Multi-metric evaluations of acute psychedelic effects on fMRI brain entropy

Psychedelics are thought to produce altered states of consciousness partly through serotonin 2A receptor agonism, and clinical trials up to Phase IIb have suggested potential benefit in some psychiatric conditions when used with psychological support. In parallel, fMRI studies have been used to investigate whether acute psychedelic effects can be captured as changes in brain entropy, but earlier work had examined a set of 14 different entropy-related metrics in only a few datasets, often with small samples, different drugs or routes of administration, and no independent re-evaluation of the full set of measures. The authors frame this as an important gap because the Entropic Brain Hypothesis proposes that psychedelic phenomenology reflects increased entropy in functional brain signals, yet it was unclear whether the various published metrics actually measure a single construct or yield consistent findings. McCulloch and colleagues therefore set out to test 14 previously reported fMRI brain entropy metrics in an independent cohort of healthy participants given psilocybin, with concurrent measures of subjective drug intensity, plasma psilocin level, and estimated 5-HT2A receptor occupancy. They hypothesised that entropy would increase after psilocybin, and they also wanted to examine whether observed associations were robust to different parcellation schemes and denoising pipelines, and how strongly the different entropy measures correlated with one another. The study is positioned as an independent replication and comparison of candidate acute psychedelic biomarkers, with direct relevance to the Entropic Brain Hypothesis.

The study used a single-blind crossover design in 28 healthy volunteers, recruited from a database of people interested in psychedelic research. Participants were screened for neurological, somatic and psychiatric illness, and the protocol was approved by the relevant Danish ethics and medicines authorities and registered at ClinicalTrials.gov. The extracted text states that the data were collected between 2018 and 2021, and that no formal sample size calculation was performed, although the sample was larger than in previous publications in this area. Participants received a single oral dose of psilocybin at 0.2-0.3 mg/kg, with a mean dose of 19.7 mg, or 20 mg ketanserin; the ketanserin scans were outside the scope of this paper and are not presented. After drug administration, resting-state fMRI scans were acquired at approximately 40, 80, 130 and 300 minutes post-dose, producing 121 scan sessions in total. After each scan, participants rated current subjective drug intensity on a 0-10 scale, blood was drawn for plasma psilocin measurement, and 5-HT2A receptor occupancy was estimated from the plasma psilocin level using a previously reported Hill-Langmuir-based relation. The researchers refer to the combined subjective, pharmacokinetic and receptor-occupancy measures as PsiFx. Pre-processing was applied uniformly across metrics using SPM12 and CONN, including slice-timing correction where appropriate, unwarping, realignment, coregistration, segmentation, normalisation, smoothing, linear detrending, aCompCor, motion regression, scrubbing of artefactual volumes, and band-pass filtering from 0.008-0.09 Hz. Because data came from two scanners, some time series were temporally downsampled, and cerebellar regions were removed from atlases when not consistently in view. The main analyses regressed each entropy metric on PPL, SDI and estimated occupancy in separate linear mixed-effects models with a subject-specific random intercept, adjusting for motion, age, sex and scanner. Regional metrics were corrected for family-wise error across regions using maxT permutation correction, while whole-brain metrics used permutation p-values. The authors also repeated analyses across two parcellation strategies and seven denoising pipelines to assess robustness, and examined pairwise correlations between entropy measures. The paper then details how each entropy metric was operationalised, including static connectivity measures, dynamic connectivity measures, dynamic activity measures, and Lempel-Ziv complexity measures.

