Banellis L., Rebollo I., Nikolova N., Allen M. Stomach–brain coupling indexes a dimensional signature of mental health. Nature Mental Health, 3, 899–908 (2025)

Авторы: Banellis L. / Rebollo I. / Nikolova N. / Allen M.

Stomach–brain coupling indexes a dimensional signature of mental health

¹ Center of Functionally Integrative Neuroscience, Aarhus University, Aarhus, Denmark.
² Department of Decision Neuroscience and Nutrition, German Institute of Human Nutrition, Nuthetal, Germany.
³ Neuroscience Research Center, Charité-Universitätsmedizin Berlin, Corporate Member of Freie Universit?t Berlin, Humboldt-Universit?t zu Berlin, and Berlin Institute of Health, Berlin, Germany.
⁴ German Center for Diabetes Research (DZD), Neuherberg, Germany.
⁵ Cambridge Psychiatry, University of Cambridge, Cambridge, UK.
⁶ These authors contributed equally: Leah Banellis, Ignacio Rebollo.
^* e-mail: leahbanellis@cfin.au.dk


Visceral rhythms orchestrate the physiological states underlying human emotion. Chronic aberrations in these brain–body interactions are implicated in a broad spectrum of mental health disorders. However, the relationship between gastric–brain coupling and affective symptoms remains poorly understood. Here we investigated the relationship between this novel interoceptive axis and mental health in 243 participants, using a cross-validated machine learning approach. We find that increased frontoparietal brain coupling to the gastric rhythm indexes a dimensional signature of poorer mental health, spanning anxiety, depression, stress and well-being. Control analyses confirm the specificity of these interactions to the gastric–brain axis. Our study proposes coupling between the stomach and brain as a factor in mental health and offers potential new targets for interventions remediating aberrant brain–body coupling.


Far from being a mere brain in a vat, the nervous system is embedded within an intricate web of visceral rhythms. While philosophy has long championed the embodiment of mind and life [1], it is only more recently that the importance of the visceral body in contextualizing brain function has gained widespread recognition [2,3]. In particular, interoceptive processes linking brain and body are thought to be critically important in mood and emotion [2,4], and their larger role in affective symptoms has become a topic of intense interest in mental health research [5].

Research investigating these links has so far focused almost exclusively on the cardiac [6,7], lower gastrointestinal [8,9] and respiratory axes [10–12]. Recent landmark findings demonstrate that, for example, anxiety responses in threatening situations rely on ascending cardiac information [6] and that respiratory rhythms modify neural patterns during emotional processing [13]. Simultaneously, the burgeoning study of gut–brain interaction has produced a plethora of new findings linking the function and biology of the lower gastrointestinal tract or gut to physical and mental health [14–16]. While these and many other findings herald a visceral turn in our understanding of the biology of the mind and its disorder, one particular domain of brain–body coupling remains notably understudied: the upper gastrointestinal tract comprising interconnections between the brain and stomach.

Gastric–brain interactions have recently emerged as a novel frontier in interoception research [17–22]. Hormones secreted by the stomach directly regulate hypothalamic mechanisms that govern satiety and hunger [23]. In addition, the stomach generates its own independent myoelectrical rhythm, in which the interstitial cells of Cajal pace muscular contractions approximately once every 20 s (ref. 24). Previously relegated to merely driving mechanical food digestion, recent discoveries indicate that the gastric rhythm is closely linked to ongoing brain activity through reciprocal vagal innervation [25–27]. This link can be directly modulated through techniques such as noninvasive vagal nerve stimulation [28], bilateral vagotomy in rodent models [29] and emerging pharmacological methods [22,30,31], offering a promising means by which to intervene upon the stomach–brain axis.

Despite the close linkage of emotion and brain–body interaction, the extent to which alterations in the gastric axis relate to mental health remains unclear. This gap is curious in part because folk psychology has long centered the stomach as a locus of stress and anxiety: difficult decisions are described as evoking ‘gut feelings’ that, in extreme cases, can make one ‘sick to their stomach’. Conversely, intense moments of love or joy are described as giving ‘butterflies in the stomach’. Aligning with these descriptions, recent empirical findings indicate that individuals often report subjective disgust, fear and anxiety as being localized in the stomach [32,33] and that pharmacological modulation of the gastric rhythm alters emotional processing [22]. On this basis, we hypothesized that interindividual patterns in gastric–brain coupling might expose unique patterns of affective mental health, in particular those relating to core mood disorders such as anxiety and depression. To test this hypothesis, we conducted a largescale neuroimaging study of simultaneous electrogastrography (EGG) and functional magnetic resonance imaging (fMRI) brain imaging in 243 participants. To assess mental health across a broad spectrum, we used a multidimensional approach to quantify highly individualistic profiles spanning a range of affective, cognitive, social and somatic health dimensions. This approach builds on research identifying mental illness as a continuum of overlapping symptoms across disorders [34– 39], rather than relying on discrete diagnostic categories with high comorbidity, high heterogeneity and poor reliability [40,41]. Utilizing a multivariate, cross-validated machine learning technique, we estimated highly robust, sensitive and specific signatures that interrelate these profiles to individual patterns of stomach–brain coupling. Our findings demonstrate that the stomach–brain axis exposes a positive-to-negative mode of mental health [37,42], revealing a previously unknown embodied target for future clinical intervention research.

Results

Mental health functional correlates
To characterize variation in the gastric–brain axis, we utilized simultaneous EGG and resting-state fMRI recordings. Following an extensive data quality control procedure (Methods), we estimated stomach–brain coupling for each participant using the phase-locking value (PLV) of the EGG and resting-state fMRI time series across 209 parcellated wholebrain regions [43] (see Fig. 1 for an example). 


Fig. 1 | CCA of stomach–brain coupling and mental health. The process and outcomes of correlating stomach–brain phase coupling with mental health, as quantified by 37 variables from 16 validated surveys. The top left quadrant presents these variables organized into their respective mental health categories (categorized for visualization only, the CCA incorporated 37 individual scores), and their distribution is visualized as histograms on the bottom left, reflecting the range of participant mental health profiles. EGG data depicted on the top right demonstrates the extraction of gastric cycle frequency from raw EGG signals, power spectra and their phase information, essential for identifying stomach–brain coupling. The middle right figure illustrates coupled versus uncoupled states in stomach–brain interaction, with the individual variability in coupling strength highlighted across three brain images from individual participants (plotted on a standard mni152 brain template using MRIcroGL: visualized Montreal Neurological Institute (MNI) coordinates plotted: 28, −19, 26, thresholded at 0.1, and small clusters <1,000 mm3 removed). For the CCA, stomach–brain phase coupling is parcellated over 209 brain regions identified using the DiFuMo atlas, shown on the bottom right. The CCA model, depicted centrally, outputs a stomach–brain signature correlating with mental health individual profiles. This pattern is represented by canonical variates, which are weighted combinations of the multidimensional mental health and stomach– brain coupling data (illustrated as the central scatter plot). These weights, depicted as bar graphs, capture the most notable relationships between gastric– brain coupling and mental health profiles. Created with BioRender.com.


