Uncovering the embodied dimension of the wandering mind

Study at a glance

• Authors: Leah Banellis, Niia Nikolova, Malthe Brændholdt, Melina Vejløe, Ignacio Rebollo, Nicolas Legrand, Francesca Fardo, Jonathan Smallwood, Micah G. Allen
• Year: 2026
• Journal: Proceedings of the National Academy of Sciences 123(13): e2520822123
• DOI: 10.1073/pnas.2520822123
• N: 536
• Design: Large-scale individual-differences resting-state fMRI (14 min) with retrospective multidimensional experience sampling; concurrent PPG/respiration/EGG; EFA + canonical correlation analysis on 216-parcel functional connectivity (CCA subsample N = 489)

Key findings

• Body-wandering is a real and reproducible dimension of the resting mind. Adding 8 interoceptive/somatomotor probes to the 14-item MDES battery reveals stable interindividual variation in how much the resting stream of thought is about the body. Split-half ICCs: 0.937 (item prevalence), 0.907 (cognitive–affect), 0.882 (body–affect).
• It is less frequent than cognitive content, and breathing dominates it. Body-item median 36.3 vs cognitive-item median 47 (W(535) = 113236.5, p < 0.001, rB = 0.574). General body and movement thoughts outrank specific organs; breath is the leading interoceptive item.
• Affective divergence. Resting thought is positive on average, but body items — heart, bladder, skin, stomach especially — correlate with reduced positive and increased negative affect (all p_FDR ≤ 0.001), while classical cognitive/descriptive items (temporal orientation, focus, self/other) correlate with more positive tone. Two thought styles, opposite valences.
• The paradox: body-wandering tracks LOWER psychopathology. Heart-, bladder- and skin-related thought correlates with lower ADHD symptoms (ASRS; rs −0.104 to −0.211) and lower depression (MDI; rs −0.105 to −0.149), all p_FDR < 0.05. Depression instead tracks mental time travel (past and future, rs 0.123 to 0.164). So a thought style that feels bad in the moment goes with fewer trait symptoms.
• Autonomic arousal splits the same way. Skin/bladder thoughts → higher heart rate (rs 0.12–0.15) and lower parasympathetic HRV (RMSSD, rs −0.15 to −0.23); cognitive contents → higher RMSSD. Convergent patterns in respiratory and electrogastrographic metrics.
• EFA (Factor 1, 9% variance): a somato-affective axis. Skin, heart, bladder, stomach and negative affect load positively; positive affect and social-cognitive thought (self, other) load negatively. Body-focused thought and pleasant socially-oriented thought sit at opposite poles of one dimension (split-half r = 0.92). Factor 2 (8%) is mental time travel plus form (split-half r = 0.68).
• CCA: one significant, cross-validated brain–thought mode (out-of-sample r = 0.354, p = 0.001; significant in 4/5 folds). Thought loadings: stomach 0.58, arousal 0.44, breath 0.35, movement 0.30, negative affect 0.29, self 0.26.
• The neural signature is thalamo-somatomotor, not default-mode. Strongest edges link right thalamus and right caudate to bilateral pre- and postcentral gyri (top 1% loadings > 0.55); right thalamus is the hub (53 edges). Insula, cingulate and paracingulate also recruited. This is not the DMN/frontoparietal architecture canonical mind-wandering research describes.
• It aligns with the principal cortical gradient. The mode’s topography maps onto Margulies’ Gradient 1 (unimodal↔transmodal; p < 0.001, spin test, 2,500 permutations) — negative poles at the somatomotor anchor, positive poles at transmodal frontal/cingulate apex. Gradients 2–5 null after Bonferroni.
• Convergent validity. Mass-univariate connectivity regression on EFA Factor 1 scores correlates r = 0.431 (p < 0.001) with the CCA connectivity loadings; Factor 2 (mental time travel) shows negligible overlap (r = 0.061). Supervised and unsupervised routes find the same substrate.

