Dynamical exploration of the repertoire of brain networks at rest is modulated by psilocybin

Lord and colleagues frame brain function as the exploration of a repertoire of metastable, large-scale functional networks that reconfigure over time. Previous work has suggested that transitions between recurrent network states can relate to cognition and that psychoactive compounds may alter these dynamical explorations; however, how a serotonergic psychedelic modulates the relative occupancy and transitions of specific whole-brain network states has not been fully characterised. The authors situate psilocybin—an agonist at serotonin 2A (5-HT2A) receptors with well‑documented subjective effects such as hyper‑associative cognition and altered selfhood—as a useful probe to link molecular pharmacodynamics to changes in large‑scale network dynamics. This study therefore aims to characterise how intravenous psilocybin rapidly modulates the brain's dynamical repertoire at rest. Using a data‑driven approach (Leading Eigenvector Dynamics Analysis, LEiDA) applied to single‑TR BOLD phase coherence, the investigators identify recurrent phase‑locking (PL) patterns, quantify their probabilities of occurrence and their state‑to‑state transition profiles before and after psilocybin infusion, and validate findings against a placebo session. They also compare recovered PL states to canonical resting‑state networks (RSNs) to assess functional correspondence.

Nine healthy adult participants (minimum 21 years, screened for psychiatric, cardiovascular and substance‑dependence history) provided datasets meeting motion and safety criteria; all had previously used psilocybin but not within 6 weeks. Ethical approval was obtained and informed consent collected. Each participant underwent two eyes‑closed resting‑state fMRI sessions (12 minutes each) on separate days in a counterbalanced design: one session with an intravenous psilocybin infusion (2 mg in 10 ml saline, delivered over 60 s) and one with placebo (10 ml saline). Infusions began 6 minutes into each 12‑minute scan; analyses used the first 5 minutes (pre‑infusion) and the last 5 minutes (post‑infusion) to create four 5‑minute conditions (pre/post psilocybin, pre/post placebo). Subjects rated subjective intensity after each session. Imaging was performed at 3T (TR/TE 3000/35 ms; 53 slices; 3 × 3 × 3 mm voxels). Preprocessing used FSL MELODIC defaults (motion correction, brain extraction, 5 mm smoothing, intensity normalisation, linear de‑trending). Data were parcellated with the AAL atlas into N = 90 regions; regional BOLD time series were band‑pass filtered 0.02–0.1 Hz, yielding T = 100 TRs per condition per subject. Dynamic phase‑locking analysis estimated instantaneous BOLD phase per region via the Hilbert transform and computed a dynamic Phase‑Locking (dPL) matrix at each TR as cos(θn(t) − θp(t)), producing an NxNxT tensor. LEiDA was applied by extracting the leading eigenvector V1(t) of the dPL matrix at each time point, reducing each NxN matrix to a 1xN vector capturing the dominant phase orientation; the leading eigenvector typically explained >50% of variance in phase coherence. To detect recurrent PL states, the authors pooled 1800 leading eigenvectors derived from the psilocybin session (9 subjects × 2 conditions × 100 TRs) and ran k‑means clustering with k ranging from 5 to 10. Cluster centroids (VC vectors) were taken as PL states. For the selected k (primarily k = 7), each TR was assigned to its nearest centroid, yielding state time courses from which probability of occurrence (fractional occupancy) and transition probabilities (switching matrices) were computed. Between‑condition differences were tested with permutation‑based paired t‑tests (1000 permutations for occupancy comparisons; 5000 for transition probabilities). PL states were compared to seven canonical RSNs by correlating centroid vectors with RSN vectors mapped into AAL space. Global synchrony and metastability were quantified using the Kuramoto Order Parameter (OP); metastability was operationalised as STD(OP) over time. Motion effects were assessed via mean framewise displacement (FD).

