Using the largest psychedelic neuroimaging dataset to date (62 adults, fMRI and EEG, before and after 19 mg psilocybin), the authors show that psilocybin reorganises brain dynamics—increasing global connectivity in associative regions and decreasing it in sensory areas—so that neural activity aligns with contextual stimuli and eyes-open versus eyes-closed states. This reorganisation, revealed by low-dimensional trajectories, produces an “embeddedness” state linking distinct neural patterns to boundary-dissolving subjective experiences and next-day mindset changes, offering a framework for how psychedelics exert context-sensitive behavioural and therapeutic effects.
Psychedelics can profoundly alter consciousness by reorganising brain connectivity; however, their effects are context-sensitive. To understand how this reorganisation depends on context, we collected and comprehensively analysed the largest psychedelic neuroimaging dataset to date. Sixty-two adults were scanned with functional MRI and EEG during rest and naturalistic stimuli (meditation, music, and movie), before and after ingesting 19 mg of psilocybin (functional MRI ~80 min post-dose; EEG ~150 min post-dose). Half the participants ranked the experience among the most meaningful of their lives. Under psilocybin, functional MRI and EEG signals recorded during eyes-closed conditions became similar to those recorded during an eyes-open condition. Global functional connectivity increased in associative regions and decreased in sensory areas. Using machine learning to represent neural activity as low-dimensional trajectories, we found that psilocybin reorganised these into structured, context-sensitive patterns of brain activity that reflected both experimental condition and the quality of subjective experience, revealing an organisation that was missed by time-averaged connectivity measures. Under psilocybin, brain networks that ordinarily segregate internal and external processing coherently integrated and aligned neural dynamics with context. This context-alignment manifested as distinct and cohesive neural trajectories in participants reporting positively felt self- and boundary-dissolving effects, corresponding to the felt experience of being part of the environment, which we refer to as embeddedness - the subjective experience of being continuous with, rather than separate from, the surrounding environment. The strength of this context-alignment was associated with next-day mindset change, bridging the neural, experiential, and therapeutic dimensions of the psychedelic state. These findings show that the organisation of brain activity covaries with the experiential coherence of the psychedelic state, and provide a systems-level framework for how context-sensitive brain dynamics link neurobiology to subjective experience and behavioural change.
Papers cited by this study that are also in Blossom
Avram, M., Müller, F., Preller, K. H. et al. · Biological Psychiatry (2024)
Stoliker and colleagues frame the study around how psychedelics reorganise brain connectivity in ways that depend on psychological and sensory context. Earlier research has shown that agents such as psilocybin act at 5-HT2A receptors to induce structural and functional plasticity, relax constraints on associative networks such as the default mode network (DMN), and can produce profound subjective effects including intensified immersion and self-boundary dissolution. However, prior human studies have often used small samples, single imaging modalities, or limited contexts, leaving open how context shapes large-scale functional integration and the neural mechanisms that relate reorganisation to subjective and therapeutic outcomes. This study set out to address those gaps by acquiring a large, multimodal dataset (fMRI and high-density EEG) in a predominantly psychedelic-naïve cohort and systematically varying contextual conditions (eyes-closed rest, guided meditation, music listening, and eyes-open movie watching). Using a fixed 19 mg oral psilocybin dose and repeated baseline (no-psilocybin) and post-dose imaging, the investigators aimed to map how psilocybin changes global and local connectivity, signal variability, directed interactions (via dynamic causal modelling), and low-dimensional neural embeddings, and how those neural changes covary with subjective experience and short-term psychological changes. The work emphasises context-sensitive mechanisms and introduces machine-learning embedding approaches to link neural dynamics with phenomenology and potential adaptive outcomes.
The study used an open-label, within-subject design with two imaging sessions per participant: a baseline (no-psilocybin) session and a session after a standardised 19 mg oral dose of psilocybin. Each session included both multi‑echo, multi‑band BOLD fMRI and 64-channel wet EEG recordings across four repeated conditions: eyes-closed resting state, guided meditation (eyes closed), music listening (eyes closed), and an eyes-open naturalistic movie (moving clouds, no audio). Scans durations were reported for each condition and the fMRI and EEG acquisitions were organised according to BIDS. Participants were healthy, psychedelic‑naïve adults recruited with screening to exclude psychiatric, neurological, substance‑use and MRI‑contraindicated conditions; the initial recruitment was 65 adults aged 18–55. After exclusions and incomplete sessions, volumetric fMRI data were available for 64 participants at baseline and 61 after psilocybin; preprocessed EEG data were available for 63 baseline and 58 psilocybin sessions. Half the cohort were randomly assigned to an 8‑week mindfulness-based cognitive therapy programme prior to dosing to assess preconditioning effects; engagement metrics were recorded. Neuroimaging preprocessing combined fMRIprep for anatomical and functional preprocessing with multi‑echo ICA denoising (tedana) and a single regression cleaning step to remove white matter, CSF, motion and non‑BOLD components, then projection to the fsLR32k cortical surface. Quality control used MRIQC metrics and principled exclusion of scans with excessive framewise displacement. Functional analyses included vertex-wise Global Functional Connectivity (GFC) computed as mean Fisher‑z correlations across vertices within each hemisphere, standard deviation (SD) maps of BOLD time series, parcel-wise FC using a 332‑parcel scheme (Schaefer cortical parcellation plus a 32‑parcel subcortical atlas), and network modularity computed with the Brain Connectivity Toolbox. The authors explicitly advised against performing global signal regression (GSR) or z‑scoring when computing GFC, providing a mathematical rationale and showing these transforms can alter interpretation. Directed (effective) connectivity was estimated with spectral Dynamic Causal Modelling (DCM) focused on six regions within the DMN and anterior hippocampus (bilateral aHip, bilateral inferior parietal cortex, mPFC, PCC), with Parametric Empirical Bayes used for group‑level inference. Two machine‑learning embedding approaches characterised low‑dimensional neural dynamics: CEBRA‑Time (contrastive embeddings tailored to temporal analyses) produced subject‑specific three‑dimensional neural trajectories across concatenated conditions, while a Temporal Attention‑enhanced Variational Graph Recurrent Neural Network (TAVRNN) modelled evolving FC matrices across 332 parcels to produce two‑dimensional ROI embeddings capturing temporal evolution. EEG preprocessing used automated RELAX with multistep artifact reduction and re-referencing; spectral power (theta, alpha, beta, gamma) and signal diversity (Lempel‑Ziv complexity) were computed per condition. Behavioural measures included the 11‑Dimensional Altered States of Consciousness (11D‑ASC) scale, MEQ30, AES‑M for music aesthetics, a novel mindset questionnaire the day after dosing, and follow‑up scales (Nature Relatedness NR‑6, LAP‑R) at one month.
