Shannon entropy of brain functional complex networks under the influence of the psychedelic Ayahuasca

Ayahuasca is an Amazonian psychoactive brew that contains N,N-dimethyltryptamine (DMT) together with harmala alkaloids that act as monoamine oxidase inhibitors (MAOIs). The beverage allows DMT to reach the brain and, like other classic psychedelics, can produce marked alterations of perception, cognition and self-experience. Interest in Ayahuasca has grown because of suggested therapeutic effects in conditions such as depression and addiction, and because neuroimaging methods now permit non-invasive study of whole-brain functional organisation in altered states of consciousness. Viol and colleagues set out to test a prediction of the entropic brain hypothesis: that psychedelics increase entropy of brain functional connectivity. Using resting-state fMRI collected before and after Ayahuasca ingestion, they applied concepts from graph theory and information theory to construct whole-brain functional networks and directly measure the Shannon entropy of the networks' degree distributions. The aim was to determine whether entropy increases under Ayahuasca and how that change relates to conventional network measures of local and global integration.

The study recruited ten healthy, right-handed adult volunteers (mean age 31.3 years, range 24–47), all experienced Ayahuasca users from the Santo Daime religious community. Volunteers were medication-free for at least 3 months, abstinent from caffeine, nicotine and alcohol prior to scanning, and had no history of neurological or psychiatric disorder by DSM-IV interview. Each participant underwent two resting-state fMRI sessions: one before ingestion and one 40 minutes after consuming Ayahuasca (dose 2.2 mL/kg, equivalent to 120–200 mL; brew contained 0.8 mg/mL DMT and 0.21 mg/mL harmine). The protocol had ethics approval and written informed consent. Images were acquired on a 1.5 T scanner with an EPI-BOLD sequence (150 volumes, TR = 1700 ms, TE = 66 ms) and a high-resolution T1. Preprocessing in FSL included slice-timing correction, motion correction, spatial smoothing (FWHM = 5 mm) and spatial normalisation to MNI152 space. A general linear model regressed out nine nuisance regressors (6 motion, white matter, CSF and global signal). One subject was excluded for excessive motion, leaving nine; subsequent steps required further exclusions so seven subjects were retained for paired before-vs-after comparisons (see network construction below). For network construction the researchers parcellated the brain with the Harvard–Oxford atlas: 110 regions were defined, six were excluded because they were not sampled for all subjects, leaving 104 regions. For each region an averaged time series was extracted, wavelet-decomposed using MODWT with a Daubechies wavelet, and scale 3 (≈0.03–0.07 Hz) was selected. Pairwise Pearson correlations between wavelet coefficients produced a 104 × 104 correlation matrix per scan; correlation elements with p ≥ 0.05 were set to zero. Correlation matrices were thresholded to produce binary, undirected adjacency matrices: an absolute correlation |c_i,j| > η became a link. Rather than a single threshold, the authors used a range chosen to produce networks that were fully connected, relatively sparse and exhibited small-world properties. The common threshold range across subjects was 0.25 ≤ η ≤ 0.43, producing networks with mean degree 〈k〉 in the range 24–39. Because two subjects had no threshold range in common with the others, and a third had excessive motion, seven subjects remained suitable for paired comparisons; for each subject the authors generated 16 networks (different mean degrees) per condition. Network measures calculated (using the Brain Connectivity Toolbox) included node degree and its distribution, mean geodesic distance, clustering coefficient, and global and local efficiencies. The Shannon entropy S[P] of the degree distribution P(k) was computed using the natural logarithm. To assess which effects followed from degree-distribution changes alone, iso-entropic randomized networks were generated using the Maslov algorithm: for each original network 30 randomized variants were produced that preserve each node's degree but rewire links, conserving the degree-distribution entropy while altering other topological features. Statistical comparisons between before and after conditions used paired-sample Student's t-tests; the null hypothesis assumed the paired differences were normally distributed with zero mean.

