LSD alters dynamic integration and segregation in the human brain

Luppi and colleagues situate their study within recent efforts to link psychedelic-induced alterations of consciousness to their neurobiological substrates using non-invasive imaging. Prior work indicates that serotonergic psychedelics such as LSD, psilocybin and DMT increase measures of brain complexity and alter functional connectivity, typically reducing within-network integrity while increasing between-network coupling. The authors note that the stream of consciousness is inherently dynamic, and that previous studies have largely neglected temporal fluctuations in functional connectivity. They argue that integration and segregation are fundamental network properties relevant to consciousness and hypothesise that LSD's effects may be temporally specific rather than uniform across time. This study therefore aimed to combine graph theory with dynamic functional connectivity (dFC) from resting-state fMRI to test whether LSD alters time-resolved network organisation, focusing on integrated versus segregated dynamic sub-states. Following the Entropic/Anarchic Brain framework, the researchers predicted that LSD would produce effects broadly opposite to those observed during loss of consciousness, and that these effects would be differentially expressed during integrated and segregated dFC sub-states. The investigation used a placebo-controlled design in healthy volunteers with prior psychedelic experience to examine network-level and regional changes and their relation to subjective experience.

Fifteen healthy volunteers (final N = 15 after one aborted experiment and four exclusions for excessive motion) underwent two imaging sessions 14 days apart in a placebo-controlled within-subject design. On one day participants received placebo (10 mL saline) and on the other 75 μg LSD in 10 mL saline, infused over 2 minutes and administered 115 minutes before fMRI scanning. Imaging comprised three seven-minute resting-state BOLD scans with eyes closed; analyses reported here used the first and third (no-music) scans, each 7:20 minutes long. Subjective ratings were obtained in-scanner using VAS and post-scan using the 11-item Altered States of Consciousness (ASC) scale. Preprocessing used the CONN toolbox and SPM12 routines: removal of the first three volumes, realignment, slice-timing correction, outlier detection (art), normalisation to MNI space (2 mm), and 6 mm FWHM smoothing. Denoising employed anatomical CompCor (aCompCor) to regress out five white-matter and five CSF principal components, six motion parameters plus derivatives, scrubbing of outlier volumes, session effects, linear detrending and band-pass filtering (0.008–0.09 Hz). Global signal regression was deliberately avoided. Excess motion was defined as >15% of volumes with mean framewise displacement >0.5; mean framewise displacement was included as a covariate because it was higher under LSD (Placebo mean = 0.079, SD = 0.026; LSD mean = 0.129, SD = 0.039; t(14) = -6.34, p < 0.001). Functional parcellation used an augmented Schaefer-232 atlas (200 cortical + 32 subcortical ROIs); robustness checks employed a finer Schaefer-454 (400 + 54) and the Brainnetome-246 (210 + 36). Region-wise BOLD timecourses were averaged and pairwise Pearson correlations computed to form FC matrices for each full scan. Dynamic functional connectivity was estimated using tapered sliding windows obtained by convolving a 22-TR (44s) rectangular window with a 3-TR Gaussian, sliding in 1-TR steps; this yielded a sequence of 232×232 FC matrices per timepoint. Dynamic sub-states were identified via the cartographic profile: within each window, nodes' participation coefficient and within-module degree Z-score were computed from graph-theoretical module assignments. A k-means clustering (k = 2, supported by silhouette analysis) of the joint histogram of these measures assigned windows to predominantly integrated or predominantly segregated sub-states. For each subject and sub-state, centroid FC matrices (median of assigned windows), the proportion of time spent in each sub-state, and the entropy of the sub-state sequence were computed. Stationarity of dynamics was assessed by comparing empirical results to surrogate data generated with pairwise 2-D VAR(4) models. Network-level measures included the prevalence of anticorrelations (proportion of negative edges), similarity between functional and structural connectivity (Spearman correlation and Hamming distance versus a group-average structural connectome from 1,021 HCP subjects), small-world propensity (ϕ, bounded 0–1, computed both on thresholded binary networks at densities 10%–25% and on weighted non-negative networks), and a measure of functional complexity based on the distribution of pairwise correlations. Statistical testing used two-tailed robust linear models with mean framewise displacement as covariate; correlations with subjective measures used Spearman's ρ and were interpreted only when reproducible across all three parcellations. Network-based Statistic (NBS) was applied to identify connected components of edges differing under LSD, with an intensity F-threshold of 10 (robustness checks at F = 12 and an extent-based threshold of 10).

