An international mega-analysis of psychedelic drug effects on brain circuit function

Girn and colleagues describe a field in which acute psychedelic effects on the brain have been studied with resting-state functional MRI, but the literature has remained fragmented, methodologically variable and sometimes inconsistent. Previous single-site studies had suggested that classic psychedelics may reduce within-network functional connectivity and increase connectivity between large-scale networks, yet the exact network-level pattern varied across datasets, drugs and analysis pipelines. As a result, it was difficult to identify which findings were genuinely reproducible across studies and which might reflect site-specific or methodological differences. The authors set out to address this problem with an international mega-analysis that pooled resting-state functional MRI data across 11 datasets from five psychedelic drugs: psilocybin, LSD, mescaline, DMT and ayahuasca. Their aim was to characterise shared and drug-specific changes in brain circuit function during the acute psychedelic state, using a uniform preprocessing pipeline and Bayesian hierarchical modelling to quantify both effect size and uncertainty. The study is framed as a consortium-based effort to provide a more convergent, probabilistic map of psychedelic brain effects and to establish a benchmark for future neuroimaging research in this area.

The researchers conducted a mega-analysis of resting-state functional MRI datasets collected during the acute effects of classic psychedelics in healthy adults. Eleven independent datasets were ultimately included, with access obtained through contact with the primary investigators of earlier studies. The extracted text indicates that the core inclusion criteria were acute resting-state fMRI under a classic psychedelic and a healthy adult sample. Most datasets were double-blind, randomised, placebo-controlled trials, usually with within-subject crossover designs and counterbalanced order; one dataset used a fixed-order design without placebo. The extracted text notes that 13 datasets were considered in total, but 11 were included in the analysis. All imaging data were processed through the same centralised fMRIPrep-based pipeline, with each site running the standardised preprocessing locally. The preprocessing included coregistration, motion correction, slice-time correction, spatial normalisation to MNI space, and generation of standard confound measures such as framewise displacement, DVARS and global signal. The authors excluded subjects and sessions with poor registration, segmentation problems or excessive motion; they removed six subjects entirely and 41 connectomes in total, leaving 519 connectomes for analysis. The final sample was reported as 267 unique participants and more than 500 scanning sessions. To reduce artefacts, the researchers applied four denoising pipelines: aCompCor with and without global signal regression (GSR), and aggressive ICA-AROMA with and without GSR. The main text focuses on the aCompCor results, while ICA-AROMA results were placed in the supplementary repository. Functional connectivity was calculated as Pearson correlations between parcels. The cortical parcellation used the Schaefer atlas grouped into Yeo 17 networks, and subcortical and cerebellar regions were extracted using standard atlases, including a dedicated thalamic parcellation. Analytically, the study first conducted descriptive comparisons of drug-placebo differences in between-network and within-network connectivity, including parcel-level measures of within-network integration and between-network integration. It then applied Bayesian hierarchical models to estimate drug effects for each network and network pair, with study- and drug-level variation explicitly modelled. The Bayesian approach generated posterior distributions rather than relying on binary significance testing, allowing the authors to quantify uncertainty and consistency across datasets and drugs.

Across the pooled datasets, the researchers analysed 267 participants and more than 500 resting-state scans after motion-related exclusions. The descriptive, non-Bayesian analyses showed a broadly similar pattern across the psychedelic conditions: increased between-network connectivity, especially between transmodal association networks such as the default mode and frontoparietal networks and unimodal or sensorimotor networks such as visual, somatomotor and dorsal attention networks. There were also prominent increases in coupling between sensorimotor networks and subcortical regions, particularly the putamen, caudate, thalamus and cerebellum, although these patterns varied somewhat depending on whether GSR was used. In contrast, within-network connectivity tended to decrease across many networks, with the strongest reductions in visual, somatomotor and some default mode subnetworks. At the drug level, psilocybin and LSD showed highly similar patterns, with widespread increases in between-network connectivity and decreases in sensorimotor connectivity. DMT produced the largest apparent response and resembled an amplified version of the psilocybin/LSD pattern, although the dataset was small. Mescaline showed a broadly similar but somewhat more selective pattern. Ayahuasca was the most idiosyncratic, with weaker or absent evidence for the same broad increase in transmodal-unimodal coupling seen for the other drugs. The extracted text notes that some mean effects were visible in matrices, but that these did not always translate into high-confidence Bayesian effects. The Bayesian hierarchical analysis sharpened the picture by identifying the most consistent cross-drug effects. The clearest and most reproducible positive effects were increases in connectivity between unimodal networks, especially visual and somatomotor systems, and transmodal networks including default mode and frontoparietal subnetworks. Strong posterior support was also found for increased coupling between the caudate and visual or somatomotor networks, and more generally between the dorsal striatum and cortical networks. By contrast, thalamic coupling was less robust: although some descriptive analyses suggested increased thalamocortical connectivity, the Bayesian posteriors were more mixed and often overlapped with zero. For within-network connectivity, the Bayesian results indicated weak-to-moderate reductions rather than widespread network disintegration. Many posterior distributions overlapped with zero, and the most reliable reductions were concentrated in visual and somatomotor subnetworks. Effects in default mode, frontoparietal, dorsal attention and salience networks were smaller and less consistent. The authors also report that GSR tended to push estimates in a more negative direction for within-network connectivity, but this was not uniform across all networks and drugs.

Girn and colleagues argue that their mega-analysis provides the most comprehensive synthesis to date of how classic psychedelics alter human brain circuit function. They interpret the main result as evidence for a reproducible cross-drug signature: psychedelics increase functional coupling between transmodal association circuits and unimodal sensorimotor circuits, and also increase coupling between these cortical systems and key subcortical regions, especially the caudate and putamen. In their view, this supports previous ideas that psychedelics relax or flatten the brain’s intrinsic hierarchical organisation, although the functional connectivity data do not establish directionality. The authors position their findings as extending earlier work while also refining it. They say the study confirms the general observation that psychedelics increase whole-brain integration, but identifies which inter-network links are most robust and shows that these effects are more specific than some prior broad claims suggested. They note that previous studies often reported widespread within-network disintegration, whereas their Bayesian analysis found only limited and selective evidence for such reductions. They also highlight the value of subnetwork and subcortical analyses, which revealed distinctions that were previously obscured at coarser scales. They discuss several drug-specific patterns. LSD and psilocybin were described as nearly identical at the network level, mescaline as broadly similar but more selective, DMT as showing the strongest perturbations in raw terms, and ayahuasca as the most idiosyncratic. However, they caution that interpretation of drug-specific differences is limited because the DMT and ayahuasca datasets were small and because the studies differed in dose, timing, route of administration and pharmacology. The authors acknowledge important limitations. These include variation in scanner field strength, voxel size and repetition time across datasets, residual site-specific noise and confounding, head motion under psychedelic conditions, and heterogeneity in study design. They also note that most datasets were double-blind randomised controlled crossover studies, but one lacked placebo and another used a fixed-order design. They emphasise that blinding can be difficult with psychedelics, though they believe expectancy effects are less likely to drive resting-state BOLD findings than subjective or clinical outcomes. Overall, they argue that the consortium-based Bayesian approach helps reconcile fragmented literature and provides a more reliable foundation for future multisite, standardised psychedelic neuroimaging studies.