Mindfulness meditation training alters cortical representations of interoceptive attention

Study at a glance

• Authors: Norman A. S. Farb, Zindel V. Segal, Adam K. Anderson
• Year: 2013
• Journal: Social Cognitive and Affective Neuroscience (SCAN)
• DOI: 10.1093/scan/nss066
• N: 36
• Design: fMRI, cross-sectional two-group: graduates of an 8-week MBSR course (MT, n=20) vs an untrained waitlist control group (n=16), scanned once. Block design contrasting interoceptive attention (IA — ‘Breathe’, monitoring respiratory sensation) against exteroceptive attention (EA — a 1-back ‘Maintain’ task and a ‘Suppress’ task on visual trait words). Anatomical insula-subregion ROI analysis, whole-brain ANOVA, psychophysiological-interaction (PPI) connectivity from an interoceptive (posterior insula) and a DMPFC seed, MBSR practice-compliance covariate analysis, and voxel-based morphometry, with respiration-belt recording modelled as a nuisance regressor

Key findings

• TRAINING shifted interoceptive attention toward the DORSAL ANTERIOR insula. A three-way ROI interaction (location × attention × group, F(7,238)=3.02, p<0.005) was driven by greater MT responses to IA in dorsal anterior insula — the accessory gyrus (F(1,34)=4.50, p=0.041) and anterior short gyrus (F(1,34)=7.14, p=0.012) — regions the untrained group was instead more responsive to during EA. Read as MT integrating interoceptive sensation with external context in anterior ‘dysgranular’ insula, distinct from the primary posterior map.
• DMPFC was the ONLY region where attention (IA vs EA) interacted with group in the whole-brain analysis (BA 8, x=-3 y=27 z=51, k=79). The untrained group activated DMPFC equally in IA and EA; the MT group DEACTIVATED it during IA (t(19)=6.82, p<0.001). MT reduced recruitment of a putative executive/midline control region specifically when attending to the body.
• MT altered DMPFC↔interoceptive-cortex connectivity. In PPI, the MT group showed IA-specific NEGATIVE DMPFC–right-posterior-insula connectivity (t(1,34)=3.45) that the untrained group lacked, and negative DMPFC–left-insula connectivity in MT but not untrained — a training-related dissociation limiting unintentional interoceptive suppression when DMPFC is engaged for executive work. Independent of attention, MT also had higher connectivity between the posterior-insula interoceptive seed and the right middle putamen / anterior short gyrus.
• Enhanced propagation of interoceptive signal, present by default in MT. Untrained participants showed IA-SELECTIVE increases in connectivity between posterior and more anterior insula zones (they had to invoke IA to raise intra-insula coupling); the MT group showed the same elevated posterior↔middle-insula connectivity as a TASK-INDEPENDENT baseline — read as MT possessing a standing ‘online’ representation of body awareness that the untrained can only voluntarily invoke.
• PRACTICE COMPLIANCE, not group membership, drove the POSTERIOR insula. Within the MT group, % of prescribed daily MBSR practice completed (mean 78.6±13.4%) predicted greater IA-related activation in right sensorimotor cortex and posterior insula, and reduced activation in left pulvinar/LGN and visual cortex. Interoceptive bias (IA−EA) in the right posterior long gyrus correlated with practice completed (r=0.61, p_corrected<0.05). So training raised anterior-insula recruitment across the MT group, but the primary posterior interoceptive cortex responded to practice DOSE.
• Respiration differed with training and was controlled. The MT group slowed respiration ~0.05 Hz (~3 breaths/min) and breathed deeper (greater respiratory volume) during IA specifically — a group×attention interaction (rate F(1,25)=6.62, p=0.016; volume F(1,25)=8.33, p=0.008) — while respiratory efficiency (RV/T) did not differ. Respiratory rate/volume were modelled as nuisance regressors so the BOLD effects are not artefacts of how the two groups breathed.
• Structural: VBM found greater gray-matter volume in the MT group in the LEFT CAUDATE nucleus (x=-15 y=8 z=22, k=362), consistent with altered habitual direction of attention; NO gray/white-matter difference in the anatomical insula ROIs or functional seeds, so the functional training effects were not driven by structural volume.