To characterize individual mental health profiles, we conducted a comprehensive assessment across 37 self-reported scores encompassing autism, attention deficit hyperactivity disorder (ADHD), empathy, insomnia, interoception, depression, fatigue, social support, somatic symptoms, stress, social anxiety, trait anxiety, well-being and quality of life (see Supplementary Table 1 for a full list of instruments and subscales). This approach successfully captured robust interindividual variability spanning a variety of mental health dimensions (Fig. 1). Individual variance spanned from completely healthy to those experiencing substantial distress, such that 30% of the sample exhibited mild depression, 19% clinically relevant levels of ADHD, 19% medium or more severe somatic symptoms, 18% trait anxiety, 9% moderate depression, 7% autism spectrum and 5% insomnia (see Supplementary Table 1 for all percentage cutoffs).

Finally, to determine latent patterns interlinking these mental health profiles to stomach–brain coupling, we conducted a crossvalidated canonical correlation analysis (CCA). This method determines maximally correlated patterns between two multidimensional variables (in this case, mental health and stomach–brain coupling data). CCA does this via linear transformation of the inputted data using weights, which produces the resulting CCA variates (that is, weighted sums) (Fig. 1).

We observed a significant latent dimension in which stomach–brain coupling was associated with a positive-to-negative mode of mental health (canonical variate in-sample r(118) = 0.886, out-sample r(77) = 0.323, P = 0.001). This was reflected behaviorally as high negative loadings for trait anxiety (State-Trait Anxiety Inventory (STAI) trait subscale ?0.827), depression (Patient Health Questionnaire-9 (PHQ9) ?0.800 and Major Depression Inventory (MDI) ?0.782), stress (Perceived Stress Scale (PSS) ?0.773) and fatigue (Multidimensional Fatigue Inventory (MFI) general fatigue subscale ?0.734), as well as high positive loadings for well-being and quality of life (highest loadings: Warwick-Edinburgh Mental Well-Being Scale (WEMWBS) 0.856, World Health Organization Quality of Life (WHOQOL) psychological subscale 0.847, World Health Organization Five Well-Being Index (WHO5) 0.776). The top stomach–brain coupling canonical loadings were found in the left superior angular gyrus (?0.317), right intermediate primus of Jensen (right supramarginal gyrus posterior division using the Harvard-Oxford Cortical Structural Atlas) (?0.284), left inferior precentral sulcus (?0.270), left posterior superior frontal gyrus (?0.269) and left posterior intraparietal sulcus (?0.245). These loadings constitute a pattern of stomach–brain coupling in which healthier mental scores (that is, improved well-being and quality of life) are associated with reduced gastric coupling to frontoparietal brain activity (Fig. 2 and Supplementary Table 2) or, conversely, in which more negative mental health scores (that is, increased anxiety, depression and fatigue) are associated with increased coupling. We did not observe significant gender or age differences with this mental health associated stomach–brain coupling result (Supplementary Table 3). 

Fig. 2 | Mental health functional correlate of stomach–brain coupling. CCA results depicting the correlation between stomach–brain coupling and mental health dimensions. This indicated diminished frontoparietal stomach– brain coupling with healthier mental health scores (that is, lower anxiety, depression and stress, and higher quality of life and well-being). Left: the CCA loadings (structure correlations: Pearson’s correlations between raw mental health and stomach–brain coupling variables and their respective canonical variate). Importantly, this represents the pattern of mental health data that is maximally correlated with the stomach–brain coupling canonical variate. High negative loadings are shown for anxiety, depression, stress, fatigue, ADHD, somatic symptoms and insomnia, while high positive loadings are shown for well-being and quality of life. Middle: the top five parcellated regions (DiFuMo parcellation) with the absolute highest stomach–brain coupling loadings (all negative), colored according to their respective CCA loading: left superior angular gyrus, right posterior supramarginal gyrus, left inferior precentral sulcus, left posterior superior frontal gyrus and left posterior intraparietal sulcus (plotted on a standard mni152 brain template using MRIcroGL: Montreal Neurological Institute (MNI) coordinates: ?34, ?3, 48). Right: the cross-validated CCA result denoting the maximally correlated psychological variate and brain–stomach coupling variate (in-sample r(118) = 0.886, out-sample r(77) = 0.323, P = 0.001, two-tailed).


To further summarize our findings, we averaged parcel-level stomach–brain canonical loadings across the Yeo 7-networks [44] and also averaged psychological canonical loadings across mental health categories (to condense the multivariate variables, and to be consistent with the mental health categorization in Fig. 1). This revealed the highest absolute network level stomach brain loadings in the dorsal attention (?0.132), frontoparietal control (?0.100) and ventral attention salience network (?0.061). This pattern of stomach–brain coupling was maximally correlated with the corresponding averaged mental health structure, in particular with strong negative loadings for depression (?0.791), stress (?0.773), anxiety (?0.709) and fatigue (?0.642) and strong positive loadings for well-being (0.815) and quality of life (0.696) (see Fig. 3 and Supplementary Fig. 1 for an averaged summary of the raw weights). 

Fig. 3 | CCA loadings averaged summary. Canonical loadings (structure correlations: Pearson’s correlations between raw inputted variables and respective canonical variate) from the mental-health-associated stomach–brain coupling CCA, summarized via averaging. Note that there are prominent negative average stomach–brain loadings in the ‘dorsal attention B’ network and the ‘control A’ network, associated with reduced average depression, stress, anxiety and fatigue and increased average well-being and quality of life (that is, better mental health). The opposite pattern is also true: increased average stomach– brain loadings in attention and control networks is associated with worse mental health (increased depression, stress, anxiety and fatigue and reduced well-being and quality of life). Left: the stomach–brain loadings averaged according to Yeo 7-networks. Top: these network-averaged stomach–brain loadings projected onto a mask of the DiFuMo atlas regions for each Yeo 7-network (from left to right: DorsAttnB, Dorsal Attention B; ContA, Control A; SalVentAttnA, Salience Ventral Attention A; SomMotA, Somatomotor A; Unassigned, no network found; VisCent, Visual A; DefaultB, Default Mode B; LimbicA, Limbic A)44, plotted on a standard mni152 brain template using MRIcroGL: Montreal Neurological Institute (MNI) coordinates: ?34, ?3, 48. Right: the psychological loadings averaged across mental health categories defined for visualization in Fig. 1.


CCA control analyses
To evaluate the specificity and robustness of these results, we conducted control CCAs predicting mental health scores from either (1) functional connectivity, (2) blood-oxygen-level dependent (BOLD) signal variability, (3) cardiac-related brain maps or (4) respiratoryrelated brain maps, instead of from stomach–brain coupling. In all cases, no significant canonical variate was found (P > 0.05, Bonferroni threshold 0.01), indicating that the dimensional index of mental health reported here is specific to stomach–brain coupling and unlikely to be explained by residual BOLD connectivity, signal variability or cardiac– brain or respiratory–brain influences.