Applying LEiDA and clustering (k = 7 chosen for key analyses) produced one global PL state and six non‑global PL states that spatially corresponded to canonical RSNs. Correlations between PL centroids and RSNs were strong: PL state 6 with the Visual network (r = 0.79, p = 2.3×10⁻²⁰), PL state 3 with Frontoparietal (r = 0.70, p = 1.1×10⁻¹⁴), PL state 5 with Somatomotor (r = 0.65, p = 6.1×10⁻¹²), PL state 4 with Limbic (r = 0.36, p = 5.5×10⁻⁴), PL state 2 with Default Mode (r = 0.35, p = 5.5×10⁻⁴), and PL state 7 showing contributions from Ventral Attention and Somatomotor networks (r = 0.43, p ≈ 2×10⁻⁵). The principal between‑condition effects involved two PL states. For k = 7, the frontoparietal PL state (PL3) showed a marked decrease in fractional occupancy after psilocybin: mean 14.3 ± 2.4% pre‑injection to 4.1 ± 1.2% post‑injection (p = 0.00034 uncorrected; p' = 0.0021 after correction). Conversely, the globally coherent PL state (PL1) increased from 26.7 ± 4.5% to 37.2 ± 6.0% (p = 0.0068 uncorrected; p' = 0.047 after correction). Occupancies of the other PL states did not change significantly (all p > 0.05). Paired‑sample t‑tests confirmed these effects at the individual level: PL3 t(8) = 3.88, p = 0.005; PL1 t(8) = 2.75, p = 0.01. The authors examined robustness across partition models (k = 5–10) and found that the PL state corresponding to the Frontoparietal network was the most consistently affected by psilocybin across k values (centroids highly correlated across solutions, CC > 0.86). A second consistent effect involved the global PL state in several k solutions. Transition (switching) analyses showed that psilocybin reduced the probability of transitions into the frontoparietal PL state from other states, while increasing transitions toward the globally coherent PL1 from most other states; permutation tests (5000 permutations) identified the most significant state‑to‑state changes. Visualisation via PCA (first three components) illustrated fewer excursions into the manifold region associated with the frontoparietal PL state after psilocybin and denser occupancy of the global PL1 region. Subjective intensity ratings (mean 6.9 ± 2.6 for psilocybin vs 0.4 ± 0.6 for placebo) correlated with state occupancies: fractional occupancy of the frontoparietal PL state correlated negatively with intensity (r = −0.56, p = 0.0083, one‑tailed). A near‑significant positive correlation was found between global PL1 occupancy and intensity (r = 0.36, p = 0.07, one‑tailed). Global synchrony analyses found no change in mean OP, but metastability (STD(OP)) increased following psilocybin (STD(OP) = 0.18 ± 0.01 post vs 0.16 ± 0.015 pre; t(8) = 2.41, p = 0.04), indicating larger fluctuations in whole‑brain synchrony. The occupancy of PL1 was strongly correlated with mean OP across subjects (cc = 0.825, p = 2.9×10⁻⁵). Validation using the placebo dataset (sessions >1 week apart) applied the same psilocybin‑derived centroids to placebo data and found no significant pre/post differences in PL state probabilities (all p > 0.05). The occupancy of PL3 and PL1 during placebo remained similar to the pre‑psilocybin baseline and significantly different from the psilocybin condition. Motion control analyses indicated no significant difference in mean framewise displacement post‑placebo versus post‑psilocybin (t(8) = 1.28, p = 0.24), and fractional occupancies of PL3 and PL1 post‑psilocybin did not correlate with mean FD (CC = −0.14, p = 0.72 and CC = 0.04, p = 0.91 respectively).

The investigators interpret their findings as evidence that psilocybin rapidly and selectively alters the brain's dynamical repertoire at rest. Specifically, they report a strong destabilisation of a PL state overlapping the lateral frontoparietal control system and a concomitant increase in occupancy of a globally coherent, functionally non‑specific PL state. Lord and colleagues relate the reduced expression of the frontoparietal PL state to prior imaging work showing decreased low‑frequency power and reduced oscillatory power in frontoparietal regions under psychedelics, and to behavioural reports of impaired goal‑directed focus and increased distractibility under the drug. Mechanistically, the authors link these network effects to psilocybin's 5‑HT2A receptor agonism. They note qualitative overlap between regions with high 5‑HT2A density and the lateral frontoparietal areas implicated here, and point to cellular evidence that 5‑HT2A stimulation increases excitability of layer‑5 pyramidal neurons. The authors suggest that preferential receptor expression in frontoparietal cortex could produce dysregulated excitation that destabilises that network, favouring transitions into a globally integrated attractor and increasing metastability. This shift is framed as a flattening of the attractor landscape, which may permit atypical interregional communication in the psychedelic state. The discussion situates the present results alongside prior reports of increased global integration and signal complexity under psychedelics, arguing that the LEiDA approach provides complementary whole‑brain, high temporal‑resolution evidence for network‑specific neuromodulation. The within‑subject design is emphasised as increasing power despite the small sample size; validation across the separate placebo session further supports the specificity of drug effects. The authors acknowledge limitations they attribute to the dataset and methods: small sample size (n = 9), a relatively coarse parcellation (AAL, N = 90) that may miss finer networks (for example the Dorsal Attention Network was not recovered), and the temporal resolution of the fMRI (TR = 3 s) being too slow to capture faster dynamics seen with MEG. They also note that LEiDA focuses on phase‑locked oscillatory states via cosine phase differences and does not assess directed causal influences. Finally, the authors highlight potential implications for understanding therapeutic mechanisms and for designing circuit‑specific pharmacological interventions, while presenting the study as a proof‑of‑concept linking molecular pharmacodynamics, dynamical network measures and subjective experience.