Meditation and psychedelic effects on brain network connections: A controlled study of DMT and harmine during a meditation retreat

This study was conducted using a double-blind, placebo-controlled, between-within subject mixed design (Fig. 1). Forty healthy meditation practitioners participated in one of two structurally identical 3-day meditation retreats (n = 19 and n = 21, respectively). Participants were screened in advance and, upon successful enrollment, randomly allocated to receive either placebo or DMT–harmine on the second retreat day (details on the study drug formulation are given below).

To accommodate scanner availability, each retreat was divided into two cohorts that started 1 day apart, allowing for MRI scanning of up to 11 participants per day. The first cohort began on Day 0 and ended on Day 4; the second cohort began on Day 1 and finished on Day 5. One day before the retreat, all participants underwent fMRI scanning at the University Hospital of Psychiatry Zurich ("Pre" assessment). On the second retreat day, the pharmacological intervention took place. On the day following the retreat ("Post" assessment), participants returned to the hospital for a follow-up MRI session, including structural and functional sequences.

Throughout the retreat, participants engaged in full days of meditation practice (e.g., sitting and walking meditation). Subjective experiences were assessed each evening using the Mystical Experience Questionnaire (MEQ) (Barrett et al., 2015). Additional questionnaires included the Nondual Awareness Dimensional Assessment-State (NADA-S), the Toronto Mindfulness Scale (TMS) (Lau et al., 2006), the Sussex-Oxford Compassion Scales for Self and Other (SOCS-S and SOCS-O) (Gu et al., 2020), the Psychological Insight Scale (PIS) (Peill et al., 2022), and the Emotional Breakthrough Inventory (EBI) (Roseman et al., 2019) to cover the four dimensions (1) mindfulness, (2) compassion, (3) insight, and (4) transcendence that were investigated in the original study. One week and 1 month after the retreat, participants additionally filled out psychometric questionnaires. For further information on psychometric assessments and retreat structure, refer to the original publication (Meling et al., 2024).

Out of the 40 participants, data from 39 participants were analyzed herein (22 male, 17 female; 19 DMT–harmine, 20 placebo, mean age 43.5 ± 10.3 years; mean meditation experience 2450 ± 1890 h, refer tofor details). One participant was excluded from the analyses due to excessive head motion during functional scans (refer to). No difference in head motion was observed between groups and timepoints in the sample (refer to). Aside from one participant who reported headaches after DMT+harmine administration, no major adverse events occurred during the study. Supplementary Table S1 Supplementary Table S2 Supplementary Tables S3 and S4

The pharmacological intervention utilized a standardized oromucosal formulation containing DMT and harmine as described elsewhere (Egger et al., 2025; Meling et al., 2024). The formulation included calcium phosphate template-inverted particles (TIP) (Kost et al., 2023), which were separately loaded with either 25% DMT or 25% harmine (freebase weight). These DMT- and harmine-loaded TIP particles were then combined into the same tablets, yielding 1:1 ratio (freebase weight) of DMT and harmine per tablet. Each tablet also contained 3% croscarmellose sodium, along with 0.5% sucralose, 0.5% menthol, and 0.5% peppermint flavor for taste masking. Each tablet contained 30 mg of DMT and 30 mg of harmine (both as freebase weight), with a total of four tablets administered at 30-min intervals (Meling et al., 2024). This resulted in a cumulative dose of 120 mg DMT and 120 mg harmine. Placebo tablets were manufactured using the same formulation process, but containing TIP particles that were not loaded with DMT or harmine. All other ingredients were identical to those in the active tablets with the addition of a small amount of Bitrex^® (denatonium benzoate) to ensure matching appearance and taste. Despite this, >80% of the participants correctly identified their drug allocation at the end of the dosing day (Meling et al., 2024), thus limiting blinding efficacy (Aday et al., 2025).

Structural and functional MR images were acquired on a 3T MR scanner (Achieva 3.0T, Philips, Amsterdam, The Netherlands) equipped with a 32-channel receive head coil and MultiTransmit parallel radio frequency transmission was used. In total, 10–11 participants were scanned per scanning day (2 pre- and post-scanning days per retreat). A T1-weighted image was obtained only on the second imaging visit with a 3D multishot Turbo Field Echo (TFE) sequence with the following specifications: repetition time (TR) = 8.2 ms, echo time (TE) = 3.8 ms, flip angle (FA) = 8°, field-of-view (FoV) = 240 × 240 mm², slices = 160, no interslice gap, voxel size = 1.0 × 1.0 × 1.0 mm³, acquisition time (TA) = 4.53 min. Functional images were acquired on both pre- and post-retreat sessions with a resting-state protocol (eyes closed) using a whole-brain gradient echo planar imaging (EPI) sequence with the following specifications: TR = 1800 ms, TE = 35 ms, FA = 70°, FoV = 220 × 220 mm², matrix size = 124 × 121 mm², slices = 54, no interslice gap, voxel size = 1.7 × 1.7 × 2.0 mm³, number of scanned volumes = 240, TA = 7.40 min.

