Paulus & Khalsa (2021) — When you don’t feel right inside
The invited American Journal of Psychiatry editorial accompanying Nord, Lawson & Dalgleish (2021) in the same issue. It carries no data, so the wiki holds it not as evidence but as theoretical framing from the field’s clinical-computational centre — Paulus and Khalsa, the roadmap and computational-models authors — writing about what the mid-insula result means. Two of its moves are worth keeping because they go beyond Nord et al.’s own reading.
The mid-cingulate turn: from comparator to action selector
Nord et al. read the dysgranular mid-insula anatomically, as the comparator where descending anterior-insular predictions meet ascending posterior-insular afferents — the natural seat of an interoceptive prediction error. Paulus & Khalsa accept that placement but push the emphasis one synapse downstream: the mid-insula’s “most crucial connection” is to the mid-cingulate cortex, a region for action selection under motivated conditions (pain, reward). So the interoceptive mismatch signal is not just detected — it is fed into feedback-mediated decision-making, generating an early set-shift signal that integrates body-relevant information with the selection of action-outcome behaviours. Their worked example: shallow breathing plus a sudden noise in a dark alley provokes fight/flight; discovering the noise was a cat generates an error signal that should revise the memory, the expectation and the motor plan.
The clinical pay-off of the framing is what the conjunction nulls become. Because “disrupted integration of body-relevant information with action-outcome behaviour selection would not simply map onto affective processing,” it is expected — not surprising — that the locus fails to overlap core affect circuitry or the regions antidepressants and psychotherapy move. The dissociation Nord et al. report as a striking finding is, on this reading, what the theory predicts. See computational-psychiatry.
The two-pathway model: hyperprecise priors vs context rigidity
The editorial’s most portable content is a compact statement of the authors’ own account of how interoception goes wrong (Paulus, Feinstein & Khalsa 2019, “An active inference approach to interoceptive psychopathology”). Interoceptive psychopathology can arise in two ways:
- Hyperprecise priors — abnormally strong expectations about the bodily states a situation elicits. The anxious person meeting a cat in a dark alley does not adjust their fear because the conviction that “shallow breathing + sudden sound = assault” is held with overwhelming precision, swamping the sensory evidence.
- Context rigidity — difficulty updating those expectations when the environment changes. Emerging into a brightly lit open space, the same person keeps feeling threatened because the model is not revised to the new context.
This is the precision-weighting story cashed out as two distinct failure modes, and it is a sharper clinical parse than the roadmap’s five-dysfunction checklist on interoceptive-psychopathology: both pathways are about the prior, but one is a precision fault (too-strong priors) and one is a plasticity fault (priors that will not move). It reframes the Nord et al. disruption map as “a reduced ability to match the model to how the body is feeling, or vice versa, at least at the level of the brain.”
What it concedes
The editorial is candid that the finding is not yet actionable, and names three gaps the wiki already tracks on interoceptive-psychopathology: cause or consequence (is the mid-insular disruption preexisting and predisposing, or a downstream consequence of illness? — longitudinal designs needed); modifiability (can the locus be moved? — a case for focused neuromodulation or real-time fMRI neurofeedback, since existing treatments miss it); and functional heterogeneity (aggregating nociceptive, respiratory, cardiac and touch tasks, the meta-analysis “cannot determine the functional role of the mid-insula” — conscious vs nonconscious processing, intensity decoding, metacognitive insight). This last is Nord et al.’s own domain-general caveat, restated approvingly.
Where it lands
On interoceptive-psychopathology it supplies the two-pathway (hyperprecise-priors / context-rigidity) parse and endorses the mid-insula as a candidate treatment target reachable only by region-specific means. Against Harrison et al. (2021) it inherits the same unsettled question the Nord page raises: Paulus & Khalsa treat the mid-insula-as-error-encoder placement as established, where Harrison’s trial-by-trial data put both prediction and error terms in the anterior insula. The editorial does not engage that tension — it is a framing piece, and it frames the mid-insula reading as the settled one. See martin-paulus, sahib-khalsa.