Nord, Lawson & Dalgleish (2021) — Disrupted dorsal mid-insula activation during interoception across psychiatric disorders
The empirical neuroimaging paper the wiki had been carrying secondhand. “Reduced dorsal-mid-insula reactivity in depression” appears on both insular-cortex and interoceptive-psychopathology via Quadt et al. (2018); this is the study that generalizes that finding across disorders and gives the transdiagnostic frame its single most direct piece of neural support — a preregistered meta-analysis of 33 neuroimaging experiments (626 patients, 610 controls) asking whether disrupted interoceptive activation across psychiatric disorders converges on one place. It does: the left dorsal mid-insula, and nowhere else.
What it did
Coordinate-based ALE meta-analysis. The authors searched bibliographic, neuroimaging and preprint databases through May 2020 for whole-brain fMRI studies comparing a psychiatric group (diagnosis, symptom-dimension score, or recovered/at-risk status) with healthy controls during an interoceptive probe. Interoception was defined by published criteria (tickle, itch, temperature, hunger, thirst, heat, pain, organ integrity), and only one contrast per study was used — the one most specific to interoception — to keep other neural systems out of the result. Thirty-three experiments survived (306 foci), all fMRI, mean ages 16–43. Tasks split into top-down interoceptive attention (heartbeat counting; attentional focus on named organs) and bottom-up perturbation (breathing load, pain, affective touch, hunger). ALE treats each peak as a 3D Gaussian scaled by sample size and tests for above-chance spatial convergence; results were cluster-FWE thresholded at p<0.05.
Three follow-up conjunction/contrast analyses tested the locus against extant meta-analytic maps: core affect circuitry (Lindquist et al. 2016), antidepressant-medication effects (Ma 2015), and psychological-therapy effects (Marwood et al. 2018).
The result: one cluster, on the left, in the middle
A single significant cluster — the left dorsal mid-insula (Z=4.47, peak MNI −36, −2, 14, 928 mm³), robust in the adult-only subsample. Studies falling inside it compared patients with unipolar/bipolar depression, anxiety, remitted anorexia and schizophrenia, using tasks as varied as heartbeat detection, hunger, pain and interoceptive focus. So the convergence is transdiagnostic (across disorders) and domain-general (across interoceptive channels and across top-down/bottom-up probes) — a shared site of disruption where the individual literatures see only their own disorder and their own task.
The two conjunction analyses matter as much as the primary result, because they say what the locus is not:
- Not core affect. Against the largest affect-processing database, there was no overlap. The mid-insula was preferentially engaged by disrupted interoception; core affect instead lit a huge left-hemisphere cluster peaking in the left anterior insula (30,064 mm³). The authors read this as the interoceptive-dysfunction locus sitting lower in the interoceptive hierarchy than the affect network — mid-insula (integration) versus anterior insula (abstract affective re-representation). This is the wiki’s posterior→mid→anterior gradient falling out of an independent meta-analysis: affect anterior, interoceptive disruption mid.
- Not what current treatments move. No overlap with regions changed by antidepressants (which converged on amygdala + globus pallidus) or by psychological therapy. Whatever the mid-insula is doing, evidence-based treatments — assessed mostly with non-interoceptive tasks — do not appear to touch it. The authors turn this into a prospective claim: the mid-insula is a putative novel treatment target, and the field needs treatment studies that actually use interoceptive probes.
Why the mid-insula, specifically: a prediction-error argument
The interpretive core is an anatomy argument. The dysgranular precentral insular gyrus has a hybrid connectivity profile — it projects to and receives from both the agranular anterior insula (which the predictive account makes the source of descending interoceptive predictions) and the granular posterior insula (primary interoceptive cortex). That places it exactly where a comparator encoding the mismatch between predicted and actual bodily state would sit. Nord et al. propose the dysgranular mid-insula as the anatomical candidate for interoceptive prediction error encoding, and a disruption there as a common computational pathway (“higher weighting of prediction errors” in some pathologies, “increased precision of prediction models” in others) that different disorders reach by different routes. See predictive-coding, computational-psychiatry.
