. Each move builds on, ignores, or argues against the prior ones, and reading the field forward registers the structure of disagreement that reading it backwards from a contemporary review article does not.

Precursor anatomy (1906–1996)§

The substantive anatomical work was done long before the modern construct was articulated under its current name.

The term interoception is from Charles Sherrington's 1906 The Integrative Action of the Nervous System, where he proposed a tripartite division of sensory function: exteroception (the body's interface with the outside world), proprioception (sense of self-position and limb-state), and interoception (sense of internal visceral state). Sherrington's coinage sat largely unused for ninety years. The visceral and homeostatic literatures of the intervening period studied the components piecemeal — vagal afferents, lamina-I projection neurons, the spinothalamic tract, nucleus tractus solitarius — without unifying them under Sherrington's term.

James R. Augustine's 1996 Brain Research Reviews synthesis on the insular lobe is the canonical pre-modern anatomical reference. The paper documents the insula's afferent and efferent connections in granular detail and the functional subdivisions (visceral sensory, visceral motor, motor association, vestibular, language) and the cytoarchitectonic distinctions (granular posterior, dysgranular middle, agranular anterior). The review is anatomical, not functional, and predates the Craig 2002 synthesis by six years. Almost every subsequent insula paper cites Augustine as the cytoarchitectonic foundation on which the functional claims rest.

Conceptual prehistory (1994)§

Antonio Damasio's Descartes' Error (Grosset/Putnam, 1994) is the conceptual prehistory of modern interoception, articulated under a different vocabulary. The somatic-marker hypothesis — that emotion is not a disturbance of reason but a constitutive component of it, that the body's somatic states are read by the brain as signals biasing decision-making — is the body-feedback frame on which Craig 2002 would later build.

Damasio's empirical substrate is the Iowa Patient Registry: focal lesion cases collected across two decades at the University of Iowa, particularly the ventromedial-prefrontal-cortex damage patients whose preserved intelligence is coupled to catastrophic decision-making. The Iowa Gambling Task (Bechara, Damasio, Damasio and Anderson 1994, Cognition 50, 7–15) is the laboratory operationalisation. The same Iowa programme is the empirical substrate of the 2007 Naqvi and Bechara insula-and-addiction paper that would later extend the somatic-marker frame from prefrontal cortex to insula and from decision-making to craving.

The 1994 book was a trade-press popularisation of a research programme running in primary venues across the 1990s. Its citation life in the cognitive-neuroscience literature has been enormous; the popular reception has been larger still.

The 2002 reformulation§

A. D. (Bud) Craig's How do you feel? Interoception: the sense of the physiological condition of the body (Nature Reviews Neuroscience 3, 655–666, 2002) is the field-founding review. Craig was a neuroanatomist at the Atkinson Pain Research Laboratory at the Barrow Neurological Institute in Phoenix; his career-long primary work was tract-tracing the spinothalamic projection in primates, particularly the lamina-I cells that carry pain, temperature, and homeostatic afferent signal. The 2002 paper synthesised two decades of that primary work into Sherrington's old term and gave it new anatomical specificity.

Two moves are doing the work. First, the construct of interoception is named as a distinct sensory modality with its own thalamocortical pathway, not a special case of somatosensation: small-diameter spinothalamic afferents project through a specific thalamic relay to the dorsal posterior insula, which re-represents homeostatic body-state. Second, the right anterior insula is named as the re-representation layer, where the felt-state substrate is generated; the lateralisation to the right hemisphere connects the construct to subjective emotional experience in a way other interoceptive frames had not.

Craig's 2009 follow-up How do you feel — now? extends the argument into consciousness territory, placing the anterior insula and its von Economo neurons as the substrate for the moment-to-moment sense of being a feeling self. The 2009 paper is the bolder of the two and the one that subsequent predictive-processing accounts have argued against.

Operationalisation (2004)§

Hugo Critchley and colleagues' 2004 Nature Neuroscience paper Neural systems supporting interoceptive awareness gave the field its operational handle. Where Craig had named the construct anatomically, Critchley, Stefan Wiens, Pia Rotshtein, Arne Öhman, and Raymond Dolan used the heartbeat- detection task — judge whether external tones are synchronous with your own heartbeats — to produce a per-subject accuracy score and to correlate that score with fMRI activity and grey-matter volume in the right anterior insula. The cross-subject prediction is what made the paper land.