Preprocessing was performed using the configurable fMRIPrep 23.0.2 (Esteban et al., 2019, 2023) pipeline, which included intensity non-uniformity correction, skull stripping, and spatial normalization to MNI152NLin2009cAsym space. Functional data preprocessing steps included motion correction, slice-timing correction, and co-registration to the T1-weighted image. Further, denoising steps were performed with CONN (Whitfield-Gabrieli & Nieto-Castanon, 2012) release 22.a (Nieto-Castanon & Whitfield-Gabrieli, 2022) and included smoothing of the data with a 6 mm FWMH Gaussian kernel and regression of motion artifacts with the Friston 24-parameter model (Friston et al., 1996), physiological noise correction with the aCompCor50 method including combined white matter and cerebrospinal fluid components explaining 50% of the variance per scan (Behzadi et al., 2007), scrubbing of volumes with a framewise displacement (FD) > 0.5 mm (Power et al., 2014), and temporal filtering (0.008–0.09 Hz) (Hallquist et al., 2013). Additionally, the regression of global signal (GSR) was done as complementary analysis, and results are reported both with (in Supplementary Material) and without GSR as the inclusion of GSR is still an ongoing debate in the (psychedelic) neuroimaging field (Avram et al., 2024; McCulloch et al., 2022). For some analyses (as described below), fully preprocessed and denoised data were transformed from MNI152NLin2009cAsym volumetric space into fsaverage5 surface space using surface.vol_to_surf (Nilearn 0.10.4). A detailed description of each step is provided in the Supplementary Methods.

The RSNs used herein were based on the templates of Yeo et al. (2011) and included the visual network (VIS), auditory–sensorimotor network (ASM), dorsal attention network (DAN), salience network (SAL), frontoparietal network (FPN), and the default-mode network (DMN) (visualized in Fig. 3, Panel A). We excluded the limbic network (LIM) from our analyses given its comparably lower reproducibility and signal-to-noise ratio (Avram et al., 2024; Gordon et al., 2016; Yeo et al., 2014). Network FC maps were computed with CONN. Seed regions included the above-mentioned RSNs (Yeo et al., 2011). Network connectivity was conducted to calculate FC between each RSN and every gray matter voxel in the brain. FC reflects Fisher-transformed bivariate correlation coefficients from a weighted general linear model (weighted-GLM (Nieto-Castanon, 2020)), estimated separately for each seed area (i.e., one of the six RSNs) and target voxel, modeling the association between their BOLD signal time series.

Within-network FC was calculated using first-level maps obtained from the network connectivity analysis described above. For each RSN, these subject-specific connectivity maps were masked with the corresponding network mask to extract the average FC within that network using fslmeants. This yielded subject- and condition-specific FC values for each RSN.

RSN-to-RSN FC (between-network FC) matrices were estimated characterizing FC between each pair of regions among the six RSNs with CONN's ROI-to-ROI FC analysis option. FC strength was estimated separately for each pair of ROIs (i.e., RSNs), characterizing the association between their BOLD signal time series.

Global FC was computed using the BrainSpace toolbox (0.1.10, Python implementation (Vos de Wael et al., 2020)), with data analysis performed in fsaverage5 surface space. To reduce data dimensionality, data in fsaverage5 space (20,484 vertices) were parcellated into 400 regions using the Schaefer atlas (Schaefer et al., 2018), by averaging the values of all vertices within each parcellation. For each scan, a 400 × 400 functional connectivity matrix was computed by calculating Pearson's correlation coefficients between the time series of all pairs of Schaefer parcels. These correlation coefficients were then Fisher z-transformed. Global FC was defined as the average of each parcel's correlation with all other parcels, resulting in one global connectivity value per region and scan. Similar calculations of global connectivity were performed with other psychedelic substances such as LSD, psilocybin, and DMT during acute states (Madsen et al., 2021; Preller et al., 2018, 2020; Timmermann et al., 2023).

