Decreased directed functional connectivity in the psychedelic state

Participants and drug administration were drawn from previously published pharmaco‑MEG experiments. Eligibility criteria excluded those under 21, pregnant individuals, anyone with personal or immediate family history of psychiatric disorder, substance dependence, cardiovascular disease, claustrophobia, blood/needle phobia, or other medically significant conditions. For LSD and psilocybin participants were required to have prior hallucinogen experience (but not within 6 weeks). The ketamine sample excluded smokers, females, and those outside a body‑mass index of 18–30. Dosing regimes were: LSD 75 μg IV (single bolus administered in under 1 minute) with MEG recorded about 4 hours later; psilocybin (PSI) 2 mg IV with recordings immediately after administration; ketamine (KET) an initial bolus 0.25 mg/kg over 1 minute followed by 0.375 mg/kg/h infusion for 40 minutes with recordings immediately after administration; tiagabine (TGB) 15 mg oral with recordings about 3 hours post‑ingestion. Placebo for IV drugs was saline and for TGB an identical inactive capsule; each participant had a separate placebo session for the corresponding drug. MEG acquisition used a CTF 275‑channel axial gradiometer system. Participants lay supine for ketamine and LSD sessions and were seated for PSI and TGB; resting‑state recordings post‑administration lasted drug‑dependently (approximately 5–14 minutes). Original sampling was 600 Hz or 1200 Hz; preprocessing included 1 Hz high‑pass and 150 Hz low‑pass filtering, downsampling to 300 Hz, removal of 50 Hz line noise and harmonics, segmentation into 2 s epochs, and visual artefact rejection. Source localisation projected sensor data to 90 cortical regions defined by the Automated Anatomical Labelling (AAL) atlas using a linearly constrained minimum variance beamformer with 5% regularisation. Epochs were treated as stationary multitrial (panel) data per drug/subject/condition and normalised per condition except for spectral power estimation. The number of usable epochs varied by drug and subject; the extracted text does not clearly report subject counts or per‑condition epoch numbers in the prose provided. Functional connectivity analysis measured both undirected and directed metrics in time and frequency domains. Time‑domain undirected FC was estimated via correlation/mutual information (MI); time‑domain directed FC via Granger causality (GC). Frequency‑domain undirected FC was estimated via coherence and directed FC via spectral Granger causality. Where appropriate, conditional (partial) measures were used to control for indirect influences mediated by other recorded sources. Three measurement scales were reported: SRC (per‑source unconditional measures versus the rest of the brain), ROI (pairwise ROI–ROI measures conditioned on all other sources), and GLO (intra‑ROI global‑conditional measures reflecting within‑ROI association density). Pairwise conditional statistics were computed at the ROI level (rather than source–source) to preserve statistical power given the high dimensionality. Many implementation details and analytic formulae were presented in appendices referenced in the extracted text. Statistical inference used asymptotic theory for time‑domain second‑order statistics (log‑likelihood ratios with F or χ2 approximations) and empirical subsampling/surrogate methods for frequency‑domain statistics. Cross‑subject comparisons of drug versus placebo used Wilcoxon signed‑rank tests (paired non‑parametric), with the normalized Wilcoxon W statistic reported as a rank correlation effect size ranging from −1 to +1. False discovery rate (FDR) correction controlled for multiple comparisons per drug/measure/condition, with significance at α = 0.05. The authors note that empirical model order variation and finite‑sample bias were negligible relative to observed effect sizes given the large number of observations per condition.