Limitations

• CROSS-SECTIONAL, not longitudinal: MBSR graduates were compared with a separate waitlist group scanned once, so ‘training’ effects are between-group differences and the design cannot show that MBSR CAUSED them (self-selection, expectancy, and pre-existing differences are uncontrolled). The authors state this outright and call for a pre/post longitudinal study. This is the same design caveat as Farb et al. (2007) and Farb et al. (2010).
• The interoceptive channel is RESPIRATION only (‘Breathe’), chosen because it is directly attendable and drivable; whether the training effects generalize to non-drivable channels (cardiac, gastric) is untested — see Respiratory interoception on why the breath is an exceptional rather than representative channel. And the MT group’s slower/deeper breathing during IA is itself a training effect, controlled statistically but confounded in principle with the attentional one.
• No objective interoceptive ACCURACY task was administered — a breath-monitoring pilot (7 non-study participants, 97% accuracy) established the task is doable, but the study measures cortical recruitment and connectivity during attention, not whether MT improved detection. So it speaks to the ‘right construct, wrong metric’ side of contemplative training increase objective interoceptive accuracy? without closing it.
• Only 7 of 16 untrained participants had complete respiration data (vs all 20 MT); analyses were run on both the full cohort and the reduced-respiration sample, with results reported only where both converged — but the respiratory control is thinner in the control group than the design assumes.
• The paradigm has no ground truth for whether or to what participants attended (Interoceptive attention task); compliance rests on instruction, a pilot and self-report. Small cells (16 untrained, 20 MT); the anterior-insula ‘integration with external context’ reading is an interpretation of a group×attention interaction, not a manipulation of the insula.

Connections

• Methods: body-focused contemplative training, Interoceptive attention task
• Concepts: Insular cortex, Respiratory interoception, Interoception, Default mode network, Narrative and experiential self-reference, Lamina I spinothalamocortical pathway, Exteroception Boundary, Salience network
• Debates: contemplative training increase objective interoceptive accuracy?, inferential?, Is more interoceptive awareness better?, Is interoception domain-general or organ-specific?
• Related studies: Farb et al. (2013), Farb et al. (2007), Farb et al. (2010), Farb et al. (2015), Haruki & Ogawa (2023), Weng et al. (2021)

The training companion the wiki had been promising itself for two ingests. When the raw/papers/Farb/ exclusion lifted, the Cerebral Cortex attention paper (Farb et al. (2013)) came in first, and its disambiguation note ended: “the companion — Farb, Segal & Anderson (2013), SCAN, ‘Mindfulness meditation training alters cortical representations of interoceptive attention’ — is the training paper, and is still pending in the ingest queue.” This is that paper. It is the source of the “mindfulness training increased posterior insula activation” phrasing the wiki inherited secondhand through Weng et al. (2021) and provisionally reattributed on Insular cortex and Respiratory interoception — and reading it first-hand sharpens that claim in two ways the secondhand note got slightly wrong (below).

Norman Farb is this wiki’s author (see Norman A. S. Farb); recorded as provenance, not weighted as authority, per the convention across the Farb corpus.

The design, and how it differs from its Cerebral Cortex twin

The two 2013 papers share a scanner paradigm — the same Breathe / Maintain / Suppress block design, contrasting interoceptive attention (IA, monitoring the breath without altering it) against exteroceptive attention (EA, a 1-back task and a suppression task on visual trait words) — but they ask different questions of it:

	[[farb-2013-attentional-modulation|Cerebral Cortex 2013]]	SCAN 2013 (this paper)
question	does attention (IA vs EA) select which cortex represents the breath?	does training (MBSR vs waitlist) alter that representation?
sample	27 adults, MBSR completion a nuisance covariate	36 adults split into MT graduates (20) vs untrained waitlist (16)
key contrast	IA vs EA, within-subject	MT vs untrained, between-groups, ×(IA vs EA)
headline structure	posterior insula, VMpo thalamus, graded posterior→anterior tuning	dorsal anterior insula, DMPFC deactivation, DMPFC–insula connectivity

The clean disambiguation matters because the wiki’s secondhand summary had merged the two. The insula↔posterior-ventromedial-thalamus (VMpo) connectivity finding belongs to the attention paper, not this one; this paper’s connectivity story is about DMPFC and the putamen, not the thalamic relay. And the “training raised posterior insula” claim is here — but as a practice-dose effect, not the group-level effect, which was in the anterior insula. See the correction filed on Insular cortex and body-focused contemplative training.

The group-level training effect is in the anterior insula

The anatomically partitioned insula ROI analysis (eight gyri per hemisphere, after Craig 2009) gave a three-way interaction (location × attention × group) driven by the MT group responding more to IA in the dorsal anterior insula — the accessory gyrus and anterior short gyrus — where the untrained group was instead more responsive during exteroceptive attention. Farb et al. read this as training recruiting the anterior “dysgranular” insula, where interoceptive afference is integrated with higher-order contextual and external information, rather than the posterior primary map. It fits their broader claim that MBSR develops interoceptive awareness that is integrated with present-moment context rather than bare sensation — the anterior insula doing the binding that the attention paper assigns it (“not an area of pure body awareness”).