Global functional connectivity and BOLD variability. Under psilocybin, eyes‑closed conditions showed a marked reorganisation in GFC: associative cortex exhibited increased GFC (percentage changes up to +72%, Cohen's d ≈ 0.53) while sensory regions, particularly occipital cortex, showed decreased GFC (percentage decreases up to 39%, Cohen's d = 0.55). These effects formed significant spatial clusters identified with TFCE. GFC during eyes‑open movie watching increased in both sensory and associative areas; however, applying GSR or z‑scoring altered the movie results, underscoring preprocessing sensitivity. Standard deviation (SD) maps revealed a spatially dependent redistribution of BOLD variability: SD increased in orbitofrontal, occipitotemporal and inferior temporal regions and right somatosensory cortex during eyes‑open states, while SD decreased in early visual areas during eyes‑closed conditions. The spatial pattern of SD changes broadly mirrored GFC changes, and group‑level effects were robust though individual variability in GFC responses was substantial. State separation, modularity and network‑level FC. Histograms of GFC showed reduced separation between eyes‑open and eyes‑closed distributions under psilocybin, a convergence that was statistically significant across multiple resting‑state networks and particularly pronounced in visual networks. Functional modularity decreased under psilocybin, indicating reduced segregation; within‑network connectivity decreased more than between‑network connectivity, consistent with reorganisation of canonical network architecture. Machine‑learning embeddings and subjective experience. CEBRA‑Time embeddings produced low‑dimensional neural trajectories whose condition‑specific clustering tightened with stronger subjective effects. Participants with higher MEQ30 and 11D‑ASC scores exhibited embeddings that were more separable by condition, and classification accuracy of condition labels from embeddings correlated positively with subscales indexed as positively felt associative effects (self‑ and boundary‑dissolving dimensions) while correlations with sensory subscales were weaker and negative/dysphoric subscales showed minimal association. Participants who reported late‑onset effects had embeddings resembling baseline scans, linking acute subjective timing to neural reorganisation within the imaging window. TAVRNN produced complementary findings: under psilocybin, ROIs showed tighter within‑network clustering alongside increased global cohesion across networks; both local and global patterns scaled with subjective intensity. Subjective and longitudinal psychological outcomes. Behavioural data showed robust subjective effects from the standardised 19 mg dose: roughly half the participants retrospectively ranked the session among their top five most meaningful experiences, and many rated high intensity. MEQ30 and 11D‑ASC distributions showed substantial group‑level responses with wide individual variability. One‑month follow‑up revealed group‑level increases in death acceptance (LAP‑R; p = 0.0006, d = 0.49) and personal meaning (LAP‑R; p = 0.0004, d = 0.51). Nature relatedness (NR‑6) also increased (t(55) = -3.37, p = 0.0014, d = 0.26). Correlations between day‑after mindset change scores and subjective subscales indicated insightfulness had the strongest association (r = 0.65, p ≈ 10^-7), with mystical/positive/blissful subdimensions showing r = 0.4–0.6 (p < 0.00003); sensory effects correlated more weakly (r = 0.18–0.4, p < 0.05) and negative experiences had minimal, non‑significant associations (r < 0.1). Directed connectivity. Spectral DCM of an anterior hippocampus–DMN circuit revealed context‑dependent modulation of effective connectivity between baseline and psilocybin sessions. Specific patterns included consistent inhibition from left inferior parietal cortex to right anterior hippocampus across tasks (but not at rest), PCC→aHip R excitation and mPFC→aHip R inhibition during rest, bilateral IPC inhibition of hippocampi during meditation, asymmetric hippocampal inhibition during music, and the strongest changes during eyes‑open movie stimuli with predominating cortical inhibition onto hippocampi and intra‑cortical excitation. These directed changes were interpreted as context‑sensitive reorganisation of cortico‑subcortical interactions. EEG findings and convergence across modalities. EEG corroborated fMRI results: under eyes‑closed psilocybin, theta/alpha/beta power decreased while gamma power increased, with movie viewing attenuating many of these shifts except for gamma increases in early visual areas. The alpha power difference between eyes‑open and eyes‑closed was reduced by 48% under psilocybin (robust Cohen's d = 0.79, Mann‑Whitney U p < 10^-19), consistent with a unification of internally and externally directed processing. Lempel‑Ziv complexity increased, particularly in eyes‑closed conditions, indicating higher signal diversity. Music was rated as more meaningful under psilocybin (15–25% increase; Cohen's d > 0.67; Mann‑Whitney p < 0.001). Interventions and adverse events. The pre‑dosing mindfulness training (8 weeks) produced no statistically significant differences in acute subjective effects, next‑day mindset scores, or connectivity metrics. Adverse events were infrequent: transient headaches (n = 3) and three participants required brief follow‑up psychological support; a small number of participants did not complete imaging due to adverse responses or incidental findings, which determined the final sample counts reported above.
The authors interpret their findings as evidence that psilocybin reorganises large‑scale brain dynamics in a context‑sensitive manner, shifting eyes‑closed activity toward patterns that engage associative networks and thereby aligning neural dynamics with environmental and psychological context. They introduce the construct of 'embeddedness' to describe a brain state in which internal (associative) and external (sensory) processing become more coherently integrated; embeddedness is linked to positively felt associative subjective effects such as unity and bliss and to short‑term adaptive changes in mindset and meaning. Machine‑learning embeddings (CEBRA and TAVRNN) are presented as tools that reveal hidden coherence in low‑dimensional neural trajectories and that covary with the quality and timing of subjective experience, suggesting potential utility for predicting or monitoring state‑dependent psychological outcomes. Relative to previous work, the investigators highlight that their large, multimodal dataset clarifies prior inconsistencies around whether sensory or associative connectivity increases under psychedelics and demonstrates that preprocessing choices such as GSR and z‑scoring can materially affect interpretations. They also note concordance across fMRI and high‑density EEG, strengthening the case that observed effects are not solely vascular. Context manipulations—eyes‑open visual stimulation, music and meditation—produced distinct patterns of directed connectivity in the anterior hippocampus–DMN circuit, suggesting that curated sensory and cognitive environments can shape acute circuit dynamics and potentially guide behavioural outcomes. The authors acknowledge several sources of uncertainty. Individual responses varied substantially despite group‑level effects, and some analyses (for example, movie results) were sensitive to preprocessing transforms; they therefore caution against indiscriminate use of GSR and z‑scoring in GFC analyses. They also note that while embeddings correlated with adaptive short‑term changes, further work is required to test generalisability across clinical populations and more diverse samples. Finally, although the standardised 19 mg dose produced robust subjective and neural effects in this healthy cohort, the open‑label design and single fixed dose limit causal inference about dose‑dependent effects and therapeutic mechanisms. In terms of implications, the authors propose that recognising the dominant role of associative processes under psychedelics reframes how context is used in research and clinical settings: structured environments and targeted stimuli may be leveraged to shape neural reorganisation and psychological outcomes. Machine‑learning embeddings are advanced as promising biomarkers for state‑dependent neural organisation that might inform precision psychiatry and consciousness research, but the authors stress the need for replication, translational work in clinical samples, and careful methodological choices when computing connectivity metrics.
The profound e↵ects of psychedelics reshape subjective responses to internal and external sensations and are frequently reported as among the most meaningful experiences in life. These states can manifest sustained therapeutic benefits, including reductions in depression, anxiety, and addiction, alongside increases in social connectedness and overall well-being. Understanding how contexts influence these changes, and the underlying brain connectivity that can be harnessed clinically, requires a comprehensive examination of how psychedelics reshape the brain's functional integration. The brain constructs perception and selfhood by integrating external sensory inputs with internal models of the environment. These interactions between sensory and associative brain regions are enabled by structural and functional connectivity. Psychedelics such as psilocybin disrupt these interactions by acting as an agonist at the serotonin 5-HT 2A receptor, inducing structural and functional plasticity that can reshape macro-level connectivity. Studies that examine e↵ective connectivity further suggest that associative network communication becomes reconfigured. Reflecting this reorganisation, individuals frequently report an intensified sense of immersion, in which context of space, time, and selfhood feel deconstructed and interconnectedthough the exact mechanisms linking network-level shifts to subjective experience remain unknown. A central mediator of these e↵ects is the default mode network (DMN), which supports the integration of information from diverse associative regions spanning spatial, temporal, and selfreferential contexts. Under psilocybin, constraints on these networks relax, permitting novel patterns of connectivity between sensory and associative regions. This reconfiguration alters crucial connectivity that underlies emotion, cognition, and perception. Notably, preclinical and recent human research suggests that reductions in DMN and anterior hippocampus connectivity may contribute to the therapeutic e↵ects of psilocybin. However, despite mounting evidence for these shifts in connectivity, existing studies often rely on smaller sample sizes or limited experimental contexts, making it di cult to generalise how psilocybin alters functional integration across a variety of naturalistic conditions. Furthermore, few investigations combine multiple imaging modalities, behavioural measures and analytical methods to capture both large-scale brain activity and detail fine-grained interactions. We acquired the largest single-site psychedelic neuroimaging dataset to date, integrating multimodal fMRI and EEG with controlled contextual manipulations across eyes-open and eyes-closed states in a predominantly psychedelic-naïve cohort (N=62). Experimental conditions of rest, mindfulness meditation, music, and movie viewing were systematically varied, and extensive behavioural measures were collected (see Fig.). Combining multimodal neuroimaging, diverse contextual manipulations, and multiple analytical methods revealed that the brain shifts towards internally generative (associative) processes over sensory processes under psilocybin. Using state-of-the-art machine learning (ML) techniques we uncovered hidden coherence in brain activity that was structured by context and covaried with the strength of self-and boundary-dissolving e↵ects. This work underscores the basic and clinical relevance of context-focused research in enhancing and tailoring responses to psychedelics, suggesting a state-dependent neural signature of 'embeddedness' emerges when connectivity and perception become flexibly integrated and aligned with context. We identify this key feature of the psychedelic brain state as psychologically adaptive and contextually modifiable, and introduce new methods for studying mental state and context-dependent neural dynamics in basic and clinical research.