Spectral power analyses showed a consistent broadband reduction in auto‑spectral power for the three psychedelics (LSD, psilocybin, ketamine), most pronounced in the δ–β range, and a small upward shift in the α‑peak frequency particularly for LSD. Tiagabine produced a slight increase in spectral power in δ–α bands. The extracted text does not provide explicit subject counts or confidence intervals for spectral effects in the prose. Time‑domain per‑source (SRC) analyses revealed a general, statistically significant decrease in directed FC (Granger causality) across cortical sources for all three psychedelics, with LSD showing the largest effect sizes. Directed measures were decomposed into inbound and outbound GC and both showed decreases under psychedelics; by contrast, tiagabine exhibited increases in directed FC. Undirected per‑source measures (mutual information) gave a more heterogeneous picture: LSD and tiagabine showed significant increases in MI, while psilocybin and ketamine produced inconclusive or weak effects on undirected FC. Frequency decomposition indicated that the LSD increase in MI was strongest in the γ‑band (noted by the authors to be interpreted cautiously because of potential muscle artefact). For psilocybin and ketamine there was suggestive but statistically weak evidence for slight γ increases; tiagabine’s MI increases were strongest in δ–θ bands, suggesting a distinct mechanism. ROI and intra‑ROI (GLO) analyses provided spatial specificity. For undirected MI, LSD increases were most marked between occipital regions and other ROIs, especially the cingulate; tiagabine showed pronounced MI increases between parietal and frontal/limbic/occipital regions, while intra‑ROI MI changes were generally weaker. Directed GC decreases under LSD were widespread across cortex, both inter‑ and intra‑ROI. Psilocybin’s GC decrease was strongest between parietal and other regions with only slight intra‑ROI increases. Ketamine’s strongest GC decreases were directed toward parietal and occipital regions from other ROIs, with significant intra‑ROI decreases in occipital, parietal and sensorimotor areas. Tiagabine showed fairly evenly distributed GC increases across cortex, except for a decrease in intra‑ROI directed connectivity in occipital cortex. Because directed and undirected measures sometimes moved in opposite directions (notably LSD: decreased GC with increased MI) and sometimes similarly (tiagabine: increases in both), the authors further probed whether observed FC changes could be explained by changes in spectral power, SNR, or VAR residual correlation. They argued that changes in auto‑spectra alone are not expected to change GC because Granger causality is scale invariant and invariant under invertible filtering; therefore broadband power reductions do not by themselves predict the GC decreases observed. Simulations adding correlated additive broadband white noise to placebo data—intended to emulate SNR reduction—showed that substantial additive noise produced decreases in both GC and MI, contrary to the LSD empirical pattern (decreased GC coupled with increased MI). Thus a simple SNR reduction is unlikely to account for the empirical observations. The authors also investigated the relationship between GC and MI empirically (cross‑subject correlations of GC and MI and of their within‑subject changes) but found no consistent global correlation: specific ROI pairs sometimes showed correlation or anti‑correlation but these were not statistically robust after FDR correction. To provide mechanistic insight they analysed a minimal stationary bivariate VAR(1) model with unidirectional coupling and correlated residuals. Theoretical and numerical exploration showed that mutual information and Granger causality may move independently under parameter changes, and in this minimal model changes in residuals correlation (κ) can produce opposite movements in MI and GC. Empirically, residuals correlation estimated from the fitted VAR models increased significantly for LSD and psilocybin, decreased significantly for tiagabine, and showed no significant change for ketamine. The authors conclude that while residuals correlation change could plausibly account for the LSD pattern (increased residuals correlation associated with increased MI and decreased GC in the minimal model), it does not readily explain the full pattern across drugs. Overall, the principal empirical finding is a generalised decrease in directed functional connectivity (information flow) in psychedelic conditions, sometimes accompanied by increases in undirected FC (notably for LSD), whereas the non‑psychedelic control tiagabine showed increases in both kinds of FC.

Barnett and colleagues interpret the main finding—a broadband, cortex‑wide decrease in directed functional connectivity under LSD, psilocybin and low‑dose ketamine—as evidence for reduced information flow and a loosening of dynamical constraints between and within brain regions in the psychedelic state. They link this reduced information flow to theoretical proposals that an increased repertoire of brain dynamics and signal diversity underlie characteristic psychedelic subjective effects such as unconstrained cognition and ego‑dissolution; this is consistent with prior analyses of the same dataset that reported increased Lempel‑Ziv complexity for the psychedelics. The discussion situates these results relative to earlier work. The authors note that fMRI studies have reported both decreases and increases in undirected connectivity across different psychedelics and methods, and that MEG/EEG studies are fewer but have shown altered synchronisation and information transfer in thalamocortical and other networks. They emphasise that the high temporal resolution and source localisation afforded by MEG enabled a rigorous, simultaneous comparison of directed (Granger causality) and undirected (correlation/coherence) measures on the same data, revealing that these metrics can move independently. Consequently, the authors argue that combining directed and undirected FC measures yields a more comprehensive picture of neural dynamics than relying on undirected metrics alone. The investigators also report analyses intended to rule out simple confounds. They emphasise that the observed GC decreases cannot be readily explained by broadband power reductions (due to GC’s scale invariance), by SNR reductions (simulated additive noise decreased both GC and MI, unlike the LSD empirical pattern), or uniformly by changes in VAR residual correlations (empirical changes in residual correlation varied across drugs and did not account for all observed patterns). Nevertheless, they acknowledge that the mechanistic processes producing the relative changes in directed and undirected FC remain to be fully elucidated. Key limitations that the authors acknowledge include potential contamination by muscle artefacts—particularly affecting high frequency (γ‑band) findings under LSD; the limited temporal coverage of the MEG recordings relative to the full duration of psychedelic experiences (MEG snapshots may not capture time‑varying dynamics across an entire ‘trip’); procedural differences such as eyes‑closed recordings for tiagabine versus eyes‑open recordings for other drugs (though comparisons were within‑drug versus placebo so not mixed); and limited statistical power due to comparatively small subject samples, which may have constrained the detection of finer‑grained FC changes. The authors conclude by recommending that future EEG/MEG studies apply both directed and undirected FC measures when exploring neural dynamics, and note that the analytic pipeline used here provides a statistically sound approach for such investigations given epoched, multivariate neurophysiological data with adequate temporal resolution.