We conducted a further control analysis to determine if a simpler univariate model would yield comparable results as to our multivariate approach. A principal component analysis (PCA) of the mental health scores yielded a highly similar latent structure as that derived from the CCA (r(35) = ?0.999, P < 0.001) (Supplementary Fig. 2). We then correlated this mental health PCA component with stomach– brain coupling values from each Dictionary of Functional Modes (DiFuMo) atlas parcellated region separately. This revealed significant correlations of stomach–brain coupling with the mental health PCA component in nine DiFuMo parcellated regions, all of which were in the top loading regions of the CCA, with the exception of the ‘lateral fissure anterior’ (Supplementary Table 4). Furthermore, across all brain regions, loadings from the stomach–brain coupling CCA were highly correlated with the univariate correlation coefficients linking mental health to stomach–brain coupling (r(207) = ?0.870, P < 0.001) (Supplementary Fig. 2). These results complement our cross-validation procedure by demonstrating that the multivariate CCA model detected effects could be reproduced with a simpler, albeit less sensitive, model, with the CCA explaining more variance (in-sample R² = 0.785, out-sample R² = 0.104) than the significant univariate correlations (mean R²= 0.033).

We also controlled for whether the mental-health-associated stomach–brain coupling result was driven by gastrointestinal symptoms by removing the somatic symptoms survey (PHQ15) from the CCA and instead including the PHQ15 survey as a nuisance regressor. The CCA result persisted without the somatic symptoms survey with highly replicable loadings (Supplementary Fig. 3). Furthermore, when completing the CCA with the somatic symptom survey items only, or the gastrointestinal symptom items only, the results were not significant (all somatic symptom survey items smallest P = 0.428, gastrointestinal symptom survey items only smallest P = 0.267).

EGG control analyses
To control for possible low-level physiological confounds, we further estimated the association between the psychological canonical variate loadings and summary EGG metrics, using nonparametric Spearman’s rank order correlation coefficients (see Supplementary Table 5 for EGG metric descriptive statistics). We did not observe a significant correlation of normogastric EGG activity measured via the proportion of normogastric power, maximum power or peak frequency with the observed mental health canonical variate (smallest false discovery rate-corrected P = 0.267). Therefore, the link between stomach–brain coupling is specific to the strength of brain–body coupling with mental health, rather than being explained by baseline differences in peripheral gastric physiology.

Discussion

Our study reveals a distinctive stomach–brain signature of mental health, established through cross-validated multivariate regression techniques and control analyses. This signature encapsulates a positive-to-negative latent dimension of mental health, with notable negative loadings on anxiety, depression, stress and fatigue, as well as enhanced well-being and quality of life. Notably, all 20 of the highest loadings for stomach–brain coupling were negative, indicating a direct correlation between diminished gastric–brain coupling and improved mental health (refer to Supplementary Table 6). Our control analyses confirm that this extensive psychological signature spanning affective, cognitive, social and somatic health dimensions is uniquely attributable to stomach–brain coupling, distinct from factors such as residual brain connectivity or variability, cardiac–brain or respiratory–brain influences, gastric activity variations, bodily mass, age or gender differences. However, future research should explore this using a more diverse sample with a balanced representation of age groups and genders.

Our main result reveals a pattern of worse mental health with increased gastric coupling in brain regions that are known transdiagnostic hotspots, such as the posterior superior frontal gyrus and the posterior intraparietal sulcus [45,46]. Notably, the left superior angular gyrus, our model’s most prominently featured region, is crucial for its integrative role in various cognitive functions [47]. This region is also associated with a range of psychiatric disorders, including schizophrenia [48], somatization disorder [49] and major depressive disorder [50,51]. Thus, gastric rhythms may covary with brain activity in neural hubs that are highly sensitive to disruptions in mental health. Recent advancements have discovered numerous transdiagnostic biotypes in affective disorders such as depression and anxiety with varying resting and emotion-evoked connectivity and brain activation profiles [52]. Our findings indicate that there may be brain–body biotypes, but further causal research is necessary. In our study, the brain areas in which psychological health dimensions were most prominently associated with the gastric axis comprised attentional and control network hubs [44]. This indicates that top-down attentional and inhibitory control mechanisms may be particularly important for the relationship of visceral-brain rhythms with mental health. Beyond these control-oriented networks, we also observed a negative association with the ventral salience network [53], weak negative loadings in the somatomotor network [54] and weak positive loadings in the limbic system, emphasizing the multidimensional nature of the signature.

This research substantially advances our understanding of the mental health implications of stomach–brain coupling [18,19,55,56]. While previous smaller-scale studies have linked stomach–brain coupling with bodily shame and weight preoccupation [57], our multivariate CCA approach leverages our extensive sample size to encompass a continuum of transdiagnostic mental health scores. Indeed, our multidimensional mental health variate is comparable to previously observed positive-to-negative axes of well-being across cognitive, affective and lifestyle dimensions [37,42], as well as observed ‘general mental health factors’ that encompass symptoms across numerous psychiatric disorders [35,38,58]. Importantly, although our analyses do not focus on clinical diagnostic categories, our multivariate, psychological dimension-based approach is advantageous in directly assessing highly individualistic mental health profiles across a broad multidimensional spectrum. This aligns with recent paradigm shifting calls for a dimensional schema in mental health with biological plausibility [34,40,41], as dichotomous psychiatric diagnoses are plagued with shortcomings, including arbitrary thresholds for binarization, poor reliability, high rates of diagnosis comorbidity, shared symptomatology across disorders and symptom heterogeneity within disorders [40,41]. Moreover, the cross-validated method we apply here is specifically optimized to estimate the continuous statistical prediction of these dimensions, while also robustly protecting against overfitting [59]. Future work could build upon these results to predict multidimensional psychiatric symptoms based on stomach–brain coupling in controlled clinical samples or longitudinal studies.

Interestingly, we identified trait anxiety as a prominent mental health feature associated with stomach–brain coupling, but a previous study found no such relationship with state anxiety [19]. A key difference between that study and ours was the use of a region-of-interest-based approach in a smaller sample size. Our whole-brain, multivariate method was optimized to detect such effects and probably yielded a substantial improvement in statistical power by estimating latent psychological dimensions directly. It may also be that there are distinct stomach–brain relationships with trait and state aspects of anxiety. Notably, the anxiogenic relationship of stomach–brain coupling is also consistent with a previous report linking state anxiety with intestinal–brain coupling in the insula [60] and rodent research of increased anxiety behaviors when activating gut-innervating vagal afferents [14], as well as research with patients with generalized anxiety disorder demonstrating an increased bodily reactivity and intensity of interoceptive sensations in response to adrenergic stimulation [7]. Furthermore, a recent pilot study revealed that stress increased gastric phase–amplitude coupling with electroencephalographic activity, in contrast to a relaxing biofeedback task [61]. Future research could similarly causally manipulate anxiety or stress to help determine its influence on stomach–brain coupling in fMRI. Complementary approaches could also directly modulate stomach–brain coupling using various emerging interventions as a potential means to remediate anxiety or stress symptoms in patients [17,22,31].