The DMPFC is where training shows up most cleanly

In the whole-brain analysis the dorsomedial prefrontal cortex was the only region where attention interacted with group. Untrained participants activated DMPFC equally whether attending in or out; MT participants deactivated it during interoceptive attention (t(19)=6.82, p<0.001). The reading Farb et al. give is continuous with the dual-mode framework their 2007 paper founds: DMPFC deactivation during IA marks a departure from both mind-wandering and effortful focal attention — a shift out of evaluative cortical-midline processing toward an “expansive and diffuse” sensory attention. It is the midline disengagement of the dual-mode story, appearing here for interoceptive attention specifically.

The connectivity analysis makes the DMPFC finding mechanistic. The MT group showed IA-specific negative connectivity between DMPFC and the right posterior insula (t(1,34)=3.45) that the untrained group did not — training builds an interoceptive network in which engaging DMPFC for executive work does not automatically suppress primary interoceptive cortex. Farb et al. propose this as “an emergent prefrontal pathway through which MT alters IA”: whether DMPFC deactivation causes sustained posterior-insula activation or vice versa is left open (PPI is correlational), but the negative coupling is present during IA and absent during EA, so it is context-specific rather than a blanket suppression.

Practice dose, not group membership, moved the posterior insula

This is the finding the wiki most needs stated precisely, because the secondhand version flattened it. The MT-vs-untrained group effect was anterior. The posterior insula — the primary interoceptive cortex of Craig and the attention paper — moved with how much people actually practised. Within the MT group, percentage of prescribed daily MBSR practice completed predicted greater IA-related activation in right sensorimotor cortex and posterior insula, and reduced activation in the left pulvinar/LGN and visual cortex (less recruitment of the visual pathway during body-focus). Interoceptive bias (IA−EA) in the right posterior long gyrus correlated with practice completion at r=0.61. So the training’s effect on primary interoceptive cortex is graded by dose — which is a stronger, more specific claim than “training raised posterior insula,” and the one the wiki should carry.

The distinction Farb et al. draw from it: participating in MBSR at all was enough to raise anterior-insula interoceptive integration (connectivity), but only daily practice raised the amplitude of the primary posterior signal. Integration is cheap; signal gain is earned.

What the wiki gains, and the two corrections

1. The training claim is now first-hand and correctly attributed. Insular cortex and Respiratory interoception had held “training raised posterior insula + VMpo connectivity (Farb 2013)” as a Weng-sourced summary, flagged as belonging to “the SCAN companion, still pending.” Read first-hand: the VMpo connectivity is the attention paper, and the posterior-insula training effect is a practice-dose effect, with the group-level effect sitting anteriorly. Both corrections are filed on those pages.
2. It supplies the training half of the dual mode’s interoceptive generalization. 2007 showed training decouples the right insula from the vmPFC self-model during self-reference; this shows training decouples (negatively couples) DMPFC from the posterior insula during breath attention, and builds a standing “online” posterior↔middle-insula connectivity the untrained can only voluntarily invoke. Same architecture — training changes the organization and default availability of interoceptive processing, not (as far as this paper can say) raw sensitivity — now demonstrated for interoceptive attention rather than trait self-reference.

Where it sits in the debates

• contemplative training increase objective interoceptive accuracy?. This is the “right construct, wrong metric” pattern in neural form: training demonstrably alters cortical recruitment and connectivity during interoceptive attention, but no accuracy task was run, and the design is cross-sectional. It is evidence that something interoceptive changes with training — attentional integration, network organization, practice-dosed posterior signal — without being evidence of an accuracy gain, exactly the position the Farb side of that debate occupies (and which Eusebio et al. 2022 later complicates by showing self-report doesn’t discriminate under randomization).
• inferential?. DMPFC deactivation with negative DMPFC–posterior-insula coupling during IA is naturally read as top-down modulation of a low-level channel — precision-weighting language — but the paper frames it as recruitment/suppression rather than prediction, and cannot adjudicate the architecture.
• Is more interoceptive awareness better? / Is interoception domain-general or organ-specific?. More training (and more practice) means more anterior-insula integration and more posterior-insula gain — a “more is better” datum on the wellbeing side — but bounded to the one voluntarily-drivable channel, so its reach to the heart and gut is exactly the open question those pages hold.

Provenance and funding

Zindel Segal (MBCT co-developer) and Adam K. Anderson recur across this group’s work. The paper declares no conflicts; funded by CIHR (#MT81164), NIMH (#MH066992), and a Mind and Life Institute Varela grant.