Greenwood & Garfinkel (2025) — Interoceptive Mechanisms and Emotional Processing
The wiki has reconstructed Garfinkel’s dimensional framework almost entirely secondhand — from Quadt et al. (2018) and from Farb’s pushback against her terminology (interoceptive-taxonomy). This Annual Review of Psychology review is her group’s current, first-hand statement of the whole programme, organized around a single question: how does each level of interoception shape emotion? It is a consolidation, not a new result, and it contradicts nothing on the wiki; its value here is that it states the constructs the wiki cites constantly, and it fills two gaps — a bottom-up transduction mechanism, and the clinical conditions (schizophrenia, EUPD) interoceptive-psychopathology had not covered. First-authored by Benedict M. Greenwood (Garfinkel’s group, Institute of Cognitive Neuroscience, UCL).
The eight dimensions, stated first-hand (Figure 2, Table 2)
The review is built on the Suksasilp & Garfinkel (2022) scheme — the “processing-depth” hierarchy the wiki had only glimpsed through Desmedt’s comparison. Here it is laid out in full, each dimension with an emotion example, running from the raw signal to its higher-order interpretation:
- Nature of afferent signals — the signal itself. Emotions manifest as coordinated physiological change across multiple bodily systems (Kreibig 2010), and different organ signatures may matter for different emotions (e.g. gastric signals for disgust).
- Neural representation of afferent signals — how the signal is centrally encoded (HEP, insular reactivity); stronger representations magnify a signal’s influence on emotion.
- Preconscious impact of afferent signals — the effect of afferent fluctuations on the processing of external stimuli without awareness — the cardiac-timing work (fear and threat amplified at systole; Garfinkel et al. 2014, 2021).
- Interoceptive accuracy — objective task performance (heartbeat-detection-task); better accuracy is linked to feeling emotions more intensely (Wiens et al. 2000).
- Self-report and interoceptive beliefs — momentary confidence and enduring trait beliefs; maladaptive beliefs bias perceptions of bodily state and can override veridical signals (a predictive-coding hyperprior).
- Interoceptive insight — the correspondence between self-report and behavioural measures (Garfinkel’s metacognitive awareness); high insight is adaptive, reduced insight tracks anxiety and dissociation.
- Attention to interoceptive sensations — directed attention enhances detection and neural representation, but habitual self-focus can be maladaptive and is tied to reduced granularity (Ventura-Bort et al. 2021).
- Attribution of interoceptive sensations — how nonspecific arousal is causally attributed (Schachter & Singer); catastrophic attribution of benign sensations is the panic engine, and chronic difficulty here is alexithymia.
This is a fifth taxonomy alongside the four already on interoceptive-taxonomy (Farb’s seven, Garfinkel’s three, Khalsa’s eight, Berntson’s seven) — and its relationship to them is instructive: it is Garfinkel’s own three (accuracy / sensibility→“self-report and beliefs” / awareness→“insight”) re-embedded in a fuller stack that foregrounds the signal and its preconscious influence as dimensions in their own right, the same move Quadt et al. made. See interoceptive-taxonomy for how the eight map onto — and collide with — the others.
The bottom-up mechanism the wiki was missing (Jammal Salameh et al. 2024)
The review’s freshest contribution is a transduction pathway. Mechanosensitive ion channels can convert heartbeat-induced pulsations of cerebral blood vessels directly into central neuronal activity — a route by which the cardiac cycle reaches the brain that is distinct from the classical ascending baroreceptor pathway, and that has prompted speculation about a brain-wide network of “heartbeat sentinel neurons” mediating interoceptive modulation of cognition and emotion (Jammal Salameh et al. 2024, Science). This is a mechanistic floor under the cardiac-cycle-effects phenomena the wiki files as “robust but unexplained”: it supplies a candidate how for pulsed influence on cortex.
Afferent manipulation is not sufficient for emotion (Hsueh et al. 2023)
The review uses a pair of rodent results to adjudicate an old question — is a bodily change enough to make an emotion? Optogenetically driving tachycardia in mice produces anxiety-like behaviour — but only in risky contexts (Hsueh et al. 2023, Nature). The heart alone does not manufacture the state; it does so in interaction with the external situation. Greenwood & Garfinkel read this as causal support for two-factor and constructionist accounts (dror-2017-two-factors, theory of constructed emotion) over a pure Jamesian reading: afferent signals are necessary contributors, not sufficient causes. It is the cleanest causal datum the wiki has for the constitution debate, because it manipulates the periphery directly rather than correlating it — the experiment Damasio said in 1996 should have “relatively little impact,” run in reverse and finding an effect that is real but context-gated.