Limitations

• Retrospective, single-probe experience sampling after a 14-min scan. This indexes stable interindividual differences in the relative salience of bodily thought, not the moment-to-moment dynamics of thought — the authors say so. No causal direction between bodily state, arousal and thought content is available.
• Effect sizes are small. The clinical and physiological correlations are rs ≈ 0.10–0.23; the two EFA factors explain 9% and 8% of variance. With 23,220 connectivity edges even trivial correlations reach significance, which the authors flag.
• The scanner is an unusual body. Lying motionless in a narrow tube with minimal sensory input, having been told to empty one’s bladder beforehand, plausibly shapes both bodily state and what one thinks about. Generalization to everyday body-wandering is untested here.
• Item specificity is uneven — ‘my thoughts were about my body’ and ‘about my bladder’ are not comparable in grain, which may drive their different rating distributions and hence their different correlations.
• This is self-reported attention to the body, not interoceptive accuracy. No perceptual task was run. Body-wandering sits on the sensibility side of the taxonomy — see the cross-link to Banellis et al. (2026), where the domain-general quantity turned out to be confidence rather than performance.
• Same cohort family as the decorrelation paper. Both draw on the Visceral Mind Project (Aarhus); the samples are not independent, so the two 2026 Banellis papers should not be treated as separate replications of anything.
• Demographically narrow: median age 24, range 18–56, ~2/3 female, Danish/English-fluent, MRI-eligible university-pool participants. Mostly subclinical symptom variation (2% above the high-depression MDI cutoff).
• The lower-symptom finding is cross-sectional and correlational; the protective interpretation offered in the discussion is explicitly labelled speculative by the authors.

Connections

• Methods: Multidimensional Experience Sampling (MDES), Ecological Momentary Assessment (EMA), Heart rate variability (HRV)
• Concepts: Body-wandering, Default mode network, Embodied selfhood, Interoceptive psychopathology, Allostasis, Salience network, Insular cortex, Core affect, Interoceptive taxonomy (Farb et al. Box 1), Respiratory interoception, Cortical somatotopy (the homunculus), Gastric network
• Debates: Is more interoceptive awareness better?, Is interoception domain-general or organ-specific?
• Related studies: Banellis et al. (2026), Allen (2026), Farb et al. (2011), Farb et al. (2010), Zeharia et al. (2019), Quigley et al. (2021), MacCormack et al. (2021), Levakov et al. (2023)

The wiki’s first study of what people think about when nothing is asked of them, and its first to find that a meaningful share of that is the body. 536 participants lay in a scanner for 14 minutes and were then asked, without warning, 22 questions about the stream of thought they had just had — 14 of them the standard multidimensional experience sampling battery, and 8 newly written to ask about breath, heart, stomach, bladder, skin, movement, body and arousal.

Same senior author and same research programme as the decorrelation paper ingested one run earlier, and a very different kind of study: that one asked how well people perceive the body, this one asks how often they think about it.

What was measured

layer	instrument	N
ongoing thought	MDES, 22 items, 0–100 VAS, retrospective, order randomized	536
brain	14-min resting-state fMRI; Schaefer-200 + 16 subcortical parcels → 216-node connectivity	489
body	photoplethysmography, respiration belt, electrogastrography	subsets
symptoms	ASRS (ADHD), MDI (depression); PHQ-15 and STAI-T as controls	536

Four analytic stages, deliberately escalating: item-level correlations → EFA → CCA against whole-brain connectivity → convergence check between the unsupervised and brain-constrained solutions.

The finding, and the paradox inside it

Body-wandering is affectively negative. Thoughts about the heart, bladder, skin and stomach travel with reduced positive and increased negative affect, and with higher heart rate and lower parasympathetic HRV. Thoughts about the past, the future, other people and one’s focus travel with the opposite of all of that. EFA turns this into a single bipolar axis: visceral-and-negative at one pole, social-and-pleasant at the other.

And body-wandering goes with fewer symptoms. Heart, bladder and skin thoughts correlate negatively with both ADHD and depression scores. Depression’s positive correlate is instead mental time travel — past and future — which is what the rumination literature would predict.

Set those side by side and the naive reading fails. A state that is unpleasant in the moment and physiologically aroused is associated with less trait psychopathology than a state that feels pleasant and calm. The authors’ own interpretation, which they label speculative: attention to bodily sensation “reflects preserved engagement with ongoing sensory input,” and such engagement may be protective relative to forms of mind-wandering that are decoupled from the present body. Negative affect during rest is not the same thing as maladaptive cognition.