This is a specific, testable placement, and it should be read against Harrison et al. (2021), the wiki’s one trial-by-trial test of the same anatomy. Harrison found the anterior insula carrying both prediction and prediction-error terms, with the anterior/posterior dissociation the theory predicts failing to appear. Nord proposes the mid-insula as the error encoder from a wholly different method (convergence of group-difference peaks, no model, no trial-by-trial estimates). Neither settles it: Nord’s is an anatomical inference from a disruption map with no computational parameters in it, and no included study fitted a learning model. The two are complementary bets about where in the insula interoceptive prediction error lives, both currently under-determined by their own designs.
Where it lands in the wiki
On interoceptive-psychopathology. This is the direct answer to the “empty biomarker cupboard” the roadmap conceded: a candidate neural marker of psychopathology and a treatment target, derived transdiagnostically rather than one disorder at a time. It does not by itself resolve the page’s cause-or-consequence problem — a convergent activation difference is as compatible with consequence as with cause — but it is the strongest neural content the transdiagnostic claim has, and it is the version of the Quadt “reduced dorsal-mid-insula reactivity” finding freed from any single diagnosis.
On insular-cortex. It puts the integrative mid-insula, not the anterior seat of feeling, at the centre of interoceptive dysfunction, and its left laterality is a striking rhyme with Haruki & Ogawa’s finding that the left dorsal middle insula is the one subdivision whose multivoxel pattern classifies which organ is attended. Two very different methods — a transdiagnostic disruption meta-analysis and a within-subject MVPA — both single out the left mid-insula as the interoceptively load-bearing part, and both locate the action at the integrative stage of the gradient rather than the anterior one. The affect-versus-interoception dissociation (affect in anterior insula, disruption in mid-insula) also fits Haruki’s deflation that the right AIC may be about arousal rather than the body in general.
On is-interoception-domain-general. Read carelessly, a single convergent locus across cardiac, respiratory, pain and hunger tasks looks like evidence for a domain-general interoceptive system — one place, many channels. But the paper is careful: it could not run domain-specific subanalyses for lack of power, and states plainly that “different interoceptive channels are not necessarily integrated.” So the convergence is at the level of a shared disruption site in patients, not a shared ability in individuals — which is compatible with Banellis et al.’s behavioural decorrelation, since a common site of pathology need not imply a common trait. The single-locus result is neural-convergence evidence, and it does not adjudicate the domain-generality debate that Banellis and Haruki frame at the individual level.
Brakes
Coordinate-based, so it has no effect sizes and cannot weight or meta-regress; the left laterality is possibly a verbal-probe artefact rather than a fact about interoception; it pools interoceptive tasks it admits may not be one thing; no included study measured interoceptive accuracy or any behaviour in the scanner, so this is a map of task-evoked activation differences, not of a measured interoceptive capacity; and by excluding healthy-only and non-psychiatric-somatic studies it cannot say whether the mid-insula is the common locus for every interoceptive-difference condition or specific to the psychiatric ones it examined. Many contributing studies were themselves underpowered (raising false-positive coordinate risk), which cluster-level thresholding mitigates but does not remove — the authors note ALE needs ≥17 experiments to control this and they have 33. And, like everything on interoceptive-psychopathology, it cannot separate cause from consequence.
The accompanying editorial
Paulus & Khalsa (2021) ran an invited editorial alongside this paper in the same AJP issue. Two of its framings go beyond Nord et al.’s own: it pushes the mid-insula’s role one synapse downstream — from prediction-error comparator to input for mid-cingulate action selection, making the conjunction nulls (no overlap with affect circuitry or treatment-responsive regions) the theory’s prediction rather than a surprise — and it reads the disruption through the authors’ two-pathway model (hyperprecise priors vs context rigidity). It treats the mid-insula-as-error-encoder placement as settled, which the Harrison tension above says it is not.