One potential limitation of our study concerns our EGG data exclusion rate, which was somewhat higher than the 20% rate reported in previous EGG literature [28,62]. This increase was driven by an exhaustive quantitative and qualitative quality-control protocol, which may have resulted in higher numbers of excluded participants than in previous, smaller-scale studies. To assuage these concerns, we conducted additional control analyses demonstrating the validity of these procedures (Supplementary Fig. 4). Furthermore, we ensured that the excluded participants did not differ in terms of mental health characteristics, gastrointestinal symptoms or under- or overeating behavior, ruling out the possibility that our quality control may have created a sampling bias that could impact our results (see ‘Electrogastrography peak selection and preprocessing’ section in Methods). While our preprocessing pipeline aligns with prior literature [18,19,55,56], emerging methods based on independent component analysis offer promising alternatives for noise reduction, particularly in datasets with high-density montages. Future work may benefit from applying such approaches to reconstruct the EGG signal from components with a high signal-to-noise ratio in the normogastric range [63].

A growing body of evidence links gastrointestinal dysfunction to various mental health conditions [15,64,65] and highlights the frequent cooccurrence of gastrointestinal dysfunction with affective disorders [15]. By elucidating the multimodal interactions between the stomach and the brain in mental health, our findings provide a starting point for future research on novel diagnostic and therapeutic strategies targeting disordered brain–stomach interactions. This includes not only innovations like noninvasive vagus nerve stimulation, which recent studies have found to modulate stomach–brain coupling [28], but also the exploration of new mechanical [17,66] and pharmacological [22,31] interventions to remedy aberrant stomach–brain interactions. Similarly, future research could leverage newly emerging technologies such as ingestible recording devices to further elucidate the physiological mechanisms linking mental health to the stomach–brain axis [67].

Conclusion

In summary, our study provides a large and comprehensive neuroimaging investigation of brain–body interaction and multidimensional features of mental health. Our results signify a link between increased stomach–brain coupling and poorer mental health across anxiety, depression, stress and well-being dimensions. This finding contributes substantially to multidisciplinary research on brain–body interaction and opens new avenues for therapeutic, diagnostic and classification strategies to improve psychological well-being and mental health.

Methods

Participants
We recruited participants as part of the Visceral Mind Project, a large brain imaging project at the Centre of Functionally Integrative Neuroscience, Aarhus University. We recorded EGG in 380 participants (230 female individuals, 149 male individuals, 1 other gender individual, median age 24 years, age range 18–56 years). As our aim was to apply machine learning to individual differences in mental health, we adopted a participant recruitment strategy that sought to maximize between individual variance from fully healthy to those with scores crossing clinical thresholds. Accordingly, we did not explicitly exclude participants for psychiatric diagnosis, and we recruited participants from a wide range of possible online communities and backgrounds (Supplementary Table 1). These participants did not report any major physical illnesses, or medication beyond over-the-counter antihistamines or contraceptives; furthermore, they reported abstinence from alcohol and drugs 48 h before participation. We acquired participants in two data collection cohorts by advertising on nation-wide participant pools, social media, newspapers and posted flyers. As additional criteria, participants had normal or corrected-to-normal vision and were fluent in Danish or English. Furthermore, we included only participants compatible with MRI scanning (not pregnant or breastfeeding, no metal implants, no claustrophobia and so on).

Participants took part in multiple sessions including fMRI scans, behavioral tasks, physiological recordings and mental health and lifestyle inventories. In this study, we focus on resting-state fMRI data, EGG recordings and a mental health and lifestyle assessment battery to evaluate individual differences in gastric–brain coupling and their link to mental health. Participants were provided with hourly monetary compensation for their participation in the study. The local Region Midtjylland Ethics Committee granted approval for the study, and all participants provided informed consent. The study was conducted in accordance with the Declaration of Helsinki. After the removal of poorquality fMRI and EGG data (see ‘Electrogastrography peak selection and preprocessing’ section), we estimated stomach–brain coupling in 243 participants. Including the mental health scores, a total of 199 full-dataset participants were included in the mental health functional correlate analysis (138 female individuals, 61 male individuals, median age 23 years, age range 18–47 years) (Supplementary Figs. 5 and 6).

Anatomical and resting-state fMRI acquisition
We acquired anatomical MRI and resting-state fMRI data using a 3 T MRI scanner (Siemens Prisma) with a 32-channel head coil. We positioned small cushions around the head to minimize head movement. The participants wore earplugs and were instructed not to move. The resting-state scan included 600 volumes acquired over 14 min using a T2*-weighted echo-planar imaging multiband accelerated sequence (repetition time (TR) 1,400 ms, echo time (TE) 29.6 ms, voxel size 1.79 × 1.79 ? 1.80 mm). An acceleration factor of 4 was used in the slice direction along with a Generalized Autocalibrating Partial Parallel Acquisition (GRAPPA) in-plane acceleration factor of 2. A set of high-resolution whole-brain T1-weighted anatomical images (0.9 mm³ isotropic) were acquired using an Magnetization-Prepared Rapid Gradient-Echo (MP-RAGE) sequence (TR 2.2 s, TE 2.51 ms, matrix size 256 ? 256 ?192 voxels, flip angle 8°, AP acquisition direction).

Physiological recording acquisition
We simultaneously recorded physiological measurements (photoplethysmography, respiratory breathing belt and EGG) during resting-state fMRI. For the EGG recordings, we cleaned the abdomen and applied abrasive gel to remove dead skin and improve the signal-tonoise ratio. Three EGG recording montages were implemented using a Brain Vision MRI-compatible ExG system and amplifier (see Supplementary Fig. 7 for each recording montage consisting of one, three or six bipolar channels). All physiological montages were acquired with a sampling rate of 1,000 Hz, a low-pass filter of 1,000 Hz (with a 450 Hz anti-aliasing filter) and no high-pass filter (d.c. recordings). EGG was recorded at a 0.5 μV per bit resolution and ±16.384 mV range, while photoplethysmography and respiratory recordings were acquired at 152.6 ?V per bit resolution and ±5,000 mV range.

MRI and fMRI preprocessing
We implemented the minimal preprocessing pipeline in fmriprep. MRI and fMRI results included in this Article come from preprocessing performed using fMRIPrep 22.1.1, which is based on Nipype 1.8.5 (see Supplementary Information for anatomical and functional MRI preprocessing details with fMRIPrep). Additional fMRI preprocessing steps following fMRIPrep included spatial smoothing with a 3-mm full width at half maximum kernel and regressing out six motion parameters, six aCompCor parameters and 13 RETROICOR components reflecting cardiac and respiratory physiological noise.

Electrogastrography peak selection and preprocessing
The EGG data were first demeaned and downsampled from 1,000 Hz to 10 Hz for computational efficiency, followed by computing the power spectrum using a Hanning-tapered fast Fourier transform incorporating 1,000 s of zero-padding in 200-s data segments with 75% overlap. For each participant, we selected the bipolar EGG channel that showed the most prominent peak within the normal frequency range of the gastric rhythm in humans (that is, normogastric range: 0.033–0.066 Hz), which is on average one cycle every 20 s (0.05 Hz) [62]. Specifically, two researchers (L.B. and I.R.) independently conducted peak selection by visually inspecting each channel to identify the EGG channel with the highest normogastric power peak, without large artifacts and with power above 5 ?V². Peak quality was rated as ‘excellent’ for Gaussian-like peaks (n = 184) and ‘good’ for shoulder-like peaks (n = 81); those not meeting these standards were deemed ‘poor quality’ (n = 115) and excluded. This visual inspection approach is consistent with previous research to account for noise in the normogastric window, or cases of multiple peaks [18,19,28]. Notably, selected electrode choice did not cause significant differences in the mental health CCA variate (F range (2–5, 86–87) 0.867–1.972, P range 0.146–0.507, n² range 0.044–0.051) or the stomach–brain coupling CCA variate (F range (2–5, 86–87) 1.376–2.863, P range 0.063–0.242, n² range 0.063–0.078). Similarly, the EGG recording montage did not cause any significant differences in the mental health CCA variate (F(2, 198) = 1.717, P = 0.182, n² = 0.017) or the stomach–brain coupling CCA variate (F(2, 198) = 1.220, P = 0.298, n² = 0.012).

