Cardiac and Gastric Interoceptive Awareness Have Distinct Neural Substrates

Study at a glance

• Authors: Yusuke Haruki, Kenji Ogawa
• Year: 2023
• Journal: eNeuro 10(1):ENEURO.0157-22.2023
• DOI: 10.1523/ENEURO.0157-22.2023
• N: 31
• Design: block-design fMRI; within-subject interoceptive attention paradigm (heart vs stomach vs visual control), mass-univariate GLM plus MVPA over anatomical insula subdivisions

Key findings

• Attending to the heart and attending to the stomach activate largely OVERLAPPING territory when each is contrasted against a visual-attention control — insula, frontal and parietal operculum, middle cingulate, SMA — replicating the standard interoceptive-attention map.
• But contrasted against each other they dissociate. The right dorsal anterior insula extending into frontal operculum was the ONLY region higher for cardiac than gastric attention (48 voxels, peak 33/17/2, t = 4.65).
• Gastric attention activated far more territory than cardiac: occipitotemporal and medial visual cortices (3793-voxel cluster, t = 8.42), bilateral primary motor and primary somatosensory cortex, left lateral and medial orbitofrontal cortex, fusiform gyrus, dorsomedial and dorsolateral prefrontal cortex, left frontal eye field, and bilateral posterior hippocampus.
• The medial visual territory recruited by gastric attention largely overlapped, by figure comparison, with the resting-state gastric network of Rebollo & Tallon-Baudry (2022) — a network defined by phase-locking to the stomach’s rhythm, here reached by simply asking people to attend to their stomachs.
• MVPA over 12 anatomical insula subdivisions: only the LEFT posterior short gyrus (PSG, dorsal middle insula) classified cardiac vs gastric attention above chance (56.32%, t(30) = 3.39, d = 0.61, FDR p = 0.024). Right ASG, left ASG and left MSG were marginal (ps = 0.089).
• The posterior insula neither discriminated the two conditions nor activated above baseline at all — read by the authors as supporting its role in coding ongoing bodily CHANGE rather than awareness, since attention was deployed without any homeostatic perturbation.
• Mean signal change showed NO main effect of condition in either insula (left F(1,30) = 0.07; right F(1,30) = 0.15, p > 0.70); the dissociation lives in the location x condition interaction. Right ASG and MSG were higher for cardiac (p = 0.021, 0.016); left and right ALG marginally higher for gastric (p = 0.059, 0.051).
• Exteroceptive (visual) attention activated the RIGHT ASG more than interoceptive attention did (F(1,30) = 10.28, p = 0.003), while left MSG and PSG were the interoception-preferring subdivisions.
• Subjective intensity differed by condition (visual highest, stomach lowest), but an ANCOVA removing it left the activation maps essentially unchanged, and no voxel anywhere covaried with rated intensity even at a lenient threshold.

Limitations

• NO physiological recording of any kind — no ECG, no electrogastrography, no respiration. For a paper whose headline gastric result is an overlap with a network defined by the gastric rhythm, not recording that rhythm is a conspicuous gap, and the authors’ defence (Petzschner et al. 2019 found attention does not change cardiac parameters) addresses a different worry than the one it needs to.
• No ground truth and no accuracy measure. This is an ATTENTION paradigm: compliance is verified only by a practice trial, a verbal check, and one post-run Likert rating. Nothing establishes that participants attended to the organ named rather than thought about it.
• The gastric-network overlap is asserted from a figure (Fig. 2B), with no Dice coefficient, no spin test and no null distribution — a weaker standard than the same comparison received in Levakov et al. the same year.
• An untested alternative reading of the central gastric finding: participants lay with eyes open viewing a word, and the gastric > cardiac contrast recruited left FRONTAL EYE FIELD and premotor cortex alongside the visual cortices. Visual imagery or oculomotor differences would produce that profile too, and the paper does not consider imagery.
• Cluster asymmetry is extreme (48 voxels cardiac-preferring vs 3793 gastric-preferring). ‘Gastric engages more brain’ may partly index that stomach attention is the harder, vaguer, less practised task — it was rated least intense of the three.
• Two internal reporting inconsistencies. The Results text twice names the left PSG as the significant classifier while Figure 4’s caption names the left MSG; and the insula ANOVA line reports the same F(5,150) = 39.18 for both hemispheres with duplicated eta-squared values.
• N = 31, single site, all right-handed Japanese adults aged 20-31, supine in a scanner. The authors flag the posture and the sample themselves.
• Cutaneous contribution to felt heartbeat not excluded (acknowledged, citing Khalsa et al. 2009).