Clinical conditions, emotion-first — including the two the wiki lacked
Where Quadt et al. worked six conditions through the inference machinery, this review works them through emotion, and adds schizophrenia and EUPD — both absent from interoceptive-psychopathology until now.
- Anxiety — reduced resting HRV, anterior insula as a key locus, altered HEP and respiratory-evoked potentials. Accuracy is inconsistent (up in panic, null or down elsewhere); what is stable is the elevated self-reported awareness (BPQ) relative to accuracy — the discrepancy again, and the Harrison et al. ranking (beliefs carry the affective variance) restated.
- Depression — somatic symptoms (appetite, sleep, fatigue), blunted physiological reactivity, reduced dorsal-mid-insula reactivity, and a U-shaped relation to cardiac accuracy (preserved in the highly depressed, possibly via medication). Lower cardiac sensitivity linked to reduced intensity of positive emotion (Dunn et al. 2010).
- Schizophrenia — new to the wiki. Reduced HRV and elevated heart rate (a resting HR > 82 bpm at conscription predicting a 21% higher later risk); reduced cardiac accuracy tracking negative and positive symptoms; altered self–other discrimination and body ownership tied to disturbed insula-based self-representation. Framed as an interoceptive-predictive-coding failure — both hypo- and hyperprecise priors relative to sensory data generating large prediction errors.
- PTSD — higher resting HR, reduced HRV, and a proposed mechanism in which interoceptive signals become conditioned stimuli that trigger avoidance and fear generalization; cardiac-timed fear learning (Garfinkel et al. 2014, 2021) as a candidate body-to-brain amplifier. Accuracy findings mixed and possibly interacting with acute stress.
- EUPD (borderline) — new to the wiki. Higher resting HR/BP, lower HRV and baroreflex sensitivity; resting HEP amplitude positively associated with emotion-regulation ability and reduced in EUPD, with remitted patients intermediate — a candidate state marker. Heterogeneous HEP findings possibly reflecting distinct interoceptive endophenotypes (alexithymia, PTSD comorbidity).
- Autism — emotional differences framed as differences, not deficits; high comorbid alexithymia as a likely driver of the mixed accuracy picture; the accuracy-low/sensibility-high ITPE signature.
The through-line: no single interoceptive failure is diagnostic, different conditions load on different dimensions, and (the review’s Limitations concede) the evidence is cross-sectional and cause-or-consequence is open — the same verdict Chen et al. pass on the whole literature.
Interventions, sorted by which level they target
The most useful synthesis for the wiki’s clinical-training thread is the treatment section, which is organized by which dimension of interoception a therapy acts on rather than by modality: exercise and slow breathing/breathwork (shifting parasympathetic balance); propranolol (dampening the afferent signal); citalopram (modifying gastric mechanosensitivity); flotation-REST (raising interoceptive salience by attenuating exteroception); cardiac-cycle-timed exposure (presenting fear stimuli at systole to potentiate extinction); CO2 / interoceptive exposure for panic (reappraising aversive sensations); CBT (with anterior-insula activity predicting CBT-vs-medication response in depression); mindfulness; and taVNS (enhancing cardiac accuracy and HEP amplitude). The candid close, which the wiki should keep: “we lack the means of predicting which treatment will be most beneficial for a specific patient” — the personalized-medicine aspiration of Khalsa et al. restated, still unmet.
The tension worth flagging
Read against Banellis et al. (2026) — from Garfinkel’s own extended circle — this review states the eight dimensions as person-level properties and rests much of its clinical argument on cardiac accuracy, exactly the two assumptions Banellis et al. found wanting (the constructs do not travel between organs; the evidence base is one channel). Greenwood & Garfinkel are not unaware — their Limitations note that accuracy research “has occurred in the cardiac domain, neglecting other bodily axes” — but the framework as presented predates that caveat’s full force. See is-interoception-domain-general and is-the-heartbeat-counting-task-valid.