As an additional check, we computed signal quality metrics using a comparison template-based procedure of ten ideal participants with very clear and prominent gastric peaks. ‘Poor quality’ participants had significantly lower signal quality as measured by cosine similarity (excellent/good quality: median 0.963, range 0.667, poor quality: median 0.595, range 0.585; U = 63,780, P < 0.001, r_rb = 3.186) and Pearson’s correlation (excellent/good quality: median 0.950, range 1.169; poor quality: median 0.054, range 1.287; U = 63,849, P < 0.001, r_rb = 3.190) (Supplementary Fig. 4). As an extra precaution, we confirmed that the mental health scores of the included and excluded EGG participants did not significantly differ when using the first PCA component of the 37 mental health scores (excellent/good quality: median 3.766, range 150.429; poor quality: median 5.745, range 106.762; U = 46,989, P = 0.964, r_rb = 2.084). Furthermore, included participants and excluded participants did not differ in reported gastrointestinal symptoms (average of PHQ15 items inquiring of ‘stomach pain’, ‘constipation, loose bowels or diarrhea’ and ‘nausea, gas or indigestion’: excellent/ good quality: median 1.333, range 2; poor quality: median 1.333, range 2; U = 14,104, P = 0.913, r_rb = ?0.0744). In addition, participants did not differ in reported under- and overeating behavior (PHQ9 item ‘poor appetite or overeating’: excellent/good quality: median 1, range 3; poor quality: median 1, range 3; U = 13,250, P = 0.361, r_rb = ?0.130). The selected EGG channel was then bandpass filtered, centered at the individual peak frequency (filter width of ±0.015 Hz, filter order of 5 or 1,470 samples), in forward and backward direction to avoid time shifts. After phase correction, the data were resampled to the fMRI rate (0.7143 Hz) and processed through a Hilbert transform to calculate the average phase per volume.

Gastric–brain coupling estimation
We followed procedures validated in previous EGG studies to estimate gastric–brain coupling [18,19]. The preprocessing of BOLD time series for all brain voxels involved bandpass filtering, using parameters identical to those applied during the EGG analysis. The initial and final 21 volumes (equivalent to 29.4 s) were excluded from both the fMRI and EGG time series. This adjustment resulted in a total signal duration of 781.2 s for further analysis. The instantaneous phases of both signals were obtained through the application of the Hilbert transform. Subsequently, the PLV was calculated as the absolute value of the average phase angle differences between the EGG and each voxel over time (equation (1)) [43]. The PLV is quantified by values ranging from 0 (representing a total absence of phase synchrony) to 1 (corresponding to absolute phase synchrony). 

where T is the number of time samples, and x and y are brain and gastric time series.

To account for any biases in PLV that arise from differences in signal amplitude, we created surrogate PLV values by disrupting the phase relationship between EGG and BOLD time series. We achieved this by shifting the EGG by at least ±60 s with respect to the BOLD time series, with concatenation at the edges. Given the 558 samples in the BOLD time series, this procedure generated 472 surrogate PLV datasets. We then took the median value of these surrogate PLV distributions as chance-level coupling, and defined coupling strength as the difference between empirical and chance-level coupling. Therefore, a higher value represents stronger stomach–brain coupling strength.

Mental health assessment battery
Participants completed a battery of mental health and lifestyle assessments. This encompassed 16 separate survey instruments spanning autism, ADHD, empathy, insomnia, interoception, depression, fatigue, social support, somatic symptoms, stress, social anxiety, trait anxiety, well-being and quality of life. All scales utilized validated Danish translations, except in cases where participants spoke English as their first language, in which case validated English versions were used. This allowed us to explore a broad range of mental health and lifestyle factors across 37 subscale scores (see Supplementary Table 1 for details of surveys, abbreviations and subscale scores).

Statistics
Statistics and reproducibility. We used a cross-validated CCA using the CCA-PLS MATLAB toolkit to link stomach–brain coupling in 209 parcellated regions with 37 mental health scores [59]. Data dimensionality was reduced via an optimized PCA–CCA approach. Statistical inference was evaluated using a fivefold cross-validation, based on out-of-sample correlations and permutation testing with 1,000 permutations. This included outlier screening and nuisance variable regression (age, gender, body mass index and data collection cohort) before inference. Control analyses tested for residual effects of neural connectivity and variability, physiological gastric activity, cardiac–brain and respiratory–brain interactions, and age and gender differences.

Canonical correlation analysis. We used the CCA-PLS toolkit to fit multivariate, cross-validated CCA models relating stomach–brain coupling (coupling strength of the BOLD and EGG time series) to mental health scores. Specifically, CCA aims to find linear combinations of each multidimensional variable (that is, canonical variates: which are weighted sums of stomach–brain coupling (V = X ? B) and mental health (U = Y ? A)) that are maximally correlated with each other, but uncorrelated with all other combinations (X and Y represent the inputted stomach–brain coupling and mental health data, while A and B represent the canonical weights) [59,68]. The toolbox incorporates various CCA and PLS models, including the cross-validated and optimized PCA–CCA techniques applied here [59]. This method importantly guards against overfitting via optimized data-reduction methods, assessing statistical inference between independent training and test sets, as well as by implementing permutation testing based on the out-of-sample correlation (see below for further details).

We first reduced the dimensions of stomach–brain coupling per fMRI voxel by parcellating with the 256-region DiFuMo atlas, excluding regions of cerebrospinal fluid, ventricles or white matter, yielding 209 relevant regions. Because CCA is very sensitive to outliers [69–72], it is important to screen for outliers in the stomach–brain coupling and mental health data, leading to the exclusion of 12 and 25 participants respectively (see Supplementary Fig. 5 for a complete flow chart of exclusions). This avoids false dependencies in the training set and distortions to the canonical projection weights [69–72]. Our final CCA sample comprised 199 participants for whom we had complete stomach–brain coupling and mental health matrices. These were standardized to have zero mean and unit variance, and nuisance regression was applied to control the estimated canonical variates for the influence of gender, age, body mass index and data collection cohort.

We applied the cross-validated CCA approach within the predictive framework (machine learning) [59]. Data were randomly split five times into 60% training or optimization and 40% test or holdout sets (‘outer splits’) for statistical inference (fivefold cross-validation). P values were determined by permutation testing (1,000 permutations), with Bonferroni correction (α = 0.05/5 = 0.01) applied to each of the five holdout sets. A significant associative CCA effect was declared if the P value was significant in at least one test set, once trained on the training or optimization set (out-of-sample correlation). Significant effects were iteratively removed via deflation to determine orthogonal CCA associations.

