Comparing neural correlates of consciousness: from psychedelics to hypnosis and meditation

Moujaes and colleagues frame altered states of consciousness (ASC) — whether pharmacologically induced (psychedelics) or produced by non-pharmacological means (hypnosis, meditation) — as increasingly relevant to psychiatric treatment. Earlier research has documented phenomenological overlap across different ASC and begun to map neural correlates of psychedelic states, but direct, whole-brain, data-driven comparisons across pharmacological and non-pharmacological ASC remain scarce. The authors note additional gaps: many hypnosis and meditation studies use seed-based analyses that vary in region selection and thus hinder cross-study comparisons, and links between neural changes and behavioural measures of ASC have been relatively neglected. This study aims to fill those gaps by directly comparing two classic psychedelics (psilocybin and LSD) with two non-pharmacological methods (an imagery-based form of hypnosis termed 'Esdaile' and open-awareness meditation). The investigators set out to (i) identify whole-brain neural correlates of each ASC using resting-state functional connectivity MRI, (ii) assess differences between pharmacological and non-pharmacological methods, (iii) test whether connectivity changes are predictive at the individual level with machine learning, and (iv) conduct a preliminary analysis linking ASC-induced behavioural changes to neural connectivity alterations.

The authors combined four existing datasets collected at the same site using the same MRI scanner and an eyes-closed resting-state paradigm; each study included an appropriate within-person control condition. The final sample comprised N=107 healthy participants (psilocybin n=23, dose 0.2 mg/kg orally; LSD n=25, dose 100 μg orally; meditation n=29; hypnosis n=30). The hypnosis condition used an 'Esdaile' protocol involving guided visualisation and deep absorption; meditation was an open-awareness practice. Groups differed in age and gender, so the main imaging analyses used each participant's intervention-versus-control contrast to mitigate between-group confounds. Resting-state fMRI preprocessing and acquisition details are provided in the supplement. Functional connectivity was estimated using ROI-to-ROI (RRC) matrices computed with the Conn19c toolbox. Time series from 132 ROIs (derived from Harvard–Oxford cortical/subcortical plus AAL cerebellar parcels) were Fisher-transformed and grouped into 22 networks. For second-level inference the team performed three primary analyses: a 2x2 ANOVA contrasting pharmacological versus non-pharmacological methods (between-subject factor) and intervention versus control (within-subject factor); direct pairwise comparisons of intervention-induced contrasts; and individual intervention versus its control. Threshold-Free Cluster Enhancement (TFCE) was used for cluster inference and results were family-wise error (FWE) corrected at p<0.05. As quality checks, correlations between mean connectivity change and motion (framewise displacement) or sleepiness were examined, and results without global signal regression (GSR) were reported in the supplement given ongoing debate about GSR. To test predictive value, ROI-to-ROI contrast matrices (lower triangle) were converted into image-format inputs for machine learning. A binary support vector machine (SVM) in PRoNTo discriminated pharmacological versus non-pharmacological ASC, while a multiclass Gaussian Process Classifier (GPC) attempted four-way classification across the interventions. Leave-one-subject-out cross-validation assessed generalisability and nested 5-fold cross-validation tuned hyperparameters, with 5,000 permutations used to test statistical significance. The authors reported total accuracy, balanced accuracy, per-class accuracy and area under the curve (AUC) as performance metrics. Kernel-based classifiers and explicit hyperparameter regularisation were used to mitigate overfitting. Finally, a preliminary behavioural–neural regression was conducted (details in the supplement), relating intervention-induced behavioural changes to connectivity changes for psilocybin, LSD and meditation; hypnosis lacked in-depth behavioural measures and was excluded from that analysis.