Psilocybin induced distinct shifts in global integration of functional connections across brain regions. This was quantified using Global Functional Connectivity (GFC), a measure of the average correlation of BOLD signals between each vertex and every other vertex in a high-resolution cortical surface map. During eyes-closed states, psilocybin reduced the GFC of sensory regions (Fig.), and these reductions formed statistically significant clusters (see Fig.). Occipital areas showed the strongest decrease in GFC (percentage changes up to 39% and Cohen's d = 0.55), with midline ventral parietal areas showing strongest decreases during music. In contrast, GFC increased in associative areas (percentage changes up to +72%, Cohens d = 0.53). This evidence indicates that, during eyes-closed conditions, psilocybin suppressed the integration of sensory regions with the broader cortical network but enhanced the integration of associative regions. In contrast, eyes-open movie-watching increased GFC in both sensory and associative areas (Fig.), enhancing global synchrony. We confirmed this was not due to head motion An open label 'baseline' (no-psilocybin) imaging session was conducted approximately one week before imaging under administration of 19 mg psilocybin. Music, mindfulness, personality and psychedelic measures were collected before and longitudinally after psilocybin. Meditators received additional assessments before and after the mindfulness course. Meditation showed no di↵erences in our connectivity and subjective e↵ect analyses, and was subsequently collapsed into the main arm for group level analyses. See Methods for details. both eyes-open and eyes-closed states, suggesting its consistent involvement in the integration of associative processes under psilocybin. Further analyses using Z-scoring and global signal regression were performed to resolve previous contradictory findings that used smaller samples, where GFC changes di↵ered across datasets and flipped sign after global signal regression. Our analyses produced consistent grouplevel findings for all eyes-closed conditions, confirming the robustness of the observed e↵ects in our large sample (see Fig.Data 3). However, the movie-watching results changed after GSR and z-scoring, demonstrating that applying these transformations can twist the interpretation of GFC. In the Methods, we provide a mathematical explanation for why it happens and recommend not performing GSR and z-scoring in GFC analysis. Despite the consistent group-level findings, substantial variability in GFC changes was observed between participants, indicating individual di↵erences in response to psilocybin (see Fig.).
Mapping the standard deviation (SD) of the BOLD signal within subjects across the cortex revealed that signal variability is spatially reorganised and contextually-dependent under psilocybin. Compared to baseline, the SD increased during eyes-open states in the orbitofrontal cortex, occipitotemporal gyri, inferior temporal lobes (particularly anterior regions), and the right hemisphere primary somatosensory cortex. In contrast, SD decreased in early visual areas, in alignment with previous findings on visual imagery. During eyes-open movie watching, we also detected spatially constrained SD increases in early visual regions, as well as the posterior cingulate cortex and precuneus (Fig.), which otherwise showed widespread increases across the brain in other contexts. SD increases and GFC increases were partially lateralised to the right hemisphere during eyes-closed rest, meditation and music and decreases in sensory regions under psilocybin during eyes-closed conditions, particularly the occipital lobe during music (Fig.). Finally, the spatially dependent redistribution of functional variability across the brain during eyes-open and eyes-closed states was broadly consistent with GFC spatial changes during psilocybin-induced states (Fig.). Increased BOLD signal variability under psychedelics has been commonly associated with increased entropy.
Complementing the spatial maps of GFC changes, we examined the frequency of GFC values across all cortical vertices, independent of their spatial organisation. Under psilocybin, the histograms of GFC values in eyes-open and eyes-closed states-which were distinct at baseline (no-psilocybin) imaging-overlapped (Fig.). This convergence was consistent and statistically significant across sensory, limbic and associative resting-state networks (RSN) (Fig.
A significant reduction in functional modularity was observed under psilocybin, suggesting decreased functional segregation. Previous studies with smaller cohorts have reported disruptions in RSN organisation, primarily focusing on default mode network connectivity. Our larger dataset confirms and extends these findings by demonstrating more robust evidence of psychedelic reorganisation of the integration and segregation of brain network architecture. All canonical RSNsshowed a widespread reduction in functional connectivity; however, the correlation decreased more for brain regions within the same networks than between them (Fig.
Dimensionality reduction methods have been adopted in neuroscience for uncovering meaningful low-dimensional structures in neural data. In ML, the term embedding refers to mathematical functions that map high-dimensional data to lower-dimensional representations while preserving essential contextual and semantic information in the lower-dimensional representations. When applied to neural data, it allows to uncover hidden patterns underlying the brain's very high-dimensional complex neural interactions, analogous to manifolds. Here, we used CEBRA to generate low-dimensional embeddings of neural activity and trained a support vector machine (SVM) classifier to test how well session-specific embeddings distinguished between rest, meditation, music, and movie conditions(see Methods for CEBRA description). We then examined the relationship between classification accuracy and subjective experiences assessed via the Mystical Experience Questionnaire (MEQ30), administered at the end of the session We observed distinct shifts in embeddings associated with subjective ratings of psychedelic e↵ects. Notably, stronger subjective e↵ects were associated with tighter clustering of time point embeddings within the same condition, making them more easily separable from embeddings of other conditions (Fig.). Importantly, this conditionspecific clustering was moderated by the timing of subjective e↵ects, as demonstrated by a participant who reported late-onset of substantial e↵ects but minimal subjective experience during the imaging session (Fig.). Their embeddings resembled baseline (no-psilocybin) scans, indicating that the observed neural reorganisation directly reflects the occurrence and intensity of subjective e↵ects within the imaging window (Fig.). Further analysis of the top five and bottom five MEQ scores reinforced these findings, revealing a gradient in neural embeddings that scaled with strength of subjective e↵ects (Fig.). This gradient reflects a reorganisation of neural trajectories under psilocybin that is both context-specific and e↵ect-dependent, suggesting that context (rest, meditation, music, or movie) shapes the emergent patterns of neural activity in unison with subjective e↵ects that amplify the cohesion of these patterns. Next, using the CEBRA-derived trajectories, we classified the condition label for each time point for each participant with a Support Vector Machine (SVM) classifier. We examined the correlation between classification accuracy and each of the 11-Dimension Altered States of Consciousness (11-D ASC)and Mystical Experiences Questionnaire (MEQ30)scores, after psilocybin administration. A stronger association was observed between classification accuracy and positively felt associative e↵ects (blue bars), while weaker correlations were found between classification accuracy and sensory e↵ects (orange bars) and low or negative correlation was found for negatively felt associative e↵ects (red bars) (Fig., left panel and Fig.). See the next section for conceptual colour-coding and see Methods: Reported measures for details. We also found a correlation between classification accuracy and average mindset changes one day after psilocybin (see Fig.Data 8, Supplementary Material, and the section on subjective e↵ect correspondence with day-after mindset changes). Subjects with higher MEQ30 mean scores generally exhibited higher classification accuracy, indicating a positive relationship between mystical experiences and model performance (Fig., right panel). These findings suggest embeddings are indicative of the abstract and existential qualities of the experience (i.e., self-and boundary-dissolving e↵ects), as well as positive psychological adjustment which were correlated (see section Subjective e↵ect correspondence with day-after mindset changes), suggesting they may predict the translation of interconnection experienced during the acute psychedelic state to the period after psilocybin. These results were further supported using another ML-based technique called Temporal Attention-enhanced Variational Graph Recurent Neural Network (TAVRNN), which e↵ectively captured a lower-dimensional representation of individual network connectivity during each of the four conditions. TAVRNN captures temporal changes in network structure by modelling sequential snapshots of neuronal connectivity, enabling the identification of key connectivity patterns and shifts in communicability (see Methods for TAVRNN details). TAVRNN revealed two distinct patterns under psilocybin. Nodes within individual networks exhibited tighter clustering, reflecting more internally cohesive neural dynamics, reducing within-network separation. See Supplementary Material and Fig.Data 9 for quantification. Simultaneously, embeddings across all networks showed tighter clustering, indicating a shift toward integration of brainwide organisation and increased global cohesion across contexts. This dual pattern of local within-network cohesion alongside global reorganisation scaled with subjective e↵ects, suggesting that integration depends on subjective experience and is broadly preserved across contexts (Fig.).