The primary finding was a consistent increase in the Shannon entropy of the degree distribution after Ayahuasca ingestion. Across the examined range of mean degrees, entropy values were higher post-ingestion, and subject-by-subject analysis showed increased entropy for every individual included in the primary comparisons. Complementing the entropy change, the degree distributions exhibited increased variance in all subjects and decreased kurtosis in most subjects (six of seven), indicating wider, less peaked distributions. Global network integration decreased after Ayahuasca: mean geodesic distance increased and global efficiency decreased in the post-ingestion condition. The authors compared these changes against the iso-entropic randomized networks (30 per original network) and found that the alterations in geodesic distance and global efficiency observed in the empirical data were not fully explained by the change in degree distribution alone, because the randomized counterparts had different values for those metrics. In contrast, local integration increased: both clustering coefficient and local efficiency rose after Ayahuasca. For these metrics the iso-entropic randomized networks showed almost identical changes, implying that the variation in degree distribution accounts for most of the observed increases in clustering and local efficiency. Thus, the pattern is of greater local cohesion but reduced global integration. Motion was examined as a potential confound: the Pearson correlation between individual changes in Frame Displacement (FD) and changes in entropy was 0.005 (p ≈ 1), leading the authors to reject head motion as an explanation for the entropy increases. Subjective state was assessed with the Clinician Administered Dissociative States Scale (CADSS) and the Brief Psychiatric Rating Scale (BPRS). A strong correlation (r = 0.91; p = 0.004) was reported between individual CADSS score variation and the entropy before Ayahuasca intake, although the authors caution interpretation given the small sample. Correlations between CADSS change and changes in geodesic distance and global efficiency were r = 0.71 and r = 0.67 respectively; these did not meet the study's predefined criterion for significance (r > 0.76) but are relatively large compared with correlations for clustering (r = 0.15) and local efficiency (r = 0.09).

Viol and colleagues interpret their results as evidence that Ayahuasca increases entropy of whole-brain functional network organisation while simultaneously increasing local integration and decreasing global integration. They argue that the increase in Shannon entropy reflects a broadening of the degree distribution: networks move away from a regular lattice-like topology (which has low degree entropy) toward a topology with a wider spread of node degrees. The authors contrast their findings with expectations from simple network models such as the Watts–Strogatz small-world transformation, where increasing randomness typically decreases both clustering and geodesic distance. Here, however, they observed increased entropy together with increased clustering and increased geodesic distance, so the changes cannot be reduced to a simple increase in randomness. The combination of higher local efficiency and lower global efficiency led the researchers to suggest a possible change in modular organisation, with greater within-module cohesion but weaker between-module connectivity; however, standard modularity-detection algorithms applied did not reveal clear differences in modular structure between conditions. These empirical results are presented as broadly consistent with the entropic brain hypothesis, which posits that psychedelic-induced 'primary' states of consciousness exhibit higher entropy in some brain organisation features relative to ordinary waking ('secondary') states. The authors note that increased entropy could reflect a temporary loosening of constraints that normally shape adult waking consciousness, permitting a wider repertoire of mental states; they emphasise, however, that correlation does not imply causation and that entropy increases may occur in parallel with, rather than cause, experiential changes. Key limitations acknowledged include the small sample size and the fact that all participants were experienced Ayahuasca users, limiting generalisability and precluding separation of acute versus long-term/familiarity effects; absence of a placebo control and thus no control for expectancy; and methodological choices that may affect networks (atlas, correlation-to-adjacency thresholding, and the restricted threshold range that enforces small-world characteristics). The authors conclude that, despite these caveats, their Shannon-entropy approach offers a low-computational-cost way to assess global changes in functional network organisation under psychedelics and that their findings align with other recent reports of entropy increases under LSD, psilocybin and ketamine. They suggest further work to explore links between entropy changes and subjective phenomena such as reported 'mind-expansion' or enhanced creativity.