Connections

• Methods: Interoceptive attention task, Electrogastrography (EGG)
• Concepts: Insular cortex, Gastric network, Interoception, Interoceptive taxonomy (Farb et al. Box 1), Interoceptive sensitivity, Gut-hippocampal memory, Craving, Core affect, Exteroception Boundary
• Debates: Is interoception domain-general or organ-specific?, inferential?, Is the heartbeat-counting task a valid measure of cardiac perception?
• Related studies: Levakov et al. (2023), Banellis et al. (2026), Craig (2009), Craig (2002), Quadt et al. (2018), Berntson & Khalsa (2021)

Thirty-one people lay in a scanner and were asked, in ten-second blocks, to attend to their heartbeat, to their stomach, or to a word slowly fading from black to grey. That is the whole design. It is the first study to put two interoceptive channels through the same attention paradigm in the same people and compare them directly — and the answer is that the brain does not treat “attending to the body” as one thing.

Why this paper matters more than its size suggests

The wiki arrived at this file immediately after Banellis et al. (2026), which showed that cardiac and respiratory interoceptive performance do not correlate across people. That was a behavioural, individual-differences null. This is the neural, within-subject counterpart: not “your heart score fails to predict your lung score” but “attending to your heart and attending to your stomach are different brain events, in the same head, minutes apart.”

The two results are logically independent — decorrelated individual differences could coexist with a shared substrate, and a shared substrate could coexist with decorrelated scores — and they point the same way. See Is interoception domain-general or organ-specific?.

The finding that costs the wiki something

The right anterior insula is cardiac, not interoceptive-general.

This wiki is built on a claim it inherited from Craig (2009) and has repeated on a dozen pages: the right anterior insular cortex is where bodily state becomes subjective feeling. The evidence for it is overwhelmingly cardiac — heartbeat-detection accuracy correlating with right AIC activation (Critchley et al. 2004; Pollatos et al. 2007; Caseras et al. 2013), a meta-analysis of heart-focused interoception finding right insula the most consistent activation (Schulz 2016).

Haruki & Ogawa put a second organ beside the first and the right AI goes with the heart. It was the only region preferring cardiac to gastric attention. And the authors assemble the corroborating negatives from elsewhere: the accuracy-to-right-AIC correlation does not appear for awareness of breathing (Wang et al. 2019) or of skin conductance (Baltazar et al. 2021).

Their proposed reason is functional rather than anatomical, and it is the best part of the paper’s argument. Cardiac awareness is how a person notices arousal (Paulus et al. 2019). The right insula is where arousal is built: resecting it diminishes physiological and emotional arousal (Terasawa et al. 2021; Holtmann et al. 2022), and false accelerated cardiac feedback both raises perceived arousal (Story & Craske 2008) and activates right AI (Gray et al. 2007; Kleint et al. 2015). So the right AI may look like the seat of interoceptive awareness because the field has been asking it about the one organ that reports arousal. See Insular cortex, False (bogus) physiological feedback paradigm.

A wrinkle that complicates it further, and that the paper reports without dwelling on: right ASG was activated MORE by the visual control task than by interoceptive attention (F(1,30) = 10.28, p = 0.003). The dorsal anterior insula is not straightforwardly an interoceptive region at all — it prefers cardiac within interoception while preferring a demanding exteroceptive task to interoception.