Participants' reports one day after psilocybin confirmed that their psychedelic experience during imaging were profoundly altered and meaningful. Half of participants retrospectively ranked it among the top five most meaningful experiences of their lives (Fig.), and most reported it as substantially intense (¿9/10) (Fig.). We also collected semi-structured reports one day after psilocybin, with key excerpts describing the experience as "one of the most peaceful and profound things I could ever experience", "...meldwith the MRI machine, floor, walls, air", "los[ing] the plot of who I was, where I was, if I was even here, what was happening" and "los[ing] all sense of self... bec[oming] at one with all my surroundings...as part of a bigger network of things" (see Supplementary Material for a table of qualitative excerpts). Our large sample enabled a view of within-group subjective reports given a standardised dose of psilocybin. We report histograms and radar plots of 11D ASC and MEQ30 scores that illustrate substantial group-level subjective e↵ect intensity across key scales and wide individual variability from the 19mg dose (Fig.). We colourcoded subscales to reflect a key theoretical and conceptual distinction in psychedelic e↵ects: blue for positively felt associative e↵ects (self-and boundary-dissolving experiences) which are more abstract and existential e↵ects (e.g., transcendence, bliss, unity); orange for sensory e↵ects (e.g., visual imagery, synaesthesia); red for dysphoric e↵ects (e.g., impaired control and cognition, and anxiety); and purple for uncategorised e↵ects. Positively felt associative e↵ects such as bliss and unity received higher ratings across our sample than cognitive insights or spiritual experiences. Negative e↵ects, particularly anxiety, were infrequent or rated as low. These subscales were analysed post hoc to assess patterns of covariation. This identified subsets of subscales showed high intercorrelations in our sample (Fig.). A complete matrix of 11D ASC and MEQ pairwise correlations is also provided (see. The Life Attitudes Profile Revised (LAP-R)was used to assess death acceptance and changes in personal meaning before and one month after psilocybin administration. Results showed statistically significant group-level improvements in death acceptance (p = 0.0006, d = 0.49) and personal meaning (p = 0.0004, d = 0.51), alongside individual variability. The Nature Relatedness Scalewas also administered before and one month after psilocybin to assess changes in participants' connection to nature, with results indicating both group-level increases (t(55) = -3.37, p = 0.0014, d = 0.26) and individual shiftsOur sample's scores were comparable to those observed in higher-dose studies (300-400µg/kg, bodyweight adjusted) (e.g., blissful state, experience of unity), supporting the potential relevance of broader contextual factors we employed, such as space design, minimising interruptions, depth-focus music, and encouraging inward focus, in priming the states examined during imaging.
We examined the relationship between each participant's post-psilocybin mindset change score (measured one day after administration, see Supplementary Material) and their subjective experience scores from the 11-D ASC scale and the MEQ30. Mindset change, assessed across dimensions such as connection to self, others, and nature, as well as patience, harmony, and inner peace, correlated moderately with associative dimensions of the psychedelic experience (see Supplementary Material). Insightfulness showed the strongest correlation (r = 0.65, p = 10 7 ), followed by mystical, positive, blissful, and spiritual subdimensions (r = 0.4 0.6, p < 0.00003). Sensory e↵ects were less correlated (r = 0.18 0.4, p < 0.05 except for Elementary Imagery), while negative experiences showed minimal associations (r < 0.1, not statistically significant) (Fig.). These findings indicate the potential primacy of positively felt associative dimensions (self-and boundary-dissolving e↵ects) in driving positive psychological shifts, with cognitive insights and transcendent states-such as unitive, blissful, mystical and spiritual experiences-being more indicative of mindset changes than sensory or challenging negative e↵ects (anxiety, impaired control and cognition). The observed relationship between these states and psychological shifts provides empirical context for ongoing debates about the necessity of the psychedelic experience for clinical e cacy.
Dynamic causal modelling (DCM)revealed context-dependent changes in e↵ective connectivity from baseline (nopsilocybin) to psilocybin administration between the anterior hippocampus (aHip) and nodes of the core Default Mode Network (DMN). While previous preclinical and post-administration studies have highlighted the role of this circuit in psychedelic-induced plasticity, our findings extend these observations by demonstrating acute, context-specific modulation of aHip-DMN connectivity across distinct sensory and cognitive states. These changes reflect corticosubcortical interactions, linking circuit-level dynamics to broader transformations in functional connectivity and subjective experience under psilocybin. Connectivity patterns varied significantly depending on experimental conditions, with left inferior parietal cortex (IPC L) consistently inhibiting the right anterior hippocampus (aHip R) across tasks but not at rest. During the resting state, excitation from the posterior cingulate cortex (PCC) to the aHip R and inhibition from the medial prefrontal cortex (mPFC) to the aHip R were observed, alongside inhibition from the left anterior hippocampus (aHip L) to the IPC L (Fig., very strong evidence, probability > 99%). These connections suggest a regulatory mechanism, with the PCC driving hippocampal activations, the mPFC attenuating them, and the aHip L suppressing parietal engagement, potentially amplifying self-reference while limiting evaluation and contextual integration with memory. During meditation, bilateral inferior parietal cortex (IPC L and IPC R) inhibited bilateral anterior hippocampi (aHip L and aHip R), consistent with reduced hippocampal-cortical integration associated with focused attentional states and diminished self-referential processing (Fig.). Under music conditions, inhibition from the aHip R to the aHip L was observed, potentially altering interhemispheric coordination of memory representations through asymmetric dampening of hippocampal connectivity (Fig.). The strongest connectivity changes were elicited during eyes-open movie stimuli, where inhibition from the IPC L, IPC R, and mPFC to the hippocampi and intra-cortical excitation predominated, in response to a shift toward externally driven visual processing. Excitation from the aHip L to the mPFC, IPC R, and IPC L during this condition may indicate heightened integration of narrative memory with evaluative processes, potentially reflecting a redistribution of influence in this network from higher-order regions to subcortical-driven memory integration (Fig.). The dynamic nature of these connectivity changes highlights how psilocybin supports neural adaptability, providing a temporal window for context-dependent neural reorganisation. These directed connectivity changes o↵er mechanistic insights that bridge preclinical and post-psilocybin imaging evidence of hippocampal plasticity in humans with acute mechanisms, illustrating how set and setting modulate brain function at the circuit level. This suggests that structuring a sensory and cognitive environment can shape neural activity and may be used to guide behavioural outcomes.
Our study included meditation training for half of the participants (Fig.). By assigning participants to meditation rather than allowing self-selection, we aimed to assess the clinical utility of structured meditation training before psilocybin while addressing selection bias inherent in meditation-psychedelic studies. Engagement metrics, such as session attendance rates ( 6 weeks) and independent practice completion (ave. 85 min per week), indicated consistent participation across the cohort. However, neither di↵erences in acute psilocybin e↵ects nor next-day mindset between meditators and non-meditators (11-D ASC, MEQ30) were statistically significant. Additionally, it did not lead to statistically significant changes in connectivity metrics (Fig. Extended Data 13). Music, a contextual element known to shape subjective experiences under psychedelics, played a central role in the study design. During MRI and EEG sequences, participants listened to ethereal music selected for its potential to enhance emotional depth and transcendence while harmonising with scanner noise. Participants rated the music as significantly more meaningful and emotionally resonant under psilocybin compared to non-psilocybin conditions (15-25% increase with a robust Cohen's d e↵ect size > 0.67 and Mann-Whitney U-test p-values p < 0.001). The e↵ect of music used throughout the study design were also verified using the Aesthetic Experiences in Music scale (AES-M) collected at the end of the administration day (Fig., see Methods: Reported measures for details).
Context sensitive psychedelic-induced neural dynamics observed using MRI were confirmed using 64-channel wet EEG, both in the power spectrum and in the signal complexity. Our findings extend human and preclincal observations of desynchronised local neural activity under psilocybin in rodent models, by revealing that these neural alterations are modulated by sensory context-manifesting robustly in both largescale MRI and high-density EEG measures. We found that psilocybin expanded the power spectrum, with decreases in theta, alpha, and beta power and increases in gamma during eyes-closed conditions, while movie-watching attenuated these changes, except for gamma increases in early visual areas (Fig.). Meditation and music recordings aligned in EEG, showing similar power profiles across frequencies during no psilocybin and under psilocybin, suggesting a shared introspective neural signature (Fig.). Psilocybin also reduced the di↵erence between eyes-open and eyes-closed alpha power by 48% (robust Cohen's d = 0.79, Mann-Whitney U-test right-tailed p-value < 10 19 ), suggesting a unification (or integration) of internallyand externally-focused neural processing (Fig.and ). Signal diversity, quantified via Lempel-Ziv complexity, was highest in eyes-closed conditions (seefor similar results), linking internally generated perceptions with enhanced complexity and reinforcing the convergence of meditation and music states (Fig.). Alpha-band activity (8-12 Hz), typically associated with suppressing visual input during eyes-closed conditions, decreased under psilocybin, particularly in visual regions. This reduced sensory filtering, combined with changes in MRI functional connectivity patterns, suggests diminished distinctions between eyes-open and eyes-closed conditions. The convergence of the MRI and EEG findings confirms that this e↵ect cannot be explained solely by vascular changes caused by psychedelics.