Cortical gradient analysis used the same Fisher z-transformed FC matrices calculated for global FC as a starting point. These individual connectivity matrices were subjected to row-wise thresholding to only maintain the top 10% of edges per row, as has been done previously (Bethlehem et al., 2020; Girn et al., 2022; Timmermann et al., 2023). Subsequently, cosine similarity was calculated to generate a similarity matrix that served as input for the diffusion map embedding algorithm, a nonlinear manifold learning method that uses graph Laplacians to identify gradient components at the individual subject level (Bethlehem et al., 2020; Coifman et al., 2005; Margulies et al., 2016). Diffusion map embedding reduces the high-dimensional similarity matrix into a low-dimensional space of embedding components (gradients).

Each gradient represents a dimension of FC pattern similarity, where vertices with many or strong connections are positioned closer together in the embedding space, and vertices with weaker or fewer connections are farther apart. The Euclidean distance in this space reflects the "difference in gradient scores" between regions. A single parameter α controls the influence of data point density on the embedding manifold (α = 0 indicates maximum influence; α = 1, no influence). Diffusion map embedding is characterized by α = 0.5 (Coifman et al., 2005), which balances the consideration of both global and local data point relationships in determining the embedding space.

To enable comparisons between subjects, Procrustes rotation was applied after gradient estimation to align individual gradient components to an all-subjects-baseline group average embedding template with 10 iterations (Bethlehem et al., 2020; Girn et al., 2022; Vos de Wael et al., 2020). This alignment ensures that gradient axes are consistent across scans and individuals. Heat maps showing Pearson correlations of individual scans of the first three gradients with the embedding template after Procrustes rotation can be found in Supplementary Figure S3. We calculated the first 10 cortical gradients with this approach, but only used the first 3 gradients for further analysis, as those already explained ~70% of the variance in the data (DMT–harmine pre-scans: 69.4%, placebo pre-scans: 71.2%, DMT–harmine post-scans: 70.5%, and placebo post-scans: 69.6%). These three main gradients were visualized and compared between groups and time points. To further characterize multi-dimensional differences in hierarchical organization in cortex, we calculated within- and between-network "dispersion" in 3D space (Bethlehem et al., 2020). Within-network dispersion was calculated as the sum squared Euclidean distance of network nodes (parcellations) to the centroid of this network at the individual scan level. Between-network dispersion represents the Euclidean distance between network centroids (Bethlehem et al., 2020).

The following four within- and between-group contrasts were defined and used as second level contrasts for each connectivity measure: (i) Synergy (DMT–harmine > Placebo: Post > Pre), representing the group-by-time interaction that reflects the added effect of the drug to the intervention vis-à-vis group differences from baseline, that is, synergistic effects; (ii) Group differences post-retreat (DMT–harmine > Placebo: Post), which evaluates the effect of DMT–harmine against placebo, including drug+meditation interactions; (iii) Meditation (Placebo: Post > Pre) reflecting the effect of the meditation retreat under placebo; and (iv) Psychedelic-augmented meditation (DMT–harmine: Post > Pre), reflecting combined effect of DMT–harmine and meditation.

For within- and between-network FC analyses, the first contrast was implemented as a two-way repeated measures ANOVA with drug allocation (DMT–harmine > placebo) as between-subject and time (post > pre) as within-subject factor, the second as a between-group ANCOVA comparison of post-intervention scores, with pre-intervention values included as covariates, and the last two as ANCOVAs restricted to the corresponding group, with baseline values of the other group included as covariates.

Network connectivity group-level analyses were performed using a General Linear Model (GLM (Nieto-Castanon, 2020)). For each individual voxel, a separate GLM was estimated, with first-level connectivity measures at this voxel as dependent variables, and group and/or session as independent variables. Voxel-level hypotheses were evaluated using multivariate parametric statistics with random-effects across subjects and sample covariance estimation across multiple measurements. Inferences were performed at the level of individual clusters. Cluster-level inferences were based on parametric statistics from Gaussian Random Field theory (Nieto-Castanon, 2020; Worsley et al., 1996).

Group contrasts for global FC and gradient analysis, including gradient dispersion, were conducted with the SLM function for surface-based linear models (BrainStat 0.4.2, (Larivière et al., 2023)).

If not stated otherwise, all results for voxel-wise analyses were thresholded using a combination of a cluster-forming p < .001 voxel-level threshold, and an FDR-adjusted q_FDR < .05 cluster-size threshold (Chumbley et al., 2010). The corresponding smoothness estimates are reported in Supplementary Table S5.