To address multicollinearity and overfitting, PCA–CCA optimized feature selection by further dividing each outer optimization set into inner 60% training and 40% validation sets (‘inner splits’). These inner splits were used for fivefold hyperparameter optimization, selecting the optimal number of PCA components the inputted stomach–brain coupling and mental health data dimensions were reduced to, by maximizing average out-of-sample correlation in the validation sets.

To aid interpretation of the networks underlying the estimated brain–stomach coupling signature, we averaged the canonical loadings across the Yeo 7-network parcellation for the stomach–brain coupling axis [44]. Moreover, we averaged across mental health categories for the psychological loading axis for clearer visualization of the CCA result (Fig. 3).

Control analyses. To control for underlying influences of neural connectivity or brain activity variability to the mental health stomach–brain coupling result, we conducted two separate control CCA analyses. For neural connectivity, we computed individual functional connectivity matrices from preprocessed fMRI data (209-DiFuMo atlas regions, using Python ‘Nilearn’ function: ‘ConnectivityMeasure. fit_transform’). For brain variability, we calculated and parcellated (209-DiFuMo regions) voxel-wise standard deviations. Both control analyses used the same CCA parameters, including age, gender, body mass index and data collection cohort as nuisance variables.

To ensure that the mental health dimension we uncovered is specific to the stomach and not cardiac and respiratory activity, we conducted two additional CCA analyses to control for cardiac–brain and respiratory–brain interactions. For the cardiac domain, we used interbeat intervals of the heartbeat, computed using identified R peaks via the ‘ppg_peaks’ function from systole, which uses a rolling average algorithm, while the respiration analysis focused on inhalation breath durations (interbreath intervals), computed using identified inhalation peaks via the ‘find_peaks’ function from scipy.signal with a distance of one sample and a peak prominence of 0.6. Both identified cardiac R peaks and respiratory inhalation peaks were visually inspected and manually corrected if necessary. To estimate instantaneous heart rate variability (HRV) regressors, we interpolated the cardiac interbeat intervals at the fMRI scanner frequency (TR 1.4 s, spline interpolation method) and band-pass filtered them at the frequencies corresponding to low (0.05–0.15 Hz) and high (0.15–0.357 Hz; upper limit constrained by Nyquist frequency of the scanner) HRV (low center frequency 0.1 ± 0.05 Hz, high center frequency 0.2535 ± 0.1035 Hz, MATLAB FIR filter) [18,73]. For the respiratory domain, we interpolated the interbreath intervals at fMRI scanner frequency and bandpass filtered at 0.1–0.357 Hz (refs. 74,75) (center frequency 0.2285 ± 0.1285 Hz; upper limit constrained by Nyquist frequency of the scanner). We obtained the amplitude envelopes of the instantaneous high- and low-frequency HRV and respiratory rate variability signal via a Hilbert transformation. These amplitude envelopes were used as regressors of interest (without convolution with HRF) [18,73] in first-level general linear models, with six motion and six acompcor noise regressors using SPM12 and a high-pass filter with a cutoff of 128 s. The fMRI had the same preprocessing as the stomach–brain phase coupling analysis. We then used T-contrasts to identify individual maps of brain activity associated with increased low-frequency HRV, high-frequency HRV or respiratory rate variability. Each of these individual heart and respiratory brain maps was parcellated and inputted into a CCA with the 37 mental health scores with the same parameters as the stomach–brain coupling CCA.

As an additional control, we conducted a separate whole-brain analysis to determine if we could identify a similar result when using a simpler mass univariate analysis. First, we computed PCA on the mental health scores to get a single component similar to the psychological canonical loadings from the CCA. This independent mental health PCA component was then correlated with the stomach–brain coupling from each of the 209-DiFuMo parcellated regions separately. Finally, these univariate Pearson correlation coefficients were correlated with the stomach–brain coupling loadings from the multivariate CCA to determine the similarity of the two analysis strategies.

Moreover, we completed control Spearman correlations of gastric physiology (EGG metrics) with the mental health canonical variate extracted from the stomach–brain coupling CCA analysis. From the computed EGG power spectra (as described in the ‘Electrogastrography peak selection and preprocessing’ section), we quantified the following normogastric EGG metrics: peak frequency, maximum power and proportion of power (Supplementary Table 5). Specifically, within the normogastric frequency range (0.033–0.067 Hz, 2–4 cycles per minute, 15–30 s), we stored the peak frequency and maximum power. Furthermore, we computed the proportion of normogastric power as the sum of the normogastric power divided by the sum of the power in all gastric frequencies (including bradygastric, normogastric and tachygastric frequencies: 0.02–0.17 Hz, 1–10 cycles per minute, 6–60 s). We input those EGG metrics into a correlation matrix with the mental health canonical variate, correcting for multiple comparisons using a false discovery rate at 5%.

To control for age or gender effects, we conducted the mentalhealth- associated stomach–brain coupling CCA with the same parameters as the main analysis but removed age and gender as nuisance regressors. We then tested for age and gender effects by performing a Pearson’s correlation with age and an independent-samples t-test with gender on the subsequent CCA variates for each stomach–brain coupling and mental health dimension.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Deidentified participant data (EGG, control neurological and stomach– brain coupling) implemented in this article are available via GitHub at https://github.com/embodied-computation-group/StomachBrain- MentalHealth. Due to Danish privacy law, mental health data and raw neurological data are available upon reasonable request, with the formation of a data sharing agreement. Researchers who wish to access these data may contact M.A. (micah@cfin.au.dk) at The Center of Functionally Integrative Neuroscience, Aarhus University, Denmark. The complete dataset (Visceral Mind Project) will be released as a data publication upon completion of the full anonymization process.

Code availability

Code for this article is available via GitHub at https://github.com/ embodied-computation-group/StomachBrain-MentalHealth.

