Navigating the chaos of psychedelic neuroimaging: A multi-metric evaluation of acute psilocybin effects on brain entropy

Psychedelic drugs produce profound alterations of consciousness mediated largely by agonism at the serotonin 2A receptor, and clinical studies combined with psychological support have shown promising therapeutic effects in mood and behavioural disorders. Functional MRI studies have sought neural correlates of these altered states, and one prominent framework is the Entropic Brain Hypothesis (EBH), which proposes that the phenomenological richness of the acute psychedelic state reflects brain-wide increases in the entropy, or information complexity, of functional brain signals. Entropy here is operationalised in multiple ways (for example Shannon entropy, Lempel–Ziv complexity, and sample entropy) and can be applied to static connectivity matrices, dynamic connectivity profiles, or regional time-series. Drummond and colleagues note that twelve prior fMRI studies have each reported a distinct brain-entropy metric, none evaluated in an independent cohort, and that inter-metric correlations have not been systematically examined. To address this gap, the current study evaluated acute psilocybin effects on those 12 previously reported entropy metrics in an independent sample of 28 healthy participants. Scans were acquired before and multiple times after a single oral psilocybin dose and were accompanied by plasma psilocin levels and subjective drug intensity ratings; the investigators tested whether each entropy metric related to these measures and explored inter-metric correlations to assess whether 'brain entropy' behaves as a unified construct.

Twenty-eight healthy volunteers (10 female, mean age 33 ± 8 years) participated in a single-blind, cross-over protocol in which each received either a single oral dose of psilocybin (0.2–0.3 mg/kg; mean dose 19.7 ± 3.6 mg) or ketanserin; analyses reported here focus only on the psilocybin condition. Resting-state BOLD fMRI scans were collected approximately 40, 80, 130 and 300 minutes after administration, yielding multiple post-drug scans per participant (121 scan sessions included after quality control). After each scan, participants provided a subjective drug intensity (SDI) rating and a venous blood sample for plasma psilocin level (PPL); 5-HT2A receptor occupancy (Occ2A) was estimated from PPL using previously reported Hill–Langmuir parameters (EC50 = 1.95 µg/L; Occmax = 76.6%). Preprocessing was harmonised across metrics: data were preprocessed in SPM12 and denoised in CONN with slice-timing (where applicable), unwarping, realignment, co-registration, segmentation, normalisation to MNI space, smoothing (6 mm FWHM for most analyses), aCompCor regression of WM/CSF components, motion regressors and scrubbing, and bandpass filtering (0.008–0.09 Hz). Data were acquired on two Siemens 3T Prisma scanners with different sequences (Scanner A: TR = 2000 ms; Scanner B: TR = 800 ms, multi-band factor 8); scanner B data were temporally downsampled to match a 2 s TR where needed. Parcellations varied by analysis and cerebellar ROIs were removed where not consistently in the field of view. Quality control removed nine scans (remaining n = 121); one scan lacked PPL and another lacked SDI and were excluded from those specific analyses. The investigators implemented 12 previously reported brain-entropy metrics grouped conceptually as: entropy of static connectivity (out-network connectivity distribution, degree distribution, path-length distribution, Von Neumann entropy), entropy of dynamic connectivity (intra-network synchrony distribution, motif-connectivity distribution, LEiDA-state Markov-rate, dynamic conditional correlation (DCC) distribution, meta-state complexity, integration/segregation-state distribution), and entropy of regional dynamics (multi-scale sample entropy and BOLD Lempel–Ziv complexity in spatial and temporal variants). Where necessary, analytic steps from original papers were re-implemented (for example thresholding strategies for graph metrics, sliding-window parameters, clustering for meta-states, and Hilbert transforms for LZ measures). Analyses were run using a uniform pipeline compiled into a Matlab toolbox (Copenhagen Brain Entropy Toolbox, CopBET). Statistical inference used linear mixed-effects models regressing each entropy metric separately on PPL, SDI, or Occ2A with subject-specific random intercepts and covariate adjustment for motion (framewise displacement), age, sex and scanner. A Wald statistic tested associations and region-wise family-wise error rates were controlled using a permutation-based max T method with 10,000 permutations; non-regional metrics used permutation p-values (p_perm). The investigators defined a finding as significant if an entropy metric related to all three psilocybin measures (SDI, PPL and Occ2A, collectively “PsiFx”) at p_perm < 0.05 for non-regional metrics or p_FWER < 0.05 for regional metrics. Whole-brain entropy metrics were correlated pairwise using Pearson correlations to examine inter-metric relationships.