We additionally evaluated associations between psychometric questionnaires that showed a significant between- or within-group effect at Day 2 (i.e., MEQ, EBI, NADA-S, PIS, and TMS) as reported in Meling et al. (2024) and the connectivity of the significant cluster from the Group differences post-retreat contrast that resulted from the network connectivity analysis using ordinary least squares regression. This cluster was chosen as it denotes consistent and specific FC changes in comparison with the placebo group (see section 3.1). These voxel-wise time series were averaged to obtain a single mean time series per subject, which was then further averaged into one FC value per subject. These FC values were used as the dependent variable and the questionnaire scores and group (i.e., drug allocation) were used as predictors, including main effects and interaction. We used FDR-correction to account for the number of questionnaires modeled. Significant interactions were visualized with a grouped scatter plot with the corresponding Person's/Spearman rank correlations.

To further explore the relationship between changes in subjective experience and FC, we repeated the association analysis using delta scores for each questionnaire to control for individual baseline differences, calculated as the difference between Day 2 (post-drug intervention) and the respective baseline values (assessed on Day 1, i.e., in the evening of the first retreat day).

Voxel-wise FC differences between groups and time points were investigated for each RSN with the whole brain (Fig. 2).

Results from complementary analyses conducted with GSR broadly overlap with these results, refer tofor details. Supplementary Figure S1

For within- and between-network connectivity, no significant changes were observed following correction for multiple comparisons for any contrast of interest (both with and without inclusion of GSR). However, several trends toward significance were observed without FDR correction.

In detail, for within-network FC, the Synergy contrast did not show significant interaction effects (q_FDR < .05), but revealed a main effect of time in VIS (F(1,74) = 6.16, p_uncorrected = .015). Analysis of the other contrasts revealed a significant increase in within-network FC in VIS following the retreat in the DMT–harmine group (t = 2.29, p_uncorrected = .025) in the Psychedelic-augmented meditation contrast (Fig. 3, Panel B). No other RSN showed changed within-network FC. By including GSR as a denoising procedure, we observed a significant effect of time in VIS (F(1,74) = 7.77, p_uncorrected = .007) for the Synergy contrast. The Meditation contrast showed increased within-network FC after the retreat (t = 2.20, p_uncorrected = .031) in VIS. No other contrast/network showed significant changes, see Supplementary Table S6 and Supplementary Figure S2 for details.

Between-network FC analysis results are shown in Figure 3, Panel C. The Synergy revealed interaction effects between VIS–SAL (F(1,74) = 5.31, p_uncorrected = .024), VIS–FPN (F(1,74) = 4.54, p_uncorrected = .037), and DAN–FPN (F(1,74) = 6.62, p_uncorrected = .012) connections. There was no significant effect in the Psychedelic-augmented meditation contrast. In the Meditation contrast, a decrease in FC between VIS–FPN (t = 2.71, p_uncorrected = .001) and DAN–FPN (t = 2.72, p_uncorrected = .001) was noted following the retreat. In the Group Differences Post-Retreat contrast, increased FC was observed between VIS–SAL (t = 2.51, p_uncorrected = .014) and VIS–FPN (t = 1.99, p_uncorrected = .050) in the DMT–harmine group compared with placebo. Including GSR did not have a strong impact on the results, with the same effects observed at the uncorrected level (Supplementary Fig. S2).

We examined global FC differences across time and groups using the four predefined contrasts (Fig. 4), based on the 400-region Schaefer parcellation of the cortical surface. None of the contrasts Synergy, Group differences post-retreat, or Psychedelic-augmented meditation showed statistically significant changes in global FC after FDR correction, regardless of whether GSR was applied. However, in the Meditation contrast, three clusters located in SAL /FPN networks showed significantly decreased global FC after the retreat in the analysis without GSR (FDR-corrected; Fig. 4). These effects were not observed when GSR was applied.

The cortical gradient analysis resulted in a hierarchical axis ranging from lower-level sensorimotor cortices (i.e., "lower" cognitive functions) to higher-level transmodal cortical regions (Gradient 1; Fig. 5), from insular to visual cortex (Gradient 2; Supplementary Fig. S4), and from visual cortex to cortices covered by the FPN regions (Gradient 3; Supplementary Fig. S5). We identified the typical cortical gradient axes previously reported (e.g., Bethlehem et al., 2020; Girn et al., 2022; Sydnor et al., 2021) in all four scan conditions (DMT–harmine Pre/Post, Placebo Pre/Post). However, no significant differences were found between conditions or sessions in any brain region for any of the four defined contrasts. This was true both in surface space and when comparing network-wise gradient scores. Complementary analysis with inclusion of GSR resulted in the same gradient axes, but similar to the standard analysis, did not indicate any statistical differences between groups and conditions for any gradient (results not shown).