Behavioural outcomes confirmed that all four interventions produced altered states: psilocybin and LSD produced significant increases across all 5D-ASC subscales (p<0.05), meditation increased all MEDEQ subscales (p<0.05), and all participants in the hypnosis condition reached sufficient hypnotic depth. Direct behavioural comparison between the two non-pharmacological conditions was not possible because different subjective scales were used. There were no significant differences between psilocybin and LSD on the 5D-ASC subscales. When each intervention was compared to its control, distinct whole-brain connectivity patterns emerged. Both psilocybin and LSD showed (i) increased connectivity between sensory and associative regions, (ii) decreased connectivity between different associative networks, and (iii) decreased connectivity within sensory networks. Hypnosis produced decreased connectivity within the primary visual (V1) network together with increased connectivity between V1 and somatomotor, superior temporal gyrus (STG), anterior default mode network (aDMN) and limbic/anterior parahippocampal (aPaHC) networks. Meditation led to decreased connectivity between posterior DMN regions and secondary visual networks (V2, e.g. fusiform cortex). Comparing pharmacological versus non-pharmacological ASC produced 22 significant cluster pairs (p<0.05 TFCE). Relative to non-pharmacological methods, pharmacological ASC showed decreased connectivity between and within associative networks (for example DMN and STG), increased connectivity between V1 and certain associative networks (e.g. inferior temporal gyrus, dorsal attention network), and decreased connectivity between V1 and other associative targets (e.g. somatomotor). Machine learning analyses demonstrated that pharmacological and non-pharmacological interventions were discriminable at the individual level: a binary SVM classified pharmacological versus non-pharmacological ASC with total accuracy 85.05% (balanced accuracy 84.89%, p=0.0002), class accuracies 83.33% (pharmacological, p=0.0002) and 86.44% (non-pharmacological, p=0.0002). A four-class GPC did not reliably distinguish all interventions (class accuracies: psilocybin 39.13%, p=0.07; LSD 40.00%, p=0.23; hypnosis 66.67%, p=0.01; meditation 44.83%, p=0.16), with confusion driven largely by similarity between psilocybin and LSD. A binary SVM could not distinguish psilocybin from LSD (accuracy 47.92%, p=0.46), but could separate hypnosis from meditation (accuracy 66.03%, balanced 66.03%, p=0.02; hypnosis class accuracy 70%, p=0.03; meditation 62.07%, p=0.07; AUC reported as 0.50). Preliminary behavioural–neural regressions showed that in the psilocybin condition the 5D-ASC subscales 'experience of unity' and 'insightfulness' were associated with significant connectivity clusters (p<0.05 FDR-corrected). In the LSD condition the 5D-ASC subscale 'elementary imagery' related to a significant cluster (p<0.05). For meditation, the MEDEQ subscale 'essential quality' produced borderline-significant clusters (p=0.06). Hypnosis was not included in these regressions due to lack of detailed behavioural data.

Moujaes and colleagues interpret their findings as addressing key gaps about ASC neurobiology by showing four principal points: (i) no single network reached significance across all four ASC methods; (ii) pharmacological and non-pharmacological methods produce distinct functional connectivity patterns that are predictive at the individual level; (iii) hypnosis and meditation differ from one another and each drives distinct differences relative to the pharmacological conditions; and (iv) psilocybin and LSD show overlapping connectivity signatures but different behavioural–neural relationships. The authors emphasise the notable absence of a common network across all interventions despite phenomenological overlap, and highlight that the non-pharmacological methods (hypnosis, meditation) do not share the same whole-brain connectivity pattern. Hypnosis-related changes concentrated on V1: decreased within-V1 connectivity alongside increased coupling of V1 with associative networks (aDMN, pSTG, limbic, somatomotor), a pattern the authors link to guided visual imagery and consistent with prior imaging studies implicating lingual gyrus/V1 in hypnotic processing. Meditation produced decreased connectivity between fusiform (V2) regions and posterior DMN nodes (precuneus, posterior cingulate), which the investigators suggest may reflect reduced visual imagery and mind-wandering during open-awareness practice. For psychedelics, both psilocybin and LSD showed decreased connectivity within sensory networks and between associative networks, together with increased V1–DAN coupling; some substance-specific differences were also noted (for example LSD showed unique V1–frontoparietal and language network increases, whereas psilocybin produced specific STG–DAN/sLOC changes). The replication of broad patterns across different preprocessing pipelines and analysis methods is reported as supporting the robustness of these observations. The authors point out a salient contrast: connectivity between V1 and somatomotor areas moved in opposite directions under hypnosis (increased) versus LSD (decreased), which they suggest may reflect different origins of visual-perceptual alterations — top-down imagery in hypnosis versus amplified internally driven excitation in psychedelics — while noting that causal inference is not possible from functional connectivity measures. Clinical and scientific implications are framed cautiously. The investigators stress that their results concern acute effects and that how these map onto longer-term therapeutic outcomes remains uncertain, although acute subjective experiences may relate to treatment response in psychedelic-assisted therapies. They propose that acute neural signatures could help develop biomarkers for patient stratification and that pharmacological and non-pharmacological interventions may engage distinct circuits that could be combined for synergistic effects. Key limitations acknowledged by the authors include combining datasets with differing scan parameters and demographic differences (addressed with within-subject contrasts and covariate checks), shorter scan time for the meditation dataset which may reduce reliability, inconsistent behavioural measures and missing detailed hypnotic behavioural data, inability to correct for physiological noise due to lack of recordings, and the unresolved methodological debate around global signal regression. The authors conclude by calling for more whole-brain, data-driven studies of hypnosis and meditation and for routine linking of behavioural and neural data to resolve distinctions between ASC methods.

The authors conclude that their comparative, whole-brain approach advances understanding of ASC mechanisms and underscores caution when equating different methods of inducing ASC. Although pharmacological and non-pharmacological interventions can be usefully categorised — and such categorisation predicted individual-level ASC type using functional connectivity — clear differences exist, especially between hypnosis and meditation. The study highlights the need for further whole-brain investigations of non-pharmacological ASC and for integrated behavioural–neural analyses to reveal distinctions that are not apparent from neural or behavioural data alone. Finally, the authors suggest these acute neural signatures may inform biomarker development and therapeutic strategies, while recognising the preliminary nature of clinical implications.