Participants showed substantial subjective drug intensity and plasma psilocin after psilocybin, as expected. In the primary analysis, using the original parcellation and a standard denoising pipeline, the authors found significant positive associations between PsiFx and five of the 14 entropy metrics, one inconsistent positive association, and null results for eight metrics. The five metrics with consistent positive relations across denoising pipelines were geodesic entropy, normalised global spatial complexity (NGSC), dynamic conditional correlation (DCC) entropy, meta-state complexity, and sample entropy at scale 1. Lempel-Ziv temporal BOLD complexity (LZct) showed inconsistent positive effects across some pipelines, whereas spatial BOLD complexity (LZcs) was not associated with PsiFx. The authors note that the metrics were generally weakly correlated with one another. Among the individual metrics, out-network connectivity distribution entropy, degree distribution entropy, von Neumann entropy, intra-network synchrony distribution entropy, motif-connectivity distribution entropy, LEiDA-state Markov-rate, integration/segregation-state distribution entropy, and spatial BOLD complexity were not significantly associated with PsiFx in the primary analysis. Geodesic entropy showed a significant positive association, with weak-to-moderate correlations for PPL, occupancy and SDI; this replicated across most denoising pipelines except global signal regression. NGSC was significantly and moderately positively associated with PsiFx and remained robust across all denoising pipelines. Meta-state complexity was weakly positively associated with occupancy and SDI, but not clearly with PPL, and remained significant across most pipelines. DCC entropy showed the strongest and broadest signal: it was positively associated with PsiFx in 35 of 36 network-to-network connections, with moderate-to-strong effect sizes, although the pattern was sensitive to temporal filtering and could weaken or reverse with some preprocessing choices. Sample entropy showed a short-scale positive pattern at scale 1 in 7 of 17 networks and a long-scale negative pattern at scale 5 in 14 of 17 networks, but these results were sensitive to denoising and did not appear at intermediate scales. Temporal Lempel-Ziv complexity was associated with occupancy and SDI but not clearly with PPL in the primary model, and its significance varied across pipelines. When the authors re-ran the analyses using a common parcellation, the overall pattern was similar: geodesic entropy, meta-state complexity, DCC entropy, and short-scale sample entropy remained the main positive findings, whereas other metrics remained null. Scanner moderation analyses suggested that some associations varied by scanner, including geodesic entropy, sample entropy at scale 5, and von Neumann entropy, whereas DCC entropy and NGSC appeared more robust. In the correlation analysis between whole-brain entropy measures, some metrics were positively correlated, some negatively correlated, and several conceptually related measures clustered together, such as NGSC with von Neumann entropy, and LZcs with LZct. The extracted text also states that the authors considered the possibility that the original studies reported values that were not always mathematically plausible for some motif-based metrics, and that several prior findings were not reproduced here.

The authors interpret the study as showing that acute psilocybin effects on fMRI entropy are real but heterogeneous rather than uniform across all entropy measures. They argue that they independently replicated increased geodesic entropy, increased NGSC, and the short- versus long-scale pattern previously reported for sample entropy, and they also identified a strong positive DCC entropy signal that had not been reported before. Meta-state complexity and temporal Lempel-Ziv complexity showed additional evidence of association across some preprocessing pipelines. At the same time, eight of the 14 metrics were not associated with psilocybin measures, which the authors say suggests that only some published entropy metrics are sensitive to acute psychedelic effects. Relative to earlier research, they note that their findings converge with some previous psilocybin, LSD and ayahuasca reports, but also fail to replicate a number of earlier positive results. They emphasise that this likely reflects the fact that the various metrics are not measuring a single underlying construct called “brain entropy”. Instead, the authors present the metrics as capturing distinct properties of functional brain activity, connectivity, or complexity. They also stress that the Entropic Brain Hypothesis is broad and not specific about which entropy measure should be expected to change, making the present mixed findings difficult to classify as simple support or refutation. The authors discuss several limitations. The study used two scanners with different sequences, requiring some temporal downsampling, although each participant was scanned on only one scanner. There was no placebo condition and no active non-psychedelic comparator, so the specificity of the observed effects for psychedelics cannot be determined. Motion increased with psilocybin intensity, and although motion was adjusted for and scrubbed, the authors could not exclude residual confounding. Scan durations were relatively short for some sessions, which may affect the stability of entropy estimates, and multi-band acquisition may have influenced signal quality. They also did not directly measure physiological variables such as respiration or heart rate, and they note that psychedelic effects on cerebral blood flow and neurovascular coupling could complicate interpretation of BOLD-based entropy measures. For implications, the authors suggest that future work should include active comparators such as stimulants, ketamine, MDMA and cannabinoids to determine whether these entropy changes are specific to serotonergic psychedelics or reflect broader altered brain states. They also recommend that future psychedelic fMRI studies measure subjective intensity at scan time and collect plasma drug levels, and they point to the need for methodological transparency, independent replication and further work on preprocessing sensitivity. They additionally mention a MATLAB-based toolbox developed for calculating the metrics studied here, which they present as a resource for future research.

The authors conclude that psilocybin produces consistent positive effects on a subset of fMRI entropy and complexity metrics, especially NGSC, DCC entropy, geodesic entropy, meta-state complexity and short-scale sample entropy, but not on many other published measures. They state that the limited inter-correlation between metrics and their differential sensitivity to denoising indicate that these measures do not represent a single unitary construct of brain entropy. Overall, they describe their findings as nuanced support for the Entropic Brain Hypothesis and as identifying candidate biomarkers for future clinical psychedelic studies, while emphasising the need for independent replication and transparent methodology.