References

1. Thompson, E. Mind in Life: Biology, Phenomenology, and the Sciences of Mind (Harvard Univ. Press, 2010).
2. Engelen, T., Solcà, M. & Tallon-Baudry, C. Interoceptive rhythms in the brain. Nat. Neurosci. 26, 1670–1684 (2023).
3. Allen, M. Unravelling the neurobiology of interoceptive inference. Trends Cogn. Sci. 24, 265–266 (2020).
4. Critchley, H. D. & Garfinkel, S. N. Interoception and emotion. Curr. Opin. Psychol. 17, 7–14 (2017).
5. Nord, C. L. & Garfinkel, S. N. Interoceptive pathways to understand and treat mental health conditions. Trends Cogn. Sci. 26, 499–513 (2022).
6. Hsueh, B. et al. Cardiogenic control of affective behavioural state. Nature 615, 292–299 (2023).
7. Teed, A. R. et al. Association of generalized anxiety disorder with autonomic hypersensitivity and blunted ventromedial prefrontal cortex activity during peripheral adrenergic stimulation: a randomized clinical trial. JAMA Psychiatry 79, 323–332 (2022).
8. Cryan, J. F. & Dinan, T. G. Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour. Nat. Rev. Neurosci. 13, 701–712 (2012).
9. Kolobaric, A. et al. Gut microbiome predicts cognitive function and depressive symptoms in late life. Mol. Psychiatry https://doi.org/10.1038/s41380-024-02551-3 (2024).
10. Allen, M., Varga, S. & Heck, D. H. Respiratory rhythms of the predictive mind. Psychol. Rev. https://doi.org/10.1037/ rev0000391 (2022).
11. Bagur, S. et al. Breathing-driven prefrontal oscillations regulate maintenance of conditioned-fear evoked freezing independently of initiation. Nat. Commun. 12, 2605 (2021).
12. Kluger, D. S. et al. Modulatory dynamics of periodic and aperiodic activity in respiration–brain coupling. Nat. Commun. 14, 4699 (2023).
13. Zelano, C. et al. Nasal respiration entrains human limbic oscillations and modulates cognitive function. J. Neurosci. 36, 12448–12467 (2016).
14. Krieger, J.-P. et al. Neural pathway for gut feelings: vagal interoceptive feedback from the gastrointestinal tract is a critical modulator of anxiety-like behavior. Biol. Psychiatry 92, 709–721 (2022).
15. Margolis, K. G., Cryan, J. F. & Mayer, E. A. The microbiota– gut–brain axis: from motility to mood. Gastroenterology 160, 1486–1501 (2021).
16. Foster, J. A. & McVey Neufeld, K.-A. Gut–brain axis: how the microbiome influences anxiety and depression. Trends Neurosci. 36, 305–312 (2013).
17. Mayeli, A. et al. Parieto-occipital ERP indicators of gut mechanosensation in humans. Nat. Commun. 14, 3398 (2023).
18. Rebollo, I., Devauchelle, A.-D., B?ranger, B. & Tallon-Baudry, C. Stomach–brain synchrony reveals a novel, delayed-connectivity resting-state network in humans. eLife 7, e33321 (2018).
19. Rebollo, I. & Tallon-Baudry, C. The sensory and motor components of the cortical hierarchy are coupled to the rhythm of the stomach during rest. J. Neurosci. 42, 2202–2220 (2021).
20. Azzalini, D., Rebollo, I. & Tallon-Baudry, C. Visceral signals shape brain dynamics and cognition. Trends Cogn. Sci. 23, 488–509 (2019).
21. Teckentrup, V. & Kroemer, N. B. Mechanisms for survival: vagal control of goal-directed behavior. Trends Cogn. Sci. 28, 237–251 (2024).
22. Nord, C. L., Dalmaijer, E. S., Armstrong, T., Baker, K. & Dalgleish, T. A causal role for gastric rhythm in human disgust avoidance. Curr. Biol. 31, 629–634 (2021).
23. Inui, A. Ghrelin: an orexigenic and somatotrophic signal from the stomach. Nat. Rev. Neurosci. 2, 551–560 (2001).
24. Sanders, K. M., Ward, S. M. & Koh, S. D. Interstitial cells: regulators of smooth muscle function. Physiol. Rev. 94, 859–907 (2014).
25. Richter, C. G., Babo-Rebelo, M., Schwartz, D. & Tallon-Baudry, C. Phase–amplitude coupling at the organism level: the amplitude of spontaneous alpha rhythm fluctuations varies with the phase of the infra-slow gastric basal rhythm. Neuroimage 146, 951–958 (2017).
26. Cao, J. et al. Gastric stimulation drives fast BOLD responses of neural origin. Neuroimage 197, 200–211 (2019).
27. Alladin, S. N. B., Judson, R., Whittaker, P., Attwood, A. S. & Dalmaijer, E. S. Review of the gastric physiology of disgust: proto-nausea as an under-explored facet of the gut–brain axis. Brain Neurosci. Adv. 8, 23982128241305890 (2024).
28. Müller, S. J., Teckentrup, V., Rebollo, I., Hallschmid, M. & Kroemer, N. B. Vagus nerve stimulation increases stomach-brain coupling via a vagal afferent pathway. Brain Stimul. 15, 1279–1289 (2022).
29. Cao, J., Wang, X., Chen, J., Zhang, N. & Liu, Z. The vagus nerve mediates the stomach–brain coherence in rats. Neuroimage 263, 119628 (2022).
30. Marso, S. P. et al. Semaglutide and cardiovascular outcomes in patients with type 2 diabetes. N. Engl. J. Med. 375, 1834–1844 (2016).
31. Wilding, J. P. H. et al. Once-weekly semaglutide in adults with overweight or obesity. N. Engl. J. Med. 384, 989–1002 (2021).
32. Nummenmaa, L., Hari, R., Hietanen, J. K. & Glerean, E. Maps of subjective feelings. Proc. Natl Acad. Sci. USA 115, 9198–9203 (2018).
33. Nummenmaa, L., Glerean, E., Hari, R. & Hietanen, J. K. Bodily maps of emotions. Proc. Natl Acad. Sci. USA 111, 646–651 (2014).
34. Insel, T. R. & Cuthbert, B. N. Brain disorders? Precisely. Science 348, 499–500 (2015).
35. Calkins, M. E. et al. The Philadelphia Neurodevelopmental Cohort: constructing a deep phenotyping collaborative. J. Child Psychol. Psychiatry 56, 1356–1369 (2015).
36. Gillan, C. M., Kosinski, M., Whelan, R., Phelps, E. A. & Daw, N. D. Characterizing a psychiatric symptom dimension related to deficits in goal-directed control. eLife 5, e11305 (2016).
37. Smith, S. M. et al. A positive-negative mode of population covariation links brain connectivity, demographics and behavior. Nat. Neurosci. 18, 1565–1567 (2015).
38. Kaczkurkin, A. N. et al. Common and dissociable regional cerebral blood flow differences associate with dimensions of psychopathology across categorical diagnoses. Mol. Psychiatry 23, 1981–1989 (2018).
39. Benwell, C. S. Y., Mohr, G., Wallberg, J., Kouadio, A. & Ince, R. A. A. Psychiatrically relevant signatures of domain-general decisionmaking and metacognition in the general population. npj Ment. Health Res. 1, 10 (2022).
40. Insel, T. et al. Research Domain Criteria (RDoC): toward a new classification framework for research on mental disorders. Am. J. Psychiatry 167, 748–751 (2010).
41. Schumann, G. et al. Stratified medicine for mental disorders. Eur. Neuropsychopharmacol. 24, 5–50 (2014).
42. Goyal, N., Moraczewski, D., Bandettini, P. A., Finn, E. S. & Thomas, A. G. The positive–negative mode link between brain connectivity, demographics and behaviour: a pre-registered replication of Smith et al. (2015). R. Soc. Open Sci. 9, 201090 (2022).
43. Lachaux, J.-P., Rodriguez, E., Martinerie, J. & Varela, F. J. Measuring phase synchrony in brain signals. Hum. Brain Mapp. 8, 194–208 (1999).
44. Yeo, B. T. T. et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011).
45. McTeague, L. M. et al. Identification of common neural circuit disruptions in cognitive control across psychiatric disorders. Am. J. Psychiatry 174, 676–685 (2017).
46. Yang, Y. et al. Common and specific functional activity features in schizophrenia, major depressive disorder, and bipolar disorder. Front. Psychiatry 10, 52 (2019).
47. Seghier, M. L. The angular gyrus. Neuroscientist 19, 43–61 (2013).
48. Gao, Y. et al. Decreased resting-state neural signal in the left angular gyrus as a potential neuroimaging biomarker of schizophrenia: an amplitude of low-frequency fluctuation and support vector machine analysis. Front. Psychiatry 13, 949512 (2022).
49. Su, Q. et al. Decreased interhemispheric functional connectivity in insula and angular gyrus/supramarginal gyrus: significant findings in first-episode, drug-naive somatization disorder. Psychiatry Res. Neuroimaging 248, 48–54 (2016).
50. Gao, J. et al. Habenula and left angular gyrus circuit contributes to response of electroconvulsive therapy in major depressive disorder. Brain Imaging Behav. 15, 2246–2253 (2021).
51. Mo, Y. et al. Bifrontal electroconvulsive therapy changed regional homogeneity and functional connectivity of left angular gyrus in major depressive disorder. Psychiatry Res. 294, 113461 (2020).
52. Tozzi, L. et al. Personalized brain circuit scores identify clinically distinct biotypes in depression and anxiety. Nat. Med. 30, 2076–2087 (2024).
53. Menon, V. & Uddin, L. Q. Saliency, switching, attention and control: a network model of insula function. Brain Struct. Funct. 214, 655–667 (2010).
54. Gordon, E. M. et al. A somato-cognitive action network alternates with effector regions in motor cortex. Nature 617, 351–359 (2023).
55. Choe, A. S. et al. Phase-locking of resting-state brain networks with the gastric basal electrical rhythm. PLoS ONE 16, e0244756 (2021).
56. Levakov, G., Ganor, S. & Avidan, G. Reliability and validity of brain–gastric phase synchronization. Hum. Brain Mapp. 44, 4956–4966 (2023).
57. Todd, J., Cardellicchio, P., Swami, V., Cardini, F. & Aspell, J. E. Weaker implicit interoception is associated with more negative body image: evidence from gastric-alpha phase amplitude coupling and the heartbeat evoked potential. Cortex 143, 254–266 (2021).
58. Sch?rmann-Vengels, J., Troche, S., Victor, P. P., Teismann, T. & Willutzki, U. Multidimensional assessment of strengths and their association with mental health in psychotherapy patients at the beginning of treatment. Clin. Psychol. Eur. 5, e8041 (2023).
59. Mihalik, A. et al. Canonical correlation analysis and partial least squares for identifying brain–behavior associations: a tutorial and a comparative study. Biol. Psychiatry Cogn. Neurosci. Neuroimaging 7, 1055–1067 (2022).
60. Hashimoto, T. et al. Neural correlates of electrointestinography: Insular activity modulated by signals recorded from the abdominal surface. Neuroscience 289, 1–8 (2015).
61. Jeanne, R. et al. Gut-Brain coupling and multilevel physiological response to biofeedback relaxation after a stressful task under virtual reality immersion: a pilot study. Appl. Psychophysiol. Biofeedback 48, 109–125 (2023).
62. Wolpert, N., Rebollo, I. & Tallon-Baudry, C. Electrogastrography for psychophysiological research: practical considerations, analysis pipeline, and normative data in a large sample. Psychophysiology 57, e13599 (2020).
63. Alladin, S. N. B. et al. Children aged 5–13 years show adult-like disgust avoidance, but not proto-nausea. Brain Neurosci. Adv. 8, 23982128241279616 (2024).
64. Chang, L., Wei, Y. & Hashimoto, K. Brain–gut–microbiota axis in depression: a historical overview and future directions. Brain Res. Bull. 182, 44–56 (2022).
65. Vindegaard, N., Speyer, H., Nordentoft, M., Rasmussen, S. & Benros, M. E. Gut microbial changes of patients with psychotic and affective disorders: a systematic review. Schizophr. Res. 234, 41–50 (2021).
66. Rao, S. S. C., Quigley, E. M. M., Chey, W. D., Sharma, A. & Lembo, A. J. Randomized placebo-controlled phase 3 trial of vibrating capsule for chronic constipation. Gastroenterology 164, 1202–1210.e6 (2023).
67. Porciello, G., Monti, A., Panasiti, M. S. & Aglioti, S. M. Ingestible pills reveal gastric correlates of emotions. eLife 13, e85567 (2024).
68. Linke, J. O. et al. Shared and anxiety-specific pediatric psychopathology dimensions manifest distributed neural correlates. Biol. Psychiatry 89, 579–587 (2021).
69. Luo, L., Wang, W., Bao, S., Peng, X. & Peng, Y. Robust and sparse canonical correlation analysis for fault detection and diagnosis using training data with outliers. Expert Syst. Appl. 236, 121434 (2024).
70. Wilms, I. & Croux, C. Robust sparse canonical correlation analysis. BMC Syst. Biol. 10, 72 (2016).
71. Okan Sakar, C. & Kursun, O. A method for combining mutual information and canonical correlation analysis: predictive mutual information and its use in feature selection. Expert Syst. Appl. 39, 3333–3344 (2012).
72. Mandal, A. & Cichocki, A. Non-linear canonical correlation analysis using alpha–beta divergence. Entropy 15, 2788–2804 (2013).
73. Critchley, H. D. et al. Human cingulate cortex and autonomic control: converging neuroimaging and clinical evidence. Brain 126, 2139–2152 (2003).
74. Natarajan, A. et al. Measurement of respiratory rate using wearable devices and applications to COVID-19 detection. npj Digit. Med. 4, 136 (2021).
75. Iqbal, T., Elahi, A., Ganly, S., Wijns, W. & Shahzad, A. Photoplethysmography-based respiratory rate estimation algorithm for health monitoring applications. J. Med. Biol. Eng. 42, 242–252 (2022).

