Psychedelic Resting-state Neuroimaging: A Review and Perspective on Balancing Replication and Novel Analyses

The investigators performed a structured literature search on PubMed using terms that included “mescaline”, “psilocybin”, “dimethyltryptamine”, “Lysergic acid diethylamide” and “ayahuasca” combined with fMRI. The initial search returned 75 records; after excluding non-classical-psychedelic papers, reviews, non-English items, non-MRI work, population studies and preclinical studies the yield was reduced to 48, and restricting to studies that included resting-state fMRI left 31 publications identified via the search. The authors then added three additional relevant papers that did not appear in the initial query and noted three further relevant manuscripts available on preprint servers at the time of writing. From these sources the review collated information on unique datasets, study samples and analysis methods. The extracted characteristics included sample sizes, drug type and dosing strategy, route of administration, timing of scans relative to drug administration (acute versus persistent effects), population type (healthy volunteers versus clinical cohorts), whether plasma drug levels were measured, scan parameters such as duration and eye condition, and the specific analytic approaches applied (seed-based, network-based, voxelwise, parcellation strategies and novel computational models). Details and specific parcellation definitions were referenced to a supplementary table in the original paper; figures in the manuscript summarised the distribution of analysis methods across the reviewed articles.

Across the assembled literature the review identified 16 unique datasets that have yielded 37 empirical resting-state fMRI articles. Dataset sizes ranged from 9 to 24 participants. In terms of compounds, nine datasets used psilocybin, four used LSD and three used ayahuasca. Dosing strategies varied: seven datasets used fixed doses while nine dosed by bodyweight; one study investigated low, possibly sub-perceptual doses whereas 15 datasets examined psychoactive doses. Route of administration was predominantly oral (14 datasets) with two datasets using intravenous (IV) delivery. Most datasets (15) were collected in healthy volunteers and one dataset involved a clinical cohort of individuals with treatment-resistant depression. Eleven datasets acquired scans while participants were acutely under the influence; five datasets investigated persistent effects at various post-administration intervals (24–48 hours, one week, one month, three months). Notably, only three studies correlated plasma drug concentrations with connectivity changes. Publication output was highly skewed: eleven datasets produced one publication each, three produced two publications each, and two datasets generated nine and thirteen publications respectively. Overall, 54% (20 of 37) of the published resting-state fMRI literature derived from those two highly re-analysed datasets. Both of these datasets were IV studies and exclusively sampled experienced psychedelic users. The review highlights the benefits of data sharing that enabled extensive reanalysis, but also emphasises the opportunity and necessity for independent replication using diverse datasets. Analytic approaches were markedly heterogeneous. Most studies performed static functional connectivity analyses estimating average correlations across whole scans. Fifteen of the 37 articles used seed-based connectivity, predominantly with small, localised seeds; the medial prefrontal cortex (mPFC), posterior cingulate cortex (PCC) and thalamus were the most commonly used seeds (each examined in five studies), though exact seed definitions varied. Twelve articles used network-based connectivity approaches; overall, six different network atlases were applied across the 12 network studies. Five studies performed whole-brain voxelwise analyses. Parcellation choices were inconsistent: 24 studies used atlases derived from functional data, 14 used anatomical atlases and two used atlases integrating structural and functional information. Specific atlases varied widely (e.g., Harvard–Oxford, AAL, Yeo 2011, HCP-ICA, Raichle), and some studies used ICA-derived parcellations from their own data. Other axes of heterogeneity included scan parameters and instructions: resting-state scan durations ranged from 6 to 12 minutes and most studies instructed participants to keep their eyes closed, though a minority specified eyes-open or did not report eye instructions. Music was included during some acquisitions in at least one influential dataset; those music sessions were handled inconsistently across publications (some analysed music and non-music scans separately, others combined or focused on particular sessions), prompting the authors to emphasise that music-listening acquisitions should be clearly demarcated and not conflated with resting-state data. The review also catalogues the application of explanatory models and novel computational methods. Prominent theoretical frameworks tested across studies include default mode network (DMN) disintegration and thalamic gating, with newer hypotheses such as claustro-cortical models emerging. Nine studies applied neurocomputational approaches (for example temporal variability measures, gradient-based connectivity, connectome-harmonic decomposition and leading eigenvector dynamics), though most of these were derived from the two heavily re-used IV datasets. Relatedly, measures labelled “entropy” or similar constructs were operationalised in diverse ways across studies, limiting comparability and motivating the authors’ call for clearer, harmonised definitions and measurement strategies.

Drummond and colleagues interpret the assembled literature as promising but methodologically fragmented: resting-state fMRI has provided important insights into psychedelic effects on brain function, yet no two published studies used identical core processing and analysis pipelines. This heterogeneity—spanning sample characteristics, dosing and route, scan instructions, parcellation choices, seeding strategies and computational metrics—complicates synthesis, replication and the adjudication of competing explanatory models such as DMN disintegration versus thalamic gating. The reviewers highlight several limitations that constrain current inference: small sample sizes, the disproportionate reliance on two IV datasets (which together account for over half of publications), inconsistent handling of music-listening sessions, variable parcellation and seed definitions, sparse measurement of plasma drug levels during acute imaging, and a paucity of clinical cohorts. These factors, in the authors’ view, increase the risk that some reported features of psychedelic effects may not replicate across independent data. To address these issues the paper advocates concrete changes to practice aimed at improving reproducibility and cross-study comparability. Key recommendations include routine measurement of plasma drug concentrations during acute scans to calibrate pharmacokinetic effects, explicit demarcation of music-listening versus resting-state acquisitions, standardising eyes-closed instructions for resting-state scans, sharing data and analysis scripts via public repositories (for example GitHub and OpenNeuro) and adopting the Brain Imaging Data Structure (BIDS) standard. The authors also encourage direct replication of unique analyses using independent datasets, broader testing of explanatory models across multiple compounds and populations, clearer operational definitions for constructs such as entropy, and agreement on standard spatial parcellations for seed and network analyses. They position these changes as necessary for the field to move beyond isolated novel analyses toward robust, clinically applicable biomarkers. Finally, the authors argue that expanding the diversity of psychedelic compounds and patient samples studied will help disentangle effects tied to 5-HT2A receptor agonism from those related to non-selective receptor binding. They note a current gap in systematic comparisons across different psychedelics and call for harmonised research parameters to facilitate such cross-drug and cross-population investigations.

The authors conclude that although the psychedelic resting-state fMRI literature is growing and has produced valuable insights, methodological heterogeneity across studies threatens reproducibility and the field’s ability to converge on stable neural markers. They urge the community to prioritise standardisation, transparent data and code sharing, replication of distinctive analyses on independent data, harmonised operational definitions (notably for entropy) and the routine collection of pharmacokinetic measures during acute imaging. The review is presented as a starting point for community discussion aimed at producing more robust and clinically relevant findings in psychedelic neuroscience.