Participants exhibited expected increases in subjective drug intensity and plasma psilocin following psilocybin dosing. Of the 12 entropy metrics evaluated, significant associations with psilocybin measures were observed for a subset, while the majority showed no significant acute effect. Entropy of static connectivity: Shannon entropy of out-network connectivity distributions showed no significant associations with PsiFx across 181 non-cerebellar regions (p_FWER > 0.07). Entropy of degree distribution (using a threshold producing mean degree 27 and across a range of mean-degrees 1–48) was not associated with PsiFx (p_perm > 0.18). By contrast, entropy of the path-length distribution was significantly positively associated with PsiFx at the pre-specified threshold producing mean degree 27 (p_perm < 0.04); effect sizes were weak-to-moderate with Pearson’s rho = 0.39 (PPL), 0.27 (Occ2A) and 0.23 (SDI). Similar positive associations were observed across thresholds producing mean degrees of 22–38. Von Neumann entropy of correlation matrices showed no significant relations with PsiFx (p_perm > 0.35). Entropy of dynamic connectivity: Intra-network synchrony distribution showed no significant association with PsiFx in any of nine networks after multiple comparison correction (p_FWER > 0.98). Motif-connectivity distribution (four-ROI partial-correlation motifs) was generally null across window lengths 15–150 s, with one weak association at 100 s (p_perm < 0.05; Pearson’s rho ≈ 0.25–0.30). LEiDA-state Markov-rate was not associated with PsiFx (p_perm > 0.7). Meta-state complexity exhibited positive associations with Occ2A (p_perm = 0.03) and SDI (p_perm = 0.003) but not with PPL (p_perm = 0.076); reported effect sizes were weak (Pearson’s rho ≈ 0.22 for Occ2A and 0.33 for SDI). Integration/segregation-state entropy did not show significant associations with PsiFx (p_perm > 0.06). Dynamic conditional correlation (DCC) distribution entropy showed a widespread and robust positive relation with psilocybin measures: 35 of 36 network–network connections exhibited significant positive associations (18/36 with p_FWER < 0.0001, 29/36 p_FWER < 0.001, 35/36 p_FWER < 0.05). Effect sizes ranged from moderate to strong (Pearson’s rho 0.35–0.78), with three edges—each involving the default-mode network—showing rho > 0.7. Scale-specific DCC associations were reported as weak–moderate for Scale 1 (rho 0.26–0.47) and negative for Scale 5 (rho −0.27 to −0.49) in some contexts. Entropy of regional dynamics: Multi-scale sample entropy replicated divergent scale-dependent effects: scale 1 (fine temporal resolution, ≈2 s) increased with psilocybin in seven of 17 networks, whereas scale 5 (≈10 s) decreased in 14 of 17 networks, consistent with a previous report. Temporal BOLD Lempel–Ziv complexity (LZct) was associated with Occ2A (p_perm = 0.03) and SDI (p_perm = 0.009) but not significantly with PPL (p_perm = 0.14); reported Pearson’s rho values were small (≈0.17–0.30) and only some associations reached significance. Spatial Lempel–Ziv complexity (LZcs) was not significantly associated with PsiFx (p_perm > 0.6). Null or non-replicated findings: Seven of the 12 previously reported entropy metrics did not show significant acute associations with psilocybin in this cohort. The investigators highlighted discrepancies with previously published motif-entropy values (hippocampal–ACC motifs), noting mathematical concerns in the original report. Correlations between whole-brain entropy metrics were heterogeneous: some metric pairs were positively correlated (for example path-length with degree distribution, LZcs with LZct), some negatively correlated, and others uncorrelated, indicating that different metrics do not uniformly measure a single underlying construct.

Drummond and colleagues interpret their findings as providing nuanced, not uniform, support for the Entropic Brain Hypothesis. They replicated increased entropy of path-length distribution reported previously and identified a novel, robust positive relation between psilocybin effects and DCC distribution entropy, suggesting DCC entropy may be a sensitive candidate biomarker of acute psychedelic action. Meta-state complexity and temporal Lempel–Ziv complexity showed associations with downstream measures of 5-HT2A occupancy and subjective intensity but not uniformly with plasma levels, and multi-scale sample entropy demonstrated replicated scale-dependent divergence (increased entropy at short time-scales, decreased at longer scales) similar to an earlier report. The investigators caution that seven of the 12 metrics previously reported did not replicate in this independent sample, and that whole-brain entropy metrics showed limited inter-correlation. They argue that these mixed results undermine treating 'brain entropy' as a single, unified construct and instead point to a plurality of distinct metrics that capture different aspects of brain dynamics. Possible reasons for non-replication are discussed, including differences in drug, dose, route of administration, imaging sequences and preprocessing choices, and statistical models; the authors also note that some previously reported numerical values raise methodological concerns. Key limitations acknowledged by the study team include use of two different MRI scanners and sequences requiring temporal downsampling, absence of a placebo arm (pre-drug scans were used as baseline), motion confounds associated with higher PPL/SDI despite motion correction and scrubbing, relatively short scan lengths for some sessions (some scans 5 minutes; many recommendations suggest >13 minutes for stable estimates), partial use of multi-band acceleration with uncertain effects on entropy measures, and absence of explicit physiological modelling (respiration, heart rate, vasoconstriction) which psilocybin can affect. They also note their statistical models assume a close temporal relation between entropy measures and adjacent PsiFx measurements and therefore would miss delayed effects. Implications and recommendations put forward by the investigators include the need for transparent reporting and independent replication of brain-entropy metrics, routine collection of concurrent subjective intensity and plasma drug measures in psychedelic fMRI studies, exploration of multimodal imaging (MEG/EEG) where higher temporal resolution may capture different entropy aspects, and further work to determine which entropy metrics reliably index psychedelic effects versus non-specific factors such as wakefulness or motion.

The study concludes that acute psilocybin effects were observed for three of the 12 previously reported brain-entropy metrics evaluated in an independent sample. Dynamic conditional correlation (DCC) distribution entropy showed particularly strong and widespread positive associations with psilocybin measures and is highlighted as a promising biomarker of acute psychedelic effects. Path-length distribution entropy and some Lempel–Ziv complexity measures provided convergent but weaker evidence. Seven metrics did not show significant associations, leading the authors to caution against treating 'brain entropy' as a single construct and to stress the importance of transparent reporting and replication in independent datasets.