Supplementary information

Acknowledgements

This research is financially supported by a Lundbeckfonden Fellowship (R272-2017-4345, M.A.) and a European Research Council Grant (ERC-2020-StG-948788, M.A.). I.R. is supported by a Marie Sk?odowska-Curie Action (MSCA) BRAINSTOM (101028203). The funding source was not involved in the study design, collection, analysis, interpretation or writing of the manuscript.

Author contributions

L.B. and IR analyzed the data, interpreted the results and wrote the paper. N.N. provided conceptual advice and contributed toward preprocessing of neuroimaging data. M.A. provided supervision, conceptual advice and wrote the paper.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s44220-025-00468-6.

Correspondence and requests for materials should be addressed to Leah Banellis.

Peer review information Nature Mental Health thanks Edwin Dalmaijer, Rebecca Todd and the other, anonymous reviewer(s) for their contribution to the peer review of this work.

Reprints and permissions information is available at www.nature.com/reprints.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

---

---

Сокращённый пересказ данной статьи на русском языке:

• для врачей: Литвинов И.А. Синхронизация между желудком и мозгом отражает многомерную картину психического здоровья.
• для пациентов: Тревога живет в животе: ученые нашли новую связь между желудком и психикой. Комсомольская правда. Здоровье. 03.09.2025 г.

Назад в раздел