Neural Systems Connecting Interoceptive Awareness and Feelings

Study at a glance

• Authors: Olga Pollatos, Klaus Gramann, Rainer Schandry
• Year: 2007
• Journal: Human Brain Mapping
• DOI: 10.1002/hbm.20258
• N: 32
• Design: Extreme-groups cross-sectional EEG/ERP study with source reconstruction. 16 high vs 16 low interoceptive-awareness subjects (split at a heartbeat-counting score of 0.85, from a whole-sample mean of 0.77) viewed 60 IAPS pictures (20 pleasant, 20 neutral, 20 unpleasant) for 6 s each while 62-channel EEG was recorded, rating valence and arousal on the SAM after every slide. P300 (290-500 ms) analysed by scalp region, then localized by spatiotemporal coupled current density reconstruction (stCDR, a LORETA-family distributed inverse over a template MNI brain) to identify the cortical generators.

Key findings

• High-interoceptive-awareness subjects rated affective pictures as more intense — as an emotion-specific effect. Between-subjects arousal main effect F(1,30) = 4.85, P < .05; Emotion Content x Interoceptive Awareness interaction F(2,60) = 4.02, P < .05. High-IA rated pleasant (5.4 vs 4.6, F(1,30) = 4.2, P < .05) and unpleasant (6.4 vs 5.3, F(1,30) = 9.57, P < .01) pictures as more arousing, but neutral pictures identically (2.6 vs 2.4, ns). The same emotion-specific arousal result as the sister study Pollatos et al. (2005), in a fresh n = 32 sample.
• Nothing for valence — the fourth first-hand appearance of the arousal/valence dissociation. No IA main effect on valence (F(1,30) = 0.20, ns) and no interaction (F(2,60) = 0.60, ns). This is the split Pollatos 2005, Dunn et al. (2010) and Ferré et al. (2024) each found independently. See Core affect.
• P300 amplitude was larger in high-IA subjects. Main effect F(1,30) = 6.03, P < .01 (high 4.47 microvolts vs low 2.98). For pleasant pictures the high-IA enhancement was at anterior-superior (3.71 vs 0.76 microvolts) and medial-superior (7.18 vs 3.78) sites; for unpleasant, at medial-superior (5.40 vs 3.43). The paper does not report whether the P300 main effect is emotion-general or emotion-specific — a gap, given that its sister study found the parallel P300 effect ran across neutral slides too, ‘independent of the affective content.’ See the note below.
• The distinctive contribution: source reconstruction places the differences in first- and second-order structures. stCDR over the P300 window returned five generator clusters — anterior cingulate (BA 32/10), prefrontal (BA 6/8), somatosensory (BA 7/40), right insula (BA 13; Talairach 33, 17, 6), and visual cortices (BA 18/19). High-IA subjects showed greater activation in the first four; the visual cluster showed no IA effect (Table I). This is the source-level anatomy the 2005 sister paper could only infer from scalp topography and other people’s imaging — here it is derived within the dataset.
• The insula effect is unpleasant-specific; the somatosensory effect is emotion-general. Right insula: overall IA effect F(1,30) = 5.10, but on post-hoc breakdown the difference appears only for unpleasant pictures (F(1,30) = 8.61, P < .01; pleasant and neutral ns). Somatosensory cortices: overall F(1,30) = 8.79, and the high-IA enhancement is present for all three contents including neutral (pleasant F = 5.25, neutral F = 4.86, unpleasant F = 8.65). So within one dataset the interoceptive-cortex difference is affect-gated while the somatosensory difference is not.
• The anterior cingulate effect is emotional; the prefrontal effect is general. ACC (second-order): overall F(1,30) = 4.58, high-IA enhancement for pleasant (F = 8.01) and unpleasant (F = 6.99) but not neutral. Prefrontal (BA 6/9): overall F(1,30) = 4.99, enhanced across all three contents (pleasant F = 7.25, neutral F = 7.28, unpleasant F = 5.26). The authors read the two affect-gated structures (insula, ACC) as the interface between interoceptive awareness and feeling, and the two content-general ones (somatosensory, prefrontal) as reflecting a habitually greater bodily/attentional engagement in good perceivers.
• Whole-sample mean heartbeat-counting score 0.77 (SD 0.20); the high group averaged 0.92 (SD 0.04), the low group 0.62 (SD 0.17); no group differences in age or sex. The 0.85 cut-off is the same inherited threshold used across the Munich programme, not derived here.