Additional within- and between-network dispersion analyses in 3D gradient space did not indicate statistical differences between groups or time points (refer to Supplementary Tables S7 and S8 for results from 2-by-2 ANOVA of within- and between-network dispersion). Parcellation-wise gradient data are shown in Figure 6 in 3D and 2D gradient space (representing gradient values of each parcellation along all three gradients and each gradient pair combination). Similar to the gradient analysis, network dispersion analysis including GSR did not result in significant differences within- or between-networks (data not shown).

We used ordinary least squares regressions to explore the relationships between the cluster found for the Group differences post-retreat contrast with the network connectivity analysis (increased connectivity between SAL and bilateral calcarine sulci (i.e., VIS), reflecting the FC change in this cluster from pre- to post-retreat) and questionnaires targeting mindfulness, emotional breakthrough, insight, and transcendence (Meling et al., 2024). We used FDR correction for the total number of questionnaires used (n = 5). Between-group differences were observed via significant interactions between SAL–VIS FC (contrast Group differences post-retreat; Fig. 2, Panel C) and the questionnaires MEQ on Day 2 (b = 0.0065, 95%CI [0.0014, 0.0117], q_FDR = .036), and TMS on Day 2 (b = 0.0089, 95%CI [0.0019, 0.0159], q_FDR = .036). Post hoc analyses revealed significant associations between SAL–VIS FC and MEQ on Day 2 for the placebo group (r = 0.52, p = .020), while the TMS scores were not significantly correlated with SAL–VIS FC for any group (Fig. 2). To complement these findings, we repeated the association analysis using delta scores (Day 2 minus baseline) for each questionnaire. However, no significant group-by-questionnaire interactions were observed in the analysis of delta questionnaire scores. Refer to Supplementary Tables S9–S12 for OLS model and exploratory correlation results for all tested questionnaires.

Our findings indicate that meditation alone induces subacute changes in several connectivity metrics, consistent with prior research. However, it is important to note that meditation studies usually do not include placebo administration, which may itself influence the subjective meditation experience and neural outcomes. Specifically, FC between multiple networks in the placebo group decreased, ranging from lower-order sensory regions (e.g., VIS) to higher-order attention and transmodal networks (e.g., SAL, DAN, FPN, and DMN). These effects remained consistent after applying GSR.

Our results are in line with previous findings (Ganesan et al., 2023; Hasenkamp & Barsalou, 2012; Kilpatrick et al., 2011; Sezer et al., 2022). For instance, focused attention meditation has been associated with reduced connectivity across multiple brain regions (Ganesan et al., 2023), mirroring the connectivity reductions we observed in the placebo group post-retreat. Decreased VIS–SAL connectivity may reflect enhanced self-awareness (Sezer et al., 2022) or increased attentional control (Hasenkamp & Barsalou, 2012; Kilpatrick et al., 2011), while increased FPN–DMN connectivity has also been linked to improved attentional regulation during meditation (Sezer et al., 2022).

However, some discrepancies with our findings remain; for example, an 8-day meditation retreat reported increased connectivity between SAL–DMN in trained meditators (Vishnubhotla et al., 2021) or other studies reported increased FPN–DMN FC after meditation training (Brewer et al., 2011; Kral et al., 2019). Differences in methodology, participant experience, meditation techniques, the use of a placebo, or retreat duration may explain such variations.

The DMT–harmine administration led to trend-level increases in network integrity in the VIS, and increased between VIS–DAN connectivity. These effects contrasted with those observed in the placebo group, where connectivity generally decreased. A prior fMRI study investigating the combined effects of psilocybin during a 5-day meditation retreat found both increased and decreased FC in the DMN during open-awareness meditation in the psilocybin group (Smigielski et al., 2019). Despite the comparable design, these results were not replicated herein, perhaps due to several differences such as the psychedelic compound used or the analytical focus (i.e., DMN only vs. all Yeo networks).