This should be read against the wiki’s most closely related result. Farb et al. (2011) found that elaborative self-referential reactivity predicts depressive relapse while sensory reactivity protects, and felt sadness predicts nothing — which is the same shape: what harms is not how bad it feels but whether the mind is elaborating away from the present or staying with it. Banellis et al. supply that dissociation at rest, in a large unselected sample, with the sensory pole finally interoceptive rather than visual. See Is more interoceptive awareness better?, where this becomes the page’s first row that separates momentary valence from trait outcome.

The brain result is the surprising one

Mind-wandering research is a DMN literature. The canonical account has self-generated thought arising from DMN dynamics with frontoparietal control and salience networks, and treats mind-wandering as decoupling from sensorimotor input.

The CCA finds something else. The single significant cross-validated mode (out-of-sample r = 0.354) is anchored on thalamus and striatum coupling to primary somatosensory and motor cortex, with right thalamus as the dominant hub (53 high-loading edges) and insula/cingulate recruited alongside. Body-wandering is not sensorimotor disengagement; it is sensorimotor integration during internal experience.

The gradient analysis makes this precise rather than merely contrarian. The mode’s topography aligns with Gradient 1 — the principal unimodal-to-transmodal axis (Margulies et al. 2016) — with its negative pole at the somatomotor anchor and its positive pole at transmodal frontal/cingulate cortex. Body-focused thought bridges the hierarchy rather than sitting at either end of it.

Two reasons to take the connectivity result more seriously than a single multivariate fit usually deserves: it held in 4 of 5 cross-validation folds, and the entirely separate unsupervised EFA factor’s connectivity map correlates r = 0.431 with it, while the mental-time-travel factor’s map does not (r = 0.061). Supervised and unsupervised routes converge on one substrate.

The methodological divergence worth recording

The EFA and the CCA do not pick the same body. Unsupervised, the loading order is skin, heart, bladder, stomach — the organs, opposed to social thought. Brain-constrained, it is stomach, arousal, breath, movement — and the opposing social pole largely disappears (other = 0.20, positive affect = 0.14, both weakly positive).

The authors treat this as partial overlap between related constructs, and the convergence analysis supports that. But it is worth naming what changed: letting the brain vote reweights body-wandering toward the gastric, respiratory and arousal channels and away from the bipolar body-versus-social structure. For a wiki that has been sceptical of single-channel generalization, this is a useful demonstration that which interoceptive channel appears “central” can be an artefact of the analysis chosen.

Where it sits in the wiki

It gives the resting state a body. Every fMRI page here treats rest as a baseline. This paper’s own framing — the Significance statement says as much — is that the resting state has been treated as a purely cognitive baseline while the participant is in fact lying in a tube having a bodily experience. Given how much of this wiki’s neuroanatomy rests on resting-state connectivity (Salience network, Default mode network, Central autonomic network), that is a general caution and not only a finding.

It is a sensibility measure, and the sibling paper says what that costs. Body-wandering is self-reported attention to the body — not accuracy, not precision, not metacognitive efficiency. Banellis et al.’s other 2026 paper found that the quantity which behaves like a person-level trait is confidence, and that performance is organ-specific. Body-wandering is on the trait-like side of that split, which is consistent with it showing stable individual differences (ICC 0.88–0.94) — and which means it cannot be assumed to track any perceptual ability. The two papers should be read as complementary halves: what generalizes across a person is how they relate to the body, not how well they read it. See Interoceptive taxonomy (Farb et al. Box 1).

It does not contradict anything on the wiki. No page here claimed body-focused thought is DMN-based, and no page claimed spontaneous body attention is a marker of psychopathology in a specific direction. What it does is add a construct — Body-wandering — and one genuinely awkward data point to Is more interoceptive awareness better?.

Provenance notes

PNAS Direct Submission, edited by Daniel L. Schacter; received August 2025, accepted February 2026, published 25 March 2026. Open under CC BY-NC-ND. Data and code public (github.com/embodied-computation-group/body_wandering_CCA). Funded by Lundbeckfonden and ERC Starting Grants to Micah Allen and Francesca Fardo; no declared conflicts.

Jonathan Smallwood — second-to-last author, the field’s central mind-wandering figure and the source of the MDES framework this paper extends — is recorded in the byline as deceased, 24 October 2025. He supervised the analysis and co-wrote the paper. No researcher page is created for him here (single appearance in raw/, per the co-author convention), but the MDES lineage the Multidimensional Experience Sampling (MDES) page records is his.