ML embeddings' reorganisation of networks into clusters under psilocybin was linked to positively felt self-and boundary-dissolving subjective e↵ects, (Fig.). The relative activation of these networks regulates the balance between internally (e.g., DMN) and externally directed processing, a dynamic that shapes perception and cognition. Their increased integration under psilocybin suggests a shift toward more flexible connectivity and altered functional interactions. The classification accuracy associating degree of integration with self-and boundary-dissolving e↵ects suggests these embeddings capture a connectivity state in which internal and external processes become less distinct. We define this heightened sense of interconnection with the environment as a state of 'embeddedness'. In perceptual terms, embeddedness describes how the self is situated within a broader context. In neurobiological terms, embeddedness describes the functional arrangement of brain regions, showing increased integration within and between networks. Our construct of embeddedness draws upon computational advances demonstrating that the brain encodes meaningful probabilistic relationships across highdimensional spaces and adapts flexibly to environmental uncertainty. We found that CEBRA embeddings formed distinct clusters corresponding to psychological and environmental contexts under psilocybin, but not during imaging with no psilocybin (Fig.). Independently, TAVRNN findings suggest that flexible, integrated connectivity aligns with the highly separable domains observed across di↵erent contexts in CEBRA (Fig.) These results not only deepen understanding of psychedelic mechanisms but also address a critical gap in systems neuroscience: how large-scale functional (neural) embeddings change under multiple contextual demands in a pharmacologically altered state. Furthermore, applying ML embedding techniques adds critical nuance to our interpretation of modularity changes. These embeddings revealed that the observed reduction in within-network connectivity, as measured by functional connectivity, is underpinned by coherent within-network clustering (Fig.). These findings indicates that hidden network integration underlies previous reports of desynchronisation. Moreover, CEBRA embeddings showed that the greatest classification accuracy covaried with positively felt associative subjective e↵ects, which express the emotional and transformative quality of embeddedness (e.g., spiritual, blissful and unitive), while less for sensory and no association with negative e↵ects (Fig.). Additionally, the link we identified between positively felt associative e↵ects and improved mindset (Fig.) was replicated with CEBRA embeddings, showing classification accuracy covaries with improved mindset scores (Fig.Data 8), suggesting these embeddings correspond with psychologically adaptive outcomes. This indicates ML-based embeddings are moderately associated with a brain state where individuals perceive themselves as more integrally connected across personal, social, and ecological spheres. The interpretation of the positively felt psychedelic state as psychologically adaptive is reinforced by earlier reports of changes to sense of interconnection between self, with others and environment within supportive setting. While future work is needed to test whether ML embeddings generalise across di↵erent populations, these methods have significant application for consciousness studies and psychiatry, suggesting embeddings as a novel diagnostic in basic and applied research that are sensitive to pharmacological, psychological and contextual conditions.
Leveraging a highly powered multimodal dataset and varied set of experimental contexts enabled us to address a major gap in our understanding of context and brain connectivity under psilocybin. Previous research has suggested psychedelics heighten sensory input, thereby increasing the influence of context. We advance this narrative by demonstrating the brain's increased reliance on associative connectivity, moving eyes-closed activity towards a pattern of sensory inputs, as the mechanism which amplifies the influence of context. Using multiple analyses methods, we provided evidence that this increased reliance on associative connectivity transforms how brain activity aligns with psychological and sensory context inputs during the psychedelic state. The GFC analysis demonstrated the rebalance of sensory and associative connectivity under psilocybin. During eyes-closed conditions, reduced sensory coupling was accompanied by an increased associative coupling, potentially enhancing higher-order processes and amplifying their roles in internally generated perception. These changes suggest psilocybin alters how sensory and associative signals are tuned, with sensory region activity increasing beyond no psilocybin levels when visual input was introduced. Our large-sample size advanced this area of research by clarifying controversial findings from earlier studies investigating whether sensory and associative connectivity would increase, decrease, or flip sign under certain preprocessing steps. Further analysis using histograms of GFC values revealed that psilocybin reduced the separation between eyes-open and eyes-closed states, both globally and across individual networks, and was particularly evident in the visual network (Fig.). Psilocybin also redistributed neural signal variability across sensory and associative networks. Increases in BOLD signal variance of ventral, temporal and somatosensory regions and decreased variance of early visual areas during eyes-closed states aligns with the transition toward associative processing that was observed in our GFC findings. These findings advance lines of evidence relating signal deviation to neural flexibility and entropy, demonstrating how psilocybin reshapes the spatial arrangement of signal dynamics in a context-structured manner. The 64-channel wet EEG provided a separate modality that confirmed the pattern of context-sensitive changes we identified in fMRI. Reduced alpha-band inhibition we observed during eyes-closed states, under psilocybin, suggests a mechanism for the spontaneous production of internally-generated visual e↵ects. Lempel-Ziv complexity increased brain-wide during eyesclosed condition, extending previous evidence of these e↵ects in occipital-parietal regions. Music and meditation tuned the power spectrum similarly by reducing alpha power, providing new evidence for a convergent influence of cognitive and environmental context and their potential synergy. Similar entropy e↵ects of external sitmulation under psychedelics have also been reported. Our analyses also advanced emerging preclinical and human evidence of anterior hippocampus-DMN neuroplasticity following psilocybin. For the first time, we demonstrated cognitive and environmental context di↵erentially tune the directed connectivity of these brain circuits beyond their baseline (no-psilocybin) comparison during the acute e↵ects. Our contextual manipulations suggest that curated eyes-open visual stimuli may help clinicians control connectivity changes. This avenue of research, along with the examination of the longitudinal behavioural e↵ects of specific connectivity changes in culturally diverse healthy and clinical populations can advance precision psychiatry and currently remains underexplored. Critically, our use of ML-based embeddings (CEBRA and TAVRNN) establishes new methods to characterise neural dynamics in neuroimaging research. These neural signatures di↵erentiated patterns of psychedelic brain activity across experimental contexts (e.g., rest, meditation, music, movie), with greater classification accuracy associated with the quality of acute subjective experience, particularly interconnectedness and self-dissolution (e.g., mystical, blissful, unity and related subdimensions). This evidence, based on the increased integration and fluidity of cortical and subcortical network connectivity (TAVRNN) and classification accuracy of ML embeddings (CEBRA), supports the construct of embeddedness, defined as the coherent integration of brain network connectivity that corresponds with positively-felt dissolution of self and interconnection between internal and external experience (i.e., sense of embeddedness), which was linked with subsequent adaptive psychological changes (e.g., mindset transformations, LAP-R scores). These findings redefine how context modulates perception under psychedelics, shifting the focus from enhanced sensory-driven processing to recognising the overarching role of altered associative processes in psychedelic transformations of perception. ML-derived neural embeddings were key to these advances, highlighting their utility as tools to measure state-dependent e↵ects and forecast personalised behavioural outcomes in precision psychiatry and consciousness research.
The study protocol was approved by the Monash University Human Research Ethics Committee. The trial was registered with the Australian New Zealand Clinical Trials Registry under the registration number ACTRN12621001375842.
The study was open-label and included two imaging sessions: a baseline (no-psilocybin) session and a session following the administration of a 19 mg dose of psilocybin. Both sessions involved MRI and EEG scans, with four conditions repeated in each: resting state, guided meditation, music listening, and movie watching. In the resting-state condition (8 minutes), participants were instructed to relax and keep their eyes closed while remaining still. During the guided meditation (6:30 minutes), participants received brief instructions via MRI-safe audio, followed by silent periods of meditation. For the music listening condition (11:24 minutes), a curated playlist was designed to evoke emotional depth and resonance. In the naturalistic movie condition (6:00 minutes), participants watched a video of moving clouds without audio. These conditions were studied in both the baseline (no-psilocybin) and psilocybin sessions and were repeated in both fMRI and EEG, allowing for comprehensive cross-modal and longitudinal analyses of brain activity and connectivity. The eligible participants were stratified based on age and sex and then randomly assigned to one of the groups. Half of the participants were assigned to an 8-week mindfulness-based cognitive therapy (MBCT) program Finding Peace in a Frantic World run by a trained and registered instructor, which involved weekly group meetings and daily independent practice; the other half were assigned to a control group with no intervention. Our decision to administer a standardised dose of 19 mg psilocybin, rather than a body-weight-adjusted dose was based on previous research showing limited benefits of body-weight-adjusted dosing. The dosage was administered as one oral capsule and selected through consultation multiple collaborators with previous psychedelic imaging experience. This dose was determined to be tolerable for the majority of healthy adults undergoing imaging procedures, while also su cient to produce significant subjective e↵ects. Several behavioural measures were collected before and during the baseline (no-psilocybin) and psilocybin scans (see Reported Measures). The follow-up conducted the day after psilocybin administration included semistructured, open-ended questions and experience ratings. Further follow-up measures were administered one week, and one, three, six and twelve months after psilocybin administration.
