An exploratory persistent-homology analysis of resting-state fMRI functional connectivity under Ayahuasca

The study used resting-state fMRI from nine healthy adults, five of whom were women, with no reported neurological or psychiatric disorders. Each participant completed two task-free scans: one before ayahuasca ingestion and one 40 minutes afterwards, when the acute effects were expected to begin. The brew dose was 120-200 mL, corresponding to 2.2 mL/kg of body weight, and was reported to contain DMT and harmine. The design was exploratory and within-subject, comparing each participant to themselves across the two sessions. The fMRI data were acquired on a 1.5 T Siemens scanner using an EPI-BOLD sequence. Preprocessing in FSL included slice timing, head motion and spatial smoothing corrections, and the data were normalised to MNI space. A general linear model included regressors for movement, white matter, cerebrospinal fluid and global signal regression. The brain was parcellated using the Harvard-Oxford atlas into 110 regions, of which six were excluded because of acquisition limitations, leaving 104 regions of interest. For each region, the BOLD signal was averaged into time series of 150 points, and a maximum overlap discrete wavelet transform was applied to retain the usual resting-state frequency band of 0.01-0.1 Hz. Pearson correlations were then calculated between all region pairs to build 104 × 104 functional connectivity matrices. From these matrices, the authors constructed Vietoris-Rips filtrations and clique complexes using the GUDHI library. Their main analysis converted correlations to dissimilarities using 1-|r(i,j)| so that both strong positive and strong negative associations counted as strong coupling. They also performed a sensitivity analysis using the signed distance 1-r(i,j), which preserves correlation sign. Persistent homology was computed over Z2 up to homology dimension 3, with the simplicial complex expanded to dimension 4. Persistent entropy was calculated from the barcodes for H0 to H3, with infinite bars excluded. The primary outcome was persistent entropy in each dimension, particularly S2, which captures the diversity of lifetimes of two-dimensional cavities. Pre/post differences were assessed at the subject level using two-sided Wilcoxon signed-rank tests, with false-discovery-rate correction across the four homology dimensions and matched-pairs rank-biserial correlation as the effect size. The authors also performed exploratory analyses of H2 lifespan distributions and a dominance ratio, and they examined temporal signal complexity using Lempel-Ziv complexity and sample entropy. They note that only global-signal-regressed connectivity matrices were available for this analysis.

In the primary absolute-correlation analysis, the authors found no statistically significant pre/post differences for persistent entropy in dimensions S0, S1 or S3. The main result was a nominal decrease in S2 after ayahuasca ingestion, corresponding to two-dimensional topological features and their associated cavities. This decrease was observed in 8 of the 9 participants. The Wilcoxon signed-rank test gave W = 4.0, an uncorrected p = 0.0273, a rank-biserial correlation of 0.8222, and an FDR-adjusted p = 0.1094, so the effect did not survive correction for multiple comparisons. Further inspection of H2 barcode structure suggested a redistribution of persistence lengths under ayahuasca, with more very short lifespans and fewer intermediate ones. The authors also tested whether the lower S2 could be explained by a single dominant long-lived cavity. The dominance ratio did not differ significantly between conditions (mean before = 0.0455, mean after = 0.0540; W = 8.0, p = 0.098), which argued against that explanation. At the subject level, the number of H2 bars decreased in 8 of 9 participants, and the median H2 lifespan decreased in 6 of 9 participants, indicating broadly concordant but not uniform individual changes. When the analysis was repeated using the signed distance 1-r(i,j), none of the persistent-entropy dimensions showed significant pre/post differences. The S2 result was therefore not reproduced under the signed representation (W = 18.0, p = 0.652). The supplementary signal-complexity measures also did not yield significant changes, although they increased descriptively in most participants: Lempel-Ziv complexity had p = 0.244 and sample entropy had p = 0.205. The authors also report a supplementary comparison using an independent healthy resting-state test-retest dataset from the Human Connectome Project, in which the same topological pipeline did not produce statistically detectable differences between test and retest sessions.

The authors interpret the main finding as a preliminary, nominal decrease in the persistent entropy of two-dimensional topological features after ayahuasca ingestion in the absolute-correlation analysis. They stress that the effect was large at the paired level but did not survive FDR correction and was not reproduced when correlation sign was preserved, so it should be treated as exploratory and specific to the chosen preprocessing and distance definition. They describe the result as evidence for a redistribution of H2 persistence lifetimes rather than definitive proof of drug-induced topological reorganisation. They position the work as complementary to, rather than a direct test of, the Entropic Brain Hypothesis. Their reasoning is that entropic brain measures usually refer to complexity or unpredictability of the underlying signal, whereas persistent entropy in this paper measures the diversity of lifetimes of topological features, which the authors describe as a different aspect of network organisation. In a supplementary exploration of temporal signal complexity, Lempel-Ziv complexity and sample entropy increased descriptively in most participants but were not statistically significant, and the authors do not claim a confirmed dissociation between temporal and topological entropy measures. The authors also compare their findings with earlier topological studies of psychedelics, including work on psilocybin that reported changes in H1 persistence structure. They argue that differences across studies are mainly methodological, because the present analysis focuses on persistent entropy of H2 features rather than scaffold-based H1 summaries or pairwise graph metrics. They suggest that the barcode analyses are more consistent with a redistribution and reduced diversity of H2 persistence lengths than with a single dominant cavity or a stable high-order scaffold. The paper’s limitations are emphasised strongly. The sample was very small, there was no placebo or time-control condition, and the two scans were separated by about 40 minutes, so the design cannot separate ayahuasca effects from scanner drift, habituation, arousal changes, nausea, motion, physiological fluctuation or ordinary test-retest variability. The authors therefore describe the findings as ayahuasca-session-associated changes rather than causal drug effects. They also note that the analysis relied on global-signal-regressed connectivity matrices only, that global signal regression is methodologically debated, and that the nominal S2 result disappeared under the signed-distance sensitivity analysis. Additional limitations include the use of a 1.5 T scanner, only 150 volumes per session, and ROI-size heterogeneity in the Harvard-Oxford atlas. The authors suggest future work should use larger samples, placebo-controlled designs, longer and higher-quality acquisitions, parallel GSR and no-GSR pipelines, alternative parcellations, time-resolved TDA, and comparisons across other psychedelic compounds such as psilocybin and LSD.