Effect of LSD and music on the time-varying brain dynamics

Psychedelic-assisted therapies are being trialled for several psychiatric conditions, and factors known as "set" (the individual's mindset) and "setting" (the environment and contextual elements) are recognised modulators of outcomes. Music is a common and influential component of setting in psychedelic sessions, as it can shape emotion, meaning-making and sensory processing, yet there is limited understanding of how music interacts with psychedelics to shape time-varying brain activity measured with neuroimaging. Previous studies have shown that psychedelics alter whole-brain network organisation and resting-state dynamics, including changes in default mode network (DMN) activity and network connectivity, but most work has focused on static or GLM-based contrasts rather than recurrent whole-brain activity patterns over time. Adamska and colleagues set out to examine how music, as an element of setting, modulates the dynamics of recurring whole-brain activity patterns—so-called brain states—during LSD administration. Using an open fMRI dataset in which 15 participants completed two sessions (LSD and placebo), each comprising two resting-state runs separated by a music-listening run, the study applied K-Means clustering to identify brain states and compared their fractional occupancy, dwell time and transition probabilities across runs and drug conditions. The investigators hypothesised that music on LSD would increase the occurrence of states linked to emotion, autobiographical memory and sensory processing, and that music would have a lasting impact on resting-state dynamics following listening.

Data came from an open dataset (OpenNeuro ds003059) containing preprocessed BOLD fMRI from 15 participants who each underwent two sessions separated by at least 14 days: one with LSD (75 μg) and one with placebo (saline). Scanning began 115 minutes post-infusion. Each session comprised three runs of roughly 7.5 minutes with eyes closed: resting-state (run 1), music listening (run 2) and resting-state (run 3). Two ambient tracks were used for the music run, balanced in order across participants. Technical problems reduced the analyzable music-run sample to 12 participants; all 15 contributed resting-state runs. BOLD data had TR = 2 s and were preprocessed with standard steps including detrending and regression of nine nuisance regressors (motion and anatomical signals). The 400-region Schaefer atlas (400 ROIs) provided parcel time series. For clustering, all timepoints from all participants, sessions and runs were concatenated into a single N × P matrix (N = 18,228 timepoints). The authors applied K-Means with Euclidean distance to identify repeating spatial activation patterns. To choose k, they evaluated gain in variance explained as k increased, running K-Means for k = 2…14 with 100 repeats (random seed = 42). They selected k = 4 because gains were <1% for k > 5; they also report replication analyses for k = 5 and 6 in supplementary materials. Each resulting brain state was correlated with the seven large-scale networks from the Schaefer parcellation to characterise network contributions, and with 11 Neurosynth uniformity-test topic maps (e.g. music, emotion, autobiographical memory) to aid functional interpretation. Three temporal measures were computed per subject and run: fractional occupancy (proportion of time spent in a state), dwell time (mean consecutive duration of a state) and transition probability (probability of moving from state i to state j). Statistical analysis used two main comparisons: run 1 vs run 2 (resting vs music) and run 1 vs run 3 (pre- and post-music resting-state), each examined across LSD and placebo. For fractional occupancy and dwell time, two-level multilevel linear models (MLM) were fitted using lmer with fixed effects for session (LSD/placebo), run, and their interaction, and random intercepts and slopes for subjects. Transition probabilities, which were not normally distributed, were tested with non-parametric permutation tests (10,000 permutations, seed = 0), using mean differences and t-statistics. Alpha was set at 0.05. Because of the small sample size, t-test results were not corrected for multiple comparisons. The authors note code availability for reproducibility and stress that analyses are preliminary given limited power.

Clustering identified four recurrent brain states (k = 4), each with distinct large-scale network profiles and Neurosynth associations. State 1 showed high somatomotor (SOM) activity with low DMN and frontoparietal (FPN) activity and correlated with music and pain-related maps. State 2 exhibited combined average activity of DMN, SOM and visual (VIS) networks and was associated with autobiographical memory and simple sensory/self-referential processing. State 3 was characterised by high DMN and FPN activity with low SOM activity, suggesting self-referential processing or mind-wandering. State 4 showed high activity in task-positive networks (FPN, dorsal attention network DAN, ventral attention network VAN) and related to attention-demanding, cognitive-control processes. Comparing resting-state (run 1) with music listening (run 2), fractional occupancy showed no significant main or interaction effects for any state. Dwell time analyses revealed a significant session effect for state 2: mean dwell time was lower during the LSD session compared with placebo (F(1,25) = 4.9953, p = 0.035). There was also a significant session × run interaction for state 4 (F(1,25) = 4.9642, p = 0.035); post hoc tests indicated that during music listening (run 2) dwell time of state 4 was lower on LSD than on placebo (paired t = -2.219, p = 0.0362). Transition-probability comparisons between run 2 and run 1 (within placebo, within LSD, and LSD vs placebo) yielded no statistically significant differences. For resting-state before versus after music (run 1 vs run 3; N = 15), fractional occupancy and dwell time showed no significant main or interaction effects across states. Transition analyses using permutation tests indicated that during the LSD session there was a higher probability of transitioning from state 3 (high DMN and FPN) to state 4 (high task-positive networks) in the post-music resting-state compared with pre-music (mean difference = 0.041, p = 0.024). Moreover, the difference between run 3 and run 1 for this transition (3 → 4) was significantly greater in the LSD session than placebo (mean difference = 0.057, p = 0.026). The authors caution that these findings derive from a small sample and were not corrected for multiple comparisons, so they should be treated as preliminary.

Adamska and colleagues interpret their findings as evidence that whole-brain activity during both LSD and placebo can be described by four recurrent brain states, which may form two meta-states with opposing sub-states. They report that LSD, independent of music, reduced the mean duration of the state characterised by combined DMN, SOM and VIS activity (state 2), and that the interaction of LSD with music most strongly affected the task-positive state (state 4), producing a shorter dwell time during music and promoting increased transitions from a DMN/FPN-dominant state to a task-positive state in the resting-state after music. The authors situate these results relative to earlier work showing altered functional connectivity of visual, sensorimotor and default mode systems under LSD. They note that their decrease in dwell time for state 2 is consistent with LSD-related changes in those subsystems, though the finding that this effect did not depend on musical stimuli contradicts the original hypothesis. Methodological differences are highlighted as a possible reason for discrepancies: GLM-based contrasts and network neuroscience approaches emphasise spatial or structural changes, whereas the present K-Means clustering emphasises temporal dynamics. Psychologically, the reduction in task-positive state duration during LSD plus music might reflect a redistribution of cognitive resources toward processing sensory and music-induced experiences at the expense of higher-order cognitive control. The increased post-music transitions from self-referential/mind-wandering states to task-positive states under LSD may reflect longer-lasting reorganisation of state dynamics after music exposure during the psychedelic experience. Key limitations acknowledged by the authors include the small sample size and consequent low statistical power, the use of K-Means as a single clustering approach when other models (hierarchical clustering, Hidden Markov models) might produce different state decompositions, and the possibility that observed differences between runs reflect pharmacokinetic timing rather than music-specific effects. The investigators recommend replication in larger samples and suggest future studies employ designs more specifically tailored to probing music effects during psychedelic sessions.

Using unsupervised K-Means clustering of fMRI timeseries, the study provides preliminary evidence that musical stimuli, as an element of setting, can influence time-varying whole-brain activity during LSD administration. The authors conclude that music may produce both short-term changes in state duration during listening and longer-term alterations in resting-state transitions after listening, particularly affecting task-positive network dynamics, but that larger, more targeted studies are required to confirm these effects.