Limitations

• The body was never measured — the same structural omission as every study in this tradition. ECG was recorded only to score the heartbeat-counting task; no cardiac (or any other visceral) response to the pictures is reported. So the paper’s Jamesian conclusion (‘the perception of bodily states is a crucial determinant for… the subjective experience of feelings’) rests on a chain — good counting → greater perceived bodily change → more felt arousal / more first-order activation — whose middle term is absent from the design. It cannot distinguish a perceptual account from one in which the heartbeat score proxies sympathetic tone. Set out in full on Pollatos et al. (2005) and Olga Pollatos; it applies unchanged here, and the source-level activations do not close it, because a louder/more-aroused heart plausibly produces both larger cortical responses and higher arousal ratings.
• The anatomy is EEG-inverse anatomy. The five generators come from a distributed source reconstruction (stCDR, a LORETA-family L2-minimum-norm solution) over a template MNI brain, not the participants’ own — the EEG inverse problem is ill-posed and low-resolution, and ‘the right insula (BA 13)’ from 62 scalp electrodes is a centroid of a smooth current estimate, not a localization in the fMRI sense. The clusters were retained only if present in >=50% of a group and >=50% of individual maximum strength, which guards against noise but not against the inverse solution’s spatial blur. The result upgrades the 2005 paper’s borrowed anatomy to modelled anatomy — a real gain — but should not be read as though the insula and ACC were measured directly. Compare the fMRI and intracranial insula evidence on Insular cortex.
• Extreme-groups design, so the sample is constructed and the group contrast is not a population estimate. Top and bottom heartbeat-perceivers were contrasted directly; this maximizes power to find a difference and any effect size read off it overstates the population relationship. This is the same design feature that makes the 2005 correlations non-commensurable with Dunn et al.’s unselected null — see that page’s ‘conflict with Dunn’ section, which governs this paper too. Here there are no correlations to inflate (the analyses are group ANOVAs), but the between-groups differences inherit the same caveat.
• Cross-sectional; no manipulation. Nobody was assigned to perceive their heartbeats well. The causal language (‘interoceptive awareness had a great impact on the observed psychophysiological reaction’) outruns the design, as across this literature.
• Heartbeat counting preceded the picture task, so the arousal-rating measure follows an interoceptive-attention induction that may differentially prime good perceivers toward bodily attention — an alternative reading of the arousal difference the paper does not address. Same order, same unremarked confound, as the 2005 study.
• The cardiodynamic confound is live and unaddressed. Stroke volume predicts heartbeat-counting performance (Schandry et al. 1993 — Schandry is this paper’s third author), so ‘high interoceptive awareness’ partly means ‘louder/more-aroused heart.’ A sympathetic-tone account predicts the whole result set — higher arousal ratings, larger P300s, and larger cortical responses in high-IA subjects — with no perception in it. See Heartbeat detection task, Is the heartbeat-counting task a valid measure of cardiac perception?.
• Whether the P300 amplitude effect is emotion-specific is not reported, and it matters. The paper gives the P300 IA main effect (F(1,30) = 6.03) and the per-content post-hocs for pleasant and unpleasant, but not for neutral, so it is impossible to tell from the text whether the scalp P300 difference is affect-gated or (as in the 2005 sister study, explicitly) present across neutral slides too. If the latter, the ‘cortical processing of emotional stimuli’ framing is again looser than the data — a group difference present for household objects is not evidence about emotion. Recorded as an unresolved reading rather than a demonstrated flaw.
• Student sample (mean age 25.3, range 18-36) from one German university, 2006 vintage: no preregistration, no correction reported across the full family of ANOVAs, and nothing here speaks to Age-related interoceptive decline.

Connections

• Methods: Heartbeat detection task, Event-related potentials, Affective picture viewing
• Concepts: Interoceptive sensitivity, Interoception, Core affect, Insular cortex, Anterior cingulate cortex (ACC), Somatic marker hypothesis, Interoceptive taxonomy (Farb et al. Box 1)
• Debates: Is more interoceptive awareness better?, parietal workspace?, inferential?
• Related studies: Pollatos et al. (2005), Dunn et al. (2010), Wiens et al. (2000), Ferré et al. (2024), Berntson et al. (2010), Craig (2002)

The sister paper to Pollatos, Kirsch & Schandry (2005) — same Munich lab, same paradigm (60 IAPS pictures, Schandry heartbeat counting, high vs low interoceptive-awareness groups, 62-channel EEG, SAM ratings) — with one design addition that is the whole reason to hold it separately: source reconstruction. Where the 2005 study had EEG with no source localization and had to import every anatomical claim from other people’s PET/fMRI, this one runs a distributed inverse (stCDR) over the P300 window and derives its own generators. The result is the wiki’s first first-hand source-level evidence that individual differences in interoceptive awareness track activation of the interoceptive cortex.

Thirty-two people counted their heartbeats, then watched 60 IAPS pictures. Good perceivers said the emotional ones were more arousing (not more or less pleasant), had larger P300s, and — on the reconstruction — showed more activation in right insula, anterior cingulate, somatosensory and prefrontal cortex, but not visual cortex.