Sixty-five healthy, psychedelic-naive adults aged 18-55 were recruited for the study. Participants were required to have no formal meditation practice and limited previous exposure to meditation. They were first screened via short online survey and then detailed screening was performed by a suitably trained sta↵ member for excluding any psychopathology using the long form SCID-V. Exclusion criteria included a history of psychiatric disorders or suicidality, a 5-year history of substance and/or alcohol use disorder, first-degree relatives with a diagnosed psychotic disorder, a history of major neurological disorders including stroke and epilepsy, use of hallucinogens within the last six months, a formal meditation practice in the last 6 months or extensive previous exposure to mindfulness mediation, and contraindicated medications. Participants were also screened for MR contraindications and provided informed consent. On the day of psilocybin administration, they were assisted by a study doctor, researchers, lab sta↵ with relevant training, and volunteers from the community drug harm reduction support organisation "Dancewize". Individual experiences varied the day after psilocybin administration, with some participants reporting transient headaches during the night (n=3). Three participants required follow-up support from a clinical psychologist familiar with psychedelic integration. These follow-up calls were conducted over the phone, and no further assistance was required. Two participants didn't complete the psilocybin imaging session (neither MRI nor EEG due to psilocybin-induced back pain in one case and an incidental finding on their baseline scan in the other, and one only completed part of the resting-state MRI scan due to adverse responses (a combination of dysphoria, panic, and distressing closed-eye visuals in the MRI), two others didn't complete the second EEG session only due to delayed onset of feeling overwhelmed, and one was excluded because they disclosed having extensive meditation experience at the time of baseline imaging, which was an exclusion criteria. Thus, fMRI raw data is available for sixty-four participants in the baseline session, and sixty-one participants in the psilocybin session. EEG raw data is available for sixty-four participants in the baseline session, and fifty-nine participants in the psilocybin session.
Reported measures were part of a broader assessment conducted before and longitudinally after psilocybin administration, with a subset integrated into the present analysis. 11-Dimensional Altered States of Consciousness (11D-ASC) scale (66 items)was administered post-EEG at baseline and at the end of the psilocybin session (approximately 320 minutes after dose) to assess subjective alterations in consciousness. The Mystical Experience Questionnaire (MEQ30)was also collected at this time to measure mysticaltype experiences, along with the Aesthetic Experience Scale-Music (AES-M), used to assess the intensity of emotional and aesthetic responses to music. The AES-M scale is derived from the Aesthetic Experience Scale (AES), refined through psychometric validation, and adapted for music-related experiences. We adapted the instructions of this measure for collection on the day of psilocybin by asking participants to respond with reference to their musical experience during the psilocybin session. The order of scale presentation was randomised. One day after psilocybin, participants provided selfreported intensity ratings assessing the perceived strength of the experience, meaningfulness ratings evaluating personal and spiritual significance, and a novel mindset measure capturing changes in psychological state. See Supplementary Material for mindset measure details. To assess the perceived meaningfulness of the experience, participants were asked: "Would you rate the experience among the most meaningful and spiritually significant experiences of your life?" If they responded yes (n=), they were then asked: "Where would you rate the experience among the most meaningful and spiritually significant experiences of your life?" Response options included top 200 (n=1), top 100 (n=0), top 50 (n=3), top 10 (n=9), and top 5 (n=24). The Nature Relatedness Scale (NR-6)and Life Attitudes Profile-Revised (LAP-R)-assessing death acceptance, coherence, and purpose-were collected before psilocybin and one month post-administration to examine changes in ecological connectedness and existential meaning.
Structural MRI data was acquired using a Siemens 3 Tesla Magnetom Skyra scanner at Monash Biomedical Imaging, Monash University, Australia. T1-weighted (T1w) anatomical images were obtained for each participant during two sessions: at baseline (no-psilocybin) and on the psilocybin administration day. The images were acquired using a 3D magnetisation-prepared rapid gradient-echo (MP-RAGE) sequence with a 32-channel head coil. The acquisition parameters were as follows: repetition time (TR) of 2300 ms, echo time (TE) of 2.07 ms, 192 slices per slab, 1 mm slice thickness, and 1 mm isotropic voxel size. The parallel acquisition technique mode used was GRAPPA, with an acceleration factor (PE) of 2 on the baseline day and 3 on the psilocybin administration day. A T2-weighted anatomical image was obtained only during the baseline session. The acquisition parameters were: TR of 3200 ms, TE of 452 ms, 176 slices per slab, 1 mm slice thickness, and 1 mm isotropic voxel size, with an acceleration factor of 2. Blood-oxygenation-level-dependent (BOLD) fMRI data was collected using a multi-echo, multi-band, echo-planar imaging, T2*-weighted sequence. The acquisition parameters were: TR of 910 ms, multi-echo TE of 12.60 ms, 29.23 ms, 45.86 ms, 62.49 ms, multi-band acceleration factor of 4, field of view of 206 mm, RL phase encoding direction, and 3.2 mm isotropic voxels. The scan durations were: resting state with eyes closed (8 minutes, 505 volumes), audio-guided meditation with eyes closed (6:30 minutes, 405 volumes), music listening with eyes closed (11:24 minutes, 728 volumes), movie watching (6:00 minutes, 372 volumes). The structural and functional MRI images acquired from the Siemens scanner were converted into the Neuroimaging Informatics Technology Initiative (NIfTI) format and organised according to the Brain Imaging Data Structure (BIDS) 1.7.0.
Quality control was performed using the MRIQC BIDS app, which uses structural and functional MRI images to compute several quality metrics. These include the temporal signal-to-noise ratio and the frame-wise displacement that quantifies head motion. Principal component analysis of these metrics revealed some outliers in the structural and functional images, which were further inspected visually. This process led to the exclusion of 6 structural T1w images from the psilocybin session, while all the T1w images from the baseline session were retained. As for the functional MRI images, 7 subjects had one or more conditions excluded from the analysis, i.e., one or more among rest, meditation, music, or movie. Most of these outliers corresponded to scans where the head motion was large (mean framewise displacement greater than 0.5, which is often used as an exclusion threshold).
Anatomical and functional MRI data was preprocessed using fMRIprep 22.0.2. The anatomical MRI preprocessing involved correcting T1-weighted images for intensity nonuniformity, skull-stripping, and segmenting brain tissues. The images were then registered, brain surfaces reconstructed, and spatial normalisation performed using the ICBM 152 Nonlinear Asymmetrical template version 2009c (MNI152NLin2009cAsym). The functional MRI preprocessing involved generating a reference volume and skullstripped version from the shortest echo of each BOLD run, estimating head-motion parameters, and performing slicetime correction. The BOLD reference was co-registered to the T1w reference, confounding time-series were calculated, and the BOLD time-series were resampled into MNI152NLin2009cAsym space. The preprocessed and optimally combined data produced by fMRIprep was cleaned via a single regression in SPM12 (Johnusing the general linear model. The regressors were: the white matter and cerebrospinal fluid signals computed by fMRIprep, the frame-wise displacement, and the non-BOLD components identified via multi-echo ICA performed using tedana (version 0.0.12). In short, multi-echo ICA identifies non-BOLD components that are independent of the echo time. Finally, to enable surface-based analyses, the cleaned volumetric data was projected to the FreeSurfer left-right-symmetric cortical surface template with 32000 vertices for each hemisphere (fsLR32k).
The standard deviation was computed for the time series of each vertex on the cortical surface. The percentage change (psilocybin vs no-psilocybin) was computed for each participant and then averaged across participants to obtain the mean standard deviation percentage change map shown in Fig..
For each cortical vertex, Pearson's correlation to all other vertices in the same brain hemisphere was computed, transformed to Fisher z-values, and averaged. This calculation yielded a global functional connectivity (GFC) map where each vertex value represents the mean correlation with all other vertices in the same hemisphere. This generated one map for each participant, in each session (baseline and psilocybin), and each condition (resting-state, meditation, music, movie). The histograms of GFC values were used to compare the sessions and conditions, ignoring the spatial distribution across the cortex (Fig.). To appreciate the spatial distribution, the GFC percentage change was computed (psilocybin day minus baseline, then divided by baseline) for each participant and then averaged across participants, yielding the mean GFC percentage change maps shown in Fig.. Threshold-free cluster enhancement (TFCE)was used to identify significant clusters in the spatial maps without defining arbitrary thresholds for cluster size. Intuitively, TFCE tests all thresholds and gives higher scores to clusters of vertices that are both large and survive increasingly stringent thresholds. TFCE was computed using the PALM software with the default parameters (see/master/palm defaults.m). The statistical TFCE maps were thresholded at p < 0.05 with 1000 permutations and a Gamma approximation for the distribution tail (computing permutations is computationally very costly; in our tests, the approximation produced nearly identical clusters to those obtained using 10000 permutations but reduced the computation time by one order of magnitude). Once the clusters of increased and decreased GFC were obtained, they were used to mask the GFC maps of the e↵ect of psilocybin (percentage change) to hide the vertices outside these clusters (Fig.).