Critchley was at UCL's Wellcome Department of Imaging Neuroscience and the Institute of Neurology at the time; he moved to Brighton and Sussex Medical School as foundation chair in psychiatry in 2006. The post-2004 collaboration between his autonomic-and-emotion programme and the Sussex Sackler Centre for Consciousness Science (Seth, Garfinkel) is one of the most productive subfields in 2025-vintage cognitive neuroscience.

The heartbeat-detection task itself has come under methodological critique. Brener and Ring 2016 in Philosophical Transactions of the Royal Society B documented that subjects' beliefs about their own heart rate predict counts more reliably than the actual heartbeats do; medical professionals and athletes score better not because of better interoception but because of better priors. The post-2016 methodological literature has explored alternatives — the heart-rate discrimination task developed at Aarhus from 2021 on, the cardiac elevation-detection task — and the original paradigm has been refined in response.

Causal evidence (2007)§

Damage to the insula disrupts addiction to cigarette smoking (Naqvi, Rudrauf, Hanna Damasio, Bechara; Science 315, 531–534, 2007) is the canonical causal-evidence citation for the insula's role in addiction and, by extension, in interoception as a felt state. The cohort: 19 smokers with focal brain damage involving the insula, matched against 50 smokers with brain damage elsewhere; the 19 insula-lesion patients had quit smoking with a degree of ease — without relapse, without persistent craving — that the comparison group had not.

The argument that ties the paper to Craig's interoception frame is that the felt urge to smoke is an interoceptive sensation, and the loss of the urge after insula damage is direct evidence that the insula is the substrate of the feeling itself. The lineage runs straight from Damasio's 1994 somatic-marker hypothesis through this paper: same Iowa registry, same lesion-method tradition, extended from prefrontal cortex to insula and from decision-making to craving.

The predictive-processing alternative (2013–2015)§

The mid-2010s saw the predictive-processing framework adapted to interoception in two consecutive moves that constitute the major theoretical alternative to Craig's direct-readout model.

Anil Seth's 2013 Trends in Cognitive Sciences paper Interoceptive inference, emotion, and the embodied self is the founding theoretical statement of the interoceptive inference approach. The argument inverts Craig's directionality: subjective feeling states are not direct readouts of body state but actively-inferred predictions about what the brain's interoceptive afferents should be saying, with prediction errors driving learning and updating. The frame adapts Karl Friston's free-energy principle from the perceptual-inference domain to the body-state-monitoring domain.

Lisa Feldman Barrett and W. Kyle Simmons's 2015 Nature Reviews Neuroscience paper Interoceptive predictions in the brain is the architectural-empirical companion to Seth's theoretical statement. The Embodied Predictive Interoception Coding (EPIC) model places the prediction-generating layer in the agranular visceromotor cortices — the anterior insula but also the anterior cingulate cortex and the ventromedial prefrontal cortex — and reads the insula as a comparator rather than a feeling-generator. The architectural commitments give the model testable specificity that the broader predictive-processing framework alone does not.

The two papers share the predictive-processing commitment but differ in emphasis: Seth emphasises the inference about expected body state as the substrate of feeling; Barrett and Simmons emphasise the visceromotor cortices as the prediction-generating network. Both place the locus of feeling across a wider network than Craig's insula-centric model. Both have generated computational-modelling follow-up, particularly from Klaas Stephan's group at Zurich (Petzschner, Iglesias, Stephan and colleagues).

The closely-related theory of constructed emotion — also Barrett's — extends the EPIC model into a wider claim about what emotion is: not a readout of a biologically pre-wired category but a construction made in the moment from interoceptive prediction, conceptual knowledge, and situational context. The constructed-emotion frame has been the principal theoretical alternative to the basic-emotion tradition (Paul Ekman, Jaak Panksepp, Joseph LeDoux) across the 2010s and 2020s, with the disagreement professional and sometimes pointed.

Methodological refinement (2015)§

Sarah Garfinkel, Seth, Adam Barrett, Keisuke Suzuki, and Critchley's Knowing your own heart: distinguishing interoceptive accuracy from interoceptive awareness (Biological Psychology 104, 65–74, 2015) is the methodological pivot of the post-2004 literature. Where Critchley 2004 gave the field a single number per subject — the heartbeat-detection-task score — Garfinkel 2015 showed that single number was hiding three orthogonal constructs: interoceptive accuracy (the behavioural-task score), interoceptive sensibility (subjective self-report), and interoceptive awareness (the metacognitive correspondence between the two).