Not a duplicate of the 2005 paper

Easy to mistake, because the abstracts read almost identically. They are distinct papers: Pollatos, Kirsch & Schandry (2005), Cognitive Brain Research 25:948-962, n = 44, no source localization; Pollatos, Gramann & Schandry (2007), Human Brain Mapping 28:9-18, n = 32, with stCDR source reconstruction and a third author (Gramann, the EEG-source methodologist) in place of Kirsch. Different sample, different journal, different DOI, different central method. The two form a matched pair rather than one study reported twice, and reading them together is the point: the 2005 paper establishes the behavioural + scalp-ERP effect and infers the anatomy; the 2007 paper reproduces the effect in a new sample and models the anatomy.

What the source reconstruction adds, and what it does not

The 2005 page ends by noting that “the anatomy is borrowed” — every structural claim there was inferred from scalp topography plus disagreeing P300-generator studies. This paper is where the Munich group tried to supply the missing step, and it half-succeeds.

The gain. Within one dataset, the P300 to emotional pictures localizes to five generators, and the interoceptive-awareness difference is carried by the four that Damasio’s (1999) model would predict — first-order insula (BA 13) and somatosensory cortices (BA 7/40), second-order anterior cingulate (BA 32/10) and prefrontal cortex (BA 6/9) — while the visual cluster (BA 18/19), a generator with no place in the feeling circuit, shows no IA effect at all. That last null is the quiet strength of the result: the group difference is not a global amplitude offset smeared across every generator, it spares the one cluster that should be spared. See Somatic marker hypothesis for the first-order/second-order scheme this maps onto, and Insular cortex for where the right-insula reading sits in the wider evidence.

The double dissociation the paper half-notices. The four affected clusters split cleanly by content-specificity (Table I):

structure	order	IA effect present for
right insula (BA 13)	first	unpleasant only
anterior cingulate (BA 32/10)	second	pleasant + unpleasant (not neutral)
somatosensory (BA 7/40)	first	all contents incl. neutral
prefrontal (BA 6/9)	second	all contents incl. neutral

So the interoceptive-cortex difference (insula) is affect-gated and specifically negative, the somatosensory difference is emotion-general. The authors’ reading is reasonable: the somatosensory enhancement reflects a habitually greater attention to the body in good perceivers (present whatever is on screen), while the insula’s negative specificity fits Craig’s and Critchley’s picture of the right insula as especially engaged by aversive interoceptive states. The unpleasant-specific right-insula finding is the datum this paper contributes to Insular cortex, and it is congruent with the later, causal Berntson et al. (2010) lesion result that the insula is broadly evaluative.

What it does not add. It is still EEG-inverse anatomy over a template brain — an ill-posed, low-resolution reconstruction, not a measurement. “The right insula (BA 13)” is a centroid of a smooth current estimate from 62 scalp electrodes; the fMRI and intracranial insula evidence the wiki holds elsewhere (Insular cortex) is a different order of spatial claim. The reconstruction upgrades the 2005 paper’s borrowed anatomy to modelled anatomy, which is a genuine step, and no further.

Same dissociation, same blind spot as its sister

Everything conceptual about this paper is already carried on Pollatos et al. (2005), and applies here unchanged:

• The arousal/valence dissociation gets a fourth independent leg. Interoceptive awareness raised arousal ratings for emotional pictures and did nothing to valence (F(1,30) = 0.20, ns) — the same split as Pollatos 2005, Dunn et al. (2010) and Ferré et al. (2024). Four methods, three labs, two decades, one result: the bodily dimension of emotion and the hedonic dimension come apart, and interoception tracks the arousal term specifically. That points at hybrid two-factor accounts over strong Jamesian ones, which makes the abstract’s Jamesian framing an overstatement of its own finding. See Core affect.

• The Dunn disagreement. This is another selected-sample study reporting an arousal main effect of interoceptive accuracy where Dunn et al. (2010) found a null in an unselected sample. The full accounting — why the samples are not commensurable, why an extreme-groups design inflates the contrast, and why this design cannot fit Dunn’s moderation model (the bodily response it turns on was never measured) — is on the 2005 page and tracked on Is more interoceptive awareness better?. This paper does not change that ledger; it adds a second selected-sample instance to the same side of it.

• The body was never measured. The central omission. ECG scored the counting task and was never analysed against the pictures. So the paper localizes cortical generators of a rating difference to the interoceptive cortex without ever showing that good perceivers’ bodies responded differently — the source-level anatomy makes the Jamesian story more vivid without making it more testable. Given the cardiodynamic confound, “these are people whose hearts are louder, whose cortices are more responsive, and who rate everything emotional as more arousing” fits the entire result set, source clusters included, with no perception in it.

In the wiki’s terms

A better-instrumented member of the same pre-Dunn consensus the 2005 paper represents — see Olga Pollatos. Its lasting value here is narrow and real: it is the wiki’s one source that derives, first-hand and within a single dataset, the claim that interoceptive-awareness differences live in the interoceptive and evaluative cortex (insula, ACC) and not in visual cortex, with the interoceptive-cortex effect gated to aversive material. That is worth having on Insular cortex even at the EEG-inverse discount. What it does not do — and cannot, without a bodily measure — is show that the perception of the body is what drives any of it.