The effect of lysergic acid diethylamide (LSD) on whole-brain functional and effective connectivity

The study pooled data from two randomised, placebo-controlled, double-blind, cross-over trials, yielding a combined sample of 45 healthy participants (20 in trial A, 25 in trial B). Participants received 100 μg oral LSD in one session and placebo in the other, with at least a 7-day washout between sessions; the actual interval varied across participants. Resting-state fMRI was acquired once per session roughly 121–200 minutes after administration, while subjective experience was assessed retrospectively with the 5-Dimensions of Altered States of Consciousness (5D-ASC) scale 11 hours after dosing. Further acquisition and preprocessing details are reported in the Supplement (not reproduced here). Functional connectivity was computed as Pearson correlations among 132 regions of interest (ROIs) from the Harvard–Oxford atlas, producing 8,648 unique correlation coefficients. Whole-brain effective connectivity was estimated with rDCM applied to the raw ROI time series, which provides directed connection estimates and allows inference on inhibitory self-connections (local gain parameters). The investigators treated self-connections as negative 'decay-rate' coefficients indicating a region's tendency to return to baseline. Statistical analysis combined mass-univariate paired t-tests with false-discovery-rate control, multivariate partial least squares correlation (PLSC) analyses to identify latent variables capturing condition differences or correlations with subjective scores, and machine learning. For PLSC, singular value decomposition was applied to mean-centred FC, EC, and self-connection matrices; significance of latent variables was assessed by permutation tests (2,000 permutations) and reliability by bootstrapping (2,000 samples). To reduce confounding by baseline physiology, connectivity features were residualised for baseline heart rate, blood pressure and body temperature before PLSC. Random-forest classifiers were trained separately on whole-brain FC and EC features using 5-fold cross-validation, covariate correction, and permutation testing (1,000 label permutations) to estimate statistical significance; balanced accuracy (BAC) was the primary classifier metric. Feature importance was assessed via mean decrease in accuracy on permuted out-of-bag samples.

Functional connectivity analyses indicated widespread changes under LSD. Mass-univariate paired tests identified about 23% (1,993/8,646) of unique FC coefficients as significantly different between LSD and placebo (p < 0.05), with the majority reflecting increased FC under LSD. Multivariate PLSC revealed a significant condition effect on the first latent variable, with stronger FC under LSD across many regions. The most reliable increases involved connections such as bilateral lingual gyrus with bilateral inferior frontal gyrus (pars opercularis), right inferior frontal gyrus with right lingual gyrus, left cuneus with right middle frontal gyrus, and left temporooccipital MTG with left intracalcarine cortex. Conversely, weaker FC under LSD was observed predominantly between occipital regions (for example left fusiform gyrus with right lateral occipital cortex and supracalcarine/occipital pole) and between bilateral putamen and cerebellum. Behavioural PLSC showed a significant correlation between FC patterns and global 5D-ASC scores (r = 0.775, p = 0.031). A random-forest classifier trained on FC achieved a cross-validated BAC of 86% (p < 0.001); feature importance highlighted predominately occipital and prefrontal-involving connections among the top predictors. Effective connectivity (EC) analyses yielded related but distinct results. Mass-univariate tests flagged about 13% (2,184/17,424) of EC coefficients as differing between conditions (p < 0.05), with mostly stronger EC under LSD but a notable exception of weakened EC among occipital regions. Thalamic connectivity was robustly affected: connectivity from thalamus to cortex and from cortex to thalamus was generally stronger under LSD. PLSC on EC showed a significant condition effect on the first latent variable, with reliable positive loadings for connections between occipital and prefrontal regions and negative loadings concentrated in occipital-to-occipital links. Behavioural PLSC similarly revealed a significant correlation between EC patterns and subjective effects (r = 0.932, p = 0.047). The EC-based random-forest classifier produced a BAC of 91.1% (p < 0.001); among the top 50 most important EC features, 39 were connections between pairs of occipital regions, indicating that occipital EC differences drove much of the classification performance. Comparative analyses of FC versus EC examined self-connections and asymmetry. rDCM-estimated inhibitory self-connections differed significantly across conditions in about 30% (39/132) of regions (p < 0.05). Occipital areas became more inhibitory (more negative self-connections) under LSD, whereas temporal, subcortical and brainstem regions—including cerebellum and bilateral pallidum—showed relative disinhibition (self-connections moving toward zero). PLSC on self-connections confirmed a significant condition effect. Tests of directed-asymmetry found that forward–backward pairs that were significantly different numbered roughly 10% under LSD, 14% in placebo, and 1.4% for the drug-by-asymmetry interaction, indicating that EC was asymmetric but that LSD had little effect on asymmetry overall. Finally, although EC produced a numerically higher BAC than FC (91% vs 86%), the difference was not statistically significant (p > 0.074), suggesting both measures can discriminate LSD from placebo with high accuracy.

Bedford and colleagues interpret their findings as showing that LSD produces widespread increases in directed and undirected connectivity across cortex, accompanied by regionally specific changes in local gain that point to perturbations of the brain's excitation/inhibition (E/I) balance. Multivariate analyses identified stronger EC among parietal, temporal and inferior frontal regions, whereas EC and FC both revealed selective weakening of connectivity among occipital regions. Changes in inhibitory self-connections in approximately 30% of regions—regional disinhibition across much of cortex and increased inhibition in occipital areas—are taken as evidence that LSD shifts local dynamics and moves parts of the system closer to dynamical criticality. The authors note that EC estimates showed face validity: thalamic connectivity changes were consistent with previous FC findings and with a prior spectral-DCM study focusing on thalamo-cortical loops. They discuss methodological differences with other directed-connectivity studies (for example Granger-causality analyses on MEG) that may account for discrepant results. In relation to mechanistic hypotheses, the results align with the thalamic gating hypothesis—data indicate stronger thalamus→cortex and cortex→thalamus connectivity suggesting reduced thalamic gating and possible positive feedback loops. The pattern of widespread disinhibition accords qualitatively with brain-entropy accounts of psychedelics, although the authors emphasise regional heterogeneity, with visual cortex showing increased local inhibition under LSD. Potential receptor-level interpretations are offered cautiously: widespread local disinhibition might relate to 5-HT2A-mediated glutamate release, while regional increases in inhibition (notably occipital) could involve different serotonergic receptors such as 5-HT1B; however, the study design does not directly test receptor mechanisms. The authors acknowledge several limitations they consider important: possible carry-over effects despite the washout period; variable inter-session intervals; the single post-drug scan per session; small sample size; potential confounding by LSD-induced physiological changes despite baseline correction; limited blinding due to salient subjective effects; retrospective timing of subjective ratings not restricted to the fMRI window; and the need for external validation of classifier performance. Future directions proposed include employing models that disentangle excitation from inhibition at a finer scale, targeted investigation of visual-region effects (including eyes-open vs eyes-closed comparisons), concurrent physiological monitoring during scanning, and studies testing whether baseline FC or EC can predict individual subjective responses to LSD.

This study is, to the authors' knowledge, the first to characterise the effects of LSD on whole-brain effective connectivity using rDCM. Compared with placebo, LSD was associated with predominantly stronger FC and EC, regionally increased self-inhibition in occipital areas and disinhibition in other cortical and subcortical regions, and stronger thalamo-cortical interactions. Whole-brain EC correlated with global subjective effects and discriminated LSD from placebo with high accuracy (91.11%), suggesting that EC preserves classification performance while offering additional mechanistic information consistent with LSD-induced disturbance of the excitation/inhibition balance. The authors conclude that whole-brain EC is a promising candidate biomarker for decoding or predicting subjective effects of LSD in future work.