Previous studies on the subacute effects of ayahuasca reported increased FC within SAL, both decreases (Pasquini et al., 2020) and increases in within-DMN FC with positive correlations to self-compassion measures (Sampedro et al., 2017), and increased FC between SAL–DMN (Pasquini et al., 2020). Decreased between-network FC between VIS and ACC has also been reported (Sampedro et al., 2017). While these studies also involved DMT and harmine administration, direct comparisons with our findings are complicated by differences in dosing and formulation. In our study, DMT and harmine were administered at a fixed 1:1 ratio (120 mg each over 90 min), whereas ayahuasca preparations in (Pasquini et al., 2020) and (Sampedro et al., 2017) involved a lower DMT-to-harmine ratio, with harmine doses approximately comparable with ours. From a pharmacological standpoint, this suggests that our formulation may have elicited stronger psychoactive effects (Egger et al., 2025). However, this was not clearly reflected in the observed connectivity patterns, which may indicate the important modulatory role of non-pharmacological factors such as the setting (meditation group retreat vs. single-person study settings). For a more detailed discussion of pharmacological differences and similarities between ayahuasca and our DMT–harmine formulation, see Egger et al. (2024). Notably, previous studies highlight even longer-term effects of psilocybin. For example, Siegel et al. (2024) reported persistent reductions in functional connectivity between the anterior hippocampus and DMN in healthy volunteers, effects that endured for several weeks post-administration. Further structural evidence comes from a recent preprint, which investigated psilocybin's long-term impact on measures such as diffusion tensor imaging (DTI); it indicated decreased axial diffusivity in prefrontal–subcortical tracts, interpreted by the authors as potential pruning within a neuroplastic context (Lyons et al., 2024). These findings, underscoring prolonged neuroplasticity, are also critically important clinically. Indeed, psilocybin therapy has been shown to alter dynamic connectivity between the anterior and posterior cingulate cortices in patients with major depression, with these changes correlating with alterations in specific metabolites measured by spectroscopy. Moreover, evidence points to reduced between-network connectivity (i.e., modularity) persisting for weeks after psilocybin treatment in patients with treatment-resistant depression (TRD), a change that correlated with clinical improvement (Daws et al., 2022).

Beyond subacute effects, increased FC in the primary visual cortex (V1) during acute ayahuasca intake has been shown previously during a mental imagery task (De Araujo et al., 2012). Together, these findings suggest that ayahuasca alters the activity and connectivity of visual, salience, and DMN areas in healthy individuals. These previous reports are partially supported by our findings for the DMT–harmine formulation combined with mindfulness practice, which appeared to specifically alter VIS connectivity, both within-network and between visual areas and attentional networks (i.e., DAN).

Conversely, the placebo group showed reduced SAL–VIS FC, consistent with prior findings linking this to increased attentional focus and diminished irrelevant visual input during rest in experienced meditators (Hasenkamp & Barsalou, 2012; Kilpatrick et al., 2011). Comparable patterns have also been observed following mindfulness-based cognitive therapy, where decreased SAL–VIS FC during rumination correlated with increased interoceptive awareness (van der Velden et al., 2023).

Contrary to our findings, acute intravenous DMT administration has been shown to reduce network integrity within the VIS and decrease segregation between VIS–FPN (Timmermann et al., 2023). However, our observed increase in voxel-wise FC between VIS and attentional networks may still be functionally related, as these networks, along with FPN, collectively facilitate attentional resource allocation (Menon, 2011; Vossel et al., 2014). As the previous study focused on acute drug effects versus placebo, direct comparisons are challenging.

Our findings revealed consistent group differences in FC, notably increased VIS–SAL connectivity in the DMT–harmine group. This result was robust across analyses comparing post-retreat scans and pre-to-post changes, with similar patterns emerging. Including GSR did not alter this primary finding, though it revealed additional group differences.

The increased VIS–SAL connectivity may reflect stronger engagement of visual and attentional processes during the psychedelic experience in participants receiving DMT–harmine, with residual effects detectable 2 days post-intervention. Another plausible mechanism involves heightened mental imagery induced by DMT–harmine during the retreat, potentially manifesting post-retreat at rest as visual recollections or "visions in reverse", where information flows top-down rather than bottom-up, as in typical perception. In support, neutral or pleasant visual phenomena have also been reported within a week of LSD or psilocybin administration in ~10% of participants (Müller et al., 2022). The SAL's role in detecting and integrating sensory, emotional, and memory-related input (Schimmelpfennig et al., 2023) supports the idea that DMT–harmine facilitates the integration of visual and emotional information—an effect that may extend into the subacute phase. Relatedly, acute DMT has also been shown to increase connectivity between regions involved in visual, emotional, and associative processing, including the amygdala, supramarginal gyrus, and orbitofrontal cortex (Soares et al., 2024). Future studies could further investigate the relationship between the VIS and SAL by quantifying the effective connectivity to assess the directionality between these brain regions.