The empirical work used a normative sample of 80 subjects and showed that the three dimensions dissociate: a person can be objectively accurate but subjectively oblivious, or subjectively confident but objectively poor. Pre-2015 papers using interoceptive awareness have to be re-read in context to know which construct each meant. The post-2015 clinical literature uses the three-way distinction routinely.

Field consensus (2018)§

The 2016 Interoception Summit at the Laureate Institute for Brain Research in Tulsa, organised by Sahib Khalsa, produced the Roadmap paper in Biological Psychiatry: Cognitive Neuroscience and Neuroimaging in 2018. The 24-author list is the field — Critchley from Brighton, Garfinkel from Sussex, Paulus from LIBR and UCSD, Stephan from Zurich, Stein from UCSD, Adolphs from Caltech, Mehling from UCSF — and the paper's authority comes from being the document the field agreed to.

Two specific contributions. First, the consensus nomenclature: the three-way accuracy / sensibility / awareness distinction from Garfinkel 2015 is canonicalised here. Second, a research agenda for psychiatric disorders organised around interoceptive dysfunction, with specific predictions for anxiety, depression, eating disorders, addiction, and somatic-symptom presentations. The post-2018 clinical interoception literature uses the Roadmap as its orienting reference.

The Roadmap also gives equal standing to the direct-readout Craig frame and the predictive-processing / inference alternative, declining to adjudicate. The field's principal theoretical disagreement is preserved rather than resolved.

Where the disagreements live§

Three substantive disagreements run through the literature.

Direct readout versus active inference. Whether interoception is the bottom-up re-representation of homeostatic afferent signal in the right anterior insula (Craig's position) or the top-down inference about expected body state, with prediction errors as the substrate of feeling (Seth; Barrett and Simmons). Both architectures accommodate much of the existing empirical data; the experiments that would adjudicate are difficult to design and have only recently begun to accumulate.

Insula-centred versus distributed. Whether the substrate of felt body-state is specifically the right anterior insula (Craig, following the lateralisation finding) or a wider visceromotor network including anterior cingulate cortex and ventromedial prefrontal cortex (Barrett and Simmons's EPIC model). The lesion evidence from Naqvi-Bechara 2007 gives the insula a privileged causal role; the network view reads the same evidence as showing one critical node within a wider system.

Accuracy versus sensibility as the construct of interest. Whether the dimension of interoception that should anchor clinical and research work is the objective behavioural-task measure (accuracy) or the subjective self-report (sensibility), with the awareness dimension as the metacognitive bridge between them. Body-based clinical interventions (mindfulness, yoga, somatic experiencing) move sensibility more reliably than accuracy; whether sensibility shifts without accuracy shifts mean something substantive about interoception or are belief-change phenomena is the operational question the awareness dimension is meant to settle.

A reading order§

For someone coming to the field cold and willing to do the work, the order that produces the best understanding rather than the fastest is roughly the following.

Start with Augustine's 1996 anatomy review to register the cytoarchitectonic substrate the rest of the field builds on. Read Craig's 2002 How do you feel? for the founding modern statement and Sherrington's reframed term. Read Craig 2009 for the extension into consciousness territory and to see what the post-2013 predictive-processing accounts are arguing against.

Read Critchley 2004 for the operationalisation, and Garfinkel 2015 immediately afterward to register what the 2004 framework was under-disambiguating. Read Naqvi and Bechara 2007 for the causal evidence that the insula matters.

For the predictive-processing alternative, read Seth 2013 first for the theoretical statement and Barrett and Simmons 2015 for the architectural-empirical companion. For the conceptual prehistory, read Descartes' Error for the somatic-marker frame on which Craig built; the journal-paper versions of the Iowa programme are more careful but less accessible.

For the consensus orientation, the 2018 Khalsa Roadmap is the field's standard reference. For the popular introductions, Seth's Being You (2021), Barrett's How Emotions Are Made (2017), and Damasio's The Strange Order of Things (2018) are the three principal trade books. Each is written from inside a different theoretical commitment, and reading them together registers the field's structure better than reading any one of them alone.