We recommend not performing global signal regression (GSR) when computing the GFC, although it has been common practice in previous research. Mathematically, GSR alters the covariance matrix of the data such that the average of each row is zero. Since GFC is computed by averaging the rows, it would become zero for each vertex across the cortex. The reason why GFC is not exactly zero in the literature using GSR is that the Pearson correlation matrix is used instead of the covariance matrix. The correlation matrix is a rescaled version of the covariance matrix where all the entries are divided by the standard deviations so that they are between 1 and 1. To achieve this normalisation, each entry ij of the covariance matrix is divided by the product of the standard deviation of vertices i and j, that is, by p ii p jj . If all the vertices had the same standard deviation, this normalisation would equally rescale all entries and the average of each row of the correlation matrix would still be zero. In practice, the standard deviation varies across cortical vertices (Fig.) so the average of each row of the correlation matrix is not exactly zero, but typically a small value. We argue that this complicates the interpretation of the GFC, confounding the intended goal of quantifying "the extent to which a cortical vertex is positively/negatively correlated with the other vertices, on average" with "the extent to which a vertex is positively/negatively correlated with vertices having large standard deviation, on average". Relatedly, plotting z-scored GFC values confounds the interpretation in two ways. If GSR is applied, z-scoring hides the fact that GSR significantly reduces the magnitude of GFC values. More generally, z-scoring is problematic when used to report GFC di↵erences between two conditions. For example, in the main text, we have reported the percentage change between the psilocybin and no psilocybin across conditions (Fig.). Positive and negative values indicate an increase or decrease in GFC, respectively. After z-scoring the changes, this interpretation is no longer possible: a positive z-score only indicates that the change is higher than average, regardless of its original sign (see Fig.). If the average were negative, negative values slightly above the average would have a positive z-score.
The denoised, volumetric data was divided into parcels for further functional connectivity analysis. The Schaefer parcellationwas used to partition the cortical voxels into 300 parcels, each belonging to one of the following 7 resting-state networks: Visual, Somatomotor, Dorsal Attention, Limbic, Salience/Ventral Attention, Default Mode, Control. The Melbourne subcortical atlaswas used to divide the subcortical voxels into 32 parcels including subdivisions of the striatum, thalamus, hippocampus, amygdala and globus pallidus. (This atlas is specified in voxel space, hence the use of volumetric rather than surface data for this analysis.) Combining the cortical and subcortical parcellations gave a total of 332 parcels. To extract a representative time series for each parcel, the first principal component of the voxel time series within each parcel was computed. The parcel-wise functional connectivity matrix was obtained by computing Pearson's correlation between the time series of each pair of parcels. The within-network FC was computed by averaging the FC values within the diagonal blocks of the FC matrix, each corresponding to the pairwise correlations between parcels belonging to the same functional network. Similarly, the between-network FC was computed by averaging the FC matrix entries outside the diagonal blocks, i.e., the correlation values between pairs of parcels belonging to di↵erent networks (Fig.). The Mann-Whitney U-test was performed in MATLAB using the 'ranksum' function and the right-tailed option. That is, the alternative hypothesis is that the median within-or between-network FC under psilocybin is lower than the baseline median.
The FC modularity was computed using the Brain Connectivity Toolboxwith the 'negative asym' option for asymmetric treatment of negative weights. The Mann-Whitney U-test was performed in MATLAB using the 'ranksum' function and the right-tailed option. That is, the alternative hypothesis is that the median modularity under psilocybin is lower than the baseline (no-psilocybin) median (Fig.).
Spectral dynamic causal modelling (DCM) is a method used to infer the e↵ective connectivity between brain regions from fMRI time series. E↵ective connectivity is defined as the influence one neural system exerts over another, as opposed to functional connectivity, which describes statistical dependencies among BOLD signals. DCM employs a forward generative model with neuronal and observation equations. The neuronal model is a linear stochastic di↵erential equation that describes the dynamics of hidden neuronal states. It models how the activity in one brain region is influenced by the activity in other regions, as well as by endogenous fluctuations in neuronal activity. The observation function maps the neuronal activity to the observed BOLD signal by modelling the biophysical processes involved in the haemodynamic response. Spectral DCM estimates the parameters of this model by fitting the generative model to the cross-spectral density of the observed BOLD signals, a second-order statistic that captures the correlations between time series at all time lags. By fitting the model to the crossspectral density, spectral DCM can estimate the e↵ective connectivity between brain regions, as well as the hemodynamic parameters and the spectrum of the endogenous fluctuations. Because DCM is a Bayesian approach, all model parameters are equipped with prior distributions, which are updated based on the observed data to produce posterior distributions over the parameters. Spectral DCM explicitly models the endogenous fluctuations in neuronal activity, which makes it ideal for applications to the resting state and conditions that do not involve strong experimental inputs or block task designs. Here, we have applied spectral DCM to infer the e↵ective connectivity of each subject in each condition. The analysis focussed on six regions belonging to the DMN, each modelled as a 6 mm sphere centred on the following coordinates: Left anterior hippocampus (aHip L; X=-26, Y=-16, Z=-20), right anterior hippocampus (aHip R; X= 28, Y=-16, Z=-20), left inferior parietal cortex (IPC L; X=-44, Y=-60, Z=24), right inferior parietal cortex (IPC R; X=54, Y=-62, Z=28), medial prefrontal cortex (mPFC; X=2, Y=56, Z=-4), and posterior cingulate cortex (PCC; X=2, Y=-58, Z=30). The subject-level estimates were combined at the group level in a Bayesian fashion, accounting for the varying degree of uncertainty in the estimates across subjects. Specifically, Parametric Empirical Bayeswas used to compute the group-level e↵ective connectivity at baseline and its change under psilocybin (Fig.). All DCM analyses were performed using SPM12.
Contrastive Embeddings for Behavioural and Neural Analysis (CEBRA) is a self-supervised learning framework that extracts interpretable and consistent low-dimensional embeddings from high-dimensional neural and behavioural datasets. By leveraging contrastive learning and auxiliary variables, such as time or behavioural labels, CEBRA enables analyses of population-level trajectories. In this study, we used CEBRA-Time, a variant of CEBRA tailored for temporal analyses, to explore neural trajectories derived from fMRI data. CEBRA-Time is a contrastive learning-based method designed to extract meaningful latent representations from time series data, such as fMRI signals. It learns a low-dimensional embedding by maximizing temporal coherence while preserving task-related variability, allowing the transformation of highdimensional neural data into structured latent trajectories. This enables the discovery of subject-specific neural dynamics in a more interpretable space. Here, for each participant, we concatenated fMRI time series spanning four consecutive conditions: resting state, meditation, music listening, and movie watching (Fig.). This comprehensive input captured temporal transitions in neural activity across conditions. Applying CEBRA-Time transformed each time point of these high-dimensional fMRI time series into a three-dimensional point in latent space, revealing neural trajectories unique to each subject (Fig.). These trajectories provided insights into how dynamic brain activity evolves over di↵erent conditions, aligning with self-reported subjective experiences.
While CEBRA captures overall neural trajectories, a deeper understanding requires analyzing how individual brain regions (ROIs) evolve over time and interact with each other. TAVRNN allows us to track these ROI-specific dynamics and their changing relationships, revealing finer details. This perspective inspired our application of Temporal Attention-enhanced Variational Graph Recurrent Neural Network (TAVRNN). TAVRNN is a framework designed for analysing the temporal dynamics of evolving connectivity networks. TAVRNN is a deep learning framework that models temporal connectivity networks between units of a system by leveraging both structural relationships and past dynamics. Through temporal attention, recurrent neural networks, and variational graph techniques, it learns a lower-dimensional latent representation for each unit (ROIs in this work), preserving meaningful temporal patterns while enhancing interpretability. This allows for tracking the evolution of ROI activities over time, revealing insights into their dynamic interactions. TAVRNN captures both local temporal patterns (how individual ROIs change over time) and global temporal patterns (how the relationships as a whole evolve dynamically). By leveraging latent information from the structure of time-varying functional connectivity networks, TAVRNN enables a robust representation of individual ROI activity and their changing relationships in a low-dimensional space (Fig.). Specifically, a temporal attention mechanism assesses the topological similarity of the network across time steps, incorporating varying time lags to capture complex network dynamics more e↵ectively. This approach is particularly e↵ective in uncovering temporal changes in network structures and their alignment with external behaviours or stimuli. In our study, we used TAVRNN to analyse functional connectivity (FC) matrices derived from fMRI time series under four consecutive conditions: rest, meditation, music and movie. For each condition, we computed the FC matrices between 332 brain parcels, which served as input nodes to TAVRNN. In this work, each scan under a specific condition (rest, meditation, music, or movie) is treated as a temporal snapshot. Therefore, the attention mechanism enables the lower-dimensional representation of one scan to inform the representation of the subsequent scans. The model then generated two-dimensional embeddings for each parcel, representing their temporal evolution across conditions (Fig.).