Additionally, this VIS–SAL cluster showed a significant interaction effect with mystical experience intensity (MEQ) and mindfulness (TMS) scores assessed on the day of drug administration. Interestingly, post hoc analyses revealed a positive correlation between MEQ scores and SAL–VIS FC in the placebo group, but no significant correlations were found in the DMT–harmine group, despite a significant increase in MEQ scores in the latter (Meling et al., 2024). This pattern might suggest that the relationship between SAL–VIS FC and strength of the mystical experience and mindfulness might follow an inverse U-shaped curve, where neither too much nor too little connectivity is optimal. However, this interpretation remains speculative, as group × questionnaire interactions were not significant when using Δ-scores (post–baseline). This suggests that some associations may reflect pre-existing individual differences rather than changes induced by the intervention. This hypothesis could be tested in a dedicated study.

Interestingly, this increased SAL–VIS FC differs from subacute changes observed after ayahuasca without meditation in a previous study, reporting increased SAL integrity and decreased SAL–DMN segregation (Pasquini et al., 2020). Sampedro et al. also reported both increased and decreased FC between DMN, VIS, and SAL regions (Sampedro et al., 2017), underscoring the importance of these networks in post-ayahuasca or DMT–harmine-induced changes.

In sum, our findings showed increased VIS integrity and enhanced VIS–DAN and VIS–SAL connectivity, diverging from our original hypotheses of increased within-SAL FC or stronger VIS–DMN and SAL–DMN coupling. These hypotheses were based on studies with different drug formulations, sample characteristics, and setting variables. For instance, prior meditation retreat studies have reported SAL–DMN connectivity increases (Vishnubhotla et al., 2021). Differences in retreat duration or other time-dependent variables may explain such discrepant findings.

From a broader perspective, the current results highlight neurobiological distinctions between the effects observed in meditation versus psychedelic-augmented meditation. While the meditation retreat with placebo mainly led to decreased FC between several RSNs that more closely reflect established neuroimaging findings of meditation alone (Hasenkamp & Barsalou, 2012; Kilpatrick et al., 2011; Moujaes et al., 2024; Sezer et al., 2022; Vishnubhotla et al., 2021), psychedelic-augmented meditation mainly resulted in increased FC within the visual network and between visual and attention-related networks. Supporting this distinction, a recent study found that FC patterns were similar between acute psilocybin and LSD administration but differed when compared with meditation practiced in the scanner (Moujaes et al., 2024). While this distinction may not be entirely unexpected, it underscores an important point: Combining psychedelics with mindfulness-based interventions may offer complementary benefits in therapeutic settings, potentially enhancing both visual processing and attentional control.

The persistence of VIS–SAL connectivity changes 2 days post-administration indicates a potential window of opportunity for therapeutic interventions such as visual imagery exercises or to explore emotionally salient aspects of the experience in a mindfulness-oriented way.

Both psychedelics and mindfulness-based practices have independently demonstrated efficacy in alleviating symptoms of various affective disorders and warrant further exploration in clinical trials (Andersen et al., 2021; Milliere et al., 2018; Sezer et al., 2022). A potential mechanism of synergy lies in the increased awareness and mindfulness cultivated through meditation paired with the capacity of psychedelics to induce profound alterations in consciousness. This combination may be particularly beneficial during psychedelic therapy sessions, where the depth of the psychedelic experience has been correlated with therapeutic outcomes (Ko et al., 2022).

A novel aspect of our study is the examination of cortical gradients and gradient dispersion in the subacute phase following psychedelic administration. The human principal cortical gradient, extending from the lower-order sensorimotor regions to the higher-order transmodal association cortex, serves as a fundamental organizational axis of the brain [69], whereas the second and third cortical gradients typically represent axes from the visual to insular cortex and from visual to areas covered by the FPN (Bethlehem et al., 2020; Girn et al., 2022), which is also evident in our data.