The stomach’s surprise: gastric attention lights up visual cortex

The largest cluster in the study — 3793 voxels, t = 8.42 — is visual cortex, during attention to the stomach. With eyes fixed on a word, no food cue, no gastric stimulation.

The authors’ case that this is real and not artefact rests on convergence:

• The medial visual territory overlaps the resting-state Gastric network (Rebollo et al. 2018; Rebollo & Tallon-Baudry 2022), which is defined by phase-locking to the stomach’s ~0.05 Hz rhythm and which likewise includes occipitotemporal cortex.
• Direct gastric stimulation — electrical or vibratory — evokes occipital activity in rats (Cao et al. 2019), cats (Pigarev et al. 2013) and humans (Mayeli et al. 2021).

Their functional reading: gastric state is for foraging and feeding, which requires integrating visuospatial information with energy status and motor coordination. Consistent with that, gastric attention also recruited left orbitofrontal cortex (most consistently activated region for visual food cues per van der Laan et al. 2011), bilateral posterior hippocampus (see Gut-hippocampal memory), and primary motor cortex.

Take this one more loosely than the insula result. The convergence is genuine but the overlap claim is eyeballed from a figure, not quantified. And the alternative the paper never raises is available: attending to a viscus you cannot feel may recruit visual imagery, and the same contrast activated left frontal eye field and premotor cortex — an oculomotor/imagery signature. Nothing here distinguishes “the stomach is wired to visual cortex” from “people asked to attend to their stomach picture their abdomen.”

Inside the insula: where the two channels come apart

The mass-univariate comparison found a right-anterior cardiac preference. The MVPA found something different and arguably more interesting: a region that does not prefer either channel but encodes which one you are attending to in its multivoxel pattern.

Insula subdivision	What it did
Left PSG (dorsal middle)	Classified cardiac vs gastric at 56.32%, d = 0.61, FDR p = 0.024 — the only significant classifier
Right ASG, left ASG, left MSG	Marginal classification (ps = 0.089); higher mean activation for cardiac in the right ASG/MSG
Left and right ALG (dorsal posterior)	Marginally higher mean activation for gastric (ps = 0.059, 0.051)
Posterior and ventral subdivisions	No classification (ps > 0.27, ds < 0.30) — and no activation above baseline

Two things follow.

First, viscera-specific coding sits in the MIDDLE insula, not the anterior. The authors claim this as a first: the left dorsal middle insula “codes viscera-specific interoceptive awareness.” It fits the posterior→mid→anterior gradient — organ identity is exactly the kind of thing the mid-insula’s integrative stage should carry — but it means the standard reading of that gradient, in which specificity is progressively discarded on the way to a unified anterior feeling, needs the middle stage to be doing more work than the model usually assigns it. It also gives the first direct evidence on the question Chen et al. (2021) posed and nobody had tested: is there one body map in the insula or several overlapping ones?

Note the caveat on effect size: 56% against a 50% chance level. Real, replicable-looking, and small.

Second, the posterior insula did nothing. Neither discriminating nor active above rest. The authors turn this into a positive claim consistent with Craig: the posterior insula codes ongoing physical change in the bodily signal, and in an attention paradigm there is no homeostatic perturbation to code. That is a coherent reading, and it bears on inferential? — but it is an argument from a null in a study with no physiological recording to confirm that nothing was in fact changing.

What it does not show

It is worth being exact, because the paper’s title is broader than its design.

• It measures attention, not awareness, and certainly not accuracy. No detection task, no ground truth, no psychophysics. “Cardiac and gastric interoceptive awareness” in the title is licensed by the assumption — standard in this paradigm, and untested in it — that directing attention to an organ generates awareness of it. See Interoceptive attention task, Interoceptive taxonomy (Farb et al. Box 1).
• It says nothing about individual differences. Every result is a group-level contrast. It cannot speak to whether a person good at one channel is good at another; that is Banellis et al.’s question and it was answered separately.
• It does not close the gastric measurement gap. Is interoception domain-general or organ-specific? needs a gastric psychophysical axis to run the decisive three-channel test. This supplies a gastric neural signature, which is not the same currency.