EEG data was recorded using a 64-channel BrainAmp MR Plus amplifier (Brain Products GmbH, Germany) with BrainVision Recorder software (version 1.22.0001, Brain Products GmbH, Germany) and Ag/AgCl electrodes embedded in an actiCAP slim cap according to the standardized 10-10 system. The FCz electrode served as the reference and the FPz electrode served as the ground for online recording. The impedances were maintained below 10k⌦ using an abrasive paste (Nuprep, Weaver and Company, USA) followed by applying a conductive gel (Easycap GmbH, Germany). EEG signals were sampled at 500 Hz with a bandpass filter of 0.01 1000Hz. Participants were comfortably seated in an acoustically absorbent room to minimise environmental noise and their chin strap was securely fastened to ensure head stability. They were instructed to remain as still as possible, avoid jaw clenching, and keep their eyes closed during the recording to minimise artifacts. The EEG recording session included the same four conditions as the fMRI described above. The naturalistic movie (5 minutes) was presented first, followed by the three eyesclosed conditions: resting state (5 minutes), audio-guided meditation (5 minutes), and music listening (7 minutes). The raw data was acquired as a single continuous recording over all conditions and then split into four separate files using the event markers for the start and end of each condition. Due to technical issues, event markers were not recorded for two participants in the baseline session and one participant in the psilocybin session, so these recordings were excluded from pre-processing and analysis. In summary, pre-processed EEG is available for 63 participants in the baseline session and 58 in the psilocybin session. The EEG data was organised according to BIDS 1.7.0.
The raw EEG signals were preprocessed using the default parameters of the automated RELAX pipelineimplemented in MATLAB using functions from EEGLABand FieldTriptoolboxes. First, the data was bandpass filtered between 0.25 and 80 Hz using a fourth-order Butterworth filter with zero phase, with a notch filter applied between 47 and 53 Hz to reduce line noise. Subsequently, bad channels were removed by applying a multistep process incorporating the "findNoisyChannels" function of the PREP pipeline. This was followed by the initial reduction of the artefacts related to eye movements, muscle activity and drift using a multichannel Wiener filter, after which the data were re-referenced to the robust average reference. Residual artefacts were then removed by Wavelet-enhanced Independent Component Analysis (WICA) with the artefactual components identified by the automated ICLabel classifier. Electrodes rejected during the cleaning process were then restored to the data through spherical interpolation. Finally, all pre-processed data were visually inspected for data quality.
The power spectrum was computed for all electrodes across all participants using the multitaper frequency transformation (1 45 Hz) with a 2-second Hanning window and 50% overlap, implemented in FieldTrip. To quantify the e↵ect of psilocybin, the power difference (psilocybin minus baseline) was computed for each channel and then averaged across subjects. The results are reported in Fig.for each condition and four power bands: theta (4 7 Hz), alpha (8 12 Hz), beta (13 30 Hz), and gamma (30 80 Hz). The Lempel-Ziv (LZ) complexity was calculated in two steps using the LZ 1976 algorithm. First, the time series for each channel was binarised by setting values above the mean to 1 and values below the mean to 0. The resulting binary sequence was then scanned sequentially to identify distinct patterns and build a dictionary of these patterns. The LZ complexity is given by the number of patterns in the constructed dictionary. Regular signals can be represented by a small number of patterns, resulting in low LZ complexity, while irregular signals have a greater diversity of patterns, leading to higher LZ complexity. E9 E11
Create a free account to open full-text PDFs.
Bogenschutz, M. P., Forcehimes, A. A., Pommy, J. A. et al. · Journal of Psychopharmacology (2015)
Barrett, F. S., Johnson, M. W., Griffiths, R. R. · Journal of Psychopharmacology (2015)
Burt, J. B., Preller, K. H., Demirtaş, M. et al. · eLife (2021)
Carhart-Harris, R. L., Bolstridge, M., Rucker, J. et al. · Lancet Psychiatry (2016)
Carhart-Harris, R. L., Erritzoe, D., Williams, T. et al. · PNAS (2012)
Calder, A. E., Hasler, G. · Neuropsychopharmacology (2022)
Carhart-Harris, R. L., Leech, R., Shanahan, M. et al. · Frontiers in Human Neuroscience (2014)
Carhart-Harris, R. L., Muthukumaraswamy, S., Roseman, L. et al. · PNAS (2016)
Garcia-Romeu, A., Barrett, F. S., Carbonaro, T. M. et al. · Journal of Psychopharmacology (2021)
Griffiths, R. R., Johnson, M. W. · Journal of Psychopharmacology (2016)
Gattuso, J. J., Perkins, D., Ruffell, S. G. D. et al. · International Journal of Neuropsychopharmacology (2022)
Griffiths, R. R., Richards, W. A., Mccann, U. et al. · Journal of Psychopharmacology (2006)
Herzog, R., Mediano, P. A. M., Rosas, F. E. et al. · Scientific Reports (2023)
Hirschfeld, T., Schmidt, T. T. · Journal of Psychopharmacology (2021)
Kaelen, M., Giribaldi, B., Raine, J. et al. · Psychopharmacology (2018)
Kwan, A. C., Olson, D. E., Preller, K. H. et al. · Nature Medicine (2022)
Kettner, H., Rosas, F. E., Timmermann, C. et al. · Frontiers in Pharmacology (2021)
Lyons, T., Carhart-Harris, R. L. · Journal of Psychopharmacology (2018)
Ly, C., Greb, A. C., Cameron, L. P. et al. · Cell Reports (2018)
Mediano, P. A. M., Rosas, F. E., Timmermann, C. et al. · ACS Chemical Neuroscience (2024)
Olson, D. E. · ACS Pharmacology and Translational Science (2020)
Palhano-Fontes, F., Andrade, K. C., Tófoli, L.F. et al. · PLOS ONE (2015)
Preller, K. H., Burt, J. B., Adkinson, B. et al. · eLife (2018)
Preller, K. H., Burt, J. B., Adkinson, B. et al. · Biological Psychiatry (2020)
Pokorny, T., Preller, K. H., Kometer, M. et al. · International Journal of Neuropsychopharmacology (2017)
Preller, K. H., Razi, A., Zeidman, P. et al. · PNAS (2019)
Preller, K. H., Vollenweider, F. X. · Behavioral Neurobiology of Psychedelic Drugs (2016)
Ross, S., Bossis, A. P., Guss, J. et al. · Journal of Psychopharmacology (2016)
Roseman, L., Nutt, D. J., Carhart-Harris, R. L. · Frontiers in Pharmacology (2018)
Schartner, M., Carhart-Harris, R. L., Barrett, A. B. et al. · Scientific Reports (2017)
Studerus, E., Gamma, A., Vollenweider, F. X. · PLOS ONE (2010)
Shao, L-X,, Liao, C., Gregg, I. et al. · Neuron (2021)
Stoliker, D., Novelli, L., Vollenweider, F. X. et al. · Biological Psychiatry (2024)
Stoliker, D., Novelli, L., Vollenweider, F. X. et al. · Biological Psychiatry (2023)
Stoliker, D., Preller, K. H., Novelli, L. et al. · Molecular Psychiatry (2024)
Nicol, G. E. · Nature (2024)
Swanson, L. R. · Frontiers in Pharmacology (2018)
Tagliazucchi, E., Roseman, L., Kaelen, M. et al. · Current Biology (2016)
Vollenweider, F. X., Kometer, M. · Nature Reviews Neuroscience (2010)
Watts, R., Day, C. M., Krzanowski, J. et al. · Journal of Humanistic Psychology (2017)
Yaden, D. B., Griffiths, R. R. · ACS Pharmacology and Translational Science (2020)
Papers in Blossom that reference this study
Novelli, L., Stoliker, D., Barta, T. et al. · Scientific Data (2026)