Recent neuroimaging studies suggest that altered states of consciousness, such as those induced by psychedelics such as psilocybin, LSD, and DMT, acutely disrupt cortical gradient hierarchy (Girn et al., 2022; Timmermann et al., 2023). Psychedelics appear to reduce the differentiation of the evolutionarily younger, transmodal association cortex from lower-level sensory regions, an effect captured by a contraction along the principal cortical gradient (gradient 1). Beyond this, psychedelics also diminish differentiation along gradient 2, which separates visual regions from the insula and somatomotor cortices, and gradient 3, which distinguishes visual areas from FPN regions—suggesting a broad disruption of cortical functional organization (Girn et al., 2022; Timmermann et al., 2023).

This "collapse" of gradient axes aligns with the immersive, boundary-dissolving experiences commonly reported during acute psychedelic states and likely reflects a transient breakdown of established hierarchical organization in the brain. However, in our study, we found no evidence of such a collapse in cortical gradient architecture 2 days after the DMT–harmine or placebo retreat experience. This absence of residual flattening in the cortical hierarchy suggests that these alterations are acute and reversible, normalizing once the drug effects subside. It is also worth noting that our preprocessing pipeline employed a relatively stringent denoising strategy, which may have reduced sensitivity to subtle lasting changes in cortical gradient architecture. To further assess post-acute reorganization, we examined multi-dimensional gradient dispersion—a measure of how functionally distributed or spread cortical regions are within the gradient space (Bethlehem et al., 2020; Pasquini et al., 2023). While gradient dispersion has been shown to increase with aging, reflecting a dedifferentiation of functional brain organization, and to decrease in major depressive disorder (Pasquini et al., 2020), we observed no significant differences between groups or time points in our dataset. This further supports the notion that large-scale cortical functional architecture returns to baseline following the acute psychedelic state.

Our study has several limitations that should be considered when interpreting the findings.

A main issue in determining the synergistic effects of meditation and psychedelic intervention is the absence of both an intervention-free control group and a meditation-free psychedelic group. To fully understand the unique synergy that emerges when combining meditation and psychedelics, it would also be necessary to include a group receiving only the pharmacological intervention, without any meditation practice. Second, our study design included only subacute scans. While this adds to the currently limited literature of subacute effects of ayahuasca, or ayahuasca-inspired formulations (Pasquini et al., 2020; Sampedro et al., 2017), future studies incorporating both acute and subacute time points would provide a more complete understanding of these effects. Another limitation is the blinding efficacy, as most participants (>80%) recognized whether they were in the active or placebo group (for further details see Meling et al. (2024)). However, we note that blinding is a ubiquitous problem in psychedelic research. The FDA has recently made some suggestions to overcome this issue, but a final solution has not yet been reached (Aday et al., 2025; Psychedelic Drugs: Considerations for Clinical Investigations Guidance for Industry, 2023). We prioritized a pharmacologically inert placebo to optimize neuroimaging contrasts, despite reduced blinding efficacy. Active or low-dose controls may better preserve blinding but risk confounding drug-specific neural effects. The specific study sample, primarily white and experienced meditators, limits the generalizability of our findings to broader populations. Notably, experienced meditators actively train to enter altered states of consciousness and may be more accustomed to such experiences than the general population. This familiarity might reduce their susceptibility to the perturbing effects of psychedelics, potentially contributing to the absence of strong effects observed 2 days after the meditation retreat in both treatment groups. Additionally, the group setting during the meditation retreat may have influenced participants' experiences. Emphasis was placed on facilitating an introspective and meditative setting, but some minor degree of indirect interactions/influences between participants cannot be ruled out. The DMT–harmine and placebo groups were placed together, which could have led to an unintentional blending of experiences across conditions, potentially impacting connectivity outcomes and subjective experiences. Also, the study was conducted in two separate retreats, which had the same structure, but the two sessions could have influenced the outcomes of individual participants. Finally, the smoothing estimates for the network connectivity analysis were on the lower side with around 4–4.5 mm in each direction.

In conclusion, our findings reveal that the combined effects of DMT–harmine and meditation differ from meditation alone, with each intervention producing distinct functional connectivity changes that reflect their unique neurobiological mechanisms. Enhanced functional connectivity in visual and attention-related areas following a psychedelic-enhanced mindfulness retreat complements previous studies on psilocybin and ayahuasca in the context of mindfulness practices. The absence of cortical gradient disruption 2 days post-retreat highlights that the reorganization observed during acute psychedelic states is transient, emphasizing the brain's ability to return to its hierarchical organization after the psychedelic experience subsides. These results suggest that psychedelic-augmented meditation may promote a more integrated brain state than meditation alone, potentially enhancing the depth and meaningfulness of subjective experiences.