Sample and validation. After exclusions, 15 participants were analysed. The cartographic profile reliably separated two dynamic sub-states (k = 2 selected by silhouette). The proportion of time spent in the predominantly integrated sub-state was approximately 0.69 (SD = 0.07) in placebo and was preserved under LSD, indicating no change in gross temporal allocation between integrated and segregated sub-states. Empirical proportions were greater than stationary VAR surrogates for both placebo (empirical M = 0.69 vs surrogate M = 0.51, t(14) = 2.42, p = 0.030) and LSD (empirical M = 0.69 vs surrogate M = 0.55, t(14) = 2.91, p = 0.014), supporting genuine dynamical structure. Entropy of sub-state alternation did not change significantly between conditions. Time-averaged and sub-state-specific FC. Time-averaged FC showed a predominance of global FC increases under LSD along with disconnections between prefrontal and posterior/temporal regions. Dynamic, sub-state-specific analyses revealed that the predominantly integrated sub-state exhibited no significant LSD-induced FC alterations, whereas the predominantly segregated sub-state displayed both increases and decreases in FC. Notably, the anterior medial prefrontal cortex (mPFC) showed reduced FC specifically during the segregated sub-state; time-averaged FC instead showed stronger reductions in medial orbitofrontal cortex. The segregated sub-state had net increases in somatomotor and auditory cortex connectivity and net decreases in key default mode network nodes (medial PFC, bilateral angular gyri, temporal poles). The posterior cingulate reduced connectivity with visual regions but increased connectivity with executive-control regions, and the thalamus increased connectivity with orbitofrontal, temporal cortices and bilateral amygdalae. These NBS findings were robust across parcellations, with the exception that time-averaged FC differences were not significant with the Brainnetome atlas though the segregated sub-state pattern remained. Structural-functional similarity and anticorrelations. There was no significant change in similarity between time-averaged FC and the structural connectome. However, during the predominantly segregated sub-state, LSD reduced the similarity of FC to structural connectivity (lower Spearman correlation and increased Hamming distance), indicating greater decoupling from anatomy; no significant effect was observed in the integrated sub-state. Regarding anticorrelations, LSD reduced the overall proportion of negative edges across the full scan, but dynamic analysis showed that this reduction was specific to the segregated sub-state; the integrated sub-state did not show significant LSD-related suppression of anticorrelations. Small-world propensity and functional complexity. Small-world propensity (ϕ) increased under LSD. For thresholded binary networks the increase was evident during the segregated sub-state and for time-averaged FC; when weighted networks including weaker (non-negative) edges were analysed, increased ϕ was detected for both integrated and segregated sub-states. The net rise in small-world propensity arose from opposing trends: ΔL (deviation of path length from random) increased under LSD (which would lower ϕ), while ΔC (deviation of clustering from lattice) decreased (which increased ϕ), with the latter effect predominating. Functional complexity (a measure derived from the distribution of pairwise correlations) increased under LSD overall, but this effect was confined to the segregated sub-state and was not observed in the integrated sub-state. Robustness and correlations with subjective experience. Key results were robust across alternative parcellations with some exceptions noted for specific metrics. Exploratory correlations reproducible across all three parcellations included: in the predominantly integrated sub-state, increases in weighted small-world propensity correlated positively with ASC blissful state (Spearman's ρ = 0.65, p = 0.009), ASC complex imagery (ρ = 0.68, p = 0.006) and VAS ego dissolution (ρ = 0.57, p = 0.027). In the predominantly segregated sub-state, changes in the prevalence of anticorrelations correlated with ASC disembodiment (ρ = 0.62, p = 0.013) and ASC elementary imagery (ρ = 0.61, p = 0.015). None of the reported brain measures correlated significantly with mean framewise displacement for the Schaefer parcellations, supporting that motion was not driving these effects.

Luppi and colleagues interpret their results as showing that LSD's effects on brain function and subjective experience are temporally specific rather than uniform. Although LSD did not change the overall fraction of time the brain spent in predominantly integrated versus segregated sub-states, it differentially altered network properties within those sub-states. The apparent paradox that both LSD and loss of consciousness reduce anticorrelations is resolved by this temporal decomposition: reduced anticorrelations in anaesthesia are most pronounced during integrated sub-states, whereas LSD suppresses anticorrelations primarily during segregated sub-states. This sub-state specificity also extended to structural-functional coupling and complexity: LSD increased functional complexity and reduced similarity to the structural connectome, but these effects were concentrated in segregated moments when functional patterns are already less constrained by anatomy. The authors situate these findings within the REBUS/Anarchic Brain framework, proposing that LSD weakens the influence of anatomical priors on functional coupling, allowing exploration of a broader repertoire of connectivity patterns. A specific regional finding—the reduced functional connectivity of anterior mPFC during segregated sub-states—is linked to possible attenuation of reality-monitoring processes; the authors note parallels between mPFC disconnection in LSD and observations in schizophrenia, and propose that transient reductions in top-down constraint may underlie hallmark psychedelic phenomena. Small-world propensity increased under LSD, reflecting an increased local clustering compensated by longer path lengths; importantly, increases in integrated-state small-world propensity predicted subjective reports of bliss, complex imagery and ego dissolution. The authors highlight that ego-dissolution correlations were temporally specific and may be relevant for therapeutic mechanisms. Several limitations are acknowledged. The dataset is relatively small and previously studied, so replication in independent and larger samples is necessary. Blinding is imperfect because LSD's effects are overt; an active placebo could address expectancy and arousal confounds. Head motion differed between conditions, and although aCompCor denoising, participant exclusion criteria and inclusion of mean framewise displacement as a covariate were employed, residual confounding cannot be fully excluded. Structural connectivity was taken from a group-average template rather than individual diffusion data, so structural-functional deviations are relative to the population average. Methodological differences between approaches to brain dynamics (e.g. sliding-window cartographic profile versus LEiDA or connectome-harmonic decomposition) mean that direct mapping between findings across methods is non-trivial, although converging evidence from different techniques was noted. Finally, the authors call for higher temporal resolution modalities (EEG/MEG) and improved experience sampling to better link dynamic brain states with subjective experience.

The primary novel conclusion is that LSD's effects on brain function and subjective experience depend on the brain's dynamic state at a given time. Specifically, LSD makes globally segregated sub-states more complex and more decoupled from anatomical constraints, and reduces functional connectivity of anterior medial prefrontal cortex during those segregated moments. These temporally specific alterations align with the REBUS framework, suggesting that LSD weakens anatomical priors on functional connectivity and thereby facilitates exploration of diverse connectivity patterns. Additionally, increased small-world propensity during predominantly integrated sub-states predicted ego dissolution, indicating that different dynamic sub-states relate to distinct subjective features of the psychedelic experience.